1. What this unit proves 1
2. What it uses 7
3. Strategy 9
4. Steps 10
4.1 The running coupling 10
4.2 Normalisation of the stress tensor (VII.5.4) 11
4.3 Exactly compatible frames (VII.5.5) 12
4.4 Finite-dimensional convergence (VII.5.6) 13
4.5 The classical stress-tensor algebra (VII.5.7) 13
5. Where the difficulty lies 14
6. Notation 15
7. What remains unexplained 15
The full account of unit VII.5: every proposition with its proof sketch, checked against the Lean in two rounds and edited. The condensed outline and the map of the Lean are on the unit page.
Unit page · All units · Overview
Part: VII, identification as Yang–Mills. Lean: YM/Pert/Asymptotics (the running coupling). Auxiliary declarations are in YM/Extra/WEAK_SCALING_ACTLAW (files Orientation, Criteria, Adjust, Analysis, Station), from YM/TraceExtension/Foundation (files Caps, Dilate), and one definition from YM/TraceExtension/Soot. Size: 232 lines in 43 declarations: 32 lines in Pert/Asymptotics (5 declarations), 162 in WEAK_SCALING_ACTLAW (24), 37 in TraceExtension/Foundation (13), 1 in TraceExtension/Soot (1).
(A) The running coupling (5 declarations). The inverse squared coupling of the constructed renormalization-group trajectory grows logarithmically, uniformly in the lattice level and the volume (VII.5.1). This gives the regularity clause (Reg1) of the main theorem with slope \(b_{\ast} = 2 b_0\) (VII.5.2). It also gives the law \(g^2 \log (1 / a) \to 1 / (2 b_0)\) (VII.5.3).
(B) Auxiliary results (38 declarations).
VII.5.4: the normalisation of the stress tensor along free-field scaling sequences (used by VI.2).
VII.5.5: exactly compatible frames, built from approximately compatible ones (used by VII.1).
VII.5.6: two elementary convergence lemmas (used by III.12).
VII.5.7–VII.5.8: the definitions of the classical stress-tensor algebra and a read-out predicate (used by VI.1, VII.2, VIII.1).
Setting for (A).
Gauge group. \(G \subset U (q)\), \(q \geqslant 1\), is a compact matrix group with Lie algebra \(\mathfrak{g}\) and inner product \(\langle X, Y \rangle_{\mathfrak{g}} = - \frac{1}{q} \text{Re} \hspace{0.17em} \text{tr} (XY)\). For an orthonormal basis \((e_i)\), define \(C_{\mathfrak{g}} := - (\dim \mathfrak{g})^{- 1} \sum_i \text{tr}_{\mathfrak{g}} ((\text{ad} \hspace{0.17em} e_i)^2)\), with value \(0\) when \(\mathfrak{g}= 0\). Here and below, simple means real simple and nonabelian. If \(\mathfrak{g}\) is simple, its Killing form equals \(- C_{\mathfrak{g}} \langle \cdot, \cdot \rangle_{\mathfrak{g}}\). For \(SU (q) \subset U (q)\) with \(q \geqslant 2\), this gives \(C_{\mathfrak{g}} = 2 q^2\). Put
Scales. The block factors \(b_k \geqslant 3\) are odd. Set \(D_m = \prod_{k < m} b_k\), \(a_m = 1 / D_m\) and
Then \(s_m \to \infty\).
Trajectory. For a lattice level \(N\), a volume index \(p\) and a stage \(m \leqslant N\), \(\beta_{N, p, m}\) is the coefficient of the normalised kinetic term in the effective action at spacing \(a_m\). The represented density is obtained from the Wilson weight by \(N - m\) block-spin steps.
\(\beta_{N, p} := \beta_{N, p, N}\) is the bare coupling of the main theorem.
\(\beta_{N, p, 0} = \bar{\beta} > 0\) for every \(N, p\). This is the renormalization condition at unit scale.
The reference line is \(\hat{\beta}_m := \bar{\beta} + \kappa_G s_m\). One has \(\beta_{N, p, m} \in [\hat{\beta}_m / 2, \hspace{0.17em} 2 \hat{\beta}_m]\).
The running coupling is \(g_{N, p, m} := \beta_{N, p, m}^{- 1 / 2}\).
The running clause (R) of the construction conditions reads: for every \(\varepsilon > 0\) there is \(m_0\) such that
Proposition VII.5.1 (uniform logarithmic growth of the inverse squared coupling). Let the family satisfy its construction conditions (II.2). For every \(\varepsilon > 0\) there is \(m_0 \geqslant 1\) such that
Status: theorem. Of the construction conditions only (R) is used. Simplicity of \(\mathfrak{g}\) is not needed. Lean: YM.Pert.Asymptotics.zeroUp_bound, with b0 and b0_normalizes.
Corollary VII.5.2 ((Reg1) for the constructed family). Let \(\mathfrak{g}\) be simple and let the family satisfy its construction conditions. Then its Wilson family satisfies (Reg1) with slope \(b_{\ast} = 2 b_0 = 11 C_{\mathfrak{g}} / (24 \pi^2) > 0\): for every \(\varepsilon > 0\) there is \(N_0 \geqslant 1\) with
Status: theorem, proved in VIII.1 as the first clause of the regularity of the constructed family. It is VII.5.1 at \(m = N\), together with \(C_{\mathfrak{g}} > 0\) (I.1). This is how the main theorem uses part (A). The main theorem asserts the existence of a positive slope without specifying its value. Lean: YangMills.regular (running clause of YangMills.WilsonRegularity), via zeroUp_bound.
Proposition VII.5.3 (asymptotic freedom of the running coupling). Let \(\mathfrak{g}\) be simple and let the family satisfy its construction conditions. Let \((N_i)\) and \((k_i)\) be sequences of lattice levels and volume indices with \(i \leqslant N_i\) for all large \(i\). Then
For \(N_i = i\) this is the law of the bare coupling, \(g_0 (a)^2 \sim (2 b_0 \log (1 / a))^{- 1}\).
For \(N_i \geqslant i\) it is the law of the effective coupling at spacing \(a_i\) of any finer lattice. No condition is placed on the volumes \(k_i\).
Status: theorem; its square gives the second limit. No other declaration cites it. Lean: YM.Pert.Asymptotics.comparison, regulatorRun.
Setting for (B).
Spaces and maps.
\(\mathcal{F}_d\) is the finite-dimensional complex space of labels of degree at most \(d\) in a correlation table \(S\) carrying a Euclidean hierarchy \(H\) (IV.1).
\(\mathcal{F}^{\circ}_d\) is the corresponding space of the free Maxwell comparison table, with its hierarchy \(H^{\circ}\), built on the same reference labels (VI.1).
A derivative plan (IV.1) on either table consists of complex-linear inclusions \(\iota_{d \to e}\) (\(d \leqslant e\)) and complex-linear partial derivatives \(\partial_{d, a} : \mathcal{F}_d \to \mathcal{F}_{d + 1}\), \(a = 1, \ldots, 4\).
The inclusion axiom says that inserting \(\iota_{d \to e} v\) in a separated correlation gives the same value as inserting \(v\). It implies injectivity, \(\iota_{d \to d} = 1\), transitivity and preservation of the unit. The derivative axiom identifies insertion of \(\partial_{d, a} v\) with differentiation of the correlation in that insertion point.
The orientation axioms say that inclusions commute with the hyperoctahedral group \(\Gamma_4\) of signed permutations (\(384\) elements) and with conjugation. Derivatives commute with conjugation and transform as vectors under \(\Gamma_4\).
Stress data. A candidate stress tensor is a family \(J = (J_{ab})\) in \(\mathcal{F}_4\). Its local Ward identities (IV.1, VI.1) involve a translation part \(p = (p_{d, a} : \mathcal{F}_d \to \mathcal{F}_{d + 1})\) and a rotation part \(Y\).
Free Maxwell data (VI.2, VIII.1): the comparison data specified in §2, including a derivative plan \(W\), a rotation action, a tensor \(J^{\circ}\) and \(\kappa \ne 0\). The plan satisfies the inclusion, derivative and orientation axioms; \(J^{\circ}\) satisfies the local Ward identities with translation part \(\kappa \hspace{0.17em} \partial^W\) and rotation part \(\kappa\) times the infinitesimal rotation action; and the family \(\partial^W_4 = (\partial^W_{4, a})_a\) is nonzero.
Scaling sequences. A scaling sequence (VI.1) consists of frames \(U_k (d) : \mathcal{F}^{\circ}_d \to \mathcal{F}_d\) (complex-linear isomorphisms), positive, strictly decreasing scales \(r_k \to 0\) and stress normalisations \(\xi_k > 0\).
Tracking. A scaling sequence tracks \((V, W, J)\) at level \((n, d, j, R)\) (VI.1) if three conditions hold:
the frames are exactly compatible with inclusions, signed permutations, conjugation and units up to degree \(\max (d, 4)\);
for each fixed list of at most \(n\) fields of degrees at most \(d\), the correlations transported by \(U_k\) and evaluated at \(r_k x\) converge, with all Fréchet derivatives of orders \(0, \ldots, j\), uniformly on the window \(\{x : \|x\| \leqslant R, \|x_a - x_b \|_{\infty} \geqslant R^{- 1} \text{for all } a \ne b\}\), to the free correlations;
\(\xi_k^{- 1} r_k^4 \hspace{0.17em} U_k (4)^{- 1} J_{ab} \to J^{\circ}_{ab}\).
Proposition VII.5.4 (normalisation of the stress tensor along free-field scaling sequences). Fix the two hierarchies and the free Maxwell comparison data above. Let \(V\) be a derivative plan on \(S\) satisfying the inclusion and derivative axioms. Let \(J\) be symmetric, real and \(\Gamma_4\)-covariant, with \(\sum_a \partial^V_{4, a} J_{ab} = 0\) for every \(b\), and let \((J, p, Y)\) satisfy the local Ward identities.
(a) Suppose \(p_{4, a} = c \hspace{0.17em} \partial^V_{4, a}\) for \(a = 1, \ldots, 4\) and some \(c \in \mathbb{R}\). Then there are \(D, N, M \in \mathbb{N}\) and \(R_0 \geqslant 1\) such that \(c / \xi_k \to \kappa\) for every scaling sequence that tracks \((V, W, J)\) at a level \((n, d, j, R)\) with \(n \geqslant N\), \(d \geqslant D\), \(j \geqslant M\) and \(R \geqslant R_0\).
(b) If \(p \neq 0\), then \(J \neq 0\).
Status: theorems. Part (a) uses A04 and A05 (VI.1); part (b) requires only the hierarchy and local Ward identities. Lean: YM.Extra.WeakScaling.tracks_ratio, scale_limit, flux_nontrivial. Used in VI.2 by YM.Pert.ratio and YM.Pert.centered_line.
Proposition VII.5.5 (exactly compatible frames). Suppose \(S\) and its free comparison table carry Euclidean hierarchies. Let \(V\) and \(W\) be derivative plans on these tables satisfying the inclusion and orientation axioms.
(a) For every \(d\) there is a \(\mathbb{C}\)-linear map \(P_d : \mathcal{F}^{\circ}_{d + 1} \to \mathcal{F}^{\circ}_d\) with \(P_d \hspace{0.17em} \iota^W_{d \to d + 1} = 1\) that commutes with \(\Gamma_4\) and with conjugation.
(b) Let the \(P_d\) be as in (a), fix \(D \in \mathbb{N}\), and let \(A_d : \mathcal{F}^{\circ}_d \to \mathcal{F}_d\) be complex-linear isomorphisms for every \(d\). Define recursively
Suppose that for every \(d \leqslant D\) the map \(A_d\) commutes with \(\Gamma_4\) and conjugation and maps unit to unit. Then:
\(X_e \hspace{0.17em} \iota^W_{d \to e} = \iota^V_{d \to e} \hspace{0.17em} X_d\) for all \(d \leqslant e \leqslant D\);
for \(d \leqslant D\), each \(X_d\) commutes with \(\Gamma_4\) and conjugation and maps unit to unit.
(c) Let \(A^{(k)}_d\) be complex-linear isomorphisms such that, for every \(d < D\) and every \(v\),
Let \(X^{(k)}\) be built from \(A^{(k)}\) as in (b). Then for every \(d \leqslant D\):
\((A^{(k)}_d)^{- 1} X^{(k)}_d \to 1\) in operator norm;
\(X^{(k)}_d\) is bijective for all large \(k\);
the isomorphisms \(\tilde{X}^{(k)}_d\) (equal to \(X^{(k)}_d\) when it is bijective and to \(A^{(k)}_d\) otherwise) satisfy \((A^{(k)}_d)^{- 1} \tilde{X}^{(k)}_d v \to v\) for every \(v\).
If the \(A^{(k)}_d\) also preserve the symmetries and units in (b), then, for all sufficiently large \(k\), the frames \(\tilde{X}^{(k)}_d\) are exactly compatible simultaneously in all degrees \(d \leqslant D\).
Status: theorem (finite-dimensional algebra and analysis). Lean: retraction, adjusted, Exact, adjusted_exact, Near, perturb_limit, eventually_unit, rotateChart, relative, Graded. Used in VII.1 by YM.SharpGram.Weak.Portfolio.tune_grades. Exact compatibility is also used in III.13 and VIII.1.
Proposition VII.5.6 (finite-dimensional convergence).
(a) Let \(f : E \to X\) be an injective linear map between finite-dimensional real normed spaces. If \(f (a_y) \to f (e)\) along a filter, then \(a_y \to e\).
(b) Let \(f_y, g : (\mathbb{R}^4)^n \to \mathbb{C}\) be continuous real-linear maps, and fix a filter on the index set of \(y\). If \(f_y (e_{i, a}) \to g (e_{i, a})\) at each of the \(4 n\) coordinate vectors, then \(f_y \to g\) in operator norm along that filter.
Status: theorem. Lean: norm_fit, norm_fit_deriv, components. Used in III.12 by YM.FullVolume.stretching.
Definition VII.5.7 (the classical stress-tensor algebra). Labels of degree \(d\) are gauge-invariant polynomials of weight at most \(d\) in radial Taylor coefficients at the base point. In the formulas below, \(F\) and \(\partial F\) denote the linear curls of the degree-0 and degree-1 Taylor coefficients used in the free Maxwell comparison. The curvature has weight 2 and each derivative adds 1 (II.1). \(\langle \cdot, \cdot \rangle\) denotes the invariant inner product on \(\mathfrak{g}\). These are polynomial identities, before passing to the quotient by null fields; they do not multiply limiting fields at coincident points.
Weight 4. The square is \(\mathfrak{s} := \frac{1}{4} \sum_{i, j} \langle F_{ij}, F_{ij} \rangle\), summed over ordered pairs. The candidate stress tensors of VI.1 form the family
The canonical tensor is \(T^{\text{can}} := T^{(0, 1, - 1 / 4, 1)}\), that is, \(T^{\text{can}}_{ab} = \sum_i \langle F_{ai}, F_{bi} \rangle - \delta_{ab} \hspace{0.17em} \mathfrak{s}\).
Weight 5. Define the vectors
and the equation-of-motion vector \(E_b := \sum_{a, i} \langle F_{bi}, \partial_a F_{ai} \rangle\) (VI.1). Let \(\mathcal{R}_1\) be the real space of real, \(\Gamma_4\)-covariant vectors of degree at most 5 (VI.1). Define \(\beta, \delta : \mathbb{R}^3 \to \mathcal{R}_1\) by
The definitions have two accompanying lemmas:
\(\mu^1, \mu^2 \in \mathcal{R}_1\);
every degree-0 radial jet is realised by a constant-curvature free Maxwell field, with an affine potential.
Status: definitions and two lemmas. Lean: square4, raw, momentum1, momentum2, momentum1_mem, momentum2_mem, momentum2_turn, basisV, bareVector (with mat3, proj3), shell_affine, sumPerm. The facts proved about these objects in VI.1 and VII.2 are listed in §4.5.
Definition VII.5.8 (sharp read-out bound). Fix the family, a nondecreasing control function \(K : \mathbb{N} \to \mathbb{R}\) with \(K (j) \geqslant 1\) for every \(j\), a degree \(d\), stages \(\ell_d \leqslant k \leqslant N\), a volume index \(p\) and \(u \in \mathbb{R}\). Write \(\|c\|_1 := \sum_{\alpha} |c_{\alpha} |\) for the coefficient norm, and let \(P\) remove the unit coordinate. Here \(\ell_d\) is the reference stage fixed by the family. The predicate quantifies over every \(w \geqslant 0\) and all insertion presentations \(A_i\) with coefficient vectors \(c_i\), remainder costs \(q_i\), inserted densities \(\mathcal{D}_i\) and insertion points at the origin. They must satisfy:
At each \(\ell_d \leqslant i \leqslant k\), \(A_i\) has the admissible regular and forced remainders of III.12, with total remainder cost at most \(q_i\), using the square root of the stage-\(i\) forced allowance.
\(c_{\ell_d} = 0\). Only this coefficient vector must vanish at the reference stage.
The full inserted densities obey \(\mathcal{D}_i = b_i^d \hspace{0.17em} \mathcal{T}^{(i)}_{\beta_{N, p, i + 1}} \mathcal{D}_{i + 1}\) for \(\ell_d \leqslant i < k\).
Put \((S_i c)_{\alpha} := b_i^{d - d_{\alpha}} c_{\alpha}\), where \(d_{\alpha}\) is the weight of the basis label, and \(K_i := K (J (i + 1))\). For each \(\ell_d \leqslant i < k\), \(q_i, q_{i + 1} \geqslant 0\) and there is a complex-linear automorphism \(B_i\) commuting with \(P\) such that \(\|B_i c\|_1, \|B_i^{- 1} c\|_1 \leqslant K_i \|c\|_1\) for every \(c\), and
The terminal cost satisfies \(q_k \leqslant w + n_k (w +\|c_k \|_1)\), where \(w \geqslant 0\) and \(n_k\) is the stage-\(k\) column number.
The sharp read-out predicate is the assertion that every such chain satisfies
Status: definition. The admissibility and remainder conditions are those of Format, remBound; the complete chain conditions are Duet.Woke and direction. Lean: YM.MergeProof.Accent.SharpRead. Used by: YM.Trace.Terrace.Sceptre and the cost lemmas of VII.2.
Inputs from other units.
I.1 (gauge group and scales).
\(C_{\mathfrak{g}}\) (YM.Gauge.casimir) and \(\kappa_G\), which is defined as \(11 C_{\mathfrak{g}} / (24 \pi^2)\) (slopeConstant).
For simple \(\mathfrak{g}\), the Killing-form identity in §1 is trace_adjoint.
Positivity \(C_{\mathfrak{g}} > 0\) and \(\kappa_G > 0\) for simple \(\mathfrak{g}\) (casimir_pos, slopeConstant_pos).
\(a_m = 1 / D_m\) (Schedule.a); \(s_m\) is defined as \(\log (1 / a_m)\) (Schedule.scale); and \(s_m \to \infty\) (Schedule.divergent).
II.2 (construction conditions) YM.Family.Makes. Its first field says that the trajectory is a selected history (III.1). The bare coupling is Family.beta, that is, \(\beta_{N, p, N}\).
III.1 (selected history) YM.RG.History.Selected. It contains:
III.2 (choice of constants, shooting).
(R) is derived in YM.RG.Shoot.trajectory. The inputs are \(\beta_{N, p, 0} = \bar{\beta}\), the per-step relation, and \(\delta_{m + 1} / \log b_m \leqslant J (m + 1)^{- 2}\), where \(J (m) \to \infty\) is the stage index of the scale sequence (Lemma VII.5.b).
The history with \(\beta_{N, p, 0} = \bar{\beta}\) is produced by shooting (Recipe.races, successful).
The per-step relation follows from the coupling reset (III.9). If the one-loop shift \(\alpha_{N, m}\) satisfies \(| \alpha_{N, m} + \kappa_G \log b_m | + 1 \leqslant \delta_{m + 1}\) and the reset changes \(\beta\) by \(\alpha_{N, m}\) with error at most \(1\), the triangle inequality gives the transfer bound. This is Directed.links, using Kin.Planned.
VIII.1: uses VII.5.2 in the assembly of the main theorem.
For VII.5.4.
The work items A04 (answer) and A05 (answer), both proved in VI.1. Along tracking scaling sequences of sufficient arity, degree, derivative order and window size, they give: \(r_k \hspace{0.17em} \partial^V_d\) conjugated by the frames converges to \(\partial^W_d\); and \((r_k / \xi_k) \hspace{0.17em} p_d\) conjugated by the frames converges to \(\kappa \hspace{0.17em} \partial^W_d\).
The tracking notion (Track) and the hypotheses on \(J\) (Pre), from VI.1.
The Ward identities (StressLaws) and derivative plans (DiffPlan), from IV.1.
The free Maxwell data (MaxwellData, Lighthouse.Goes). Besides the data in §1, the full package includes finite separating tests and the boundary, polynomial and internal-rotation conditions of VI.2. The ratio argument uses the free hierarchy, derivative and Ward identities, and the nonzero degree-4 derivative.
For VII.5.5: the inclusion and orientation axioms of derivative plans (IV.1), and the free comparison table (Spin.comparator, VI.1).
For VII.5.7:
the label polynomials \(\langle F_{ab}, F_{cd} \rangle\) (pair) and \(\langle F_{ab}, \partial_i F_{cd} \rangle\) (pairDeriv);
\(E\) (eom), the family \(T^{(t)}\) (stress, weyl) and \(\mathcal{R}_1\) (RawOne), all from VI.1;
the ideal of labels vanishing on free Maxwell solutions (Maxwell.equationIdeal), from III.16.
For VII.5.8: the comparison-chain definitions of III.2 and the insertion estimates of III.12. For the constructed family, attain (III.2) supplies the read-out field of the trace-analysis assumptions under its scale and one-step hypotheses.
Standard mathematics.
Asymptotic freedom. Gross–Wilczek and Politzer (Phys. Rev. Lett. 30, 1973). In the conventional normalisation for \(SU (q)\), with \(\beta = 2 q / g^2\), the one-loop law is \(1 / g^2 \sim 2 b_0^{\text{conv}} \log (1 / a)\) with \(b_0^{\text{conv}} = 11 q / (48 \pi^2)\).
Here \(\beta = 1 / g_0^2\) in the normalisation \(\langle X, Y \rangle_{\mathfrak{g}} = - \frac{1}{q} \text{Re} \hspace{0.17em} \text{tr} (XY)\), and \(C_{\mathfrak{g}} = 2 q^2\). So \(b_{\ast} = 2 b_0 = 11 q^2 / (12 \pi^2) = 2 q \cdot 2 b_0^{\text{conv}}\), which agrees with the conventional law.
For \(SU (2)\), \(b_{\ast} = 11 / (3 \pi^2)\).
For the relation between the bare coupling and the lattice spacing see Creutz, Quarks, Gluons and Lattices (1983).
Block-spin RG for lattice gauge fields. Balaban's renormalization group (Comm. Math. Phys. 109 (1987), 116 (1988), and related papers) provides the background for the marginal-flow estimate. The specific input here is the per-step bound of III.9; (R) follows by summation.
Exactly compatible frames. Averaging over a finite group gives equivariant projections (Maschke). In finite dimension, injective linear maps form an open set, and linear maps are continuous.
Classical stress tensor. The symmetric gauge-field stress tensor \(T^{\text{can}}\) is conserved on the equations of motion and is traceless in four dimensions (classical scale invariance; Callan–Coleman–Jackiw, Ann. Phys. 59, 1970). The breaking of tracelessness in the quantum theory is the trace anomaly (Adler–Collins–Duncan 1977; Collins–Duncan–Joglekar 1977), treated in VII.2.
Imported constants and assumptions.
The value \(\kappa_G = 11 C_{\mathfrak{g}} / (24 \pi^2)\) is a definition (I.1), and \(b_0 := \kappa_G / 2\), so \(\kappa_G = 2 b_0\) holds by unfolding. That \(\kappa_G\) is the actual slope of the trajectory is the content of the per-step relation, proved in Part III.
\(\bar{\beta} > 0\) is the fixed reference coupling (III.2). The limiting slope is independent of \(\bar{\beta}\); the convergence thresholds may depend on it.
VII.5.1 assumes the construction conditions. VII.5.2–VII.5.3 additionally assume simplicity. The auxiliary hypotheses are stated in §1.
Part (A). The coupling law follows from the one-step bound by summation and division by the logarithmic scale.
Marginal flow (III.9). Blocking from stage \(m + 1\) to stage \(m\) gives \(\beta_{N, p, m} = \beta_{N, p, m + 1} - \kappa_G \log b_m + e_{N, p, m}\), with \(|e_{N, p, m} | \leqslant \delta_{m + 1}\) and \(\delta_{m + 1} / \log b_m \leqslant J (m + 1)^{- 2} \to 0\). The reset uses the one-loop estimate of III.7.
Shooting and summation (III.2). For each \(N, p\) the bare coupling \(\beta_{N, p, N}\) is chosen so that the flow reaches \(\beta_{N, p, 0} = \bar{\beta}\); this is the renormalization condition at unit scale. Summing the per-step relation from step 0 to step \(m\) gives
Since \(\delta_{i + 1} = o (\log b_i)\) and \(s_m = \sum_{i < m} \log b_i \to \infty\), the right side is \(o (s_m)\) by a Cesàro argument. The bound is uniform in \(N \geqslant m\) and in \(p\), because the per-step bounds are.
Division by \(s_m\) (this unit). Since \(\bar{\beta} / s_m \to 0\), one gets \(\beta_{N, p, m} / s_m \to \kappa_G = 2 b_0\), uniformly. Taking \(m = N\) gives (Reg1); taking square roots gives the coupling law.
The result concerns the actual inverse coupling of the trajectory, including its bare endpoint. It does not identify this quantity with a coupling defined from continuum correlations.
Part (B). The auxiliary arguments use finite-dimensional limits and equivariant linear algebra.
VII.5.4 compares two convergent sequences in a finite-dimensional space. Along a free-field scaling sequence, the rescaled derivative \(r_k \partial^V_4\) tends to \(\partial^W_4 \neq 0\). The rescaled flux \((r_k / \xi_k) p_4 = (c / \xi_k) \hspace{0.17em} r_k \partial^V_4\) tends to \(\kappa \hspace{0.17em} \partial^W_4\). Hence the scalar \(c / \xi_k\) tends to \(\kappa\).
VII.5.5 uses equivariant averaging and a recursive correction:
an equivariant retraction of each inclusion, obtained by averaging over \(\Gamma_4\) and conjugation;
a recursive correction, degree by degree, that turns approximately compatible frames into exactly compatible ones;
a perturbation argument showing the correction is eventually invertible and asymptotically equal to the original frames.
VII.5.7 is the algebra of stress tensors quadratic in \(F\) under the hyperoctahedral group. On the lattice only \(\Gamma_4\) acts, so the candidate family has four parameters and its divergences have three coordinates. For simple \(\mathfrak{g}\), conservation on free Maxwell fields singles out the canonical tensor up to scale and addition of a constant multiple of \(\delta_{ab} \boldsymbol{1}\). The canonical tensor is traceless.
Lemma VII.5.a (normalisation). \(2 b_0 = \kappa_G\). Proof. Unfold both definitions. Lean: b0_normalizes.
Lemma VII.5.b (the running clause; input from III.2). Let \(\delta : \mathbb{N} \to [0, \infty)\) and \(J : \mathbb{N} \to \mathbb{N}\) satisfy \(J (m) \geqslant 1\) and \(J (m) \to \infty\). With the scales and reference line of §1, suppose:
\(\beta_{N, p, 0} = \bar{\beta}\) for all \(N, p\);
\(| \beta_{N, p, m + 1} - \beta_{N, p, m} - \kappa_G \log b_m | \leqslant \delta_{m + 1}\) for \(m < N\);
\(\delta_{m + 1} / \log b_m \leqslant J (m + 1)^{- 2}\) for every \(m\).
Then (R) holds.
Proof sketch.
Since \(s_{m + 1} = s_m + \log b_m\), induction on \(m\) gives \(| \beta_{N, p, m} - \hat{\beta}_m | \leqslant \Delta_m := \sum_{i < m} \delta_{i + 1}\) for \(m \leqslant N\).
Given \(\varepsilon\), choose \(j\) with \(J (i + 1)^{- 2} \leqslant \varepsilon / 2\) for \(i \geqslant j\). Then \(\Delta_m \leqslant \Delta_j + \frac{\varepsilon}{2} (s_m - s_j)\) for \(m \geqslant j\).
For \(m\) large, \(\Delta_j \leqslant \frac{\varepsilon}{2} s_m\), hence \(\Delta_m \leqslant \varepsilon s_m\).
The constants and thresholds do not involve \(N\) or \(p\). Lean: YM.RG.Shoot.trajectory (III.2).
Lemma VII.5.c (= Proposition VII.5.1). Proof sketch.
Apply (R) with \(\varepsilon / 4\). Because \(s_m \to \infty\), one may also require \(s_m \geqslant 4 (| \bar{\beta} | + 1) / \varepsilon\).
Then \(s_m > 0\) and \(| \bar{\beta} | < \frac{\varepsilon}{4} s_m\).
With \(\kappa_G = 2 b_0\), the triangle inequality gives
Divide by \(s_m\).
Lean: zeroUp_bound.
Lemma VII.5.d (= Corollary VII.5.2). Proof sketch. The slope is \(2 b_0\), which is positive because \(C_{\mathfrak{g}} > 0\) for simple \(\mathfrak{g}\). Given \(\varepsilon\), take \(N_0 := m_0\) from VII.5.1 and specialise to \(m = N\). The bare coupling of the public family is \(\beta_{N, p, N}\), and its \(\log (1 / a_N)\) is \(s_N\) by definition. Lean: YangMills.regular (VIII.1).
Lemma VII.5.e (= Proposition VII.5.3). Proof sketch.
For each \(\varepsilon\), eventually both \(i \geqslant m_0 (\varepsilon)\) and \(i \leqslant N_i\). So VII.5.1 gives \(\beta_{N_i, k_i, i} / s_i \to 2 b_0\).
Since \(\kappa_G > 0\) for simple \(\mathfrak{g}\), \(2 b_0 > 0\). Then \(x \mapsto \sqrt{x}\) is continuous at \(2 b_0\) and \(y \mapsto y^{- 1}\) is continuous at \(\sqrt{2 b_0} \ne 0\). Hence \((\beta_{N_i, k_i, i} / s_i)^{- 1 / 2} \to (2 b_0)^{- 1 / 2}\).
For all sufficiently large \(i\), both \(s_i\) and \(\beta_{N_i, k_i, i}\) are positive, so \(\beta_{N_i, k_i, i}^{- 1 / 2} s_i^{1 / 2} = (\beta_{N_i, k_i, i} / s_i)^{- 1 / 2}\), and \(s_i = \log (1 / a_i)\) by definition.
Lean: comparison.
Lemma VII.5.f (ratio lemma). Let \(Z\) be a finite-dimensional real normed space. If \(X_k \to x \neq 0\) in \(Z\) and \(f_k X_k \to c \hspace{0.17em} x\) with \(f_k \in \mathbb{R}\), then \(f_k \to c\). Proof sketch. Choose a linear functional \(\ell\) with \(\ell (x) \neq 0\). Then \(\ell (X_k) \ne 0\) eventually, and \(f_k = \ell (f_k X_k) / \ell (X_k) \to c\). Lean: scale_limit.
Lemma VII.5.g (= VII.5.4(a)). Proof sketch.
A04 at degree 4 gives thresholds beyond which \(X_k := r_k \hspace{0.17em} U_k (5)^{- 1} \partial^V_4 U_k (4) \to x := \partial^W_4\).
A05 at degree 4, applied to \((J, p, Y)\), gives thresholds beyond which \((r_k / \xi_k) \hspace{0.17em} U_k (5)^{- 1} p_4 U_k (4) \to \kappa \hspace{0.17em} \partial^W_4\).
Take the componentwise maximum of the two sets of thresholds.
Substituting \(p_4 = c \hspace{0.17em} \partial^V_4\) in step 2 gives \((c / \xi_k) X_k \to \kappa x\), where \(x \ne 0\) is part of the free data. Lemma VII.5.f gives \(c / \xi_k \to \kappa\).
The transport is \(U_k (d + 1)^{- 1} (\hspace{0.17em} \cdot \hspace{0.17em}) U_k (d)\), as defined by carry and aisle.
Lean: tracks_ratio.
Lemma VII.5.h (= VII.5.4(b)). Proof sketch. Suppose \(J = 0\). The translation Ward identity writes every correlation with an insertion of \(p_{d, a} v\) as a flux of \(J\) through a shell, which is then \(0\). Correlations separate the elements of \(\mathcal{F}_{d + 1}\), so \(p = 0\). Lean: flux_nontrivial.
Remark (sign of \(c \kappa\)). Suppose in addition that tracking sequences exist at every level. Then \(c \kappa > 0\): since \(\xi_k > 0\), \(c / \xi_k\) has the sign of \(c\), and its limit is \(\kappa \ne 0\). This is kappa_sign. It is not assigned to this unit and is not used by the main theorem. VI.2 uses only (a), and obtains \(c \neq 0\) from \(\kappa \ne 0\).
Lemma VII.5.i (equivariant retraction, = VII.5.5(a)). Proof sketch.
\(\iota^W\) is injective, so it has a linear left inverse \(R\).
Average over \(\Gamma_4\): \(Q := | \Gamma_4 |^{- 1} \sum_{s \in \Gamma_4} s^{- 1} Rs\). Since \(\iota^W\) commutes with \(\Gamma_4\), \(Q\) is again a left inverse, and \(Q\) is \(\Gamma_4\)-equivariant.
Let \(\theta\) be the conjugation, an antilinear involution. Put \(P := \frac{1}{2} (Q + \theta Q \theta)\). Then \(P\) is \(\mathbb{C}\)-linear. It is a left inverse, since \(\iota^W\) commutes with \(\theta\). It commutes with \(\theta\). It stays \(\Gamma_4\)-equivariant, since \(\theta\) commutes with \(\Gamma_4\).
Lean: retraction, averaged_cov.
Lemma VII.5.j (exactness of the correction, = VII.5.5(b)). Proof sketch.
One step. \(X_{d + 1} \iota^W v = \iota^V X_d P_d \iota^W v + A_{d + 1} (\iota^W v - \iota^W P_d \iota^W v) = \iota^V X_d v\).
Intertwining for \(d \leqslant e\). Induction on \(e\), using \(\iota_{d \to e + 1} = \iota_{e \to e + 1} \iota_{d \to e}\) and \(\iota_{d \to d} = 1\).
Symmetries. Induction on \(d\). Let \(g\) be a signed permutation or the conjugation. Then \(g\) commutes with \(P_d\), \(\iota^W\), \(\iota^V\), \(A_{d + 1}\) and, by induction, with \(X_d\). Hence it commutes with both terms of \(X_{d + 1}\).
Units. \(\boldsymbol{1}_{d + 1} = \iota^W \boldsymbol{1}_d\), so \(X_{d + 1} \boldsymbol{1}_{d + 1} = \iota^V X_d \boldsymbol{1}_d = \iota^V \boldsymbol{1}_d =\boldsymbol{1}_{d + 1}\).
Lean: adjusted_succ, adjusted_exact.
Lemma VII.5.k (asymptotic identity). Under the hypotheses of VII.5.5(c), for \(d \leqslant D\), \(\Pi^{(k)}_d := (A^{(k)}_d)^{- 1} X^{(k)}_d \to 1\) in operator norm. Proof sketch.
Put \(\Phi^{(k)}_d := (A^{(k)}_{d + 1})^{- 1} \iota^V A^{(k)}_d\). Then
\(\Phi^{(k)}_d \to \iota^W\) pointwise by hypothesis, hence in norm (finite dimension).
By induction, \(\Pi^{(k)}_d P_d v \to P_d v\).
Therefore \(\Pi^{(k)}_{d + 1} v \to \iota^W P_d v + v - \iota^W P_d v = v\). Finite dimensionality converts pointwise convergence to operator-norm convergence.
Lean: perturb, perturb_limit.
Lemma VII.5.l (bijectivity and the new frames, VII.5.5(c)). Proof sketch.
\(\Pi^{(k)}_d \to 1\) in norm, and injective operators form an open set. So \(\Pi^{(k)}_d\), and hence \(X^{(k)}_d\), is injective for large \(k\).
\(\dim \mathcal{F}^{\circ}_d = \dim \mathcal{F}_d\) (by \(A^{(k)}_d\)), so \(X^{(k)}_d\) is bijective.
Hence \(\tilde{X}^{(k)}_d = X^{(k)}_d\) eventually, and \((A^{(k)}_d)^{- 1} \tilde{X}^{(k)}_d v \to v\) for every \(v\).
Intersect the eventual sets for \(d = 0, \ldots, D\). If the original frames satisfy the symmetry and unit conditions of (b), all corrected frames then agree with the exactly compatible maps \(X^{(k)}_d\) on this degree range.
Lean: eventually_unit, rotate_eq, relative.
Use. In VII.1 the frames of the free-field scaling sequence intertwine the inclusions only asymptotically. VII.5.5 replaces them by frames that are exactly graded for all large \(k\) and asymptotically equal to the original ones, as tracking requires (tune_grades).
Lemma VII.5.m (= Proposition VII.5.6). Proof sketch.
(a) A linear left inverse \(g\) of \(f\) is continuous in finite dimension, so \(a_y = gf (a_y) \to gf (e) = e\).
(b) Every point of \((\mathbb{R}^4)^n\) is \(\sum_{i, a} v_{i, a} e_{i, a}\). Hence evaluation at the \(4 n\) coordinate vectors is an injective real-linear map from the space of complex-valued real-linear maps to \(\mathbb{C}^{4 n}\), and (a) applies.
Lean: norm_fit, components, norm_fit_deriv.
Lemma VII.5.n (covariance of \(\mu^1, \mu^2\)). For a signed permutation \(t\) with signs \(\varepsilon_t (\cdot)\):
Both are real. Proof sketch. \(t\) multiplies \(\langle F_{ij}, \partial_b F_{ij} \rangle\) by \(\varepsilon (i) \varepsilon (j) \hspace{0.17em} \varepsilon (b) \varepsilon (i) \varepsilon (j) = \varepsilon (b)\) and relabels the indices. The double sum over \((i, j)\) is invariant under reindexing by \(t\). For \(\mu^1_b\), the sign is \(\varepsilon (b)^3 \varepsilon (i)^2 = \varepsilon (b)\), and the remaining index is reindexed in the same way. Reality follows from the real inner products. Lean: momentum2_turn, momentum1_mem, momentum2_mem, sumPerm.
Lemma VII.5.o (affine shells). Every degree-0 radial jet is realised by a free Maxwell field with constant curvature and an affine potential. Proof sketch. The associated potential has constant curvature, so \(\sum_a \partial_a F_{ab} = 0\). Lean: shell_affine. VII.2 uses this, with its degree-1 analogue shell_quadratic, to build test jets on Maxwell solutions.
Consequences in VI.1 and VII.2. Write \(T^{(0, w)} := T^{(0, w_0, w_1, w_2)}\) for \(w \in \mathbb{R}^3\).
Trace (VI.1). \(\sum_a T^{(0, w)}_{aa} = 4 (w_0 + 4 w_1) \hspace{0.17em} \mathfrak{s}\). In particular \(T^{\text{can}}\) is traceless (mechanical_trace, conformal).
Divergence (VII.2). \(\delta (w)_b = \sum_a \partial_a T^{(0, w)}_{ab}\) (euler). The key identity is the Bianchi consequence
With it, the shear part \(\sum_{a \ne b} \partial_a \sum_i \langle F_{ai}, F_{bi} \rangle\) equals \(E_b + \frac{1}{2} \mu^2_b - 2 \mu^1_b\).
Coordinates (VII.2). For simple \(\mathfrak{g}\), \(\beta\) and \(\delta\) are bijections \(\mathbb{R}^3 \to \mathcal{R}_1\) (coordinates1, bareCoordinates). Under the same simplicity hypothesis, \(\beta (w)\) vanishes on free Maxwell solutions if and only if \(w_0 = w_1 = 0\) (through_eq).
Conservation of the canonical tensor (VII.2). \(\delta (1, - \tfrac{1}{4}, 1) = E\): \(T^{\text{can}}\) is conserved on the equations of motion (eom_vector_raw).
The functional \(\rho\) (VII.2). For simple \(\mathfrak{g}\), define \(\rho : \mathcal{R}_1 \to \mathbb{R}\) by \(\rho (\delta (w)) := - 4 (w_0 + 4 w_1)\). Then \(\sum_a T^{(0, w)}_{aa} = - \rho (\delta (w)) \hspace{0.17em} \mathfrak{s}\). The functional \(\rho\) vanishes on vectors whose components vanish on free Maxwell solutions (rho, rho_eom).
For simple \(\mathfrak{g}\), a tensor in this four-parameter family is conserved on free Maxwell fields precisely when it is a multiple of \(T^{\text{can}}\) plus a constant multiple of \(\delta_{ab} \boldsymbol{1}\). With the unit term removed, its trace is determined by its divergence through \(\rho\). VII.2 uses this functional in Lease and its limiting comparison mere.
The logarithmic coupling law uses only the uniform running clause and elementary real analysis. The analytic work lies in the upstream one-loop estimate, the coupling reset and the construction of a trajectory reaching \(\bar{\beta}\) at unit scale. The auxiliary results use finite-dimensional linear algebra, equivariant averaging and scalar-ratio limits.
The result is uniform both in the volume index \(p\) and in the finer lattice level \(N \geqslant m\). Thus VII.5.3 allows arbitrary volume sequences. If inverse couplings converge as \(N, p \to \infty\) at fixed \(m\), their limits inherit the non-strict bound \(| \beta / s_m - 2 b_0 | \leqslant \varepsilon\) for \(m \geqslant m_0 (\varepsilon)\). Existence of those limits and comparison with a coupling defined by continuum correlations require separate results.
Only the leading logarithm is established here, for the coefficients of the lattice effective actions. No two-loop correction or \(\Lambda\)-parameter is asserted. The main theorem uses VII.5.1 through (Reg1), retaining only the existence of a positive slope. Proposition VII.5.3 is not used in the assembly.
The 38 auxiliary declarations serve other parts of the proof: stress normalisation in VI.2, frame correction in VII.1, convergence in III.12, and classical stress algebra and read-out estimates in VI.1 and VII.2.
| object | code name | proposed notation |
| Casimir constant | YM.Gauge.casimir | \(C_{\mathfrak{g}}\) (\(= 2 q^2\) for \(SU (q)\)) |
| slope constant | YM.Gauge.slopeConstant | \(\kappa_G = 11 C_{\mathfrak{g}} / (24 \pi^2)\) |
| one-loop coefficient | b0 | \(b_0 = \kappa_G / 2\); slope \(b_{\ast} = 2 b_0\) |
| lattice spacing at stage \(m\) | Schedule.a | \(a_m = 1 / D_m\) |
| logarithmic scale | Schedule.scale | \(s_m = \log (1 / a_m)\) |
| coupling at stage \(m\) | C.path.β N p m | \(\beta_{N, p, m}\); bare coupling \(\beta_{N, p} = \beta_{N, p, N}\) |
| reference line | hatBeta | \(\hat{\beta}_m = \bar{\beta} + \kappa_G s_m\) |
| running clause | Running | (R) |
| running coupling | regulatorRun | \(g_{N, p, m} = \beta_{N, p, m}^{- 1 / 2}\) |
| free comparison table | Free | \(\mathcal{F}^{\circ}_d\) |
| equivariant retraction | retraction | \(P_d\) |
| corrected frames | adjusted, rotateChart | \(X_d\), \(\tilde{X}^{(k)}_d\) |
| exact compatibility up to degree \(D\) | Exact, Graded | "exactly graded" |
| asymptotic intertwining | Near | hypothesis of VII.5.5(c) |
| square of the curvature | square4 | \(\mathfrak{s}= \frac{1}{4} \sum_{i, j} \langle F_{ij}, F_{ij} \rangle\) |
| canonical stress tensor | raw | \(T^{\text{can}}\) |
| weight-5 vectors | momentum1, momentum2 | \(\mu^1\), \(\mu^2\) |
| coordinates on \(\mathcal{R}_1\) | basisV | \(\beta (w)\) |
| divergence of \(T^{(0, w)}\) | bareVector | \(\delta (w)\) |
| sharp read-out bound | SharpRead | Definition VII.5.8 |
The upstream one-loop calculation. The transfer bound is traced through the coupling reset and the one-loop shift bound in §2. The evaluation of the momentum integral giving \(11 C_{\mathfrak{g}} / (24 \pi^2)\) belongs to III.7; its analytic proof is not verified here. The coupling results in this unit take the construction conditions as hypotheses.
The coupling after the limits. The assessment's Theorem 7.2 concerns \(g_W (r)\) after removal of the initial cutoff and a subsequential infinite-volume limit, together with comparison to a coupling read from continuum correlations. VII.5.3 proves the lattice-stage law. A declaration establishing the additional limits and comparison has not been identified here (tentative). The files Beacon, Hymnal and Musicology in YM/RemainderComparison are possible starting points.
Existence of the comparison inputs. VII.5.4 assumes the free Maxwell data of VI.2 and a Euclidean hierarchy on \(S\); it gives a conclusion for every tracking sequence meeting its finite thresholds. It does not construct such sequences. Their existence belongs to VI.1 and VII.1. Likewise, the supply of the read-out bound in III.2 uses the hypotheses of the theorem linked in §2; the underlying one-step estimates are outside this unit.