1. What this unit proves 1
2. What it uses 9
3. Strategy 10
4. Steps 11
4.1 Torus, subdivision, schedule 11
4.2 Cover numbers 12
4.3 Links, transport, plaquettes 13
4.4 Gauge group 14
4.5 The Wilson measure 16
4.6 Continuum symmetry groups 16
5. Where the difficulty lies 17
6. Notation 18
7. What remains unexplained 19
The full account of unit I.1: 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: I, Setting. Lean: YM/Lattice (1,215 lines), YM/Gauge (1,131), YM/Space (430), YM/Corona (405), YM/Mink (363), YM/Wilson (157). Size: 3,701 lines in 595 declarations used by the main theorem.
The unit fixes the setting of the construction:
the periodic lattices and their symmetry group;
parallel transport and gauge transformations;
the subdivision of a torus into blocks and the schedule of lattice spacings;
a combinatorial measure of the size of a lattice set;
the gauge group in its defining representation, with its Lie algebra, Casimir constant, local logarithm and nearest-point retraction;
the Wilson measure and its invariance;
the continuum symmetry groups of \(\mathbb{R}^4\).
The results require no input theorem from another unit. Simplicity of \(\mathfrak{g}\) is assumed only in Proposition I.1.6(d)–(f); the other hypotheses are stated locally.
Conventions. \(G\) is a closed subgroup of \(U (N)\), with \(N \geqslant 1\) fixed. For an integer \(\ell \geqslant 1\), write \(\Lambda_{\ell} = (\mathbb{Z}/ \ell \mathbb{Z})^4\). Axes \(0, 1, 2\) are spatial and axis \(3\) is time; \(e_i\) are the unit vectors. A link \((x, i)\) joins \(x\) to \(x + e_i\). A configuration is a map \(U : \Lambda_{\ell} \times \{0, 1, 2, 3\} \to G\); the same transport and gauge definitions apply on the cover \(\mathbb{Z}^4\). The norm \(\| \cdot \|_F\) is the Frobenius norm on \(M_N (\mathbb{C})\); norms on real coordinate spaces are Euclidean unless marked otherwise.
Proposition I.1.1 (torus geometry). For \(y \in \mathbb{Z}^4\) let \([y]_{\ell} \in \Lambda_{\ell}\) be its reduction mod \(\ell\). For \(x \in \Lambda_{\ell}\) let \(r_{\ell} (x) \in \mathbb{Z}^4\) be the representative with coordinates in \((- \ell / 2, \ell / 2]\) (the centred lift; the tie \(x_i = \ell / 2\) lifts to \(+ \ell / 2\)). Then:
\([r_{\ell} (x)]_{\ell} = x\), and \(r_{\ell} (x)\) is the unique \(y\) with \([y]_{\ell} = x\) and \(- \ell < 2 y_i \leqslant \ell\) for all \(i\);
reduction is injective on this centred box, and \(| \Lambda_{\ell} | = \ell^4\).
Put \(|x|_{\ell} = \max_i |r_{\ell} (x)_i |\). The separation \(|x - y|_{\ell}\) is a metric on the torus:
it is symmetric and is zero exactly when \(x = y\);
it satisfies the triangle inequality;
it is invariant under translations and under the hyperoctahedral group \(B_4\) of Proposition I.1.2;
reduction does not increase distances: \(| [y]_{\ell} |_{\ell} \leqslant \max_i |y_i |\).
Status: theorem. Lean: Lattice.Site, wrap, lift, lift_eq_iff, wrap_injOn, site_card, separation, separation_comm, separation_triangle, separation_translate, separation_signed, Corona.lap.
Proposition I.1.2 (lattice symmetries, transport, gauge transformations). Let \(B_4\) be the group of signed permutations of the four axes; it has order \(2^4 \cdot 4! = 384\). An element \(\rho = (\pi, \epsilon)\), with \(\epsilon_i \in \{\pm 1\}\), acts on \(\mathbb{Z}^4\) and on \(\Lambda_{\ell}\) by \((\rho x)_{\pi (i)} = \epsilon_i x_i\). On either lattice \(\Lambda\), an affine map is \(A = (\rho, b)\), \(Ax = \rho x + b\), with \(b \in \Lambda\).
A walk \(w\) is a finite word of steps \(\pm e_i\). The transport \(U (x ; w) \in G\) is the ordered product along \(w\) from \(x\): a forward step along the link \((y, i)\) contributes \(U (y, i)\), and a backward step from \(y\) contributes \(U (y - e_i, i)^{- 1}\).
The pullback \(A^{\ast} U\) is
that is, the transport of \(U\) along the image of the link \((x, i)\). Then:
(a) \(1^{\ast} = \text{id}\) and \(B^{\ast} A^{\ast} = (A \circ B)^{\ast}\), i.e. \((\rho', b')^{\ast} (\rho, b)^{\ast} = (\rho \rho', \rho b' + b)^{\ast}\). For \(A = (1, b)\) the pullback is the translation \((\tau_b U) (x, i) = U (x + b, i)\).
(b) \((A^{\ast} U) (x ; w) = U (Ax ; \rho w)\), where \(\rho w\) is the image word.
(c) For \(g : \Lambda \to G\) put \(U^g (x, i) = g (x) U (x, i) g (x + e_i)^{- 1}\). Then \(U^g (x ; w) = g (x) \hspace{0.17em} U (x ; w) \hspace{0.17em} g (x_{\text{end}})^{- 1}\); for a closed walk this is conjugation by \(g (x)\).
(d) (locality) If \(U\) and \(V\) agree on every link whose two endpoints are visited by \(w\) from \(x\), then \(U (x ; w) = V (x ; w)\).
(e) (plaquettes) A plaquette is \(p = (x ; i < j)\) with holonomy \(U_p = U (x ; e_i, e_j, - e_i, - e_j)\). Each \(A\) induces an injective map $A_\sharp $ of plaquettes, bijective on \(\Lambda_{\ell}\). It sends the axis pair to \(\{\pi (i), \pi (j)\}\) in increasing order and the base point to \(Ax - \sum_{k \in \{i, j\} : \hspace{0.17em} \epsilon_k = - 1} e_{\pi (k)}\). For every set \(M\) and every \(\phi : G \to M\) satisfying \(\phi (u^{- 1}) = \phi (u)\) and \(\phi (vuv^{- 1}) = \phi (u)\) for all \(u, v \in G\), one has \(\phi ((A^{\ast} U)_p) = \phi (U_{A_{\sharp} p})\).
(f) (periodic extension) For \(b \in \Lambda_{\ell}\) let \((E_b U) (y, i) = U ([y]_{\ell} + b, i)\) on \(\mathbb{Z}^4\). Then:
\((E_b U) (y ; w) = U ([y]_{\ell} + b ; w)\);
\(E_b ((\rho, a)^{\ast} U) = (\rho, 0)^{\ast} (E_{\rho b + a} U)\) for \(a \in \Lambda_{\ell}\);
\(E_b (U^g) = (E_b U)^{g ([\cdot]_{\ell} + b)}\);
if \(F\) depends only on links inside \(X \subset \mathbb{Z}^4\), then \(F \circ E_b\) depends only on links inside \([X]_{\ell} + b\).
A function \(F\) of configurations is gauge invariant if \(F (U^g) = F (U)\) for all \(g\). It is local on \(X\) if \(F (U) = F (V)\) whenever \(U, V\) agree on all links with both endpoints in \(X\). These are the notions of (Reg3)–(Reg4) in the main statement. Status: theorem. Lean: Signed with its group law, Signed.act, travel, pull, pull_pull, pull_one, travel_pull, gauge, travel_gauge, GaugeInvariant, Agrees, LocalOn, travel_local, Face.send, Face.injective_send, Face.send_price, expandAt, expandAt_pull, travel_at, expandAt_gauge, LocalOn.expandAt.
Proposition I.1.3 (schedule of spacings; subdivision and blocks).
(a) Schedule. A schedule consists of odd integers \(L_m \geqslant 3\) and a strictly increasing sequence of positive integers \(P_p\) with \(4 \mid P_p\), indexed by \(m, p \in \mathbb{N}\). Put
Then:
\(D_m\) is odd and strictly increasing;
\(a_0 = 1\), \(a_{m + 1} = a_m / L_m\), and \(a_m \downarrow 0\) strictly;
\(s_0 = 0\), \(s_{m + 1} = s_m + \log L_m\), and \(s_m \to \infty\);
\(4 \mid \ell (m, p)\) and \(a_m \ell (m, p) = P_p\);
the placement \(x \mapsto a_m \hspace{0.17em} r_{\ell (m, p)} (x)\) maps \(\Lambda_{\ell (m, p)}\) into \((- P_p / 2, P_p / 2]^4\).
For \(m \leqslant m'\), \(\Lambda_{\ell (m, p)} \to \Lambda_{\ell (m', p)}\) is a subdivision of odd ratio \(L_m \cdots L_{m' - 1}\) that preserves placements.
(b) Mesh inequality. If \(\log L_m / s_m \to 0\) as \(m \to \infty\) (the schedule is gentle), then there is \(c > 0\), depending on the schedule, with \(a_{m + 1} \geqslant c \hspace{0.17em} a_m^2\) for all \(m \in \mathbb{N}\). The ratio defining gentleness is considered for \(m \geqslant 1\), when \(s_m > 0\).
(c) Subdivision. Let \(L \geqslant 1\) be an integer and \(\ell' = \ell L\). Put \(\iota (x) = [L \hspace{0.17em} r_{\ell} (x)]_{\ell'}\). Then:
\(r_{\ell'} (\iota x) = L \hspace{0.17em} r_{\ell} (x)\);
\(\iota : \Lambda_{\ell} \to \Lambda_{\ell'}\) is an injective group homomorphism commuting with \(B_4\).
The residue map \(\text{res} : \Lambda_{\ell'} \to \Lambda_L\) (reduction mod \(L\)) is additive with \(\text{res} \circ \iota = 0\). The map \((x, y) \mapsto \iota (x) + [r_L (y)]_{\ell'}\) is a bijection \(\Lambda_{\ell} \times \Lambda_L \to \Lambda_{\ell'}\), with inverse \(z \mapsto (\text{par} \hspace{0.17em} z, \text{res} \hspace{0.17em} z)\); moreover \(\text{par} \circ \iota = \text{id}\).
(d) Odd ratio. If \(L\) is odd, then:
\(r_L\) commutes with \(B_4\) and \(\text{par} (\rho z) = \rho \hspace{0.17em} \text{par} (z)\);
the block \(\text{par}^{- 1} (x)\) is \(\iota (x) + [\{- \tfrac{L - 1}{2}, \ldots, \tfrac{L - 1}{2} \}^4]_{\ell'}\);
\(\text{par} (z + e_i) = \text{par} (z) + e_i\) if \(r_L (\text{res} \hspace{0.17em} z)_i = \tfrac{L - 1}{2}\), and \(\text{par} (z + e_i) = \text{par} (z)\) otherwise.
(e) Metric comparison. For every integer \(L \geqslant 1\), all \(x, x' \in \Lambda_{\ell}\) and all \(z, z' \in \Lambda_{\ell'}\):
\(|z - \iota \hspace{0.17em} \text{par} \hspace{0.17em} z|_{\ell'} \leqslant L / 2\);
\(L \hspace{0.17em} |x - x' |_{\ell} \leqslant | \iota x - \iota x' |_{\ell'}\);
\(L \hspace{0.17em} | \text{par} \hspace{0.17em} z - \text{par} \hspace{0.17em} z' |_{\ell} \leqslant |z - z' |_{\ell'} + L\) (foreshortening).
Status: (a) the schedule is a definition (data with conditions); the schedule of the construction is chosen in III.2. The consequences are theorems. Lean: Schedule, denominator, side, side_four, a, strictAnti, shrinks, scale, divergent, span_side, place, places_period, toDepth, toDepth_odd, embeds_place, Gentle, mesh_inequality, Ratio, Ratio.embed, Ratio.root, Ratio.signed, Ratio.residue, Ratio.assemble_bijective, Ratio.parent, Ratio.parent_signed, Ratio.half_box, Ratio.parents_hop, Ratio.embed_parent, Corona.near, Corona.below, Corona.foreshorten.
Proposition I.1.4 (cover numbers of lattice sets). For \(X \subset \Lambda_{\ell}\) and \(M \in \mathbb{N}\) let \(\text{halo}_M (X) = \{y : \exists x \in X, |x - y|_{\ell} \leqslant M\}\). Use the graph joining sites at sup-distance at most \(1\); diagonal moves are allowed. Connectivity means that every two vertices can be joined, so the empty set is allowed. A set \(D\) covers \(X\) at scale \(M\) if \(X \subset \text{halo}_M (D)\) and \(\text{halo}_M (D)\) is connected in this graph. Let \(\text{cov}_M (X)\) be the least \(|D|\) over such \(D\); it is finite since \(D = \Lambda_{\ell}\) covers. For all \(X, Y, D \subset \Lambda_{\ell}\) and \(M \in \mathbb{N}\), the following bounds hold independently of \(\ell\):
(a) \(X \subset Y\) implies \(\text{cov}_M (X) \leqslant \text{cov}_M (Y)\), and \(\text{cov}_M (X \cup Y) \leqslant \text{cov}_M (X) + \text{cov}_M (Y)\) if \(X \cap Y \neq \emptyset\).
(b) \(| \text{halo}_M (D) | \leqslant (2 M + 1)^4 |D|\) and \(|X| \leqslant (2 M + 1)^4 \hspace{0.17em} \text{cov}_M (X)\).
(c) \(\text{halo}_M (D)\) is connected in this graph if and only if any two points of \(D\) are joined by a chain in \(D\) with steps of sup-distance at most \(2 M + 1\).
(d) (diameter) For \(x, y \in X\): \(|x - y|_{\ell} \leqslant (4 M + 2) \hspace{0.17em} \text{cov}_M (X) + 2 M\).
(e) (change of scale) For \(b, k, m \in \mathbb{N}\) with \(b \geqslant 1\) and \(x \in \Lambda_{\ell}\), \(\text{cov}_b (\text{halo}_{kb} \{x\}) \leqslant (2 k + 1)^4\). If \(D\) covers \(X\) at scale \(mb\), then \(\text{cov}_b (X) \leqslant (2 m + 3)^4 |D|\).
(f) (coarsening) Let \(\text{par} : \Lambda_{\ell L} \to \Lambda_{\ell}\) be the parent map of a subdivision of integer ratio \(L \geqslant 1\). Let \(X \subset \Lambda_{\ell L}\) and \(M, B, s \in \mathbb{N}\) with \(B \geqslant 1\) and \(M \leqslant 2 B\). Then
with the left side computed on the coarse torus. Thus a fine set of cover number \(k\) has coarse cover number \(\leqslant (2 s + 13)^4 (1 + 10 k / L)\).
Status: theorem. Lean: halo, covers, coverNumber, Corona.cover_mono, Corona.meets_sum, Corona.halo_size, Corona.pays, Corona.ship_halo, Corona.halo_ship, Corona.extent, Corona.tile, Corona.tiling, Corona.collapse.
Proposition I.1.5 (gauge group, plaquette cost, chord, Haar measure). \(G\) is compact and second countable, with its Borel structure. For \(u \in G\) put
Then:
\(\chi^2 = 2 \sigma\), \(0 \leqslant \chi \leqslant 2\), \(0 \leqslant \sigma \leqslant 2\), and \(\sigma (1) = \chi (1) = 0\);
\(\chi (uv) \leqslant \chi (u) + \chi (v)\);
\(\sigma\) and \(\chi\) are continuous and invariant under \(u \mapsto u^{- 1}\) and \(u \mapsto vuv^{- 1}\).
The Frobenius norm is invariant under left and right multiplication by unitaries, and \(\|u\|_F = \sqrt{N}\). The normalized Haar measure \(\mu_G\) is invariant under left and right translations and under inversion; hence \(u \mapsto vuw\) preserves \(\mu_G\). Status: theorem. Lean: Gauge, compactGaugeU, price, chord, chord_price, chord_le_two, price_range, chord_triangle, price_inv, price_conj, chord_inv, chord_conj, mul_frob_eq, probHaar, probHaar_rightInv, preserve_inv, preserve_two_sided.
Proposition I.1.6 (Lie algebra, colours, Casimir constant, slope constant).
(a) \(\mathfrak{g}= \{X \in M_N (\mathbb{C}) : e^{tX} \in G \forall t \in \mathbb{R}\}\) is a closed real subspace of skew-Hermitian matrices. It is closed under \([X, Y] = XY - YX\) and under \(X \mapsto uXu^{- 1}\), \(u \in G\). Tangent criterion: if \(f : \mathbb{R} \to M_N (\mathbb{C})\) takes values in \(G\), \(f (0) = 1\), and \(f\) is differentiable at \(0\) with \(f' (0) = X\), then \(X \in \mathfrak{g}\).
(b) Colours. \(\langle X, Y \rangle_{\mathfrak{g}} = - \frac{1}{N} \text{tr} (XY)\) is real on \(\mathfrak{g}\) and is an \(\text{Ad} (G)\)-invariant inner product. Fix an orthonormal basis and write \(\mathfrak{g} \cong \mathbb{R}^d\), \(c \mapsto X_c\), \(d = \dim \mathfrak{g}\). Then:
\(\|X_c \|_F = \sqrt{N} \hspace{0.17em} |c|\);
if \(P_{\mathfrak{g}}\) is the orthogonal projection onto \(\mathfrak{g}\) for the real Frobenius inner product \(\text{Re} \hspace{0.17em} \text{tr} (XY^{\ast})\), define \(\Pi : M_N (\mathbb{C}) \to \mathbb{R}^d\) by \(X_{\Pi A} = P_{\mathfrak{g}} A\). Then \(\Pi X_c = c\) and \(| \Pi A| \leqslant \|A\|_F / \sqrt{N}\);
\(\text{Ad} (u) c = \Pi (uX_c u^{- 1})\) is an orthogonal representation of \(G\);
\(\text{ad} (c) c' = \Pi [X_c, X_{c'}]\) satisfies \(X_{\text{ad} (c) c'} = [X_c, X_{c'}]\);
\(c \mapsto \exp c := e^{X_c} \in G\) is continuous.
(c) \(G\) is simple if two conditions hold: there are \(X, Y \in \mathfrak{g}\) with \([X, Y] \neq 0\), and every real subspace \(I \subset \mathfrak{g}\) with \([\mathfrak{g}, I] \subset I\) is \(0\) or \(\mathfrak{g}\). This agrees with the simplicity hypothesis of the main theorem: its Lie algebra is characterized by the same exponential condition (GaugeGroup.coreSimple, unit VIII.1).
If \(G\) is simple, then:
(d) for every symmetric real bilinear form \(B\) on \(\mathbb{R}^d\) satisfying \(B (\text{ad} (x) y, z) = - B (y, \text{ad} (x) z)\) for all \(x, y, z\), there is \(c \in \mathbb{R}\) such that \(B (x, y) = c \langle x, y \rangle\) for all \(x, y\);
(e) \(- \text{tr}_{\mathbb{R}^d} (\text{ad} (x) \text{ad} (y)) = C_{\mathfrak{g}} \langle x, y \rangle\) for all \(x, y \in \mathbb{R}^d\), where \(C_{\mathfrak{g}} := - \frac{1}{d} \sum_{i = 1}^d \text{tr}_{\mathbb{R}^d} (\text{ad} (b_i)^2) > 0\) and \((b_i)_{i = 1}^d\) is the standard orthonormal basis;
(f) the slope constant \(\kappa := 11 C_{\mathfrak{g}} / (24 \pi^2)\) is positive.
Status: theorem; \(C_{\mathfrak{g}}\) and \(\kappa\) are definitions. Lean: lie, closed_lie, generates_skew, tangent_generates, bracket_mem, conjugate_mem, Simple, colorIso, colorMat, trace_color, trace_color_real, colorMat_norm, projectColor, colorMat_project, project_colorMat, project_norm, adjoint, adjoint_mul, norm_adjoint, ad, colorMat_ad, lieExp, lieExp_cont, casimir, slopeConstant, invariant_scalar, trace_adjoint, casimir_pos, slopeConstant_pos.
Proposition I.1.7 (Haar mass of small balls around \(1\)). For \(0 < r \leqslant 1\), \(\mu_G \{u : \chi (u) \leqslant r\} \geqslant (r / 4)^{2 N^2}\). The exponent is the real dimension of \(M_N (\mathbb{C})\), not \(\dim G\). Status: theorem (not sharp). Lean: shell, shell_ge.
Proposition I.1.8 (local logarithm, nearest point, curvature of a group element). Let \(\mathfrak{g}^{\perp}\) be the orthogonal complement of \(\mathfrak{g}\) in \(M_N (\mathbb{C})\) for \(\text{Re} \hspace{0.17em} \text{tr} (XY^{\ast})\).
(a) There are a map \(\log_G : M_N (\mathbb{C}) \to \mathbb{R}^d\), an open neighbourhood \(V\) of \(1\) in \(M_N (\mathbb{C})\) and an open neighbourhood \(W\) of \(0\) in \(\mathbb{R}^d\) such that:
\(\log_G\) is real-analytic on \(V\) and \(\log_G (1) = 0\);
\(\log_G (e^{X_c}) = c\) for all \(c \in W\);
\(\exp (\log_G u) = u\) for all \(u \in G \cap V\).
It is the first component of the local inverse of \((c, \nu) \mapsto e^{X_c} e^{\nu}\) on \(\mathbb{R}^d \times \mathfrak{g}^{\perp}\). Connectedness of \(G\) is not assumed, and nothing is claimed away from \(1\).
(b) \(u \in G\) is a nearest point to \(X \in M_N (\mathbb{C})\) if \(\|X - u\|_F \leqslant \|X - v\|_F\) for all \(v \in G\). Then:
nearest points exist;
for every \(v \in G\), \(u\) is nearest to \(X\) iff \(vu\) is nearest to \(vX\), iff \(uv\) is nearest to \(Xv\);
if \(u\) is nearest to \(X\), then \(u^{\ast} X - 1 \in \mathfrak{g}^{\perp}\);
there is \(\delta > 0\), depending only on \(G\), such that \(X\) has exactly one nearest point whenever \(\|X - v\|_F < \delta\) for some \(v \in G\).
On a neighbourhood of \(1\) the nearest point is \(\text{foc} (X) = \exp (c)\), where \((c, \nu)\) is the local inverse of \(\Psi (c, \nu) = e^{X_c} (1 + \nu)\) at \(X\). As a matrix-valued map, \(X \mapsto \text{foc} (X)\) is real-analytic at \(1\), with strict derivative the orthogonal projection \(P_{\mathfrak{g}}\) onto \(\mathfrak{g}\). Consequently, for every \(\varepsilon > 0\) there is \(\delta > 0\) such that, for all \(X, Y \in M_N (\mathbb{C})\) and \(u, v \in G\), the conditions \(\|X - 1\|_F, \|Y - 1\|_F < \delta\) and \(u, v\) nearest to \(X, Y\) imply
(c) The curvature of \(u \in G\) is \(F (u) = \Pi \left( \tfrac{1}{2} (u - u^{- 1}) \right) \in \mathbb{R}^d\), the projected anti-Hermitian part. It is continuous and satisfies:
\(F (1) = 0\) and \(F (u^{- 1}) = - F (u)\);
\(F (vuv^{- 1}) = \text{Ad} (v) F (u)\);
\(|F (u) | \leqslant \chi (u)\);
\(\frac{d}{dt} F (\exp (tc)) |_{t = 0} = c\).
Status: theorem. The neighbourhoods and \(\delta\) come from the inverse function theorem and compactness and are not explicit. Lean: plog, splitNormal, plog_diff, plog_left, plog_right, Nearest, nearest_exists, nearest_left, nearest_right, nearest_normal, focus, nearest_is_focus, focus_diff, nearest_tube, focus_deriv, nearest_approximate, curvature, curvature_inv, curvature_conjugate, curvature_norm, leading_curvature.
Proposition I.1.9 (the Wilson measure on the torus). Let \(\mu_{\ell} = \mu_G^{\otimes (\Lambda_{\ell} \times \{0, \ldots, 3\})}\) be the product Haar measure on configurations. Let
be the Wilson energy: \(6 \ell^4\) terms, each unoriented plaquette counted once. For every \(\beta \in \mathbb{R}\) put
For integrable complex \(F\), write \(\langle F \rangle_{\ell, \beta} = \int F \hspace{0.17em} d\mathbb{P}_{\ell, \beta}\). Then:
\(E_{\ell}\) is continuous, with \(0 \leqslant E_{\ell} \leqslant 12 \ell^4\);
\(Z_{\ell} (\beta) > 0\) and \(\mathbb{P}_{\ell, \beta}\) is a probability measure;
every continuous complex function, and every bounded measurable complex function, is \(\mathbb{P}_{\ell, \beta}\)-integrable;
for integrable complex \(F\), \(\langle F \rangle_{\ell, \beta} = Z_{\ell} (\beta)^{- 1} \int e^{- \beta E_{\ell}} F \hspace{0.17em} d \mu_{\ell}\).
These are the energy and measure of Definitions 1.3–1.4 of the main statement, with \(h = \mu_G\); the identification is proved in VIII.1 (energy_eq, law_eq). Status: definition, with theorems. Lean: Wilson.Config, energy, energy_nonneg, energy_cont, prior, bare, partition, partition_pos, law, moment, continuous_integrable, bounded_integrable, moment_formula.
Proposition I.1.10 (invariance of the Wilson measure). For every \(\beta \in \mathbb{R}\), every lattice affine map \(A = (\rho, b)\) and every \(g : \Lambda_{\ell} \to G\), the maps \(U \mapsto A^{\ast} U\) and \(U \mapsto U^g\) preserve \(\mu_{\ell}\), \(E_{\ell}\) and \(\mathbb{P}_{\ell, \beta}\).
This covers translations, axis permutations, and reflections \(x_a \mapsto b_a - x_a\). When \(\ell\) is even, the reflection planes pass through sites for even \(b_a\) and bisect links for odd \(b_a\). Hence \(\langle F \circ A^{\ast} \rangle_{\ell, \beta} = \langle F \rangle_{\ell, \beta}\) for every integrable complex \(F\). More generally, any measurable map preserving \(\mu_{\ell}\) and \(E_{\ell}\) preserves \(\mathbb{P}_{\ell, \beta}\). Reflection positivity is a separate assertion. Status: theorem. Lean: energy_pull, energy_gauge, prior_pull, prior_gauge, law_preserved, law_pull, law_gauge, law_translate, moment_pull.
Proposition I.1.11 (continuum symmetry groups). \(\mathbb{R}^4\) carries the Euclidean norm and Lebesgue measure. Write \(x = (\vec{x}, x_t)\), with time the last coordinate. Let \(\textsf{m} (x) = x_t^2 - | \vec{x} |^2\), \(\bar{V}_+ = \{| \vec{x} | \leqslant x_t \}\) and \(V_+ = \{| \vec{x} | < x_t \}\).
(a) Euclidean. There is a homomorphism \(B_4 \to O (4)\) whose image acts on coordinates as in Proposition I.1.2. The mirror in axis \(a\) maps to the reflection \(x_a \mapsto - x_a\). The real skew matrices form \(\mathfrak{so} (4)\), with plane generators \(\Omega_{ij} = E_{ij} - E_{ji}\) for \(i < j\), and \(t \mapsto e^{t \Omega} \in O (4)\).
(b) Lorentz group. \(\mathcal{L}= \{g \in GL (\mathbb{R}^4) : \det g = 1, \textsf{m} \circ g = \textsf{m}, g \bar{V}_+ = \bar{V}_+ \}\); this is \(SO^+ (1, 3)\). It is a topological group of Lebesgue-preserving maps that preserves \(V_+\), and it contains every element of \(SO (4)\) fixing the time coordinate. For a unit spatial vector \(e\), the boost
is a continuous one-parameter subgroup of \(\mathcal{L}\). Its generator at \(r = 0\) is \(K_e : (\vec{x}, x_t) \mapsto \left( x_t \hspace{0.17em} e, \vec{x} \cdot e \right)\).
(c) Dual action and cone. With \(R (\vec{x}, x_t) = (- \vec{x}, x_t)\), put \(g^{\vee} = RgR\). Then \(g^{\vee} \in \mathcal{L}\), \((g^{\vee})^{\vee} = g\), \(g^{\vee} = (g^{- 1})^T\), and \(\langle gx, g^{\vee} p \rangle = \langle x, p \rangle\) for the Euclidean pairing. Moreover:
\(x \in \bar{V}_+\) iff \(\langle x, p \rangle \geqslant 0\) for all \(p \in \bar{V}_+\);
if \(x \in V_+\) and \(p \in \bar{V}_+ \setminus 0\), then \(\langle x, p \rangle > 0\);
\(x \in \bar{V}_+\) iff \((g^{\vee} x)_t \geqslant 0\) for every \(g \in \mathcal{L}\).
(d) Poincaré group. \(\mathcal{P}=\mathcal{L} \ltimes \mathbb{R}^4\) consists of pairs \((\Lambda, b)\) and acts by \(x \mapsto \Lambda x + b\), with \((\Lambda, b) (\Lambda', b') = (\Lambda \Lambda', \Lambda b' + b)\). It is a topological group, determined by its action on points, and preserves Lebesgue measure on \(\mathbb{R}^4\) and, diagonally, on \((\mathbb{R}^4)^n\) for every \(n \in \mathbb{N}\).
(e) Configurations. For \(n \in \mathbb{N}\), \(r \in \mathbb{R}\) and \(x \in (\mathbb{R}^4)^n\), call \(x\):
separated (\(x \in \text{Conf}_n\)) if the \(x_i\) are pairwise distinct; \(\text{Conf}_n\) is open;
generic if for every axis the \(n\) coordinates on that axis are pairwise distinct (generic implies separated);
\(r\)-spaced if \(|x_i - x_j |_{\infty} \geqslant r\) for \(i \ne j\) (\(r > 0\) implies separated).
Lebesgue measure on \((\mathbb{R}^4)^n\) is the product of the four-dimensional ones. Status: theorem, apart from the definitions. Lean: Signed.ortho, Signed.ortho_move, Signed.inclusionO, Signed.mirrored, Ortho.skewSpace, Ortho.oneParameter, Ortho.plane, Mink.Proper, Lorentz.preserves, Lorentz.stays, Lorentz.room, Ray.boost, Ray.generator, Lorentz.dual, Lorentz.dual_sent, Lorentz.dual_adjoint, cone_test, strict_pair, Lorentz.from_half, Poincare, Poincare.by_placement, Poincare.together, conf, generic, spaced, conf_open, Points.volume_product.
Proposition I.1.12 (boxes, axis routes and slabs on the cover). For \(r \in \mathbb{N}\) let \(B_r = \{- r, \ldots, r\}^4 \subset \mathbb{Z}^4\); then \(|B_r | = (2 r + 1)^4\) and \(\rho B_r = B_r\) for \(\rho \in B_4\). For \(x, y \in \mathbb{Z}^4\) and an ordering \(o\) of the four axes, the axis route \(\text{route}_o (x, y)\) goes from \(x\) in straight legs along the axes in the order \(o\), adjusting one coordinate at a time. It ends at \(y\), and:
reversing the order gives the reverse of \(\text{route}_o (y, x)\);
the image of \(\text{route}_o (x, y)\) under \(A = (\rho, b)\) is \(\text{route}_{\pi \circ o} (Ax, Ay)\).
For an anchor \(a \in \mathbb{Z}^4\) and a set \(J\) of axes, define \(\text{slab} (r, a, J) = \{y \in B_r : y_i = a_i \text{ for } i \not\in J\}\). If \(i \not\in J\), then, for any function \(f\) with values in a commutative additive monoid,
For \(z, w \in \{- r, \ldots, r\}\), translation by \((w - z) e_i\) maps the layer with pinned coordinate \(z\) bijectively onto the layer with pinned coordinate \(w\). Write \(\text{straight}_j (d)\) for the signed leg of displacement \(de_j\), \(d \in \mathbb{Z}\). For every cover configuration \(U\),
Status: theorem. Lean: box, box_covariant, Corona.box_size, route, route_end, back_route, send_route, slab, slab_sum, slab_update_image, carry_succ.
No result from another unit is needed. The definitions and estimates are used throughout the proof. Main consumers include:
the nearest-point retraction (I.1.8(b)): gauge fixing and alignment in III.1;
the invariant-form lemma (I.1.6(d)): the one-loop computation in III.7 and the stress-tensor forms in VI.1–VI.2;
the coarsening estimate (I.1.4(f)): III.5 and III.10;
the mesh inequality: V.1;
invariance of the law: IV.1, III.13 and VII.2;
positivity of \(C_{\mathfrak{g}}\): VIII.1 (the slope in (Reg1)) and VII.2;
the Haar-ball bound (I.1.7): a product lower bound for configurations whose links are near \(1\) (YM.RG.space), a lower bound for the block-kernel noise normalizer (YM.SectorNext.Vacuum.shell_pZ), and positivity of the ball mass used in corrected-field tests (YM.Push.Corrected.Tests.Offer.stamp).
Standard mathematics. The underlying results from Mathlib, together with references for the mathematical context, are:
Existence and uniqueness of Haar measure on a compact group, and its right and inversion invariance (Halmos, Measure Theory, §§58–60; Folland, A Course in Abstract Harmonic Analysis, Ch. 2).
Product measures and invariance of product measures under coordinate permutations; change of variables for measure-preserving maps; Bochner integration.
Matrix exponential; the inverse function theorem in the real-analytic class.
The local structure of a closed matrix group near \(1\): the Lie algebra as the set of generators of one-parameter subgroups, the Lie–Trotter limit, and von Neumann's argument that a closed subgroup meets \(\exp (\mathfrak{g}^{\perp})\) only at \(1\) near the identity (B. C. Hall, Lie Groups, Lie Algebras, and Representations, Ch. 2–3). The local structure is derived here from the exponential definition of the Lie algebra.
Spectral theorem for symmetric operators on a finite-dimensional Euclidean space; the Schur-type uniqueness of invariant forms on a simple real Lie algebra (Knapp, Lie Groups Beyond an Introduction, Ch. I).
Scaling of Lebesgue measure of balls in a finite-dimensional space.
Lattice gauge theory: Wilson's action (K. Wilson, Phys. Rev. D 10 (1974) 2445); gauge-invariant lattice measures and their symmetries (Osterwalder–Seiler, Ann. Phys. 110 (1978) 440; Seiler, Lecture Notes in Physics 159).
Minkowski geometry and the restricted Lorentz and Poincaré groups (Streater–Wightman, PCT, Spin and Statistics, and All That, Ch. 1).
Imported constants and assumptions. The group \(G\) is an arbitrary closed subgroup of \(U (N)\), possibly disconnected. Simplicity is assumed only in I.1.6(d)–(f), and agrees with the main theorem's hypothesis by the bridge cited in I.1.6(c). The local radii depend on \(G\); the mesh constant also depends on the schedule.
The torus metric comes from the absolute values of centred representatives. Odd subdivision ratios give symmetric residue windows, so the parent map commutes with signed permutations even when the coarse torus has even side. The schedule separates the physical period from the mesh; gentle growth of the block factors gives the mesh inequality. Cover numbers measure the size of lattice sets by connected unions of balls. Sampling a path through a fine cover at intervals of \(LB\) yields the coarsening bound \(O (1 + k / L)\) for a set of fine cover number \(k\), with constants independent of the period.
Affine pullback assigns to each link the transport along its image. Reflections reverse links and shift their base points. The pullback therefore permutes link variables and inverts some of them, preserving product Haar measure. Image plaquette holonomies differ by conjugation and possibly inversion; the cost \(\sigma\) is invariant under both. Summing over the plaquette bijection proves invariance of the energy and hence of the Wilson measure.
For the closed matrix group, the tangent criterion follows by approximating \(e^X\) with \(f (1 / n)^n\). It gives the linear and bracket closure of \(\mathfrak{g}\). The inverse function theorem, together with isolation of \(\exp (\mathfrak{g}^{\perp})\) from \(G\) near \(1\), gives a local logarithm. Applying the same theorem to \(e^{X_c} (1 + \nu)\) identifies the nearest-point retraction and its derivative. Finally, the spectral theorem reduces an invariant symmetric form to an eigenspace; real simplicity forces that eigenspace to be all of \(\mathfrak{g}\). Applied to the negative Killing form, this gives \(C_{\mathfrak{g}} > 0\).
Lemma I.1.a (centred lift and separation). Proposition I.1.1. Proof sketch. Coordinatewise, \(r_{\ell}\) is the representative of minimal absolute value, with the tie sent to \(+ \ell / 2\). This gives uniqueness, \(|r_{\ell} (a + b)_i | \leqslant |r_{\ell} (a)_i | + |r_{\ell} (b)_i |\) (hence the triangle inequality for the maximum), and \(|r_{\ell} ([y]_{\ell})_i | \leqslant |y_i |\). A zero lift reduces to the zero site, so zero separation implies equality. Signed permutations permute coordinates and change signs, and the centred absolute value is even, including at the tie. Thus \(| \rho x|_{\ell} = |x|_{\ell}\). There are \(\ell\) choices in each coordinate, giving \(\ell^4\) sites. Lean: lift_eq_iff, pdist_triangle, pdist_invariant, Corona.lap.
Lemma I.1.b (subdivision). Proposition I.1.3(c). Proof sketch.
\(L \hspace{0.17em} r_{\ell} (x)\) lies in the centred window of \(\Lambda_{\ell L}\), so \(r_{\ell L} (\iota x) = L \hspace{0.17em} r_{\ell} (x)\), and \(\iota\) is injective.
For every \(y \in \mathbb{Z}^4\), \(\iota ([y]_{\ell}) = [Ly]_{\ell L}\), because changing \(y\) by \(\ell z\) changes \(Ly\) by \(\ell Lz\). This gives additivity and \(B_4\)-covariance.
Applying \(\text{res}\) to an equality of assembled points gives equality of their residue coordinates. Injectivity of \(\iota\) then gives equality of the coarse coordinates. The count \(\ell^4 L^4 = (\ell L)^4\) makes the assembly map bijective. Since \(\iota x\) is assembled from \((x, 0)\), its parent is \(x\).
Lean: Ratio.lift_embed, Ratio.embed_injective, Ratio.root, Ratio.signed, Ratio.residue_embed, Ratio.assemble_bijective, Ratio.unpack_at.
Lemma I.1.c (odd blocks and their symmetry). Proposition I.1.3(d). Proof sketch. For odd \(L\), the window \((- L / 2, L / 2] \cap \mathbb{Z}= [- \frac{L - 1}{2}, \frac{L - 1}{2}] \cap \mathbb{Z}\) is symmetric. Hence \(r_L (\rho y) = \rho \hspace{0.17em} r_L (y)\), the assembly map intertwines \((\rho, \rho)\) with \(\rho\), and \(\text{par}\) commutes with \(\rho\). Every residue has exactly one representative in \(B_{(L - 1) / 2}\), so the assembly bijection identifies each parent fibre with the stated translated box. For the crossing rule, write \(z = \iota (\text{par} \hspace{0.17em} z) + [q]\) with \(q = r_L (\text{res} \hspace{0.17em} z) \in B_{(L - 1) / 2}\).
If \(q_i < \frac{L - 1}{2}\), then \(q + e_i\) stays in the block.
If \(q_i = \frac{L - 1}{2}\), then \(z + e_i = \iota (\text{par} \hspace{0.17em} z + e_i) + [q - (L - 1) e_i]\), because \(\iota (x + e_i) = \iota (x) + Le_i\).
Lean: Ratio.lift_signed_of_odd, Ratio.assemble_signed, Ratio.parent_signed, Ratio.half_box, Ratio.half_lift, Ratio.parents_hop, Ratio.internal.
Lemma I.1.d (metric comparison across scales). Proposition I.1.3(e). Proof sketch.
The vector \(r_L (\text{res} \hspace{0.17em} z)\) has coordinates bounded by \(L / 2\). Its reduction is \(z - \iota (\text{par} \hspace{0.17em} z)\), and reduction cannot increase the torus norm.
The lift of \(\iota (x - x')\) is \(L \hspace{0.17em} r_{\ell} (x - x')\), which gives the second inequality.
Foreshortening follows from the second inequality and two applications of the first.
Lean: Corona.near, Corona.below, Corona.foreshorten.
Lemma I.1.e (schedule and mesh inequality). Proposition I.1.3(a),(b). Proof sketch. The product \(D_m\) is odd, strictly increasing and at least \(3^m\), which gives \(a_m \to 0\) and \(s_m \to \infty\). Placement follows by multiplying \(- \ell < 2 r_{\ell} \leqslant \ell\) by \(a_m\) and using \(a_m \ell (m, p) = P_p\). The scaled-lift identity gives compatibility under subdivision. Gentleness gives \(\log L_m < s_m\) for all \(m \geqslant m_0\), hence \(a_{m + 1} > a_m^2\). Take \(c\) to be the minimum of the finite nonempty set \(\{1\} \cup \{a_{m + 1} / a_m^2 : m < m_0 \}\); every member is positive. Lean: den_strict, a_step, scale_succ, embeds_scale, mesh_inequality.
Lemma I.1.f (elementary properties). Proposition I.1.4(a),(b). Proof sketch. A cover of a set covers its subsets. For the union, take optimal covers \(A\) of \(X\) and \(B\) of \(Y\). The halo of \(A \cup B\) is the union of the two haloes, which is connected through a common point of \(X \cap Y\). Counting: \(\text{halo}_M (D)\) is the image of \(D \times B_M\) under \((x, q) \mapsto x + [q]\), and \(X \subset \text{halo}_M (D)\) for an optimal \(D\). Lean: Corona.cover_mono, Corona.meets_sum, Corona.halo_size, Corona.pays.
Lemma I.1.g (connectivity of haloes). Proposition I.1.4(c). Proof sketch. Use lattice segments \(t \mapsto x + [\text{clamp}_{[- t, t]} r_{\ell} (y - x)]\), which move by \(\leqslant 1\) per unit time and reach \(y\) at \(t = |x - y|_{\ell}\). If \(|x - y|_{\ell} \leqslant 2 M + 1\), the segment stays in \(\text{halo}_M \{x, y\}\). Conversely, a path with steps of sup-distance at most \(1\) in \(\text{halo}_M (D)\) yields a chain of centres with consecutive distances \(\leqslant 2 M + 1\), by the triangle inequality. Lean: Corona.ship_halo, Corona.halo_ship.
Lemma I.1.h (a path through a chained set; diameter). Let \(D \subset \Lambda_{\ell}\) and \(b \in \mathbb{N}\). Suppose any two points of \(D\) can be joined in \(D\) by a chain with steps of sup-distance at most \(b\). Then there are \(n \leqslant 2 b |D|\) and sites \(R_0, \ldots, R_n\) visiting every point of \(D\), with \(|R_{k + 1} - R_k |_{\ell} \leqslant 1\) for \(k < n\). Consequently Proposition I.1.4(d) holds. Proof sketch. For nonempty \(D\), start at one point. The chain condition gives a point of \(D\) outside the visited set within distance \(b\) of a visited point. Insert a detour out and back along a coordinatewise-clamped segment, costing at most \(2 b\) steps. For \(D = \emptyset\), take a constant path of length \(0\). A cover has chain bound \(b = 2 M + 1\), so the resulting path has length at most \((4 M + 2) \hspace{0.17em} \text{cov}_M (X)\). Distances along it are bounded by its length, and the two halo radii add \(2 M\). Lean: Corona.ring_path, Corona.path_span, Corona.extent.
Lemma I.1.i (change of scale). Proposition I.1.4(e). Proof sketch. The grid \(x + [b \hspace{0.17em} q]\), \(q \in B_k\), covers \(\text{halo}_{kb} \{x\}\) at scale \(b\):
division with remainder puts every point within \(b\) of a grid point;
consecutive grid points along coordinatewise-clamped paths are at most \(b\) apart on the torus, so the grid is \(b\)-chained.
For a general \(D\) covering \(X\) at scale \(mb\), I.1.g gives chains of centres with steps at most \(2 mb + 1 \leqslant 2 (m + 1) b\). The balls \(\text{halo}_{(m + 1) b} \{x\}\) along such a chain overlap. For each centre, the grid above gives a cover of its ball whose \(b\)-halo is connected and contains that ball. These connected haloes meet successively, so their union covers \(X\) and is connected. The union of their centres has at most \((2 m + 3)^4 |D|\) points. Lean: Corona.tile, Corona.tiling.
Lemma I.1.j (coarsening). Proposition I.1.4(f). Proof sketch.
Take an optimal cover \(A\) of \(X\) at scale \(M\). By I.1.h there is a fine path \(R_0, \ldots, R_n\) visiting \(A\), with steps of sup-distance at most \(1\) and \(n \leqslant (4 M + 2) |A|\).
Sample its parents every \(LB\) steps: \(D' = \{ \text{par} \hspace{0.17em} R_{jLB} : 0 \leqslant j \leqslant \lfloor n / (LB) \rfloor\}\). By foreshortening, parents of path points at most \(LB\) steps apart are within \(2 B\) on the coarse torus. Thus \(D'\) is \(2 B\)-chained, every \(\text{par} \hspace{0.17em} R_k\) with \(0 \leqslant k \leqslant n\) lies in \(\text{halo}_{2 B} (D')\), and \(LB |D' | \leqslant n + LB\).
A point of \(\text{halo}_{sB} (\text{par} \hspace{0.17em} X)\) lies within \(sB + 3 B + 2 B\) of \(D'\). The middle term uses \(L | \text{par} \hspace{0.17em} c - \text{par} \hspace{0.17em} v|_{\ell} \leqslant M + L \leqslant 3 BL\) for \(c \in A\) within \(M\) of \(v \in X\). The \((s + 5) B\)-halo of \(D'\) is connected by I.1.g, since \(D'\) is \(2 B\)-chained, so \(D'\) is a cover at that scale.
Apply I.1.i with \(m = s + 5\), giving a factor \((2 s + 13)^4\). Since \(M \leqslant 2 B\) and \(B \geqslant 1\), \(4 M + 2 \leqslant 10 B\). Thus \(LB |D' | \leqslant LB + 10 B |A|\), and division by \(B\) gives \(L |D' | \leqslant L + 10 |A|\).
Lean: Corona.fuel, Corona.collapse.
Lemma I.1.k (pullback is a right action). Proposition I.1.2(a). Proof sketch. Evaluate at a link. The image step of \(\pm e_i\) under \(\rho'\) and then \(\rho\) is its image under \(\rho \rho'\), and the affine maps compose as \((\rho, b) \circ (\rho', b') = (\rho \rho', \rho b' + b)\). Lean: Step.rotate_mul, pull_id, pull_pull, pull_one.
Lemma I.1.l (transport under pullback and gauge; locality). Proposition I.1.2(b)–(d). Proof sketch. Induction on the walk. One step: the variable that \(A^{\ast} U\) assigns to a step equals the variable that \(U\) assigns to the image step at the image point. For a backward step, this is the inverse of the forward variable at the shifted base point, which is why base points move under reflections. Gauge: the product telescopes, \(\prod_k g (x_k) U_k g (x_{k + 1})^{- 1} = g (x_0) \left( \prod_k U_k \right) g (x_n)^{- 1}\). Locality: each factor reads one link whose endpoints are visited. Lean: Step.pull_read, travel_pull, Step.gauge_read, travel_gauge, travel_local.
Lemma I.1.m (plaquettes under symmetries). Proposition I.1.2(e). Proof sketch.
By I.1.l, \((A^{\ast} U)_p\) is the transport of \(U\) around the image square, read from the image of the base point along the image steps.
Reversing the orientation of a square inverts its holonomy, and starting it at the next corner conjugates its holonomy.
Hence \(\phi\) of the holonomy equals \(\phi\) of the holonomy of the positive-axis square based at \(Ax - \sum_{k \in \{i, j\} : \hspace{0.17em} \epsilon_k = - 1} e_{\pi (k)}\), with axes in increasing order.
Injectivity of $A_\sharp $: the axis pair is permuted bijectively, and the shifted base point and the axis pair determine \(Ax\). On the finite torus, injective implies bijective.
Lean: oriented_swap, oriented_next, price_flip, price_shift_pos, Face.lower, Face.injective_send, Face.send_price.
Lemma I.1.n (periodic extension). Proposition I.1.2(f). Proof sketch. \(E_b U\) evaluates \(U\) at the reduced, shifted point. Reduction is additive, commutes with \(B_4\) and sends unit steps to unit steps. So transport, pullback and gauge action commute with \(E_b\) step by step, and an agreement set on the cover maps to its image. Lean: expandAt_pull, travel_at, expandAt_gauge, LocalOn.expandAt.
Lemma I.1.o (boxes, routes, slabs). Proposition I.1.12. Proof sketch. The endpoint of a route is computed axis by axis: each leg changes one coordinate by \(y_a - x_a\), and the axes are distinct. Reversal and covariance hold leg by leg, since the image of a straight leg along \(a\) is a straight leg along \(\pi (a)\) of signed length \(\epsilon_a (y_a - x_a)\). When \(i \not\in J\), the slab sum is a disjoint decomposition by \(y_i\); translating a pinned layer by \((w - z) e_i\) changes only that coordinate. The transport identity follows by appending a forward step for \(d \geqslant 0\) and cancelling a final backward step for \(d < 0\). Box covariance follows coordinatewise from the symmetric interval \([- r, r]\), and its cardinality is the product of the four interval cardinalities. Lean: route_end, back_route, send_route, slab_sum, carry_succ.
Lemma I.1.p (cost, chord, Haar). Proposition I.1.5. Proof sketch.
\(\|u - 1\|_F^2 = \text{tr} ((u - 1) (u - 1)^{\ast}) = 2 N - 2 \hspace{0.17em} \text{Re} \hspace{0.17em} \text{tr} \hspace{0.17em} u\), using \(uu^{\ast} = 1\).
\(\|u - 1\|_F \leqslant \|u\|_F + \|1\|_F = 2 \sqrt{N}\).
\(uv - 1 = (u - 1) + u (v - 1)\) and \(\|u (v - 1)\|_F = \|v - 1\|_F\) give the triangle inequality.
The trace is conjugation invariant, and \(\text{tr} \hspace{0.17em} u^{- 1} = \overline{\text{tr} \hspace{0.17em} u}\).
A right translate of \(\mu_G\) is again a left Haar probability measure, hence equals \(\mu_G\) by uniqueness. Right invariance then makes the image under inversion a left Haar probability measure as well. Continuity of \(\sigma\) and \(\chi\) follows from continuity of the trace and norm.
Lean: frob_sq, unitary_frob, chord_price, chord_triangle, price_inv, probHaar_rightInv, probHaar_invInv.
Lemma I.1.q (the Lie algebra). Proposition I.1.6(a). Proof sketch.
Skewness: differentiate \(f (t) f (t)^{\ast} = 1\) at \(0\).
Tangent criterion: if \(f\) takes values in \(G\) with \(f (0) = 1\) and \(f' (0) = X\), then \(X\) is skew and
by the telescoping \(\|A^n - B^n \|_F \leqslant n \|A - B\|_F\) for unitary \(A, B\). Unitary multiplication is isometric, so the bound has no exponential factor. Closedness of \(G\) gives \(e^X \in G\); apply this to \(f (t \hspace{0.17em} \cdot)\).
Additivity: apply the criterion to \(e^{tX} e^{tY}\).
Scalar multiplication follows by rescaling the one-parameter subgroup. The resulting real subspace is closed because it is finite dimensional. Conjugation preserves it since \(e^{s (uXu^{- 1})} = ue^{sX} u^{- 1}\) for \(u \in G\).
Brackets: \(e^{tX} Ye^{- tX} \in \mathfrak{g}\) and the closed subspace \(\mathfrak{g}\) contains the limit \([X, Y]\) of its difference quotients.
Lean: tangent_skew, frob_pow_difference, tangent_approx, tangent_generates, generates_add, conjugate_mem, bracket_mem.
Lemma I.1.r (colours, invariant forms, Casimir). Proposition I.1.6(b)–(f). Proof sketch.
The colour frame is a Frobenius-orthonormal basis of \(\mathfrak{g}\) scaled by \(\sqrt{N}\). For skew \(X, Y\), \(\text{Re} \hspace{0.17em} \text{tr} (XY^{\ast}) = - \text{tr} (XY)\) and \(\text{tr} (XY)\) is real. Orthogonal projection followed by the inverse colour frame gives \(\Pi\) and its norm bound. Unitary conjugation preserves \(\mathfrak{g}\) and the Frobenius inner product, so it induces the orthogonal adjoint representation. The exponential is continuous by the matrix exponential and the induced topology on \(G\).
(d): let \(A\) be the symmetric operator with \(\langle Ax, y \rangle = B (x, y)\). Invariance of \(B\) and skewness of \(\text{ad}\) give \(A \hspace{0.17em} \text{ad}_x = \text{ad}_x A\). Nonabelianity gives \(d > 0\). Choose a real eigenvalue \(c\) of \(A\); its nonzero eigenspace is an \(\text{ad}\)-invariant subspace, hence an ideal. Simplicity makes it the whole space, so \(A = c \cdot \text{id}\).
(e): \(K (x, y) = - \text{tr} (\text{ad}_x \text{ad}_y)\) is symmetric and invariant, by \(\text{ad}_{[x, y]} = [\text{ad}_x, \text{ad}_y]\) and cyclicity, so \(K = c \langle \cdot, \cdot \rangle\). Taking the trace over an orthonormal basis gives \(c = C_{\mathfrak{g}}\).
Positivity: \(K (x, x) = \sum_i | \text{ad}_x b_i |^2\), which is \(> 0\) for some \(x\) because \(\mathfrak{g}\) is nonabelian. This forces \(C_{\mathfrak{g}} > 0\), and the positive factor \(11 / (24 \pi^2)\) gives \(\kappa > 0\).
Lean: inner_frob_color, Simple.ideals, ad_jacobi, ad_skew, minusKilling_invariant, invariant_scalar, minusKilling_sq, trace_adjoint, casimir_pos.
Lemma I.1.s (Haar mass of small balls). Proposition I.1.7. Proof sketch. Points of \(G\) lie on the Frobenius sphere of radius \(\sqrt{N}\) in \(M_N (\mathbb{C}) \cong \mathbb{R}^{2 N^2}\). Take a finite set \(p \subset G\) whose distinct points have Frobenius distance strictly greater than \(r \sqrt{N}\):
the balls of radius \(r \sqrt{N} / 2\) around its points are disjoint and lie in the ball of radius \(2 \sqrt{N}\), so \(|p| (r / 4)^{2 N^2} \leqslant 1\); in particular a maximal such set exists;
if \(p\) is maximal, then \(G = \bigcup_{i \in p} i \cdot \{\chi \leqslant r\}\), and left invariance gives \(1 \leqslant |p| \hspace{0.17em} \mu_G \{\chi \leqslant r\}\).
Lean: shell_ge.
Lemma I.1.t (local logarithm). Proposition I.1.8(a). Proof sketch.
\(\Phi (c, \nu) = e^{X_c} e^{\nu}\) on \(\mathbb{R}^d \times \mathfrak{g}^{\perp}\) is analytic, with \(D \Phi (0) (c, \nu) = X_c + \nu\) invertible, so it has a local analytic inverse.
Isolation: near \(0\), \(e^{\nu} \in G\) with \(\nu \in \mathfrak{g}^{\perp}\) forces \(\nu = 0\). Otherwise take \(\nu_k \to 0\), \(\nu_k \neq 0\), \(e^{\nu_k} \in G\), with \(\nu_k / \| \nu_k \|_F \to Y\) along a subsequence. For \(t \geqslant 0\), \(e^{tY} = \lim (e^{\nu_k})^{\lfloor t /\| \nu_k \|_F \rfloor} \in G\) by closedness; inverses handle \(t < 0\). So \(Y \in \mathfrak{g} \cap \mathfrak{g}^{\perp} = 0\), contradicting \(\|Y\|_F = 1\).
For \(u \in G\) near \(1\), write \(u = \Phi (c, \nu)\); then \(e^{\nu} = e^{- X_c} u \in G\) forces \(\nu = 0\).
Lean: direction_in_lie, isolates, splitDeriv, plog_diff, plog_left, plog_right.
Lemma I.1.u (nearest point). Proposition I.1.8(b). Proof sketch.
Existence by compactness; equivariance because left and right unitary multiplication are Frobenius isometries.
First-order condition: if \(1\) is nearest to \(X\), then \(t \mapsto \|e^{tY} - X\|_F^2\) is minimal at \(0\), so \(\langle X - 1, Y \rangle = 0\) for \(Y \in \mathfrak{g}\). The general case follows by left translation.
Identification near \(1\): for any choice \(b (X)\) of nearest points, \(\|b (X) - 1\|_F \leqslant 2 \|X - 1\|_F\). Hence \(\nu (X) = b (X)^{\ast} X - 1 \in \mathfrak{g}^{\perp}\) tends to \(0\), and \(X = \Psi (\log_G b (X), \nu (X))\) with both arguments near \(0\). Injectivity of the local diffeomorphism \(\Psi\) gives \(b (X) = \text{foc} (X)\). To obtain one neighbourhood valid for every nearest point, choose \(b (X)\) to differ from \(\text{foc} (X)\) whenever such a nearest point exists. The same argument forces this selection to equal \(\text{foc}\) on a neighbourhood, excluding all such exceptions there.
Uniqueness in a tube: translate by \(v^{- 1}\).
The derivative of \(\text{foc}\) at \(1\) is \(X_{(\cdot)} \circ \text{pr}_1 \circ D \Psi (0)^{- 1} = P_{\mathfrak{g}}\), and the approximation bound is the strict differentiability of \(\text{foc}\) at \(1\).
Lean: nearest_exists, nearest_left, nearest_normal, tubular, tubularDeriv, nearest_is_focus, nearest_tube, focus_deriv, nearest_approximate.
Lemma I.1.v (curvature of a group element). Proposition I.1.8(c). Proof sketch. Conjugation by \(v \in G\) is a Frobenius isometry preserving \(\mathfrak{g}\), hence also \(\mathfrak{g}^{\perp}\), so \(\Pi (vAv^{- 1}) = \text{Ad} (v) \Pi A\). For the bound:
The derivative of \(\frac{1}{2} (e^{tX_c} - e^{- tX_c})\) at \(0\) is \(X_c\), and \(\Pi X_c = c\). Lean: project_conjugate, curvature_conjugate, curvature_norm, leading_curvature.
Lemma I.1.w (the law). Proposition I.1.9. Proof sketch. \(E_{\ell}\) is a finite sum of continuous functions on a compact space. There are \(6 \ell^4\) terms, each in \([0, 2]\), giving the energy bounds. The weight \(e^{- \beta E_{\ell}}\) is continuous and strictly positive, so it is integrable and \(Z_{\ell} > 0\). The density \(e^{- \beta E_{\ell}} / Z_{\ell}\) has integral \(1\). Continuous complex functions are bounded by compactness, and bounded measurable functions are integrable against a probability measure. The moment formula follows by integration against the density. Lean: bare_integrable, partition_pos, moment_formula.
Lemma I.1.x (invariance of Haar measure, energy and law). Proposition I.1.10. Proof sketch.
Haar. \(A^{\ast}\) is the composite of two maps. The first is the relabelling \(U \mapsto U \circ e_A\), where \(e_A\) maps the link \((x, i)\) to the positively oriented link carrying its image; this is injective, hence a permutation of the finite link set. The second inverts the variables on links with \(\epsilon_i = - 1\). Permutations of factors preserve a product measure, and inversion preserves \(\mu_G\). A gauge transformation acts on each factor by \(u \mapsto g (x) ug (x + e_i)^{- 1}\), which preserves \(\mu_G\).
Energy. Gauge: plaquette holonomies are conjugated. Pullback: by I.1.m the energy is re-summed over the bijection $A_\sharp $.
Law. A map preserving \(\mu_{\ell}\) and \(E_{\ell}\) preserves the density, hence the law.
Lean: sendEdge, sendEdge_injective, prior_pull, prior_gauge, face_gauge, energy_pull, law_preserved, moment_pull.
Lemma I.1.y (Euclidean group and configurations). Proposition I.1.11(a),(e). Proof sketch. The matrix of \(\rho\) has entry \(\epsilon_a\) at \((\pi (a), a)\); it is orthogonal and multiplies like \(B_4\). The exponential of a skew matrix is orthogonal. The set \(\text{Conf}_n\) is a finite intersection of open sets \(\{x_i \neq x_j \}\). Distinct time coordinates imply distinct points, as does a positive lower bound on pairwise sup-distances. Splitting off one Euclidean four-dimensional factor at a time identifies Lebesgue measure on \((\mathbb{R}^4)^n\) with the product measure. Lean: Signed.ortho_move, Signed.inclusionO, generic_conf, spaced_conf, conf_open.
Lemma I.1.z (Lorentz group, boosts, cone, Poincaré group). Proposition I.1.11(b)–(d). Proof sketch.
The three defining conditions of \(\mathcal{L}\) are stable under products and inverses, and \(\det = 1\) gives Lebesgue invariance.
Boosts: the addition law \(B_e (r) B_e (r') = B_e (r + r')\) follows from the addition formulas for \(\cosh, \sinh\), and \(\textsf{m}\) is preserved by \(\cosh^2 - \sinh^2 = 1\). For the determinant: \(RB_e (r) R = B_e (- r)\) gives \(\det B_e (r) = \det B_e (- r) = \det B_e (r)^{- 1}\), and \(\det B_e (r) = \det B_e (r / 2)^2 > 0\), so \(\det = 1\). The cone is preserved by \(| \sinh r| \leqslant \cosh r\). Differentiating the boost formula at \(r = 0\) gives \(K_e\); continuity follows from \(\cosh\) and \(\sinh\).
Dual action: the identity \(\langle gx, g^{\vee} p \rangle = \langle x, p \rangle\) is polarization of \(\textsf{m} (x) = \langle x, Rx \rangle\).
Cone duality: Cauchy–Schwarz gives \(\langle x, p \rangle \geqslant x_t p_t - | \vec{x} | \hspace{0.17em} | \vec{p} | \geqslant 0\) for \(x, p \in \bar{V}_+\). Conversely, test against \(p = (- \vec{x} / | \vec{x} |, 1)\) when \(\vec{x} \ne 0\), and against \((0, 1)\) otherwise. If \(x \in V_+\) and \(p \in \bar{V}_+ \setminus \{0\}\), then \(p_t > 0\) makes the bound strict.
Half-space characterization: for \(x \in \bar{V}_+\), every \(g^{\vee} x\) remains in that cone and has nonnegative time coordinate. If \(x_t < 0\), take \(g = 1\). Otherwise, if \(x \not\in \bar{V}_+\), put \(e = \vec{x} / | \vec{x} |\), choose \(x_t / | \vec{x} | < s < 1\), and set \(r = \text{artanh} \hspace{0.17em} s > 0\) and \(g = B_e (r)\). Then \(g^{\vee} = B_e (- r)\) and \((g^{\vee} x)_t = \cosh r \hspace{0.17em} (x_t - s| \vec{x} |) < 0\).
The open cone is characterized by \(x_t > 0\) and \(\textsf{m} (x) > 0\), and is preserved by the cone and quadratic-form identities. A Euclidean rotation fixing time preserves \(| \vec{x} |\) and hence both cones and \(\textsf{m}\); determinant \(1\) places it in \(\mathcal{L}\). Affine composition gives the Poincaré group law. Evaluation at \(0\) recovers the translation, and subtracting it recovers the Lorentz map. Joint continuity follows from linear evaluation; the diagonal measure statement follows from preservation of each factor.
Lean: Ray.adds, Ray.positive, Ray.staysM, Lorentz.dual_sent, cone_test, Lorentz.from_half, Poincare.preserving, Poincare.together.
The local logarithm and nearest-point retraction require a closed-subgroup argument that also handles disconnected groups. The coarsening estimate must control arbitrary periodic sets uniformly in the torus side. The invariant-form argument uses simplicity over \(\mathbb{R}\); irreducibility of the defining representation is not assumed. Invariance of the Wilson law reduces to Haar invariance, but reflections require the correct shifts of link and plaquette base points.
The representation \(G \subset U (N)\) fixes the cost and the colour inner product, and hence \(C_{\mathfrak{g}}\) and \(\kappa\). Each positive-axis plaquette is counted once. The later running-coupling estimate says that the ratio of the level-\(m\) coupling to \(s_m\) tends to \(\kappa\), uniformly in the ultraviolet cutoff \(N' \geqslant m\) and the volume index \(p\), for a family satisfying the construction conditions (zeroUp_bound). Its normalization is \(\kappa = 2 b_0\), where \(b_0 = 11 C_{\mathfrak{g}} / (48 \pi^2)\) (b0_normalizes).
Even torus sides create tie points in the centred lift. Absolute values, and hence separation, remain invariant under reflections. Placements commute with signed permutations on the strict interior of the fundamental box. Odd subdivision ratios remove this ambiguity from the residue window, which gives equivariance of the parent map.
| object | code name | proposed notation |
| torus, side, cover | Site, Side, Cover | \(\Lambda_{\ell} = (\mathbb{Z}/ \ell \mathbb{Z})^4\), \(\ell\), \(\mathbb{Z}^4\) |
| reduction, centred lift | wrap, lift | \([y]_{\ell}\), \(r_{\ell} (x)\) |
| torus norm, separation | pdist, separation | \(\lvert x \rvert_{\ell}\), \(\lvert x - y \rvert_{\ell}\) |
| signed permutations, affine maps | Signed, affine | \(\rho = (\pi, \epsilon) \in B_4\), \(A = (\rho, b)\) |
| step, walk, transport | Step, Walk, travel | \(\pm e_i\), \(w\), \(U (x ; w)\) |
| configuration | Links, Wilson.Config | \(U\) |
| pullback, translation | pull, translate | \(A^{\ast} U\), \(\tau_b U\) |
| gauge transformation; invariance; locality | gauge, GaugeInvariant, LocalOn | \(U^g\) |
| plaquette, holonomy, image plaquette | Face, faceValue, Face.send | \(p = (x ; i < j)\), \(U_p\), \(A_{\sharp} p\) |
| periodic extension | expandAt | \(E_b U\) |
| box, axis route, slab | box, route, slab | \(B_r\), \(\text{route}_o (x, y)\), \(\text{slab} (r, a, J)\) |
| schedule: block factors, periods | Schedule | \((L_m)\), \((P_p)\); the statement writes \(L_p\) for \(P_p\) and \(b_k\) for \(L_k\) |
| denominator, spacing, side, log-scale | denominator, a, side, scale | \(D_m\), \(a_m\), \(\ell (m, p)\), \(s_m = \log (1 / a_m)\) |
| placement | place | \(a_m \hspace{0.17em} r_{\ell} (x)\) |
| subdivision | Ratio, embed, residue, parent | ratio \(L\), \(\iota\), \(\text{res}\), \(\text{par}\) |
| halo, cover number | halo, coverNumber | \(\text{halo}_M (X)\), \(\text{cov}_M (X)\) (as in III.5) |
| gauge group | Gauge, El, mat | \(G \subset U (N)\), \(N = n + 1\) (the statement writes \(m\)) |
| plaquette cost, chord | price, chord | \(\sigma (u) = 1 - \frac{1}{N} \text{Re} \hspace{0.17em} \text{tr} \hspace{0.17em} u\), \(\chi (u)\) |
| Haar probability | probHaar | \(\mu_G\) |
| Lie algebra, colours, projection | lie, Color, colorMat, projectColor | \(\mathfrak{g} \cong \mathbb{R}^d\), \(c \mapsto X_c\), \(\Pi\); \(X_{\Pi A} = P_{\mathfrak{g}} A\) |
| adjoint action, bracket, exponential | adjoint, ad, lieExp | \(\text{Ad} (u)\), \(\text{ad} (c)\), \(\exp c\) |
| simplicity | Simple | \(G\) simple |
| Casimir, slope constant | casimir, slopeConstant | \(C_{\mathfrak{g}}\), \(\kappa = 11 C_{\mathfrak{g}} / (24 \pi^2)\) (\(= b_{\ast} = 2 b_0\) of the statement) |
| local logarithm, nearest point | plog, Nearest, focus | \(\log_G\), nearest point, \(\text{foc} (X)\) |
| curvature of a group element | curvature | \(F (u)\) |
| Wilson energy, weight, partition function | energy, bare, partition | \(E_{\ell} (U)\), \(e^{- \beta E_{\ell}}\), \(Z_{\ell} (\beta)\) |
| product Haar, Wilson law, expectation | prior, law, moment | \(\mu_{\ell}\), \(\mathbb{P}_{\ell, \beta}\), \(\langle \cdot \rangle_{\ell, \beta}\) |
| spacetime, time, space, sup-distance | E4, time, spatial, cubedist | \(\mathbb{R}^4\), \(x_t\), \(\vec{x}\), \(\lvert x - y \rvert_{\infty}\) |
| \(n\)-point configurations | Points, conf, generic, spaced | \((\mathbb{R}^4)^n\), \(\text{Conf}_n\), generic, \(r\)-spaced |
| orthogonal group, rotation generators | Ortho, skewSpace, plane | \(O (4)\), \(\mathfrak{so} (4)\), \(\Omega_{ij}\) |
| Minkowski square, cones | m, cone, inside | \(\textsf{m} (x)\), \(\bar{V}_+\), \(V_+\) |
| Lorentz group, dual, boost, Poincaré group | Mink.Lorentz, dual, Ray.boost, Poincare | \(\mathcal{L}= SO^+ (1, 3)\), \(g^{\vee}\), \(B_e (r)\), \(\mathcal{P}\) |
Additional helper lemmas. The assignment of declarations to I.1 includes helpers outside the six principal folders, including in Cells, Extra, Further and Next. Their statements and proofs are not covered here; their elementary character should not be assumed.
Coarsening along the selected schedule. Direct consumers retain \(M \leqslant 2 B\) explicitly, for example YM.Next.Scheduling.Buy.footage, or include it in a conjunction, as in YM.Extra.Iteration.Shapes.raw. The derivation of this condition for the particular consecutive cover scales used in III.5 and III.10 remains to be traced through their schedule hypotheses.
Bibliographic precision. The references in §2 give context. Exact theorem locations and agreement of their conventions with the statements here remain to be checked.