Four-dimensional Yang–Mills existence and mass gap

Canonical Clay question (Jaffe–Witten, §4). A compact simple gauge group here means a compact connected finite-dimensional Lie group $G$ whose Lie algebra is nonabelian and has no nonzero proper ideals; the distinct global groups, including central quotients, are included in the quantifier. A Lie group is a smooth manifold with smooth multiplication and inversion, and its Lie algebra is its tangent space at the identity with the bracket induced by left-invariant vector fields. For every such $G$, prove that a nontrivial pure quantum Yang–Mills theory exists on four-dimensional spacetime ${\mathbb{R}}^4$ and has a mass gap. “Pure” excludes matter fields. Existence must establish axiomatic properties at least as strong as the Wightman and Osterwalder–Schrader frameworks cited by Jaffe and Witten. In a Minkowski realization, with metric $\eta =\mathrm{diag}(-1,1,1,1)$, this entails local operator-valued distributions on a common dense domain in a positive Hilbert space, covariant under a strongly continuous unitary Poincare representation, with positive energy, a vacuum unique up to phase, and locality: bosonic observables with spacelike separated supports commute. Spacelike separation means $\eta (x-y,x-y)>0$. A Euclidean construction must establish the corresponding axiomatic properties and its passage to such a quantum theory; reflection positivity alone is insufficient.

The local fields must correspond, with the qualifications imposed by renormalization, to gauge-invariant local polynomials in the curvature and its covariant derivatives. In a local bundle trivialization a smooth connection is a Lie-algebra-valued one-form $A$, its curvature is $F_A=dA+\frac{1}{2}[A\wedge A]$, and its covariant derivative is $D_A=d+[A,\cdot ]$, with the bracket extended by exterior multiplication. A local polynomial uses finitely many components of $F_A,D_A{F}_A,\dots$ at one point and contracts Lie-algebra indices by invariant tensors; free spacetime indices give tensor fields. Gauge invariance means invariance under smooth changes of bundle trivialization. Footnote 1 of the source expressly denies a natural one-to-one correspondence between classical differential polynomials and quantum operators because of renormalization; it describes an expected correspondence of spaces through a dimension bound. The source also requires short-distance agreement of correlation functions with asymptotic freedom and perturbative renormalization theory, including a local stress-energy tensor and an operator product expansion with the predicted local singularities. Here correlation functions are vacuum expectation distributions of field products; asymptotic freedom means that the renormalized interaction strength tends to zero at short distances; an operator product expansion expresses a product of nearby fields asymptotically in local fields with coefficient distributions. These are the source's short-distance requirements: §4 does not specify the particular all-order remainder topology, dimension-truncation estimate, or stress-tensor cutoff limits prescribed in the variant below. A solution must give a mathematical meaning to, and establish, the stated short-distance agreement; the absence of a uniquely prescribed topology in this source is not an equivalence theorem for a particular choice.

More precisely, let $H$ be the self-adjoint generator of time translations, let $\mathrm{\Omega }$ be the normalized vacuum, and let $\sigma (H)$ denote the spectrum. In units $c=\mathrm{\hslash }=1$, require $\sigma (H)\subseteq [0,\mathrm{\infty })$, $H\mathrm{\Omega }=0$, and

$m:=\sup \{\delta >0:\sigma (H)\cap (0,\delta )=\mathrm{\varnothing }\},0<m<\mathrm{\infty }\mathrm{.}$

Thus some $\mathrm{\Delta }>0$ is a spectral gap and the supremal gap is finite, as explicitly required in §4. Confinement, chiral symmetry breaking, and existence of an isolated one-particle state are separate extensions in §5. This is an unresolved mathematical existence question; experimental observations and simulations are not its axiomatic premises or a proof.

Stronger analytic variant (additional requirements). Retain the canonical task just stated and additionally require all the following detailed conditions. They specify a more restrictive target for a construction. They are not attributed to Jaffe–Witten as their exact formulation, and no equivalence with, or necessity for, a solution of the canonical Clay problem is asserted. In particular, the field-space spanning condition, explicit regularity bounds, formal reference scheme, all-order little-$o$ comparisons, OPE truncation estimate, and cutoff-independent stress-tensor limits below belong to this additional variant.

A Lie group is a finite-dimensional smooth manifold with smooth multiplication and inversion; its Lie algebra is the tangent space at the identity with the bracket induced by left-invariant vector fields. A Lie algebra is simple when it is nonabelian and has no nonzero proper ideals. A topological space is compact if every open cover has a finite subcover, and connected if it cannot be written as a union of two disjoint nonempty open subsets. For every compact connected finite-dimensional Lie group $G$ with simple Lie algebra $\mathfrak{g}$, construct a pure four-dimensional quantum Yang–Mills theory with gauge group $G$ and no matter fields. The quantifier includes distinct central quotients $G\mathrm{/}Z$ whenever they have isomorphic Lie algebras; here $Z(G)=\{z\in G:zg=gz\text{ for every }g\in G\}$ is the center, $Z$ is a subgroup of it, and an isomorphism of Lie algebras is a linear bijection preserving brackets. The theory is on Minkowski spacetime $M={\mathbb{R}}^4$ with metric $\eta =\mathrm{diag}(-1,1,1,1)$, or is constructed from Euclidean data on ${\mathbb{R}}^4$ that meet the reconstruction criterion below. Write $S{O}^{+}(1,3)=\{\mathrm{\Lambda }\in GL(4,\mathbb{R}):{\mathrm{\Lambda }}^T\eta \mathrm{\Lambda }=\eta ,\det \mathrm{\Lambda }=1,\mathrm{\Lambda }\{x:x^0>\mathrm{\mid }\mathbf{x}\mathrm{\mid }\}=\{x:x^0>\mathrm{\mid }\mathbf{x}\mathrm{\mid }\}\}$ for the proper, orthochronous Lorentz group; the proper Poincare group is ${\mathbb{R}}^{1,3}⋊S{O}^{+}(1,3)$, acting by $x\mapsto \mathrm{\Lambda }x+a$.

For the required gauge-field content, a principal $G$-bundle is a smooth fiber bundle $\pi :\mathcal{P}\to M$ with a smooth right $G$-action that preserves $\pi$ and acts freely and transitively on each fiber. In a local trivialization over an open set $U\subseteq M$, a connection is represented by a smooth $\mathfrak{g}$-valued one-form $A$. The adjoint action ${\mathrm{Ad}}_g$ is the differential at the identity of $h\mapsto gh{g}^{-1}$; under a smooth gauge map $g:U\to G$, $A$ transforms as $A^g={\mathrm{Ad}}_{g^{-1}}A+g^{-1}dg$, where $(g^{-1}dg{)}_x(v)=d(L_{g(x{)}^{-1}}{)}_{g(x)}(d{g}_x(v))$ and $L_h(z)=hz$. Its curvature is $F_A=dA+\frac{1}{2}[A\wedge A]$, with $[A\wedge A](X,Y)=2[A(X),A(Y)]$, and its covariant derivative on $\mathfrak{g}$-valued forms is $D_A=d+[A,\cdot ]$, where the bracket on forms is induced by the Lie bracket and exterior product. The finite-dimensional Lorentz tensor representations on free spacetime indices are induced by tensor powers of the defining representation of $S{O}^{+}(1,3)$ and its dual, with invariant subspaces allowed; the orientation tensor ${ϵ}_{\mu \nu \rho \sigma }$ is the volume form determined by $\eta$ and the chosen orientation. A gauge-invariant local differential polynomial is a finite polynomial in components of $F_A$ and finitely many iterated covariant derivatives $D_A^k{F}_A$, with spacetime indices either retained as indices in finite-dimensional Lorentz tensor representations or contracted using $\eta$ or ${ϵ}_{\mu \nu \rho \sigma }$, and Lie-algebra indices contracted using tensors fixed by simultaneous application of ${\mathrm{Ad}}_g$ to all slots; its value is unchanged under every such gauge map. These polynomials are the field labels $P$. A renormalization prescription is a fixed rule for subtracting short-distance singularities and choosing finite local counterterms. It fixes a dimensionless renormalized coupling $g_R(\mu )$, the coefficient of the pure Yang–Mills interaction at inverse-length scale $\mu >0$; in the reference subtraction convention below write $g_R=g$. The construction must assign renormalized local quantum fields to these labels using the chosen prescription. It must also state the normalization of the positive-definite inner product $\kappa$ on $\mathfrak{g}$, invariant in the sense that $\kappa ({\mathrm{Ad}}_{g}X,{\mathrm{Ad}}_{g}Y)=\kappa (X,Y)$. Renormalization may mix labels, so no canonical one-to-one map is required; for each canonical engineering-dimension bound $d$, the fields assigned to labels of dimension at most $d$ span the corresponding local-operator space through that dimension, with mixing allowed inside that filtered space. The canonical dimensions are $[A]=1$, $[F_A]=2$, and $[D_A]=1$; monomial dimensions add, and $[P]$ is the maximum monomial dimension of a nonzero polynomial label $P$, with $[0]=0$ by convention.

A Wightman realization consists of a complex Hilbert space $\mathcal{H}$, a common dense domain $\mathcal{D}\subseteq \mathcal{H}$, and a family of gauge-invariant bosonic fields ${\mathrm{\Phi }}_P$ indexed by the polynomial labels $P$ defined above, with $\mathrm{\Omega }\in \mathcal{D}$. Let $\mathcal{S}(M)$ be the Schwartz space of smooth test functions whose derivatives decrease faster than every inverse power. Each ${\mathrm{\Phi }}_P$ is an operator-valued tempered distribution: for every appropriate scalar- or tensor-valued $f\in \mathcal{S}(M)$, ${\mathrm{\Phi }}_P(f)$ maps $\mathcal{D}$ into $\mathcal{D}$, and every matrix element $⟨\psi ,{\mathrm{\Phi }}_P(f)\phi ⟩$ is continuous in $f$ for $\psi ,\phi \in \mathcal{D}$. The domain is invariant under all fields and the symmetry representation, and the field family is closed under adjoints. A strongly continuous unitary representation $U$ of the proper Poincare group defined above acts covariantly: if ${\mathrm{\Phi }}_P$ carries Lorentz indices in the finite-dimensional tensor representation ${\rho }_P$, then $U(a,\mathrm{\Lambda }){\mathrm{\Phi }}_P(x)U(a,\mathrm{\Lambda }{)}^{-1}={\rho }_P(\mathrm{\Lambda }){\mathrm{\Phi }}_P(\mathrm{\Lambda }x+a)$ as an equality of operator-valued distributions, with ${\rho }_P=1$ for scalar labels. The four self-adjoint Stone generators $P^{\mu }$ of translations are fixed by $U(a,I)=e^{i{a}_{\mu }{P}^{\mu }}$ and have joint spectrum in the closed future cone ${\bar{V}}_{+}=\{(p^0,\mathbf{p}):p^0\ge \mathrm{\mid }\mathbf{p}\mathrm{\mid }\}$. A normalized vector $\mathrm{\Omega }$ is the unique Poincare-invariant vector up to phase and is cyclic, meaning the linear span of finite products of smeared fields applied to $\mathrm{\Omega }$ is dense in $\mathcal{H}$. For every pair of fields smeared with test functions whose supports are spacelike separated (every difference $x-y$ has $\eta (x-y,x-y)>0$), the corresponding operators commute on $\mathcal{D}$. For every finite family of labels, the vacuum expectations extend to tempered distributions on the corresponding product of Minkowski spaces and cluster when groups of insertions are separated by a spacelike translation tending to infinity. “Nontrivial” here requires both that for some label $P$ and test function $f$, ${\mathrm{\Phi }}_P(f)\mathrm{\Omega }\mathrm{\notin }\mathbb{C}\mathrm{\Omega }$, and that $g_R({\mu }_0)\mathrm{\ne }0$ at some finite inverse-length scale ${\mu }_0>0$ under the stated prescription.

The Euclidean alternative consists of bosonic Schwinger distributions $S_{n;P_1,\dots ,P_n}$, meaning Euclidean vacuum correlation distributions for the polynomial field labels defined above, for $n\ge 0$ with $S_0=1$, on configurations of $n$ points in Euclidean ${\mathbb{R}}^4$; the Euclidean metric is the standard positive-definite inner product. They are covariant under the Euclidean group $E(4)={\mathbb{R}}^4⋊SO(4)$: $SO(4)=\{R:R^{T}R=I,\det R=1\}$, and the tensor labels transform by their induced finite-dimensional representations ${\tau }_{P_i}$, so the covariance identity is

$$S_{n;{\tau }_{P_1}(R)P_1,\dots ,{\tau }_{P_n}(R)P_n}(R{x}_1+a,\dots ,R{x}_n+a)=S_{n;P_1,\dots ,P_n}(x_1,\dots ,x_n)$$

for every $R\in SO(4)$ and $a\in {\mathbb{R}}^4$. They are symmetric under permutations of bosonic field insertions, reflection positive, and clustering: when two groups of test-function supports are separated by a Euclidean translation $a$ with $\mathrm{\mid }a\mathrm{\mid }\to \mathrm{\infty }$, their joint correlation tends to the product of the separate correlations. For reflection positivity put $\theta x=Jx$, $J=\mathrm{diag}(-1,1,1,1)$. Choose real tensor labels first and extend by complex linearity, writing $P^{†}$ for the conjugate label. On a rank-$r$ ordinary tensor field define

$\mathrm{\Theta }\text{\hspace{0.17em}⁣}(c{\mathrm{\Phi }}_{P;a_1\cdots a_r}(x))=\bar{c}{J}_{a_1{b}_1}\cdots J_{a_r{b}_r}{\mathrm{\Phi }}_{P^{†};b_1\cdots b_r}(\theta x)\mathrm{.}$

Repeated tensor indices are summed; upper and lower Euclidean indices have the same reflection matrix. For a pseudotensor there is also the factor $\det J=-1$ for each orientation-tensor factor in its transformation law. Thus an ordinary scalar has no index factor and a pseudoscalar has an additional minus sign. On smeared fields this formula conjugates the test function and composes it with $\theta$, as well as applying the displayed index factors. Extend $\mathrm{\Theta }$ antilinearly to field polynomials, reversing the order in a monomial (irrelevant after evaluation by the symmetric bosonic Schwinger distributions). The positive-time test algebra uses finite test-function sequences supported where all times are positive and vanishing to every order on coincidence diagonals when only ${\mathcal{S}}_0^{\mathrm{\prime }}$ data are available. Reflection positivity is the real inequality $⟨\mathrm{\Theta }(F)F{⟩}_S\ge 0$ for every $F$ in this algebra. There is no further adjoint after $\mathrm{\Theta }$: it already includes coefficient conjugation. For this variant, impose one of the following two explicit regularity bounds; their sufficiency for a reconstruction theorem is not assumed. For $n\ge 1$, let ${\mathcal{S}}_0(({\mathbb{R}}^4{)}^n)$ be the Schwartz functions whose derivatives of every order vanish on each coincidence diagonal $x_i=x_j$, and let ${\mathcal{S}}_0^{\mathrm{\prime }}$ be its continuous dual. For an integer $N\ge 0$, define the Schwartz seminorm $\mathrm{\parallel }f{\mathrm{\parallel }}_{N,n}={\max }_{\mathrm{\mid }\alpha \mathrm{\mid }\le N}{\sup }_{x\in {\mathbb{R}}^{4n}}(1+\mathrm{\mid }x\mathrm{\mid }{)}^N\mathrm{\mid }{\mathrm{\partial }}^{\alpha }f(x)\mathrm{\mid }$. The first bound requires $S_n\in {\mathcal{S}}_0^{\mathrm{\prime }}$ and $\mathrm{\mid }{S}_n(f)\mathrm{\mid }\le {\sigma }_n\mathrm{\parallel }f{\mathrm{\parallel }}_{sn,n}$ for every $f\in {\mathcal{S}}_0$, while the second bound requires $S_n\in {\mathcal{S}}^{\mathrm{\prime }}(({\mathbb{R}}^4{)}^n)$ and the same bound for every $f\in \mathcal{S}(({\mathbb{R}}^4{)}^n)$. Here $s$ is one fixed positive integer and ${\sigma }_n\le A^n(n!{)}^B$ for fixed constants $A\ge 1$ and $B>0$. A Euclidean construction must also establish a Wightman realization with all the properties specified here, including the necessary analytic continuation and reconstruction. The displayed bounds are retained as requirements of this variant; no identification with the exact regularity hypotheses of the corrected Osterwalder–Schrader theorem, or proof of their sufficiency, is asserted. A Euclidean hierarchy without that reconstructed theory does not meet this alternative.

The perturbative comparison data are fixed as follows; they are formal series, not an asserted construction of the required theory. In a $\kappa$-orthonormal real basis write $[T_a,T_b]=f_{ab}^c{T}_c$ and define $C_A>0$ by ${\sum }_{c,d}{f}_{ac}^d{f}_{bc}^d=C_A{\delta }_{ab}$. Use the geometric connection $A=gB$, with canonical gluon variable $B$, and Euclidean gauge-fixed density

$\mathcal{L}=\frac{1}{4}\kappa ({\mathcal{F}}_{\mu \nu },{\mathcal{F}}_{\mu \nu })+\frac{1}{2\xi }\kappa ({\mathrm{\partial }}_{\mu }{B}_{\mu },{\mathrm{\partial }}_{\nu }{B}_{\nu })+\kappa (\bar{c},-{\mathrm{\partial }}_{\mu }({\mathrm{\partial }}_{\mu }c+g[B_{\mu },c])),{\mathcal{F}}_{\mu \nu }={\mathrm{\partial }}_{\mu }{B}_{\nu }-{\mathrm{\partial }}_{\nu }{B}_{\mu }+g[B_{\mu },B_{\nu }]\mathrm{.}$

Here $c,\bar{c}$ are anticommuting adjoint scalar variables used only in the formal calculation, not physical fields in $\mathcal{H}$. Compute with generic covariant gauge parameter $\xi$ and take $\xi =1$ after renormalization. The coefficient rule is Wick expansion of $\exp (-\int {\mathcal{L}}_{\mathrm{i}\mathrm{n}\mathrm{t}})$ about the quadratic part, with gluon propagator ${\delta }_{ab}({\delta }_{\mu \nu }-(1-\xi )p_{\mu }{p}_{\nu }\mathrm{/}{p}^2)\mathrm{/}{p}^2$, ghost propagator ${\delta }_{ab}\mathrm{/}{p}^2$, a minus sign per closed ghost loop, and loop measure $d^{D}p\mathrm{/}(2\pi {)}^D$, continued meromorphically from $D=4-2ϵ$. Use modified minimal subtraction at inverse-length scale $\mu$: write $g_B={\mu }^{ϵ}{Z}_{g}g$, and subtract poles with the associated $-{\gamma }_E+\log (4\pi )$ constants, equivalently use ${\bar{\mu }}^2=4\pi e^{-{\gamma }_E}{\mu }^2$ and pole subtraction. These are equivalent conventions, used once, not simultaneously. ${\gamma }_E={\lim }_{n\to \mathrm{\infty }}({\sum }_{j=1}^{n}1\mathrm{/}j-\log n)$. At each loop order subtract proper ultraviolet subdivergences using the already computed counterterm vertices, then subtract the remaining local pole part; add no finite counterterms. More explicitly, for a regularized graph amplitude $I_{\mathrm{\Gamma }}$ let $K$ extract its pole part in the equivalent pole-subtraction convention. Set

$\bar{R}{I}_{\mathrm{\Gamma }}=I_{\mathrm{\Gamma }}+{\sum }_{S\mathrm{\ne }\mathrm{\varnothing }}\left({\prod }_{\gamma \in S}{\mathrm{\Delta }}_{\gamma }\right)I_{\mathrm{\Gamma }\mathrm{/}S},{\mathrm{\Delta }}_{\mathrm{\Gamma }}=-K\bar{R}{I}_{\mathrm{\Gamma }},R{I}_{\mathrm{\Gamma }}=(1-K)\bar{R}{I}_{\mathrm{\Gamma }}\mathrm{.}$

The sum runs over sets $S$ of mutually disjoint proper ultraviolet-divergent one-particle-irreducible subgraphs; $\mathrm{\Gamma }\mathrm{/}S$ contracts each member to its local counterterm vertex. Nested subdivergences enter recursively in ${\mathrm{\Delta }}_{\gamma }$, and overlapping ones occur in separate allowed sets. At loop order zero there is no subtraction. A counterterm is the local polynomial in external momenta represented by the displayed pole part, with composite-source vertices included. This recursive rule, applied to every graph with its Wick-expansion symmetry factor and then taking its finite $ϵ\to 0$ value, defines the formal coefficients to every finite order. For ultraviolet counterterms use nonexceptional Euclidean external momenta (no vanishing proper partial sum); infrared poles are not subtracted as ultraviolet poles. Composite insertions are generated by adding sources $J_{P}P(gB)$ before Wick expansion, differentiating in the sources, and applying the same subtraction rule including counterterms polynomial in the sources. Choose and explicitly specify a basis of local tensor monomials at each engineering dimension, including total derivatives, equation-of-motion terms, ghost terms and evanescent tensors needed for this subtraction; retain their mixing before projecting to gauge-invariant physical labels. For labels involving the four-dimensional orientation tensor, use a four-dimensional external-index subspace with its orientation tensor, orthogonal to the formal $(D-4)$-dimensional complement, and subtract in this enlarged basis before taking $ϵ\to 0$. These conventions fix a reference scheme; another scheme is admissible only with an explicit finite coupling redefinition and finite local-field mixing and the corresponding transformed comparison coefficients.

For clarity, the higher beta coefficients are defined by this graph rule rather than by an unknown list of numbers. Let $Z_B,Z_c,Z_1$ be the multiplicative factors making respectively the one-particle-irreducible gluon two-point function, ghost two-point function, and ghost–gluon vertex finite under that recursion; one-particle-irreducible means connected after removal of any one internal edge. Define $Z_g=Z_1\mathrm{/}(Z_c\sqrt{Z_B})$, with the formal square root having constant term one, and let $z_1(g)$ be its simple $1\mathrm{/}ϵ$ pole coefficient in the equivalent pole-subtraction convention. The formal four-dimensional beta function is $g^{2}d{z}_1\mathrm{/}dg$. In the notation

${\beta }_{\mathrm{p}\mathrm{e}\mathrm{r}\mathrm{t}}(g)=-{\sum }_{\mathrm{\ell }\ge 0}\frac{b_{\mathrm{\ell }}{g}^{2\mathrm{\ell }+3}}{(16{\pi }^2{)}^{\mathrm{\ell }+1}},b_{\mathrm{\ell }}=-2(\mathrm{\ell }+1)(16{\pi }^2{)}^{\mathrm{\ell }+1}[g^{2\mathrm{\ell }+2}]z_1(g),$

$[g^k]$ extracts the formal coefficient, $b_0=11C_A\mathrm{/}3$ and $b_1=34C_A^2\mathrm{/}3$. No closed formula for all $b_{\mathrm{\ell }}$, nor convergence of this series, is required or asserted. The constructed theory must supply a positive dimensionless running coupling $g(\mu )$, nonzero at a finite scale, with $g(\mu )\to 0$ as $\mu \to \mathrm{\infty }$ and $\beta (g)=\mu dg\mathrm{/}d\mu$ satisfying, for every integer $L\ge 0$,

$\beta (g)+{\sum }_{\mathrm{\ell }=0}^L\frac{b_{\mathrm{\ell }}{g}^{2\mathrm{\ell }+3}}{(16{\pi }^2{)}^{\mathrm{\ell }+1}}=o(g^{2L+3})(g\downarrow 0)\mathrm{.}$

The correlation comparison also has a finite-order meaning, with no summation of a divergent series presumed. Use a homogeneous local-field basis and its scale-dependent mixing in the reference scheme just defined. At fixed reference inverse length ${\mu }_{\ast }>0$, put ${\mu }_{\lambda }={\mu }_{\ast }\mathrm{/}\lambda$ and $g_{\lambda }=g({\mu }_{\lambda })$ for dimensionless $\lambda \downarrow 0$. For labels $P_1,\dots ,P_n$ of dimensions $d_1,\dots ,d_n$, let ${\mathcal{W}}_{\lambda }(y)={\lambda }^{{\sum }_i{d}_i}⟨{\prod }_i{\mathrm{\Phi }}_{P_i}^{({\mu }_{\lambda })}(\lambda y_i){⟩}_E$ be the Euclidean continuation of the constructed vacuum correlation with fields renormalized at ${\mu }_{\lambda }$. The superscript specifies the full field mixing, not just replacement of a coupling in fields fixed at another scale. Let $w_k(y;{\mu }_{\ast })$ be the coefficient of $g^k$ in the vacuum-normalized Wick expansion from the preceding graph rule, with insertions $P_i(gB)$ at $y_i$ and scale ${\mu }_{\ast }$. Vacuum normalization divides the insertion series by its vacuum series, whose constant term is one. Require for every $N\ge 0$ and every smooth compactly supported tensor test function $h$ on the configuration space $y_i\mathrm{\ne }{y}_j$,

$$⟨{\mathcal{W}}_{\lambda }-\sum _{k=0}^N{g}_{\lambda }^k{w}_k,h⟩=o(g_{\lambda }^N)\mathrm{.}$$

This is convergence in the weak distributional sense on each such configuration space. It specifies the remainder, tensor coefficients and scale convention; it does not assert an expansion uniform at collision diagonals or at infinite separation. The same requirement applies to Wilson coefficients: define their formal coefficients by expanding the renormalized insertion graphs for $P(x)Q(0)$ in the hard separation $x$, Taylor expanding their soft external momenta in the local monomial basis and recursively subtracting the local-insertion contributions already matched at lower orders. Matching includes the full operator mixing above, so it is coefficientwise matching of all insertion Green functions, not a choice of scalar leading singularities. For a coefficient $C_{PQ}^R$, replace ${\mathcal{W}}_{\lambda }$ in the displayed remainder condition by ${\lambda }^{[P]+[Q]-[R]}{C}_{PQ}^R(\lambda y,{\mu }_{\lambda })$ and use its formal matching coefficients and tests on a fixed annulus $0<a<\mathrm{\mid }y\mathrm{\mid }<b<\mathrm{\infty }$. Thus scale dependence and logarithmic singularities are those of the specified renormalized graphs, to every finite perturbative order.

In addition, the actual fields must admit an operator product expansion ${\mathrm{\Phi }}_P(x){\mathrm{\Phi }}_Q(0)\sim {\sum }_R{C}_{PQ}^R(x,\mu ){\mathrm{\Phi }}_R(0)$ as $x\to 0$. After insertion in vacuum correlations with other insertions kept away from the origin and after smearing the rescaled separation in a fixed annulus, truncation to labels of dimension at most $d$ must leave a remainder whose scaling degree at $0$ is at most $[P]+[Q]-(d+1)$. For a distribution $t$ near $0$, its scaling degree is $\inf \{\omega :{\lambda }^{\omega }t(\lambda \cdot )\to 0\text{ distributionally as }\lambda \downarrow 0\}$. This dimension-truncation requirement is separate from the finite-order perturbative remainder above; neither follows merely from specifying formal graphs.

The theory also has a local symmetric stress-energy tensor $T_{\mu \nu }$, conserved as a spacetime distribution, ${\mathrm{\partial }}^{\mu }{T}_{\mu \nu }=0$. Require $\mathcal{D}\subseteq {\bigcap }_{\nu }\mathrm{Dom}(P^{\nu })$. For real $h\in C_c^{\mathrm{\infty }}(\mathbb{R})$ with $\int h(t)dt=1$, set $h_{\epsilon }(t)={\epsilon }^{-1}h(t\mathrm{/}\epsilon )$, $\epsilon >0$. For any real $\chi \in C_c^{\mathrm{\infty }}({\mathbb{R}}^3)$ equal to one near the origin, put ${\chi }_R(\mathbf{x})=\chi (\mathbf{x}\mathrm{/}R)$ and define the genuine spacetime smearing $Q_{R,\epsilon }^{\nu }=T^{0\nu }(h_{\epsilon }\otimes {\chi }_R)$. For every fixed $\epsilon >0$ require

${\lim }_{R\to \mathrm{\infty }}⟨\psi ,Q_{R,\epsilon }^{\nu }\phi ⟩=⟨\psi ,P^{\nu }\phi ⟩(\psi ,\phi \in \mathcal{D})\mathrm{.}$

The topology here is pointwise convergence of sesquilinear forms on $\mathcal{D}\times \mathcal{D}$; its limiting form must be represented on $\mathcal{D}$ by the specified translation generator. The limit must exist for every such $h,\chi$ and be independent of both. Taking $\epsilon \downarrow 0$ only after $R\to \mathrm{\infty }$ therefore gives the same generator in the same topology, independently of $h$. No restriction of the distribution $T$ to a time-zero hypersurface or interchange of these limits is assumed.

Writing $H=P^0$, require the canonical spectral condition $H\mathrm{\Omega }=0$ and $0<m<\mathrm{\infty }$ with $m$ defined above; in particular there is a finite $\mathrm{\Delta }>0$ with $\sigma (H)\cap (0,\mathrm{\Delta })=\mathrm{\varnothing }$. Units are $c=\mathrm{\hslash }=1$. Confinement and chiral symmetry breaking are separate questions and are not part of this problem.

Clay Mathematics Institute, “Yang-Mills & the Mass Gap” (current problem and status page; checked 2026-10-25)

Arthur Jaffe and Edward Witten, “Quantum Yang-Mills Theory,” in The Millennium Prize Problems, §§3–6, printed pp. 134–150

Arthur Jaffe, “Constructive Quantum Field Theory,” §8, “For the Millennium: Gauge Theory in Four Dimensions,” PDF p. 14

Konrad Osterwalder and Robert Schrader, “Axioms for Euclidean Green’s Functions,” Communications in Mathematical Physics 31 (1973), 83–112

Konrad Osterwalder and Robert Schrader, “Axioms for Euclidean Green’s Functions II,” Communications in Mathematical Physics 42 (1975), 281–305

Jaffe and Witten, Clay-hosted standalone text, §4 and footnote 1, PDF p. 6; §§5–6.5, PDF pp. 6–12 (canonical formulation, renormalization qualification, separate extensions and construction limits)

Van Ritbergen, Vermaseren and Larin, pp. 2–6, equations (1)–(8), with zero fermion flavors (coupling convention, graph counterterms and first beta coefficients)

The canonical Clay question remains unresolved. The separately stated analytic variant asks for that canonical task together with additional quantitative conditions; whether those extras are necessary or equivalent to the canonical requirements is not established here. Its detailed subtraction and remainder conditions are additional requirements, not established estimates for a constructed four-dimensional theory. The graph prescription defines formal comparison data only. It supplies no positive Hilbert space, continuum measure, convergence of perturbation theory, or mass gap.

The Clay Mathematics Institute page was checked on 2026-10-25 and labels the problem unsolved, stating that no proof of the mass-gap property is known. Jaffe and Witten distinguish this existence-and-gap task from confinement and chiral symmetry breaking, which they describe as separate extensions. Their mathematical perspective records established constructive results mainly in lower dimensions or compact approximations; it says that the four-dimensional Yang–Mills construction on a compact four-torus is incomplete and that a uniform gap and the infinite-volume limit to ${\mathbb{R}}^4$ remain required. Jaffe's earlier lecture notes (§8, PDF p. 14) give an $SU(2)$ formulation with lowest-order perturbative agreement and describe partial progress while leaving the infrared limit unresolved. These statements do not promote finite-regulator models, formal perturbation series, simulations, or lower-dimensional constructions to a continuum solution on ${\mathbb{R}}^4$.