The mathematics
behind Yang–Mills.

01 / THE PRECISE TARGET

Existence is a statement about a quantum theory.

The Clay problem asks, for every compact simple gauge group \(G\), for a nontrivial quantum Yang–Mills theory on four-dimensional spacetime with a strictly positive mass gap. The group quantifier matters: constructing one model for \(SU(2)\) would not settle the full statement. The required axiomatic properties must be at least as strong as the Wightman or Osterwalder–Schrader frameworks cited in the official formulation. “Existence” therefore includes physical structure, not merely a finite value assigned to a formal functional integral. [1, §4, p. 6]

Here \(\mathcal H\) is the physical Hilbert space, \(\Omega\) is the vacuum, \(U\) represents spacetime symmetries, and the \(\mathcal O_i\) are local, gauge-invariant operator-valued distributions. In the Lorentzian formulation, the representation has positive energy and its generators satisfy the spectrum condition. Local observables commute at spacelike separation. Their short-distance behavior must also identify the intended Yang–Mills theory; the official statement discusses gauge-invariant curvature polynomials, asymptotic freedom, and the subtleties of their renormalization. [1, §§3–4, pp. 5–6]

The Euclidean construction studied below is a route to that Lorentzian theory. Its \(\mathbb R^4\) has Euclidean signature; reconstruction supplies real-time dynamics. These signatures should not be silently interchanged in an argument.

Positive energy alone allows spectrum accumulating at zero. A gap excludes that accumulation in the physical vacuum representation. The nontriviality requirement also rules out satisfying the display with a Hilbert space containing only the vacuum. The actual spectral threshold must be finite. Confinement is a related dynamical question, but it is listed separately among extensions of the Millennium problem. [1, §§4–5, pp. 6–7]

The target is pure Yang–Mills: no dynamical quarks or Higgs field are introduced to generate a mass. The gap belongs to the spectrum of physical states created by gauge-invariant observables. A mass parameter inferred from a gauge-dependent gluon propagator is consequently not, by itself, the requested spectral statement. Nor does the definition demand an isolated one-particle eigenvalue at the threshold. The official discussion lists such a stronger spectral property separately. A proposed construction must make clear which of these logically different assertions it establishes.

02 / FIX THE CONVENTIONS

One connection, with the coupling in the action.

We work at \(\theta=0\). For explicit coefficients, specialize to \(SU(N)\) and use Hermitian generators in the fundamental representation. Repeated spacetime and colour indices are summed unless a restriction is displayed.

We absorb the coupling into the connection variable. Thus the covariant derivative contains \(A\), while an overall factor in the action carries \(g\). Expanding the commutator of two covariant derivatives fixes every sign:

In differential-form notation this becomes the following expression. The factor of one half belongs to the graded-commutator notation; dropping it would double the nonlinear term.

For a gauge transformation \(h(x)\), covariance of \(D_\mu\psi\) determines the inhomogeneous transformation of the connection. Curvature transforms by conjugation, so traces of its products are gauge invariant.

With these choices, the positive Euclidean action has equivalent matrix and component forms. The distinction between summing all ordered index pairs and only \(\mu<\nu\) explains the factor of two:

For general compact simple \(G\), a positive \(\operatorname{Ad}\)-invariant inner product replaces the chosen matrix trace. Its normalization is part of the coupling convention. Taking that inner product to be \(2\operatorname{tr}(XY)\) in the present Hermitian realization recovers the preceding action. This is the invariant content behind superficially different normalizations. [1, §1, pp. 1–2]

In natural units, the engineering dimensions make four dimensions special: the coupling is dimensionless. Composite operators subsequently acquire scale-dependent renormalizations, so engineering dimensions alone do not determine their quantum correlations.

The action also fixes the classical equation. For compactly supported variations, integration by parts and invariance of the trace give the following expression, with the derivative acting in the adjoint representation on curvature:

This variation concerns smooth classical fields. Solving the resulting nonlinear partial differential equation is not the same problem as constructing its quantum theory: the latter requires control of all fluctuations and renormalized observables. The distinction persists even if particular classical solutions are known exactly.

03 / A FINITE, EXACT STARTING POINT

The Wilson measure and its classical normalization.

Take a periodic box of side \(R\), with \(R/a\) an integer. A link matrix transports a colour vector from its terminal site back to its initial site. This orientation fixes its small-spacing sign consistently with our covariant derivative.

Gauge transformations act independently at lattice sites. Opposite orientations use inverse matrices. The closed product around a plaquette therefore transforms only at its basepoint:

The Wilson action sums each elementary square once. Its gauge invariance follows directly from conjugation invariance of the trace; no gauge-fixing term is required to define this finite integral. [2]

Normalized Haar measure supplies one integration per independent link. Its left and right invariance preserves gauge symmetry in the measure as well as in the action.

Every plaquette term lies between zero and two. If \(N_p\) is the finite number of plaquettes, compactness gives a simple existence estimate:

This estimate deteriorates with the number of plaquettes. It establishes finite-lattice existence, not the uniform estimates needed for a continuum theory.

Dividing by the partition function removes an overall normalization, including field-independent contributions to the action. It does not make the limiting observables automatically finite. Numerator and denominator can each contain substantial regulator dependence, while composite insertions need their own renormalization. Conversely, a vanishing unnormalized partition function along a sequence of regulators would not alone prove the failure of normalized expectations. The relevant issue is control of the ratios and the resulting hierarchy, not the numerical size of the partition function considered in isolation.

To check the normalization, evaluate links on a fixed smooth classical connection. Expanding the holonomy and using traceless curvature yields

Multiplication by \(2N/g_0^2\), followed by the Riemann-sum limit, reproduces the continuum coefficient:

This is a classical consistency calculation at fixed coupling and controlled smooth background. Typical quantum fields need not be smooth, and the quantum continuum limit requires coupling renormalization. The expansion cannot justify exchanging a limit with the full path integral. [1, §6.5, pp. 11–12]

04 / WHAT MUST ACTUALLY CONVERGE

Renormalized observables, tested as distributions.

A continuum construction should identify local gauge-invariant observables, such as curvature composites, rather than demand pointwise convergence of a gauge potential. Composite operators generally mix with other local operators allowed by the symmetries, including the identity where appropriate. Schematically, after fixing a renormalization prescription and scale \(\mu\),

The coefficients may contain dimensional powers of the cutoff as well as logarithms. Products at coincident points require additional contact prescriptions. Accordingly, the appropriate convergence language is that of distributions. For a smooth compactly supported test function, the lattice approximation to a smeared field is

Test functions have fixed physical support as the regulators change. This prevents an apparent convergence obtained merely by shrinking every observable to an irrelevant scale. The desired hierarchy contains all suitable products of these smeared observables, with compatible prescriptions:

A proposed theorem must specify a limiting procedure. One possible target is a continuum limit at fixed volume, followed by an infinite-volume limit; another uses an explicitly controlled joint sequence. Neither order is automatic, and interchanging them requires a separate argument.

The displayed limit is a construction target, not a theorem being asserted here. Useful estimates would bound distributions by test-function seminorms uniformly in the regulators, in a stated range of parameters. For example, a bound of the schematic form

can provide compactness or continuity information when its constants and topology are specified. It does not by itself prove convergence, nontriviality, or the required short-distance behavior. Convergence away from coincident points likewise leaves the distributional extensions unresolved. The official problem explicitly warns that weak existence alone is insufficient without establishing the physical properties of the limit. [1, §4 and §6.5, including p. 12, footnote 2]

There are two further distinctions to keep visible. First, the existence of a convergent subsequence does not establish independence of the chosen regulator sequence or renormalization conditions; the construction must state the sense in which its continuum object is identified. Second, finite bounds at each correlation order do not specify how their constants grow with that order. A reconstruction theorem may require quantitative growth control across the entire hierarchy. These are separate analytic obligations, rather than consequences of writing a formal generating functional.

05 / FROM EUCLIDEAN DATA TO STATES

Reflection positivity builds the physical inner product.

Let \(\mathfrak A_+\) be an algebra of suitable gauge-invariant Euclidean observables supported at strictly positive times. Write \(\mathcal S\) for their Schwinger functional; this notation does not assume a continuum measure on ordinary smooth connections. For scalar observables, time reflection and complex conjugation define an antilinear operation. Tensor observables require the corresponding reflected index transformation.

Reflection positivity is positivity of a particular sesquilinear form, not ordinary positivity of a probability density. It must hold for every finite collection of allowed positive-time observables:

The finite-dimensional positivity condition implies a Cauchy–Schwarz inequality. In particular, an observable with zero norm has zero inner product with every other allowed observable. This is why quotienting out null vectors is mathematically consistent, rather than an arbitrary removal of inconvenient states:

The reconstructed norm is determined by correlations across the reflection plane. It is generally different from an ordinary Euclidean square-integrability norm on field configurations.

The null vectors must be removed before completing the space. The constant observable represents the vacuum:

Under the remaining reconstruction hypotheses, positive Euclidean time translations descend to a strongly continuous self-adjoint contraction semigroup. Its generator is the physical Hamiltonian:

Reflection positivity alone is insufficient. Reconstruction also needs the applicable Euclidean covariance, permutation symmetry, and regularity or growth conditions; clustering is associated with the unique-vacuum requirement. The precise hypotheses depend on the field framework. The 1975 Osterwalder–Schrader paper corrects and extends the 1973 treatment, so both are relevant. [3]

Wilson lattice regularization has reflection positivity in the appropriate lattice formulation. Passing it to a limit requires convergence of the very reflected products defining this form; arbitrary approximations or composite-field prescriptions cannot be assumed to preserve it. Restoring continuous Euclidean symmetries and controlling growth remain additional obligations. [1, §6.5, p. 11]

Even successful reconstruction yields only a nonnegative Hamiltonian. It supplies the setting in which the separate mass-gap question becomes precise; it does not supply a positive lower bound above the vacuum.

06 / RUNNING COUPLING AND PHYSICAL SCALE

Asymptotic freedom does not determine the infrared spectrum

Use the convention \(\beta(g)=\mu\,dg/d\mu\), with no matter fields and \(f^{acd}f^{bcd}=C_A\delta^{ab}\). For \(SU(N)\), \(C_A=N\). The first two coefficients are universal under analytic changes between mass-independent coupling schemes with the same leading normalization. They describe a weak-coupling expansion; neither coefficient is a nonperturbative spectral estimate. The original asymptotic-freedom papers and the PDG review give the relevant conventions. [6] [7] [8]

At one loop, differentiating the inverse squared coupling makes the equation linear. Integrating between two high-energy scales gives logarithmic weakening, provided both couplings remain in the perturbative regime:

Integrating the reciprocal beta function at small coupling yields the conventional scheme-dependent scale \(\Lambda_{\mathsf s}\). The power prefactor is the two-loop correction to the essential exponential; higher orders enter the indicated factor.

The prefactor follows by expanding the reciprocal beta function, rather than assuming that the one-loop solution remains exact. The logarithm in its integral produces the power of the coupling. This calculation is local to the ultraviolet regime: integrating a truncated beta function into strong coupling does not control the omitted terms there.

The normalization of this integration constant is a convention, not an observable. For example, the following analytic redefinition leaves \(b_0,b_1\) unchanged but rescales \(\Lambda\). A comparison between a lattice coupling and \(\overline{\mathrm{MS}}\) therefore requires matching.

For a continuum pure theory at fixed theta angle, dimensional analysis permits masses and string tensions to be expressed in units of this scale. It supplies neither their existence nor the positivity of their dimensionless coefficients. In particular, the perturbative formula becoming unreliable near \(\Lambda\) is not a derivation of confinement or a mass gap.

Finally, distinguish physical and lattice units. In a massive correlation channel, retaining a finite physical correlation length while removing the cutoff requires its length in lattice sites to diverge. The vanishing of the dimensionless lattice mass is therefore compatible with a positive continuum mass:

07 / CONSTRUCTIVE RENORMALIZATION

What a multiscale construction must control

Wilson regularization makes the finite-lattice theory an integral over compact-group variables. Bałaban’s program organizes that integral into successive scales, using averaging maps compatible with gauge transformations and gauges adapted to local geometry. This is substantially more structure than replacing neighboring matrices by their arithmetic average. A schematic normalized blocking kernel gives the exact marginalization identity below; it is an explanatory model, not a transcription of Bałaban’s particular transformation. [2] [4]

With invariant integration measures, covariance of the kernel transfers gauge invariance to the coarse density. The coarse gauge transformation \(\bar h\) must be defined consistently with the fine transformation \(h\); arbitrary independent transformations of the two arguments are not intended.

After blocking, the action does not stay inside a one-parameter Wilson family. An exact description generates all symmetry-allowed interactions, including extended loops. One can organize local operators and the remaining contributions on connected collections of blocks, called polymers. With overlaps assigned consistently, a schematic organization is

The operator sum is generally infinite. Discarding it produces an approximation unless a theorem bounds the discarded terms. In a convergent polymer analysis, decay in polymer size must compensate for the growing number of possible polymers. A representative type of weighted estimate is shown here; its norm, domain, and scale dependence must be specified in an actual proof.

Small-field regions permit expansions around controlled background configurations. Large-field regions need separate bounds from the action's suppression, together with control of boundaries and combinatorics. Small bare coupling does not make every configuration small. Bałaban’s 1987 analysis constructs localized effective actions and recursive coupling renormalization in the small-field region; the later large-field papers extend the effective-density analysis. [4]

Gauge fixing here is an analytical device, not a physical observable. A construction must show that gauge choices do not change the gauge-invariant quantities eventually reconstructed. Similarly, an exponential decay bound for a single-scale fluctuation covariance is not the mass-gap estimate: after all scales are combined, arbitrarily long-wavelength physical modes may still contribute. The constants and their dependence on scale must remain visible throughout this comparison.

08 / SPECTRAL THEORY AND CORRELATIONS

A uniform statement about the physical Hilbert space

Assume reconstruction has produced a nonnegative self-adjoint Hamiltonian \(H\) and a unique normalized vacuum \(\Omega\). Write \(P_0=|\Omega\rangle\langle\Omega|\) and \(Q=1-P_0\). The gap is a spectral exclusion, equivalently a quadratic-form inequality on the form domain. It need not correspond to an isolated one-particle eigenstate. [1]

Functional calculus turns spectral support into a contraction estimate, valid at every positive Euclidean time. Conversely, the operator-norm estimate excludes spectrum below \(\Delta\) on the vacuum-orthogonal subspace:

Use a properly smeared physical observable \(A\), with \(A\Omega\) in the Hilbert space, and remove its vacuum expectation. Point fields are distributions and cannot be substituted without domain control. The spectral theorem supplies a positive finite measure for the resulting state:

The converse requires enough states. Suppose a linear dense set in \(Q\mathcal H\) obeys exponential bounds with one common rate \(\Delta\); the finite prefactor may depend on the state. Positivity, rather than cancellation, then forces each spectral measure to vanish below that rate:

For every \(0<\epsilon<\Delta\), the associated spectral projection annihilates a dense set and hence vanishes. A single glueball correlator only tests states to which its operator couples. Different positive rates in different channels, with infimum zero, also fail to prove a global gap. Finite-time numerical fits do not supply the all-time inequality.

The logarithmic slope makes the channel limitation explicit. For a nonzero state and positive time, exponentially weighted spectral moments exist, so the effective mass is a mean energy with a time-dependent positive weight. Its derivative is minus the corresponding variance. It approaches the bottom of this state's spectral support, which may sit above the bottom of the full physical spectrum.

A continuum threshold can therefore produce an exponential multiplied by a power of time; an isolated pole is unnecessary for a positive spectral lower bound. Positivity is crucial to these deductions. A general cross-correlator can exhibit cancellations, whereas the autocorrelation above is the Laplace transform of a positive measure.

A positive, injective, self-adjoint lattice transfer operator gives the analogous relation at a cutoff. On its physical Hilbert space, normalize by its largest eigenvalue \(\lambda_0\); for a nonzero excited eigenvalue \(\lambda_n\),

The ratio is dimensionless; division by the temporal spacing restores energy units. Merely showing \(\lambda_1<\lambda_0\) at each finite cutoff and box does not prevent the gap from collapsing:

Here \(m_1(a,R)\) is the smallest positive excitation energy and \(R\) is the physical spatial box size; transfer-matrix evolution runs in Euclidean time. Even a uniform lower bound must be transferred to the reconstructed limit using proved convergence of states or correlation functions. Both cutoff removal and growth of the box must respect the chosen physical normalization.

09 / STATIC SOURCES AND LONG-DISTANCE DYNAMICS

Area laws ask a different spectral question

A Wilson loop probes parallel transport around a closed curve in a specified representation \(\mathcal R\). Retaining this page's Hermitian connection and \(D=d-iA\), parallel transport includes a factor of \(i\). Its continuum renormalization must handle perimeter and, for nonsmooth loops, cusp divergences.

For a rectangle of spatial length \(r\) and Euclidean duration \(T\), transfer-matrix evolution involves a sector containing two external static sources. At a reflection-positive cutoff, the positive spectral weights project onto the lowest source-sector energy when the trial state overlaps it. Subtracting the static self-energies defines a renormalized potential. [2] [5]

An asymptotic area law, with sufficiently controlled limits and subleading perimeter terms, gives a linearly rising potential. Its coefficient has dimensions of mass squared:

The vacuum mass gap concerns gauge-invariant excitations without these external sources. It does not automatically give a string tension. In pure \(SU(N)\), gluons can completely screen sources with zero center charge, also called zero N-ality. Sources with nonzero center charge cannot be completely screened by gluons alone. [5]

Dynamical fundamental matter can break the string. Moreover, the Clay problem includes simple groups with trivial center: in \(G_2\), gluons can screen fundamental sources, so their asymptotic string can break even in the pure theory. An area law for every representation cannot be made part of the universal mass-gap requirement. [5] Holland et al., exceptional confinement ↗

10 / READING AN EXISTENCE PROOF

Identify the theorem before auditing the estimates

An existence proof should be assessed through its exact theorem statements and dependencies. The November 10 assessment connects the released construction to Wilson regularization and Bałaban’s methods, and explicitly leaves the mass gap and uniqueness of the infinite-volume limit unresolved. The checklist here explains the logical distinctions; Part II applies them to the released sources. Assessment, pp. 1, 6–7 ↗

Specify the limit. Record the gauge group, boundary conditions, coupling trajectory, theta angle, observable algebra, and topology of convergence. Local composite fields can mix and need counterterms; convergence of the partition function alone does not settle their correlations. A representative target is convergence of suitably renormalized smeared Schwinger functions:

The limit notation abbreviates a stated joint or iterated procedure, not permission to interchange limits. At finite spacetime volume, tests can first have compact support away from the boundary; extension to Schwartz tests then needs estimates on growth and tails. If only subsequential convergence is proved, identify whether the claimed properties hold for every selected limit and whether uniqueness is actually needed for the stated result.

Check reconstruction. Prove the required covariance, symmetry, positivity, clustering, and growth estimates for the full hierarchy. Reflection positivity must hold for all admissible finite combinations of positive-time observables, not just one two-point function:

Here \(\Theta\) includes complex conjugation and Euclidean time reflection. The corrected Osterwalder–Schrader reconstruction theorem needs an appropriate growth condition; merely listing the original 1973 axioms with ordinary temperedness overlooks the correction in the 1975 paper. [3]

Identify Yang–Mills. A nonzero Hilbert space is insufficient: verify the local observable correspondence and short-distance behavior required by the official problem. Check which gauge groups the theorem covers, how continuum symmetries emerge, and what is actually established about uniqueness and nontriviality. [1]

Audit formal dependencies. Inspect imported axioms, theorem assumptions, placeholders, and the connection between formal definitions and physical quantities. A machine-checked implication retains its hypotheses. Establishing existence would still leave a separate, quantitative demand on the resulting Hamiltonian:

No uniform lower bound over different groups is required. Within each theory, however, the bound must survive the continuum and infinite-volume limits in physical units. That is the additional statement a complete Millennium solution must establish.

11 / THE THEOREM AND ITS QUANTIFIERS

Read the quantifiers before the proof.

The release constructs a family of Wilson lattice theories, removes the ultraviolet cutoff at fixed physical volume, extracts an infinite-volume limit, and reconstructs relativistic quantum fields from its clustering components. The committee’s assessment treats this as an existence construction. It explicitly does not identify a positive spectral gap or a unique infinite-volume theory. The following is a technical reading of the assessment and selected definitions and theorem statements. Assessment, pp. 1–3 ↗

The concrete input is a closed matrix subgroup \(G\subset U(m)\) with nonabelian simple real Lie algebra \(\mathfrak g\), together with a positive parameter \(b\) in the blocking kernel. This is the submission’s matrix realization of the compact simple gauge-group setting. The penalty \(b\) is fixed before the family is chosen; it is not a physical gluon mass. Let \(C\) denote the resulting family, including its cutoff schedule, local lattice representatives, and renormalization data. A useful compressed reading of the committee’s CombinedAcceptance statement is

Here \(p\) is a requested extraction order, \(q\) is its refinement, \(S\) is a continuum correlation hierarchy, and \(F\) contains the fields, phase construction, stress tensor, and relativistic reconstruction associated with that hierarchy. The lattice clause replaces the ultraviolet extraction by the full chosen ultraviolet sequence while retaining a refined volume sequence. These choices are coordinated: the same family and the same reconstructed object serve all the clauses. Acceptance specification ↗ · Combined theorem ↗

The last line matters. A short public theorem can conceal connections between its abstract labels and the actual Wilson observables. The stronger construction retains C.Makes, which links the local representatives to classical curvature data, and Spec.Final, which carries the selected reconstruction. The combined acceptance theorem preserves these connections while adding independently formulated checks. It does not turn every retained source clause into an independent semantic verification. Throughout this account, “the source proves” and “the committee separately checks” are distinguished where their scopes differ. Joint source construction ↗

12 / THE STARTING WILSON LAW

A Wilson law with fixed conventions.

Convention switch. Part I used Hermitian matrices. The released submission uses an anti-Hermitian connection: \(A^{\mathrm{ah}}=-iA^{\mathrm{h}}\). We follow its convention in Part II, and do not transfer numerical coupling normalizations between the two parts without converting the trace normalization. The curvature and invariant inner product used in the assessment are

At cutoff \((N,p)\), the physical torus has side length \(L_p\) and spacing \(a_N\). A link variable \(U_e\in G\) is assigned to each positively oriented edge, with \(U_{-e}=U_e^{-1}\). The plaquette holonomy \(U_\square\) is the ordered product around its boundary. Compactness of \(G\) and normalized Haar measure give an ordinary finite-dimensional probability law:

Each positively oriented plaquette is counted once. Expanding a smooth-link plaquette explains the classical normalization, although the proof uses the exact lattice law rather than this expansion as a replacement for it. In these conventions the cutoff coupling satisfies \(\beta=1/g_0^2\), with formal continuum action

This \(\beta\) differs from frequently used lattice conventions such as \(2N/g^2\). In the committee’s bridge to an independently formalized lattice-gauge library, the parameter passed to that library is \(\beta_{\mathrm{LGT}}=\beta_{\mathrm{submission}}/m\). The bridge proves equality of the finite probability laws under the configuration identification, and the combined theorem ties the same law to the subsequent moments. Such a normalization check is essential: merely naming a measure “Wilson” would not establish that connection. Assessment, pp. 4–5 ↗ · Finite-law bridge ↗

The continuum output is a hierarchy of correlations and its reconstructed local quantum fields. The submission does not assert that it has constructed a probability measure \(e^{-S_E[A]}DA\) on a space of continuum gauge potentials. That distinction avoids assigning a literal measure-theoretic meaning to an undefined \(DA\). The precise relationship to Yang–Mills comes through the starting lattice law, the field dictionary, and the quantitative continuum properties described below.

13 / EXACT RENORMALIZATION GROUP

Exact blocking and a bounded RG trajectory.

The stability statement concerns a sequence of represented densities, with quantitative bounds on their local coefficients. It has three ingredients: an exact integral identity, analytic estimates on the output activities, and numerical choices making those estimates fit the next scale's bounds. The last ingredient does not replace the analytic estimates. The construction also retains the history of admissible Wilson descendants; it does not claim that every density satisfying a few norm bounds admits the required next step.

1. Exact blocking retains the large fields

Writing \(m+1\) for the finer scale and absorbing the chosen coupling into the blocking kernel,

All Haar measures here are probability measures. Once absolute integrability is proved, Fubini and the second identity give \(\int\rho_m=\int\rho_{m+1}\). Iteration therefore preserves the original partition integral exactly. Suppressing the extensive indexing, the representation is

The canonical action contains the actual inverse coupling, with fixed kinetic normalization. The regular term \(E_m\) is a sum of masked local activities; \(\mathcal G_m\) is a gas of forced activities, not the gauge group. Let \(\mathfrak F_t(U)\) be the set of sites touching a plaquette whose group chord exceeds threshold \(t\). A support \(X\) is healthy if \(X\cap\mathfrak F_t(U)=\varnothing\). In this notation the activity formula reads

Admissibility means that the forced supports are disjoint and their marked inventories cover exactly the flagged sites. Each nonzero forced coefficient must obey its eligibility test. Thus bad-field regions are accounted for, not discarded. If there are no flagged sites, only the empty family contributes and \(\mathcal G_m(U)=1\). The regular representatives vanish at the flat configuration and have zero Hessian on the specified affine directions on nonwrapping supports. This separates the regular remainder from the canonical quadratic term. Activity representation and admissibility ↗

2. The bounds carry derivatives and support size

At a fixed scale let \(r\) be the derivative order, \(q\) the growth power, \(\sigma>0\) the support exponent and \(B\) the cover size. For a regular activity anchored at \(o_i\), a majorant \(b_i\) bounds every ordered left-link derivative of order at most \(r\) on healthy configurations. The tested directions satisfy

These are bounds on actual unmasked representatives and their derivatives; differentiating a discontinuous indicator is not being assumed. Let \(c_B(X)\) denote the source's support cover number. Define the support weight \(w_\sigma(X)=e^{\sigma c_B(X)}\). The regular and forced budget conditions are

The source includes nonnegativity and absolute summability in these conditions. The distinction matters: regular terms are charged at their anchor; forced terms are charged at every support site. Increasing support incurs an exponential penalty. Locality also includes the neighboring plaquettes inspected when a flag is tested, so an apparent short support cannot hide additional physical reads. Derivative and support weights ↗ · Exact budget definitions ↗

These bounds already explain why the represented sums make sense. Since \(w_\sigma\ge1\) and every forced support is nonempty, summing over the finite lattice gives \(\sum_i b_i\le|\Lambda_m|\delta_m\) and \(\sum_i k_i\le|\Lambda_m|\omega_m\). Dropping the compatibility restriction supplies a positive majorant:

The infinite-product identity follows from the finite identities and monotone convergence; \(\log(1+k_i)\le k_i\) gives the last bound. Along with the canonical-action bounds, this produces a measurable bounded envelope and hence integrability at each finite lattice. This envelope can grow with volume; it is not the later volume-uniform correlation estimate. Absolute summability proof ↗ · Density envelopes and exact partition transfer ↗

3. Why a step fits the next bounds

The numerical scale profile uses the hatted inverse coupling \(x=\widehat\beta_m\), distinct from the actual trial coupling. Write

For a fixed stage transition, let \(L\) be the block ratio, \(y=x+c_G\log L\), and \(g=\beta^{-1/2}\) for the actual incoming coupling. It stays in a prescribed band comparable to \(y\). The stage chooses constants \(C,H,D,\eta,\theta\), with \(H\ge1\) and \(\eta>0\), before delaying the entrance coupling. Define the following condensed version of the source's numerical prices:

The analytic output estimates, in the same weighted activity bounds just defined, have the price structure

The norm notation abbreviates the existence of the corresponding majorants, rather than introducing a new unproved Banach norm. The displayed inequalities are the output obligations consumed by numerical closure. Their analytic producer constructs a common activity family throughout each trial-coupling interval; the reset argument then attaches its estimates to the returned density. This requires local integration and connected-activity bounds. Merely choosing a schedule would not prove the inequalities. Exact one-step prices ↗ · Output price obligations ↗ · Construction and bounds of the activity family ↗ · Actual reset output and its estimates ↗

The closure mechanism can now be seen explicitly. Choose \(L\) large enough relative to the pre-block constant \(C\), making \(C^2/L\) small. At a fixed transition, \(P(x)=x^{o(1)}\), \(B(x)=x^{o(1)}\), and \(y=x+\text{constant}\). Thus the cubic terms, relative to the allowed regular error, obey

The two exponential remainder terms dominate their polynomial prefactors as the entrance coupling grows. Also \(\delta(y)/\delta(x)\to1\). These facts leave strict room below \(\delta(x)\). For the forced contribution, the enlarged input band gives \(g\le2\widehat g(y)\), hence \(t(y)/g\ge P(y)/2\). Absorbing the fixed prefactor at a sufficiently late entrance yields

where the schedule supplies a suitable positive \(\gamma\). The intermediate strong forced bound is retained for later calls, not weakened immediately to \(\omega(x)\). The source's recloses proves these implications from the paid numerical gates; reload applies them to the actual returned activities and preserves the certified history. Canonical path bounds, locality and continuity along trial-coupling intervals are additional induction data. The same-stage transitions must work repeatedly without silently spending derivative order at every step. Numerical closure ↗ · Closure on represented descendants ↗

4. Inside one local RG integration

5. Select a trajectory through the bands

The histories here are finite. More precisely, \(\beta_m\) abbreviates \(\beta_{N,p,m}\), with \(0\le m\le N\). Even at fixed \(m\), the selected value may depend on the terminal cutoff \(N\) and volume \(p\); consistency of one infinite trajectory across all cutoffs is not being assumed.

A valid output representation need not automatically have its coupling in the next admissible band. The construction varies the bare coupling along continuous families and restricts to surviving intervals. With \(s_m=\sum_{j<m}\log L_j\), the inner barriers are \(\bar\beta+\tfrac12c_Gs_m\) and \(\bar\beta+2c_Gs_m\). They lie inside the allowed band \([\widehat\beta_m/2,2\widehat\beta_m]\).

The exact transfer records

Write these barriers as \(\ell_m\) and \(u_m\). If \(d_{m+1}\le(c_G/4)\log L_m\), the step \(\beta_m=\beta_{m+1}-c_G\log L_m+e_m\), with \(|e_m|\le d_{m+1}\), gives

Continuity and the intermediate value theorem select a closed subinterval covering the full next barrier interval and staying between its endpoints. Restricting to this subinterval preserves the certificates at all already-constructed scales. Induction down from the cutoff therefore produces a trajectory reaching \(\beta_0=\bar\beta\), with every intermediate representation valid and banded. This is the shooting argument; it does not assert uniqueness of the selected bare coupling. Barrier and interval-selection proof ↗

6. The accumulated drift is subleading

The stage index \(J_m\) increases slowly but tends to infinity. Its budget condition includes

Set \(\widehat\beta_m=\bar\beta+c_Gs_m\). Since the trajectory reaches \(\bar\beta\), telescoping (64) gives

For any \(\varepsilon>0\), choose \(j_0\) so that \(J_{j+1}^{-2}\le\varepsilon/2\) for \(j\ge j_0\). The finite initial cost is \(D_0=\sum_{j<j_0}d_{j+1}\). Then

Since \(s_m\to\infty\) and \(\varepsilon\) is arbitrary, the ratio tends to zero. All quantities on the right are independent of the volume and terminal cutoff. At the Wilson endpoint, \(s_N=\log(a_N^{-1})\) in the fixed reference units, and the finite \(\bar\beta/s_N\) also vanishes. Consequently,

This is the source's explicit trajectory argument, not an inference from the formal beta function alone. \(C_G\) is the adjoint Casimir in the chosen invariant metric. Telescoping and common-history assembly ↗

Where Bałaban enters. The assessment identifies the marginal RG analysis with Bałaban's general program, augmented by symmetry-preserving gauge fixing and propagated irrelevant-term control. The detailed estimates above show the role of small-/large-field separation, support weights, nonlinear error budgets and kinetic normalization. They do not establish a mass gap or agreement with perturbation theory at every loop order. The local analytic integration estimates are stated here at their usable output strength; their connected-activity and gauge-analysis subproofs remain in the linked implementation. Assessment, p. 6 ↗

14 / RENORMALIZED LOCAL FIELDS

A common dictionary of local fields.

The proof needs more than one well-behaved observable. It supplies a compatible family of local field labels built from gauge-invariant curvature polynomials and covariant derivatives. Curvature has weight two; each derivative raises the weight by one. Products add weights. At any finite degree cutoff, the raw label space \(V_d\) is finite-dimensional:

A local section associates a label with a gauge-invariant polynomial \(L_v\) of finitely many nearby links. Its smooth-connection expansion has to reproduce the prescribed curvature data to the required order. This condition is part of C.Makes; a convenient-looking plaquette expression cannot simply be substituted for the actual symmetry-compatible local section. Quantum renormalization then mixes labels within the finite-degree space. For a basis \(v_1,\ldots,v_M\), the renormalized insertion is

The coefficients are linear in the label, depend on both cutoffs, and can diverge as the spacing vanishes. Their role is to make the correlations converge. They are defined using the constructed RG transformations, rather than supplied as explicit Feynman-integral formulae. Assessment, pp. 5–6 ↗

The committee’s source walkthrough makes the actual formula more explicit. At degree \(d\), let \(B_{d,N,p}\) denote the inverse terminal renormalization followed by the inverse threshold scaling. Both maps are linear equivalences at finite cutoff. With \(\mathcal L_d\) denoting the translated local section,

Finite-cutoff invertibility says no raw coefficient direction is thrown away by this change of basis. It does not bound its inverse uniformly in \(N\), and it does not prove that every raw direction gives a nonzero continuum field. The latter question requires the null-space and faithfulness results. OctoberChain, insertion formula and its limitations ↗

The quotient accounts for relations such as those associated with Bianchi identities and equations of motion, formulated through correlations. The construction identifies this finite-degree field content with a corresponding gauge-invariant free-curvature polynomial space modulo its relations. This is a statement about the vector space of local fields, not equality of interacting and free correlation functions. The same correlation-null spaces occur in every component phase. The accepted reconstruction also supplies dense field words and faithfulness: a nonzero quotient label acts nontrivially for some test function. Assessment, pp. 4, 8 ↗ · Accepted operator content ↗

18 / SPACETIME SYMMETRY AND NORMALIZATION

Recovering Lorentz symmetry from the lattice.

What the regulator breaks

The lattice regulator is not Lorentz invariant. More precisely, the construction starts in Euclidean signature: the symmetry to recover there is continuous Euclidean covariance, before reconstructing a Minkowski theory. At nonzero spacing, a hypercubic lattice singles out coordinate axes. Coordinate permutations, sign changes and lattice translations remain, but arbitrary rotations and arbitrarily small translations do not preserve its sites. Exact lattice gauge invariance concerns the internal description of the fields; it does not supply these missing spacetime symmetries.

Nor does the continuum limit alone settle the issue. Taking \(a\to0\) at fixed physical size \(L\) gives a theory on a finite torus, whose identifications still distinguish directions. The target is continuous Euclidean covariance of the continuum, infinite-volume correlations. For scalar labels invariant under \(O(4)\), its orthogonal-and-translation covariance can be written

For fields with tensor indices, the corresponding rotation matrices also act on those indices. This is a property to prove for the limiting hierarchy. Convergence of regulated correlations, by itself, says nothing about invariance under transformations absent at finite cutoff.

How the new construction restores the symmetry

The argument preserves discrete covariance, relates local angular variations to an exterior stress integral, proves that integral vanishes, and integrates the resulting infinitesimal identity. Below, the difficult spectral estimates are stated with their hypotheses and linked to their proofs. The derivation keeps the initial shell charges separate from their later normalization. Assessment §6.3 ↗

1. Preserve the lattice symmetries

The averaged object is a coarse-graining kernel with auxiliary frame and bond choices. Let \(U\) and \(V\) be fine and coarse lattice configurations, \(\tau\) a complete set of compatible choices, \(Q_U(\tau)\) the predicted coarse configuration, and \(N_\beta(Q,V)\) the normalized noise kernel. In abbreviated notation,

On the small-field branch, a unique normalized minimizing frame is used. On the fallback branch, normalized axial frames are averaged uniformly over eligible root-and-coordinate-order tags. These choices are shared consistently across the configuration. This is not an arithmetic average of group-valued gauge transformations.

A signed coordinate permutation \(s\) induces a bijection \(r_s\) on the complete choice arrays. Write \(sU\) for the corresponding pullback. The frame, weight and noise constructions obey

For minimizing frames, transformed minimizers are identified using uniqueness. For the axial branch, the symmetry bijects the eligible sets and preserves their cardinality. Once these facts are proved, covariance of the full kernel follows by finite-sum reindexing:

The same construction handles compatible fine/coarse lattice translations. The proof of StepData.channel uses the branch hypotheses; Axial.viable supplies universal threshold choices for them. These results preserve the regulator's existing symmetries, without yet establishing continuous rotations. Frame transport ↗ · Kernel covariance proof ↗ · Branch thresholds ↗

2. From local Ward identities to exterior flux

For distinct insertion points, write \(S(\mathbf v;\mathbf x)=\langle\mathcal O\rangle\), with \(\mathcal O=\prod_{i=1}^{m}\Phi_{v_i}(x_i)\). At this stage \(J_{\mu\nu}\) denotes the conserved symmetric stress field and \(p_\mu,Y_A\) its raw translation and angular shell-charge maps. They must still be normalized. Here \(A^T=-A\) is a spacetime rotation generator, unrelated to the gauge connection denoted \(A\) elsewhere.

Choose a smooth window \(\chi_i\), equal to one near \(x_i\), with compact support avoiding the other insertions. The local identities have the form

Repeated spacetime indices are summed from zero to three. Splitting \(Ay=A(y-x_i)+Ax_i\) combines the two charges. Define the total angular variation by

Let \(F:\mathbb R^4\to\mathbb R^4\) be smooth and compactly supported, agreeing with \(Ay\) in a neighborhood of every insertion. Subtracting the local windows and integrating by parts on supports separated from the insertions gives the exact identity

Near an insertion, the contracted integrand cancels because \(\partial_\mu F^\nu=A^\nu{}_\mu\) and

The remaining integral measures the exterior obstruction to invariance. Conservation is used away from collisions; the argument does not assume an unrestricted contact-term formula. Euclid/Stress.lean defines the shell identities, while River.flux_value proves this local-to-exterior implication from the stress and shell hypotheses. Shell definitions ↗ · Exterior identity proof ↗

3. Make the exterior contribution vanish

Sending a boundary to infinity is not sufficient by itself: a nonzero exterior term would obstruct symmetry of the state. The proof first treats distinct insertions lying on a coordinate line. For each rotation-plane generator it constructs smooth compactly supported vector fields \(F_n\) that eventually agree with \(Ay\) near every fixed insertion. Opposing cross-derivatives cancel exactly, leaving exterior contributions that can be estimated.

The cutoffs are deliberately anisotropic. With \(\lambda=e^n\), the construction uses reciprocal length scales such as \(\lambda^{-100}\) in the rotation plane, \(\lambda^{-150}\) along a remaining insertion-line direction, and \(\lambda^{-300}\) in other directions. Their purpose is to make the different spectral error terms compatible; a spherical cutoff is not being assumed.

For these cutoffs and fixed admissible field insertions, the estimate takes the form

The constant and starting index may depend on the fixed data. In the source, Budget(a,f) means an eventual bound \(\lVert f_n\rVert\le C e^{an}\). The lemmas blows and drained establish exponent \(-1\) for the exterior pieces. The local-to-exterior identity makes the same, fixed angular variation equal to \(B_n\) for every sufficiently large \(n\). Therefore,

This is the conclusion of axisTurn; linearity extends it from rotation planes to every skew generator. At this stage it is a collinear Ward identity. The extension to general configurations is another step below. Meaning of the bound ↗ · Cutoff construction and vanishing proof ↗

Where reflection positivity enters the estimate

Each coordinate reflection \(\Theta_i\) supplies an OS positive form for observables supported on its positive side. After quotienting its null vectors, Cauchy–Schwarz gives

The associated spectral measures are positive and supported on \(E_i=p_i\ge0\). No continuous rotational covariance is used to obtain the coordinate-plane forms. The proof compares the same tested correlation using different axes. For the clusters with the required opposite reflection parities, this comparison bounds a positive spectral contribution by correction terms. High-frequency remainders are controlled by moment estimates; the low-energy region requires separate cluster and stress estimates.

The two-plane cancellation can be written explicitly. For a fixed cluster \(A\), let \(\nu_i^A\) be its positive spectral measure for reflection axis \(i\). Write \(R_i(h;E_i)\) for the real Laplace response of an admissible profile \(h\), the source's Reel. With transverse profiles \(f=(f_0,\ldots,f_3)\), define

Suppose \(A\) has opposite reflection parities in axes \(i\ne j\). For profiles satisfying the source's contact-separation and integrability conditions, the same tested distribution read in the two directions gives \(I_i[f_i;f]=-I_j[f_j;f]\). Choose admissible subtraction profiles \(a,d\) and apply this identity after replacing \(f_i\) by \(f_i-a\) and \(f_j\) by \(f_j-d\). Linearity gives

If the transverse \(f_k\) are nonnegative and \(R_j(f_j-d;E_j)\ge0\) for \(E_j\ge0\), then \(X_j\ge0\), hence \(X_i\le C_i+C_j\). This is the algebra behind alternate: positivity in the second plane bounds the first-plane contribution by controlled replacements. The two spectral measures are not assumed equal, and the identity is not asserted for arbitrary clusters or tests. Opposite-parity identity and subtraction proof ↗

A representative diagonal-stress estimate is the following. Let \(W_n=e^{an}\), \(a\le0\), and let \(\Pi_n(p_\perp)=\prod_{j\ne i}f_{j,n}(p_j)\) be the product of the admissible smooth, compactly supported, nonnegative, even transverse test profiles, whose widths \(e^{w_jn}\) obey \(w_j\le a\) for \(j\ne i\). With a reflection-even, pointlike representative of \(J_{ii}\), its positive spectral measure obeys, for every \(\delta>0\),

This uses conservation, stress symmetry and reality, discrete tensor covariance, and compatibility of field derivatives. It is a specific tested estimate, not a universal assertion about arbitrary states or profiles. Together with the collinear-cluster bounds, weighted Cauchy–Schwarz and Young inequalities control mixed stress–field contributions; the cutoff choices then yield the negative exponent in \(B_n\). Zero-energy atoms are included in the analysis. Comparison of reflection axes ↗ · Cluster spectral bounds ↗ · Diagonal-stress estimate ↗ · Combining the bounds ↗

One can also see how the exponents combine. Suppose two nonnegative diagonal spectral integrals obey \(X_n\le C e^{\alpha n}\) and \(Y_n\le D e^{\beta n}\). Under the mixed-integrand domination and integrability hypotheses, the weighted Cauchy–Schwarz estimate gives, for every \(\lambda>0\),

A prefactor bounded by \(M e^{\gamma n}\) adds \(\gamma\) to the resulting exponent. These are the source's balanced and balanced_mul rules. They explain the conversion of diagonal spectral estimates into a mixed flux bound. The harder inputs remain the stated stress and cluster estimates and the admissible cutoff widths; the averaging of exponents alone would not establish their values.

4. From infinitesimal invariance to finite rotations

The normalization argument identifies \(p_\mu=cD_\mu\) with a common nonzero scalar \(c\), where inserting \(D_\mu v_i\) differentiates the correlation at \(x_i\). The normalized angular map is \(\rho(A)=c^{-1}Y_A\). This identification uses the construction's short-distance comparison; it does not follow from stress conservation alone.

The collinear identity is extended using analytic continuation along admissible configuration-space paths, compatibility with field derivatives, and smoothness. The result for all separated configurations is

Analyticity is essential here: vanishing on collinear configurations alone, a lower-dimensional set, would not imply this result for an arbitrary smooth function. The source proves the continuation through its Ward identities and derivative relations. It also constructs a compatible group action on the finite-dimensional filtered field spaces; Lie brackets alone are not substituted for that construction. Normalization and assembly ↗ · Continuation along paths ↗ · Global angular identity ↗

Write \(R_t=e^{tA}\) and let \(D(R_t)\) act on the field labels with infinitesimal generator \(\rho(A)\). Consider the joint position-and-label orbit

The derivative includes both the moving positions and moving labels; the normalized angular identity sets their sum to zero at every point of the orbit. Thus covariance holds for every one-parameter rotation group. Together with the already preserved signed coordinate transformations, these generate \(O(4)\). The implementation proves joint differentiability before applying the chain rule, then establishes covariance under products and inverses. Proof of finite rotation covariance ↗

5. Covariance of each component theory

Invariance of a mixture does not imply invariance of every component: a symmetric average can conceal symmetry-breaking phases. The construction therefore supplies a separate phase-level argument for the compressed whole-word operators \(M(w)\) in the vacuum algebra. Covariance of widely time-separated pairs of field words, together with individual time-reflection invariance, lets the proof conjugate that reflection into arbitrary hyperplane reflections. Testing against arbitrary words and using separation in the vacuum algebra promotes the tested equalities to operator identities. Hyperplane reflections generate \(O(4)\), so evaluating at a character gives

Here \(Qw\) transforms both the positions and field labels of the admissible word. This extra argument establishes orthogonal covariance of each \(S_\xi\), rather than only of \(S_{\mathrm{mix}}\). Individual component invariance ↗

The stress tensor must generate translations

The tensor belongs to the degree-four local-field sector. Its symmetry, reality, discrete covariance and conservation identify a restricted tensor space; the normalization must still be fixed. At separated insertions, conservation gives the schematic kernel identity

Conservation alone would survive \(T\mapsto cT\), so it cannot normalize the operator. Choose a smooth compactly supported window \(\chi\), equal to one near a central insertion \(x_0\), whose support avoids all spectator positions. Its derivative lies in a shell separated from every insertion. The checked translation identity is

The minus sign and unit coefficient are part of the theorem. It applies to every central quotient field, every finite spectator word and every character phase, using the same tensor. The proof first identifies the shell insertion with the abstract field derivative, then identifies that derivative with differentiation of the correlation kernel. This second step connects the algebraic field dictionary to actual spacetime translations.

The committee's adapter separately checks existence of admissible windows and integrability of the shell integrand. Smoothness away from coincidences and compact support make the integral well defined. Calling the left-hand side \(Q^\chi_\nu[v;x]\), any two admissible windows give

Physically, the flux through a surrounding shell generates translation of the enclosed field. This fixes its translation charge and prevents an arbitrary rescaling of the stress tensor when comparing short-distance amplitudes. It does not eliminate every possible improvement: an improvement may leave that charge unchanged. Nor does a separated-insertion shell identity supply an unrestricted contact-term prescription or a dilation Ward identity. Those require additional arguments. Assessment §§6.3–7.1; exact Ward identity and scope.

From Euclidean covariance to Lorentz covariance

Euclidean rotations are not Lorentz boosts on the original lattice. After Euclidean covariance, reflection positivity and the required regularity and growth properties are established, analytic continuation and OS reconstruction produce the relativistic theory. Its Hilbert space carries a strongly continuous unitary action of the proper orthochronous Poincaré group. For a scalar field, the resulting covariance reads, in the distributional sense,

Tensor fields transform with their Lorentz representations as well. Thus Lorentz covariance belongs to the reconstructed Minkowski fields, and depends on the proved Euclidean properties. A formal substitution of imaginary time, without those properties, would not establish a positive Hilbert space or a unitary relativistic theory. This recovery of spacetime symmetry is part of the existence construction; it does not establish a mass gap. Assessment §6.1: relativistic reconstruction ↗

19 / IDENTIFYING THE INTERACTION

A quantitative logarithm in the stress-trace correlator.

The axioms of a relativistic field theory do not identify Yang–Mills by themselves. The released construction starts from its Wilson law and retains its invariant-polynomial field labels. A more quantitative discriminator is the centered trace of the normalized stress tensor in the selected phase \(\xi_*\):

In fixed reference length units, the theorem proves the full positive-distance limit

This is not merely convergence on a selected sequence of separations. The proof combines a renormalization-group comparison with the free curvature-square field and a logarithmic clock, then uses spectral monotonicity to upgrade the controlled comparison to the full limit. The committee also identifies the radial kernel, giving the corresponding limit for arbitrary nonzero displacement tending to zero.

Centering removes the phase's finite one-point function. This changes the disconnected constant in the two-point function, but cannot change the leading singularity after multiplication by \(r^8\log^2(1/r)\), which tends to zero.

The nonperturbative rescaling comparison

Let \(C(r)=\langle\Theta(-re_0/2)\Theta(re_0/2)\rangle_{\xi_*}\) and \(\ell(r)=\log(1/r)\). The proof follows the actual trace field through a controlled sequence of zooms \(r_i\downarrow0\). Let \(B_i\) denote its diverging inverse-coupling clock and \(\mathcal U_i\) the comparison map from the free degree-four field space to the interacting field space. The source proves both a clock limit and a field-label limit:

Here \(b_0=11C_G/(48\pi^2)>0\) in the chosen invariant metric; \(2b_0\) is the leading inverse-coupling slope in §13. The scalar \(d_2\) is obtained from the trace comparison, not fitted to its final two-point coefficient. The field convergence is in the finite-dimensional label comparison, accompanied by a theorem comparing correlations in the selected phase. It is not asserted to be convergence of unbounded field operators in operator norm. The record Scribe collects these results, including the negligible identity component; Store_l supplies it from the earlier construction. Field comparison and logarithmic clock ↗

In the source's normalization, the Wick-renormalized free curvature-square label \(\mathcal O_{\mathrm{free}}\) satisfies

This normalization differs from the conventional \(Q\) used in the perturbative check below. Multilinearity supplies a factor \((B_i r_i^4)^2\) for two trace insertions. Multiplying the correlation comparison by \((\ell(r_i)/B_i)^2\) therefore yields, at controlled generic configurations with rescaled separation tending to \(s>0\),

The nearby separations \(s_i\) include the rounding needed to compare the lattice-derived configurations with continuum points. The selected-phase correlation comparison is an essential hypothesis in this calculation; convergence of label coordinates alone would not justify replacing the two-point function by its free value. The implementation tagline combines that comparison with wick_norm and the clock limit. Free coefficient ↗ · Combining the limits ↗

Why this gives a full short-distance limit

The spectral representation provides a positive measure supported at nonnegative energy. After projecting to energy, write its time-axis two-point function as

The exponentially damped integral is finite at every positive separation; finite total mass of the undamped measure is not required. Monotonicity follows pointwise from \(e^{-rE}\le e^{-sE}\) and positivity. Choose comparison separations \(s_-<1<s_+\), close to one, with rounded \(s_i^\pm\to s_\pm\). Eventually \(r_i s_i^-<r_i<r_i s_i^+\), so positivity of \(r_i^8\ell(r_i)^2\) gives

Letting \(s_-\uparrow1\) and \(s_+\downarrow1\) removes the rounding and gives the claimed coefficient on the controlled zoom sequence. The final step is a subsequence criterion: the construction can extract such a controlled comparison from any proposed positive sequence tending to zero. If the full limit failed, a sequence staying outside a fixed neighborhood of \(A_G\) would exist; its controlled subsequence would converge to \(A_G\), a contradiction. Thus the auxiliary subsequences used in the argument do not weaken the final conclusion to a subsequential limit. Euclidean scalar covariance then replaces time-axis separation by \(|x|\). These steps are implemented by monotone_axis, border, and outside. Monotonicity, squeeze and full-limit proof ↗

A separate perturbative normalization check

The numerical coefficient has a useful perturbative explanation, explicitly separate from that analytic proof. For this comparison only, use canonically normalized component fields, with the coupling in the nonlinear curvature term. This differs from the release's anti-Hermitian connection with the coupling outside the action. With ordered spacetime indices summed, free Wick contraction gives

Writing \(\ell=\log(1/(\Lambda r))\), one-loop running and the trace-anomaly normalization then predict

A fixed \(\Lambda>0\) shifts the logarithm by a constant and leaves the leading limit unchanged. Notice that \(b_0\) cancels: this amplitude alone does not independently verify the numerical Yang–Mills beta-function coefficient. The separate effective-coupling result addresses that coefficient.

The positive amplitude proves that this trace observable is nontrivial. Its inverse logarithm squared also excludes the formalized class of finite Wick-polynomial constructions from fixed-mass free fields. The scope matters: this is not an exclusion of every continuous-mass generalized free field. Likewise, the leading ultraviolet comparison is not a proof of agreement with every perturbative coefficient for every local field. Trace theorem, coefficient derivation and limitations; finite-free obstruction.

20 / LOCAL PRODUCTS AND COLLISIONS

An OPE with controlled finite remainders.

Let \(v_i\) be a cluster of fields with filtration degrees \(d_i\), total degree \(w=\sum_i d_i\), and let \(u_j\) be separated spectator fields. The construction supplies output fields \(o_\alpha\), finitely many at each degree \(d_\alpha\), and coefficient functions \(C_\alpha\). These are chosen before the phase, spectators or truncation degree.

Construct the coefficients in fixed field spaces

At finite degree the field space is finite-dimensional, but an arbitrary projection would not give a coherent OPE. The source first chooses separating physical readouts and dual field vectors. They define a matching field from correlations against these fixed detectors. Rotation averaging and conjugation symmetrization make the readout covariant and real. These auxiliary detectors are fixed by the prescription; they are not the spectator fields later inserted to test the OPE.

Write \(i_{j,m}:\mathcal F_j\to\mathcal F_m\) for the filtration inclusion and \(L_{j,m}:\mathcal F_m\to\mathcal F_j\) for the readout of a single field at the expansion center. The recursively corrected projection is

The subtraction removes the lower-degree pieces already identified by the same detectors. Coherence of the readouts proves that higher strata annihilate lower fields and that the finite pieces reconstruct a field exactly:

For a cluster, let \(\mathcal R_m(\mathbf v;\mathbf x,o)\) be its covariant real matching field and set \(F_m=P_m\mathcal R_m\in E_m\). Choose one basis \(e_{m,\alpha}\) of each \(E_m\). The coefficient functions are its actual coordinates:

There is no phase or spectator argument on \(F_m\) or \(C_{m,\alpha}\). Their independence is built into the construction, not inferred by comparing separately fitted expansions. The lemmas reconstruct, kills_lower, assemble, and packet_single verify these algebraic claims. Compatible projections ↗ · Coefficient assembly ↗

Derive the finite-remainder power

For every \(\varepsilon>0\), the coefficient estimate has the form

The cluster shape stays away from additional collisions while all its distances shrink proportionally to \(r\). Spectators stay a fixed positive distance from the expansion center and from one another. For every finite degree cutoff \(D\), the correlation expansion and remainder satisfy

Here \(s\) can be the mixture or any character phase. The radius can be chosen uniformly over those states and the allowed geometries, once the labels, cutoff and geometric bounds are fixed. The arbitrarily small exponent loss accommodates logarithms. Filtration degree is not an assertion of an exact interacting scaling dimension.

The remainder estimate requires two comparisons. Let \(q_D\in\mathcal F_D\) be the earlier finite matching vector supplied by the physical prescription, and abbreviate insertion at the center with spectators by \(S^s(z;u)\). Exact reconstruction of \(q_D\) gives the identity

The first bracket is the original matching error. Each remaining term measures the replacement of that match by the common covariant degree pieces. Both contributions must be bounded before the final coefficient expansion is asserted. The source's balance_rate proves precisely this decomposition and combines their estimates.

The general scale bound keeps track of moving spectators as well. Introduce a logarithmic scale \(t\to\infty\): the cluster shrinks like \(e^{-at}\), its distance to spectators may shrink like \(e^{-bt}\), and spectator separations like \(e^{-ct}\), with \(0\le b<a\) and \(b\le c\). If \(w_u\) is the total spectator degree, the proved remainder rate is

For the geometry used above, set \(a=1\), \(b=c=0\), and \(r=e^{-t}\). Substitution yields \(e^{(w-D-1+\varepsilon)t}=r^{D+1-w-\varepsilon}\). Similarly the degree-\(m\) field estimate has exponent \(w-m\), giving \(r^{m-w-\varepsilon}\) for its coordinates. Thus the stated powers follow from the two comparison bounds and the finite-dimensional projections, rather than dimensional analysis alone. Matching-error decomposition and rates ↗ · Conversion to separated power bounds ↗

The proof first allows the phase to vary arbitrarily with scale. A selection argument makes the estimate uniform: if no common threshold existed, one could choose a failing phase at each scale and contradict the bound along that sequence. The character norm formula then turns this into an error bound in the common vacuum algebra. Character evaluations are contractive, \(|\xi(E)|\le\|E\|\), so the algebra estimate controls all components and the mixed state. Allowing the admissible geometry to vary along arbitrary shrinking sequences gives the common-radius conclusion by the same bad-choice argument. This is the mechanism behind universalize; it is stronger than an eventual bound separately for each fixed phase.

Common coefficients leave room for different phases: the expectation values of the output fields can still differ. Independence from the choice of extracted infinite-volume limit is not proved either. Crucially, “every finite \(D\)” is a controlled asymptotic expansion, not convergence of an infinite operator series. \(D\) counts field degree, not loop order; this theorem does not compute all perturbative OPE coefficients. Nested expansions require their stated separation and scale conditions.

What a contact extension does

The OPE also supports extensions across collision diagonals. With \(\Delta_n=\bigcup_{i<j}\{x_i=x_j\}\), the extension has the schematic distributional meaning

The committee's contactDistribution endpoint gives a continuous linear functional on all compactly supported smooth joint tests, with this separated restriction. It does not establish uniqueness of the extension, its full OS axioms, or direct lattice convergence on tests crossing the diagonals. Full contact extension and separated restriction ↗ Contact distributions supported on \(\Delta_n\) are invisible away from collisions. This also explains a limitation of the independent scalar OS comparison, which uses diagonal-flat Schwartz tests:

Such tests cannot determine contact counterterms. Their role in this Euclidean comparison should not be confused with the unrestricted Schwartz domain of the reconstructed Wightman boundary distributions. Coefficient, remainder and phase-uniformity checks; assessment §7.5.

A simple illustrative subtraction makes the distinction concrete. It is not a proposed Yang–Mills correlator. The function \(|x|^{-4}\) in four dimensions is locally nonintegrable at zero. For a smooth compactly supported \(\chi\) equal to one near zero, define

Near zero the numerator is \(O(|x|)\). In radial coordinates the absolute bound is proportional to \(r^3r^{-4}r\,dr=dr\), so the subtraction makes the integral finite. Away from zero it agrees with the original kernel. Yet another allowed subtraction gives

The difference is supported only at the collision. Stronger singularities require more Taylor coefficients to be subtracted and allow derivatives of delta functions. This explains why controlled separated-point asymptotics can support an extension without fixing every contact term or proving lattice convergence on tests crossing collisions.

21 / VALIDATION AND THE REMAINING PROBLEM

What was checked—and what remains to prove.

The committee's combined acceptance target retains one native construction witness for its physical checks, with the group, penalty and extraction quantifiers stated in §11. The logical distinction is that one object must satisfy the conjunction:

Here \(F\) is attached to the same Wilson family, extraction and field table throughout. This prevents different requirements from being satisfied by unrelated theories. The combined target includes the lattice-law bridge, scalar OS comparison, operator dictionary, normalized Ward identity, leading trace law and source conclusions. The independent OS bridge specifically checks the selected centered trace-scalar hierarchy, including arity-linear growth; it is not an independently specified all-field OS interface.

The additional explicit OPE target has its own endpoint. It combines public properties and OPE bounds on one hierarchy, but its existential witness is not automatically identified with the combined target's witness. Its adapter can instead be applied directly when a particular hierarchy's source OPE is available.

The released records report a clean Lean 4.32.0 rebuild of 9,470 project modules with zero errors, an axiom audit, and Comparator passes for the public, combined and OPE endpoints on October 30, November 3 and November 5. Comparator checked statements, definitions, permitted axioms and proofs with its built-in Lean kernel. The macOS runs did not test Linux process isolation or use an additional external kernel implementation. The permitted proof dependencies were propext, Classical.choice and Quot.sound. Deliberate reference stubs and negative controls are distinguished from accepted solution dependencies.

These are the committee's recorded checks. Kernel acceptance checks implications from formal definitions. The committee's human review addresses whether those definitions capture the intended mathematics, and explicitly describes that review as extensive rather than exhaustive. Check scope and witness distinctions; validation records and review limits.

In particular, retaining a source conclusion in a larger accepted theorem does not independently validate every definition it contains. The separate adapters make selected comparisons explicit; their success should be read at that stated scope.

The central missing infrared estimate is still a strictly positive spectral gap. With the vacuum projector removed, one would need

Clustering alone is weaker. As an elementary spectral illustration, a measure with support arbitrarily close to zero can give a decaying correlation without any exponential gap:

The release also leaves open uniqueness of the infinite-volume limit, a continuum probability measure, construction and area laws for Wilson loops, single-particle spectral analysis, and full perturbative agreement. The positive short-distance probe is not a substitute for an intrinsic spectral mass. The achievement described here is the reported existence construction with these precise limitations; the full Millennium existence-and-mass-gap problem remains open. Assessment §8: open questions.

THE READING DESK

Primary sources & further reading

Read the original hypotheses alongside the explanations. Part II links to fixed revisions of both release repositories, so the claims remain traceable if their default branches change. The assessment is also available locally.

1. [1] Arthur Jaffe & Edward Witten · Quantum Yang–Mills TheoryRead the supplied official exposition ↗

   §§3–4 give the axiomatic setting and prize problem (PDF pp. 5–6). §§5–6 separate the mass gap from further physical properties and explain the constructive program.

2. [2] Kenneth G. Wilson · Confinement of quarks (1974)Physical Review D 10, 2445–2459 ↗

   The lattice formulation, loop observables, and the relation between confinement and the area law.

3. [3] Konrad Osterwalder & Robert Schrader · Axioms for Euclidean Green’s functionsPart I (1973) ↗ · Part II (1975) ↗

   Read both papers: Part II corrects and extends the reconstruction results. Positivity must be accompanied by the appropriate regularity and growth assumptions.

4. [4] Tadeusz Bałaban, John Imbrie & Arthur Jaffe · Exact Renormalization Group for Gauge Theories (1984)Read the supplied chapter ↗

   An introduction to exact block transformations and ultraviolet control. Its main worked setting is the abelian Higgs model in two or three dimensions; it is not itself a four-dimensional pure Yang–Mills existence proof.

   For the later four-dimensional analysis, see Bałaban, Renormalization group approach to lattice gauge field theories I (1987) ↗, and Large field renormalization II (1989) ↗. Read the stated effective-coupling hypothesis in Theorem 1 of the latter.

5. [5] Jeff Greensite · The Confinement Problem in Lattice Gauge Theory (2003)Read the supplied review ↗

   Start with §2 for confinement criteria and then the discussions of center symmetry, Wilson loops, screening, and string breaking.

6. [6] David J. Gross & Frank Wilczek · Ultraviolet Behavior of Non-Abelian Gauge Theories (1973)Physical Review Letters 30, 1343–1346 ↗

   The original asymptotic-freedom result.

7. [7] H. David Politzer · Reliable Perturbative Results for Strong Interactions? (1973)Physical Review Letters 30, 1346–1349 ↗

   The independent original calculation establishing asymptotic freedom.

8. [8] Particle Data Group · Quantum ChromodynamicsReview of Particle Physics, QCD review (2025) ↗

   Conventions for the beta function, running coupling, and the scale parameter. The notes specialize to pure gauge theory with no dynamical quarks.

9. [9] Yang–Mills group · Validation of the OpenAI Yang–Mills constructionRead the November 10, 2026 assessment ↗ · Pinned review repository ↗

   Part I states the construction and its limits; Part II records the formal verification, independent checks, and their scope. Revision 55ba9fc36293b3f4d52c6b1f813169a0be76ee98. The committee describes this as an initial assessment and does not claim to fully understand the proof.

10. [10] OpenAI · October 28, 2026 Yang–Mills submissionPinned original source repository ↗

    The released Lean source underlying this walkthrough, revision 3456e5e75cf8d4cd558955a5c7638fba2d68f464. The explanations here are based on the assessment and source reading.

11. [11] Scott Armstrong’s informalization project · A readable reconstruction of the Yang–Mills sourceRead the manuscript draft ↗ · Project description and review status ↗

    The local integration discussion compares §§11 and 13 with the original implementation; the ultraviolet convergence discussion compares §14; the infinite-volume and field-calculus discussion compares §15; the phase and reconstruction discussion compares §17. This is a developing, AI-assisted reconstruction from the released Lean source. The phase comparison uses revision f958ee21f4739af0f249af26a62bbeb807dbb27b, consulted 12 November 2026; earlier discussions retain their links to revision aa8feb83b7d52d2b9fde5abb911a4fabd68e50fc. It should be read as a draft explanation, not as the original OpenAI prose manuscript or a completed independent verification.