Yang–Mills Existence and Mass Gap

Yang–Mills Existence and Mass Gap

Definitions and mathematical statement

A classical $G$-connection has curvature $F_A=dA+A\wedge A$ and Yang--Mills action proportional to $\int_{{\mathbb{R}}^4}\langle F_A,F_A\rangle\,dx$. The question concerns a rigorous, nontrivial quantum theory associated with this action, not only classical solutions or a formal functional integral. In its vacuum representation, let $\Omega$ be the vacuum and $H\ge0$ the self-adjoint generator of time translations, with $H\Omega=0$. A positive finite mass gap means \[ 0<m:=\inf\left(\operatorname{spec}(H)\setminus\{0\}\right)<\infty. \] For every compact simple $G$, construct such a theory with the local observable fields, positivity, covariance, locality, and short-distance conditions required in Jaffe--Witten, Sections 3--4. Those axioms are part of the target and are not replaced by the displayed spectral condition alone.

Formulation and concept references: [S1].

Short English statement

Can every compact simple gauge symmetry produce a rigorously defined four-dimensional Yang--Mills quantum theory with a nonzero energy gap above its vacuum?

Research attempt

Scope and source audit (27 September 2026).

This attempt treats every compact simple gauge group separately; constants may depend on the group. It requires a nontrivial four-dimensional continuum theory with the observable fields and ultraviolet properties in [S1]. A finite lattice, a classical Yang--Mills solution, or a spectral estimate for one gauge-fixed propagator does not meet those requirements.

The catalogue's recent references need explicit qualification. Nair's [S3] describes an outline for three-dimensional Yang--Mills theory. The arXiv record for Jacobsen's [S4] marks version 2 as withdrawn by arXiv Admin on 13 June 2025, with a research-content quality notice. Neither source is used here as a proved four-dimensional construction. The lattice positivity framework of Osterwalder--Seiler [E1] supplies useful finite-cutoff structure; Chatterjee's own lecture notes [E2] distinguish fixed-coupling lattice estimates from the weak-coupling continuum problem. The deductions below are elementary spectral and limiting arguments, not claims of new results in constructive field theory.

Attack map and the first object that actually exists.

The constructive route has three tasks: construct renormalized Euclidean correlations, obtain estimates uniform as the cutoff and volume are removed, and reconstruct the physical Hilbert space. A direct Hamiltonian route would instead require a well-defined renormalized operator and its spectral gap. I pursue the first route, while testing whether finite-cutoff positivity or a single massive correlation channel could close its missing step.

Choose a faithful unitary representation $\rho:G\to U(d_\rho)$ and a finite hypercubic lattice. Assign $U_e\in G$ to one orientation of each edge, put $U_{\bar e}=U_e^{-1}$, and let $U_p$ be the ordered plaquette product. For $\beta\geq0$ define \[ S_\beta(U)=\beta\sum_p \left(1-d_\rho^{-1}\operatorname{Re}\operatorname{Tr}\rho(U_p)\right), \qquad d\mu_\beta=Z_\beta^{-1}e^{-S_\beta(U)}\prod_e dU_e. \] Here $dU_e$ is normalized Haar measure. Since every summand is in $[0,2]$, \[ e^{-2\beta N_p}\leq Z_\beta\leq1. \] Thus this measure exists without perturbation theory. Haar invariance and invariance of the trace under conjugation prove gauge invariance. Bounded observables have finite moments. These facts are exact, but the bound on $Z_\beta$ degenerates with the plaquette count and gives no compactness for renormalized local fields as the lattice spacing decreases.

Reflection positivity for the appropriate Wilson lattice construction is a separate theorem [E1]. It is not a consequence of the ordinary positivity of $d\mu_\beta$: reflection positivity requires $\langle\overline{F(\theta U)}F(U)\rangle\geq0$ for positive-time observables $F$. I use the resulting transfer-operator picture only where its hypotheses are explicitly assumed below. No global gauge fixing is needed for the finite measure or the spectral argument.

Lemma 3.1: what exponential decay really proves.

Let $H\geq0$ be self-adjoint on a Hilbert space, let $\Omega$ be a unit vector with $H\Omega=0$, and let $\mathcal D$ be a dense linear subspace of $\Omega^\perp$. Suppose there is one $m_0>0$ such that for every $v\in\mathcal D$ there are finite $K_v$ and $t_v$ with \begin{equation}\label{wsref-0-79-6d-3a-64-65-63-61-79} 0\leq\langle v,e^{-tH}v\rangle\leq K_v e^{-m_0t} \quad(t\geq t_v). \end{equation} Then $\ker H=\mathbb C\Omega$ and $\operatorname{spec}(H)\subset\{0\}\cup[m_0,\infty)$.

Proof. Write $E_H$ for the spectral projection measure and $\mu_v(B)=\langle v,E_H(B)v\rangle$. For $0\leq b<m_0$, \[ e^{-bt}\mu_v([0,b]) \leq\int_{[0,\infty)}e^{-t\lambda}\,d\mu_v(\lambda) \leq K_v e^{-m_0t}. \] Letting $t\to\infty$ gives $\mu_v([0,b])=0$. Equivalently, $E_H([0,b])v=0$. Boundedness of this projection extends the equality to $\Omega^\perp$. Taking a countable increasing sequence $b\uparrow m_0$ excludes all spectrum in $[0,m_0)$ on that subspace. Since $H\Omega=0$ and $H$ is self-adjoint, $\Omega^\perp$ reduces $H$. The conclusions follow. $\square$

The common exponent matters; the constants need not be uniform in $v$. Diagonal matrix elements are nonnegative by the spectral theorem, so this argument does not silently discard cancellations in signed correlations. Conversely, the asserted gap and unique vacuum imply (ym:decay) with $K_v=\|v\|^2$ for every $v\perp\Omega$.

There is also a nontriviality check. If a nonzero $v\perp\Omega$ survives, then $C_v(0)=\|v\|^2>0$ and $C_v(t)=\langle v,e^{-tH}v\rangle>0$ for every finite $t>0$. The latter is strict because $e^{-t\lambda}>0$ at every finite spectral value. The spectral gap $m$ consequently satisfies \begin{equation}\label{wsref-0-79-6d-3a-75-70-70-65-72} m_0\leq m\leq-\frac1t\log\frac{C_v(t)}{C_v(0)}<\infty. \end{equation} Indeed $C_v(t)\leq e^{-mt}C_v(0)$. This upper bound concerns the bottom of the excitation spectrum, not the existence of an isolated particle state.

A precise conditional continuum reduction.

Take lattice spacings $a_j\downarrow0$ and spatial side lengths $L_j$ with $a_jL_j\to\infty$. Time must also become unbounded in physical units; a fixed periodic time circle would describe a thermal problem. Suppose the following properties can be proved for renormalized gauge-invariant fields.

1. Their complete family of smeared Euclidean correlations converges and satisfies the hypotheses needed for Osterwalder--Schrader reconstruction, including regularity, reflection positivity, Euclidean covariance, and the required compatibility conditions. The reconstructed observable fields have the short-distance behavior and field content required by [S1].
2. Positive-time observable polynomials, after subtracting their vacuum components and quotienting null vectors, give a dense family $v_F\in\Omega^\perp$. Their lattice representatives $v_{F,j}$ have diagonal transfer correlations $C_{F,j}(n)$ such that for every fixed $t>0$, \[ C_{F,j}(\lfloor t/a_j\rfloor) \longrightarrow\langle v_F,e^{-tH}v_F\rangle. \] The limiting norm is finite, and at least one such $v_F$ is nonzero.
3. There exists $m_0(G)>0$ such that, for each $F$, constants $K_F,t_F$ independent of $j$ satisfy \begin{equation}\label{wsref-0-79-6d-3a-75-6e-69-66-6f-72-6d} 0\leq C_{F,j}(n)\leq K_F e^{-m_0(G)a_jn} \quad\text{whenever }a_jn\geq t_F. \end{equation}

For any fixed $t>t_F$, passage to the limit in (ym:uniform) proves (ym:decay). Lemma 3.1 then gives the mass gap and uniqueness of the vacuum, while the surviving vector and (ym:upper) give a finite mass. This proves the stated implication. It does not prove any of the three assumptions for four-dimensional Yang--Mills theory. In particular, reconstructing an arbitrary massive field theory would not identify it as the Yang--Mills theory specified by the catalogue.

These assumptions separate two logically different limiting operations. Compactness of bounded Wilson-loop expectations could produce subsequential numbers, but local fields require normalization, smearing, and control of coincident-point singularities. Also, estimates established only after taking volume to infinity at each fixed $a$ need cutoff-uniform constants before they imply the displayed diagonal limit. Neither order of limits is licensed by boundedness of a finite partition function.

Failed shortcut A: a lattice gap without physical units.

If a positive normalized transfer operator has largest eigenvalue $1$ and next spectral edge $\lambda_1(a)<1$, its dimensionless gap is $\delta(a)=-\log\lambda_1(a)$. A step advances physical time by $a$, so \[ H_a=-a^{-1}\log T_a,\qquad m_a=\frac{\delta(a)}a. \] On the orthogonal complement of the vacuum, the three diagonal examples \[ \lambda_1(a)=e^{-a^2},\qquad e^{-ma},\qquad e^{-c} \quad(m,c>0) \] all have strictly positive gaps at every cutoff, but their physical gaps tend respectively to $0$, $m$, and $\infty$. Consequently finite-cutoff positivity cannot determine the continuum answer. The last example has $C_a(t)=e^{-c\lfloor t/a\rfloor}$ for a unit excited vector. Its limit is zero at every $t>0$, yet is one at $t=0$; it cannot be the correlation of a fixed unit vector under a strongly continuous continuum semigroup.

This explains why a fixed strong-coupling estimate, even if excellent in lattice units, is insufficient. The coupling trajectory, renormalized observables, and a finite physical scale must all be controlled together. The distinction agrees with the weak-coupling continuum discussion in [E2]. It does not assert that every strong-coupling method must fail; it identifies the extra estimate that would be needed to use one here.

Failed shortcut B: one observable misses light sectors.

Consider \[ \mathcal H=\mathbb C\Omega\oplus\mathbb Cv \oplus\ell^2(\mathbb N),\qquad H\Omega=0,\quad Hv=v,\quad He_n=n^{-1}e_n. \] This is a bounded positive self-adjoint operator with a unique zero eigenvector and no positive gap. Nevertheless $\langle v,e^{-tH}v\rangle=e^{-t}$. Thus an exactly massive channel, not just a numerical approximation to one, cannot prove the target. Each $e_n$ also has exponential decay, but with exponent $1/n$; allowing the exponent to depend on the observable would invalidate Lemma 3.1. Physical local observables must generate a dense family, and the decay exponent must be common to that family.

A further limit stress test takes $\mathcal H_j=\mathbb C\Omega\oplus\operatorname{span}(e_1,\ldots,e_j)$ with energies $1,1/2,\ldots,1/j$. Every approximant has a nontrivial vacuum gap, while the embedded strong limit is gapless. A volume-dependent gap tending to zero is therefore not repaired merely by convergence of the Hamiltonians on finite-support vectors.

Verification and exact remaining gap.

The proofs above were checked against a zero-dimensional excitation space, extra zero modes, a continuum of positive spectral values, and constants depending on the observable. The nonzero-vector requirement excludes the vacuum-only theory. The projection argument excludes extra zero modes and does not require eigenvectors above the vacuum. The units calculation uses physical time $an$, rather than replacing it by the number of steps. The diagonal examples give explicit counterexamples to each weakened spectral assertion. They are Hilbert-space models, not counterexamples to Yang--Mills existence.

The first unproved analytic task on this route is to produce one renormalized continuum-limit family with the full required field content and a nonzero physical observable norm. Even conditional on that construction, (ym:uniform) for a dense observable family remains unproved. The gap lemma solves only the spectral inference after those inputs are supplied.

Three focused next steps are to establish uniform bounds for a chosen family of smeared curvature-polynomial correlations; to track the transfer-correlation exponent in physical units along the same cutoff trajectory; and to exhibit a normalized observable with a nonzero limiting connected correlation. Each must retain control as the physical volume grows. A hypothetical obstruction could arise from low-energy states missed by the tested observables, or from loss of all nonvacuum vectors in the continuum limit. The examples show how to detect these failures without mistaking them for a disproof of the original conjecture.

Final status and provenance.

UNRESOLVED. This attempt proves a conditional spectral criterion and a finite-mass test, and supplies explicit counterexamples to several tempting shortcuts. It constructs no four-dimensional continuum Yang--Mills theory. Written by GPT-6 Astra Ultra from the numbered target, using direct primary-source checks and an adversarial audit of cutoff, density, and nontriviality assumptions. No theorem-prover verification or numerical lattice simulation is claimed.

Sources

• S1 Arthur Jaffe and Edward Witten, Quantum Yang-Mills Theory, official Clay problem description. https://www.claymath.org/wp-content/uploads/2022/06/yangmills.pdf. Formulation source. Consulted 24 September 2026: Sections 3--4, especially page 6.
• S2 Yang-Mills & the Mass Gap. https://www.claymath.org/millennium/yang-mills-the-maths-gap/. Further reference; not a proof endorsement.
• S3 Towards a Proof of Mass Gap in 3d Yang-Mills Theory — V. P. Nair. https://arxiv.org/abs/2608.10133. Further reference; not a proof endorsement.
• S4 A Constructive Proof of Existence and Mass Gap for Pure SU$3$ Yang-Mills in Four-Dimensional Space-Time — D. C. Jacobsen. https://arxiv.org/abs/2506.00284. Further reference; not a proof endorsement.
• E1 Konrad Osterwalder and Erhard Seiler, Gauge field theories on a lattice, Annals of Physics 110 $1978$, 440--471. https://doi.org/10.1016/0003-4916(78)90039-8. Primary lattice-positivity reference; publisher abstract checked 27 September 2026. The complete constructive argument was not independently reverified.
• E2 Sourav Chatterjee, Yang--Mills on the lattice: New results and open problems, author-hosted lecture slides, especially slides 14--19 and 33--35. https://souravchatterjee.su.domains/beam-owps-trans.pdf. Author's statements about lattice estimates and their continuum scope; consulted 27 September 2026.