12.18.16 · quantum / gauge-and-symmetry

Lattice gauge theory and confinement (QFT pointer)

shipped3 tiersLean: none

Anchor (Master): Weinberg, The Quantum Theory of Fields, Vol. II (1996), §15.9; Wilson, Phys. Rev. D 10, 2445 (1974); Kogut, Rev. Mod. Phys. 51, 659 (1979); Coleman, Aspects of Symmetry (1985), Ch. 5 §3 (1/N and confinement); Montvay-Munster, Quantum Fields on a Lattice (1994)

Intuition Beginner

Quarks are never seen on their own. Smash protons together at the highest energies we can reach, and out come more bound clumps — never a single free quark drifting away. The strong force does something a force like gravity or electricity never does: it does not fade with distance. Pull two quarks apart and the energy stored between them keeps growing, as if they were joined by an unbreakable rubber band. This is confinement, and it is the defining puzzle of the strong interaction.

Why is it hard? Because the usual calculation tools assume the force is weak enough to treat as a small correction. At short range the strong force really is weak, which is the companion fact that quarks behave almost freely when very close. But confinement lives at long range, where the force is strong and the small-correction method breaks down. We need a way to compute that does not assume weakness.

Kenneth Wilson's answer was to replace continuous space with a grid of points, a lattice, and put the gauge field on the links between neighbouring points. On a grid every quantity becomes a finite sum, the strong-force regime becomes a high-temperature regime of a model you can expand by hand, and confinement turns into a sharp, checkable statement about a loop drawn on the grid.

Visual Beginner

A cube of lattice points with little arrows on each link joining neighbours, and one highlighted rectangular loop traced through the links. The loop is the path a quark would take out to a distance, across, and back. The size of the flat region the loop encloses — its area — is the quantity that controls how hard it is to separate the quark pair.

The picture captures the whole idea. The links carry the gauge field. The loop carries a quark-antiquark pair around a rectangle. When the loop's response is governed by its area rather than its perimeter, the pair feels a force that never lets go, and that is confinement.

Worked example Beginner

Take the simplest possible gauge group, the one used by electromagnetism, where each link carries an angle and the field is a single number. Suppose we measure how a small square loop responds in two different regimes and read off the rule.

Step 1. Draw the smallest loop, a single square (a plaquette) with side length one grid spacing. Its perimeter is $4$ and its enclosed area is $1$ .

Step 2. In the strongly coupled regime the loop's value shrinks by a fixed factor for every unit of enclosed area. Call that factor $f = 0.1$ . A loop enclosing area $1$ then has value $0.1$ . A loop enclosing area $4$ has value $0.1 \times 0.1 \times 0.1 \times 0.1 = 0.0001$ .

Step 3. Compare with a rule that shrinks by a factor for every unit of perimeter instead. A loop of perimeter $4$ would give one factor, a loop of perimeter $8$ a larger but slower-growing penalty. Area grows faster than perimeter as a loop gets big.

Step 4. Convert the area rule into energy. The loop value behaving as $f$ raised to the area means the energy of separating the quark pair grows in proportion to the distance between them. A force whose stored energy grows with distance is a constant pull that never weakens.

What this tells us: the area rule and a constant confining pull are the same statement. Reading whether a loop responds to its area or its perimeter is how we decide, on the grid, whether a theory confines.

Check your understanding Beginner

Formal definition Intermediate+

Fix a compact gauge group $G$ (the cases of interest are $SU (N)$ and $U (1)$ ) and a four-dimensional hypercubic lattice $Λ = a Z^{4}$ with spacing $a$ . The dynamical variables are link variables $U_{ℓ} = U_{x, μ} \in G$ , one for each oriented nearest-neighbour link $ℓ$ from a site $x$ to $x + a \overset{μ}{^}$ , with the reversed link carrying $U_{x, μ}^{- 1}$ . A link variable is the lattice stand-in for the parallel transport $P exp (i \int A)$ of the continuum connection along that edge.

A gauge transformation is an assignment $g_{x} \in G$ of a group element to each site, acting by $$ U_{x,\mu} \longmapsto g_x, U_{x,\mu}, g_{x+a\hat\mu}^{-1}. $$ The smallest gauge-invariant object is the plaquette, the ordered product of the four links around an elementary square in the $μν$ plane: $$ U_{\square} = U_{x,\mu}, U_{x+a\hat\mu,\nu}, U_{x+a\hat\nu,\mu}^{-1}, U_{x,\nu}^{-1}. $$ Under a gauge transformation $U_{□}$ conjugates, $U_{□} \mapsto g_{x} U_{□} g_{x}^{- 1}$ , so its trace is invariant. The Wilson action is $$ S[U] = \beta \sum_{\square}\Big(1 - \tfrac{1}{N}\operatorname{Re}\operatorname{tr} U_{\square}\Big), \qquad \beta = \frac{2N}{g^2}, $$ summed over all unoriented plaquettes, with $g$ the bare coupling. Expanding $U_{□} = exp (i a^{2} F_{μν} + \dots)$ for small $a$ recovers $S \to \frac{1}{2 g ^{2}} \int tr F_{μν} F^{μν} d^{4} x$ , the Yang-Mills action of 03.07.05.

The partition function and expectation values use the product Haar measure $d μ (U) = \prod_{ℓ} d U_{ℓ}$ , which is finite because $G$ is compact and gauge-invariant because Haar measure is bi-invariant: $$ Z = \int \prod_\ell \mathrm{d}U_\ell; e^{-S[U]}, \qquad \langle \mathcal{O}\rangle = \frac{1}{Z}\int \prod_\ell \mathrm{d}U_\ell; \mathcal{O}[U], e^{-S[U]}. $$ No gauge fixing and no Faddeev-Popov ghosts are needed: the measure is finite per orbit, in contrast to the continuum, where the gauge volume diverges (cross-reference 03.07.31).

The central observable is the Wilson loop: for a closed lattice loop $C$ , $$ W(C) = \Big\langle \operatorname{tr} \prod_{\ell \in C} U_\ell \Big\rangle. $$ A theory confines static quarks when, for large rectangular loops $C$ of spatial extent $R$ and temporal extent $T$ , $W (C)$ obeys the area law $$ W(C) \sim e^{-\sigma, R, T}, $$ with the string tension $σ > 0$ . The static quark-antiquark potential extracted from $W (R \times T) \sim e^{- V (R) T}$ is then $V (R) = σ R$ , a linearly rising potential. The non-confining alternative is the perimeter law $W (C) \sim e^{- c (R + T)}$ , giving a potential that flattens to a constant. Elitzur's theorem (Phys. Rev. D 12, 3978, 1975) shows that gauge-non-invariant order parameters average to zero, which is why the gauge-invariant Wilson loop, not a local field expectation, is the order parameter for confinement.

Counterexamples to common slips

The area law is a strong-coupling (large $g^{2}$ , small $β$ ) statement. In four dimensions $SU (N)$ shows no bulk phase transition separating strong and weak coupling (Creutz 1980), so the strong-coupling area law connects continuously to the continuum confining phase. Compact $U (1)$ is different: it confines at strong coupling but has a genuine phase transition to a non-confining Coulomb phase at weak coupling (Guth, Banks-Myerson-Kogut), matching the fact that continuum QED does not confine.
A perimeter contribution always coexists with the area term; it is a self-energy of the static sources and can be subtracted. Confinement is the statement that an area term survives after the perimeter piece is removed, so $σ > 0$ strictly.
The Wilson-loop criterion detects confinement only without dynamical light quarks. With light quarks the flux tube breaks by pair creation at large $R$ , the potential flattens, and the area law is replaced by string breaking. The pure-gauge area law is the clean theoretical statement; the full theory needs the heavy-quark approximation for the criterion to apply.

Key theorem with proof Intermediate+

Theorem (strong-coupling area law; Wilson 1974). Let $G = SU (N)$ with the Wilson action on $a Z^{4}$ . For the coupling small enough (equivalently $β = 2 N / g^{2}$ small enough), the Wilson loop $W (C)$ of a planar rectangular loop $C$ enclosing area $A (C)$ measured in plaquettes satisfies an area law $$ W(C) = e^{-\sigma A(C),(1 + o(1))}, \qquad \sigma = -\log!\Big(\frac{\beta}{2N^2}\Big) + O(\beta), $$ so the string tension $σ > 0$ and the static potential is linear, $V (R) = σ R / a$ .

Proof. Work to leading order in $β$ via the character expansion of the Boltzmann weight on each plaquette. Writing the per-plaquette weight as $e^{- β (1 - \frac{1}{N} Re tr U_{□})}$ and using the Peter-Weyl decomposition of a class function on $G$ , $$ e^{\frac{\beta}{N}\operatorname{Re}\operatorname{tr} U_{\square}} = \sum_{r} d_r, c_r(\beta), \chi_r(U_{\square}), $$ where the sum runs over irreducible representations $r$ , $χ_{r}$ is the character, $d_{r} = dim r$ , and the coefficients are $c_{r} (β) = \frac{1}{d _{r}} \int_{G} χ_{r} (U) e^{\frac{β}{N} Re tr U} d U$ . The identity representation gives $c_{0} = 1 + O (β^{2})$ , and the fundamental representation gives $c_{F} (β) = \frac{β}{2 N ^{2}} + O (β^{3})$ ; every other coefficient is higher order in $β$ .

Insert this expansion on every plaquette of the lattice and integrate over all link variables using the orthogonality relations of Peter-Weyl: $$ \int_G \chi_r(AU),\chi_{r'}(U^{-1}B),\mathrm{d}U = \frac{\delta_{r r'}}{d_r},\chi_r(AB), \qquad \int_G \chi_r(AU B U^{-1}),\mathrm{d}U = \frac{\chi_r(A)\chi_r(B)}{d_r}. $$ A link variable not lying on the loop $C$ is integrated freely. Each such integral forces the representations on the plaquettes sharing that link to match and, for a link touched by the identity representation on one side, sets the others to the identity as well. The only configurations that survive the integration and produce a nonzero contribution to the fundamental-representation Wilson loop are those tiling the enclosed surface of $C$ with plaquettes each carrying the fundamental representation, so that the representation lines from the loop are screened by a connected sheet of excited plaquettes.

The minimal such tiling is the flat rectangle bounded by $C$ , with $A (C)$ plaquettes, each contributing a factor $c_{F} (β) = \frac{β}{2 N ^{2}}$ . The link integrations on the interior collapse the product to a single overall character of the identity, normalised against $Z$ . Hence $$ W(C) = \Big(\frac{\beta}{2N^2}\Big)^{A(C)} \big(1 + O(\beta)\big) = e^{-\sigma A(C)}\big(1 + o(1)\big), \qquad \sigma = -\log\frac{\beta}{2N^2} + O(\beta). $$ For $β$ small, $σ > 0$ . Non-minimal tilings (bumpy surfaces) carry extra plaquette factors and are suppressed by additional powers of $β$ , so the minimal-area surface dominates and the area law holds. $□$

Bridge. This proof builds toward the modern understanding of confinement as a property of the strong-coupling vacuum, and the same character-expansion bookkeeping appears again in 08.08.02 on the stat-mech side, where the plaquette weight is read as a Boltzmann factor of a spin-like model. The foundational reason the area law emerges is that the cheapest way to screen the representation flux carried around the loop is to fill in a surface of excited plaquettes, and the cheapest surface is the minimal one — this is exactly the lattice incarnation of a confining flux tube of fixed tension. The central insight is that confinement on the lattice is a statement of surface combinatorics, and it identifies the string tension with the per-plaquette suppression factor; putting these together, the area law and a linearly rising potential are dual to one another, so the strong-coupling expansion proves both at once. The bridge is the continuity, established numerically by Creutz, between this strong-coupling regime and the asymptotically free continuum limit governed by 12.18.03.

Exercises Intermediate+

Exercise 4 (medium, short-answer).

Show that the Wilson action reduces to the continuum Yang-Mills action as $a \to 0$ , to leading order in the lattice spacing.

Hint

Write $U_{x, μ} = exp (ia A_{μ} (x))$ , expand the plaquette using the Baker-Campbell-Hausdorff formula, and keep the leading non-vanishing order in $a$ .

Answer

Write each link as $U_{x, μ} = exp (ia A_{μ} (x))$ for a Lie-algebra-valued field $A_{μ}$ . The plaquette product is $U_{□} = exp (i a^{2} F_{μν} + O (a^{3}))$ with $F_{μν} = \partial_{μ} A_{ν} - \partial_{ν} A_{μ} + i [A_{μ}, A_{ν}]$ , the Baker-Campbell-Hausdorff commutators reproducing exactly the non-Abelian field strength. Then $Re tr U_{□} = N - \frac{1}{2} a^{4} tr (F_{μν} F_{μν}) + O (a^{6})$ , so $1 - \frac{1}{N} Re tr U_{□} = \frac{a ^{4}}{2 N} tr (F_{μν} F_{μν})$ . Summing over plaquettes, $\sum_{□} \to \frac{1}{2 a ^{4}} \int d^{4} x \sum_{μν}$ , and with $β = 2 N / g^{2}$ the action becomes $S \to \frac{1}{2 g ^{2}} \int d^{4} x tr (F_{μν} F_{μν})$ , the Yang-Mills action of 03.07.05. Rubric: full credit for the field-strength identification plus the correct continuum normalisation.

Exercise 5 (medium, short-answer).

Explain why the lattice path integral needs no gauge fixing, whereas the continuum path integral does.

Hint

Compare the volume of a single gauge orbit in the two cases: a product of compact-group Haar measures versus an infinite-dimensional non-compact gauge volume.

Answer

On the lattice each link variable lives in the compact group $G$ , so the gauge orbit through any configuration is a product of finitely many copies of $G$ , of finite Haar volume. The integrand $e^{- S}$ is gauge invariant, so integrating over the orbit just multiplies $Z$ by this finite, configuration-independent volume, which cancels in every expectation value. In the continuum the gauge group is the infinite-dimensional non-compact group of $G$ -valued functions on spacetime; its volume diverges and does not cancel, forcing the Faddeev-Popov procedure of 03.07.31 to factor it out. The lattice regularises the gauge volume to a harmless finite constant. Rubric: full credit for the compact-orbit / finite-Haar-volume argument plus the contrast with the divergent continuum gauge volume.

Exercise 7 (hard, short-answer).

State Elitzur's theorem and explain why it forces the order parameter for confinement to be the Wilson loop rather than a local field expectation.

Hint

A local gauge transformation at a single site costs no action in the unfixed measure; average a gauge-variant quantity over that local symmetry.

Answer

Elitzur's theorem (Phys. Rev. D 12, 3978, 1975) states that in a lattice gauge theory with the gauge-invariant Haar measure and no gauge fixing, the expectation value of any gauge-non-invariant local operator vanishes, even in the presence of a symmetry-breaking perturbation that is then removed. The reason: a gauge transformation supported at a single site leaves the action invariant but rotates a local field through the whole group; averaging over that exact local symmetry sends the field expectation to zero. Consequently a local order parameter analogous to a magnetisation cannot distinguish phases. The confined and deconfined phases must instead be told apart by a gauge-invariant nonlocal observable, and the Wilson loop is the canonical choice: its area-versus-perimeter behaviour is a genuine, gauge-invariant diagnostic. Rubric: full credit for the statement plus the local-averaging argument plus the conclusion that a nonlocal gauge-invariant order parameter is required.

Exercise 8 (hard, short-answer).

Sketch the Kogut-Susskind Hamiltonian lattice gauge theory and identify the strong-coupling vacuum and the origin of the string tension in that picture.

Hint

Take the continuous-time limit so links carry electric-field operators conjugate to the group variables; the Hamiltonian is a sum of an electric (Casimir) term and a magnetic (plaquette) term.

Answer

In the Hamiltonian formulation (Kogut-Susskind, Phys. Rev. D 11, 395, 1975) time stays continuous and only space is discretised. Each spatial link carries a group-element operator $U_{ℓ}$ and a conjugate left/right electric-field operator $E_{ℓ}^{a}$ generating the group action, with the Hamiltonian $$ H = \frac{g^2}{2a}\sum_\ell E^a_\ell E^a_\ell - \frac{1}{a g^2}\sum_\square \operatorname{Re}\operatorname{tr} U_\square. $$ At strong coupling ( $g^{2}$ large) the electric term dominates; its ground state is the unique state annihilated by every $E_{ℓ}^{a}$ , the strong-coupling vacuum with no electric flux. A static quark-antiquark pair forces a unit of electric flux along a connected path of links joining them, and each excited link costs energy $\propto g^{2} / a$ . The minimal-energy flux configuration is a single string of links along the shortest path, of length proportional to the separation $R$ , giving energy $V (R) \propto (g^{2} / a) R$ : a linearly confining potential whose slope is the string tension. The magnetic plaquette term, treated perturbatively, gives the flux tube transverse fluctuations and corrections to $σ$ . Rubric: full credit for the electric/magnetic Hamiltonian, the flux-free strong-coupling vacuum, and the flux-string origin of the linear potential.

Advanced results Master

Theorem (Osterwalder-Seiler reflection positivity and the transfer matrix). The lattice gauge theory with the Wilson action and Haar measure satisfies reflection positivity with respect to reflection in a lattice hyperplane. Consequently there is a positive self-adjoint transfer matrix $T$ on a physical Hilbert space $H$ , a Hamiltonian $H = - a^{- 1} lo g T \geq 0$ , and Euclidean correlators analytically continue to a relativistic quantum theory. (Osterwalder-Seiler, Ann. Phys. 110, 440, 1978.)

Reflection positivity is the lattice form of the Osterwalder-Schrader axiom that yields a genuine Hilbert space and a positive Hamiltonian. It is what makes the area law a statement about a real spectrum: the string tension is the energy per unit length of the lowest flux state, and reflection positivity guarantees this energy is real and bounded below. The construction also legitimises the strong-coupling expansion as an asymptotic expansion of true expectation values, not a formal series.

Theorem (convergence of the strong-coupling expansion). For $β$ below a model-dependent radius of convergence, the character expansion of $lo g Z$ and of $W (C)$ converges absolutely, and the area law $W (C) = e^{- σ A (C) (1 + o (1))}$ holds with an analytic string tension $σ (β) = - lo g (β /2 N^{2}) + O (β)$ . The convergence follows from a polymer (cluster) expansion: the excited plaquettes organise into connected clusters whose weights are summable by a Kotecký-Preiss criterion for $β$ small. The cluster bound also controls the corrections to the leading minimal-surface result, establishing that the string tension is analytic in $β$ throughout the strong-coupling phase.

Theorem (absence of a bulk transition for $SU (2)$ and $SU (3)$ ; Creutz 1980). Monte Carlo simulation of the Wilson action for $SU (2)$ and $SU (3)$ shows no bulk (first- or second-order) phase transition separating the strong-coupling region from the weak-coupling region; the average plaquette interpolates smoothly. This is the decisive evidence that the strong-coupling area law and the weak-coupling continuum limit lie in one analytically connected phase, so confinement proved at strong coupling persists to the asymptotically free continuum theory. Creutz further extracted a string tension whose dependence on $β$ matched the two-loop asymptotic-freedom scaling $a σ \propto exp (- \frac{1}{2 b _{0} g ^{2}})$ of 12.18.03, the first quantitative confinement-scaling test.

Theorem (Coulomb phase of compact $U (1)$ ). Compact lattice $U (1)$ in four dimensions confines at strong coupling (area law from the same character expansion) but undergoes a phase transition at a finite $β_{c}$ to a weak-coupling Coulomb phase with a perimeter law and a massless photon. The mechanism is a condensation of magnetic monopoles in the strong-coupling phase that disorders the Wilson loop (Banks-Myerson-Kogut 1977; Polyakov's three-dimensional confinement mechanism is the lower-dimensional cousin). The contrast with $SU (N)$ , which has no such transition, is exactly the statement that continuum QED does not confine while QCD does.

Theorem (large- $N$ factorisation and the master field). In the 't Hooft limit $N \to \infty$ with $λ = g^{2} N$ fixed, connected correlators of gauge-invariant operators are suppressed by powers of $1/ N^{2}$ , so Wilson loops factorise, $⟨ W (C_{1}) W (C_{2})⟩ = ⟨ W (C_{1})⟩ ⟨ W (C_{2})⟩ + O (1/ N^{2})$ . Factorisation means the large- $N$ theory behaves classically in the space of gauge-invariant observables, governed by a master field, and the loop equations (Makeenko-Migdal) close on the single-loop expectation. This is the lattice/loop-space face of the planar expansion and the structural origin of the gauge-string correspondence.

Synthesis. Lattice gauge theory is the foundational reason confinement can be stated and proved rather than only conjectured: discretising spacetime turns the strong-coupling regime into a convergent character expansion, and the area law of the Wilson loop is exactly the surface-combinatorics statement that the cheapest screening of a quark pair is a flux tube of fixed tension. The central insight is that confinement is a property of the gauge-invariant Wilson loop, forced to be the order parameter by Elitzur's theorem, and the strong-coupling area law identifies the string tension with the per-plaquette suppression factor. Putting these together, reflection positivity makes the construction a genuine quantum theory with a positive Hamiltonian, the cluster expansion makes the area law analytic in the coupling, and Creutz's numerics show this strong-coupling phase is dual to — analytically continuous with — the asymptotically free continuum governed by 12.18.03. The same loop observable generalises: the large- $N$ factorisation identifies Wilson loops with a classical master field, the bridge to the gauge-string correspondence, while the compact- $U (1)$ Coulomb transition shows that the area law is a genuine dynamical distinction, not a kinematic accident, since it can fail. Confinement, asymptotic freedom, and the continuum limit are three faces of one renormalisation-group flow, and the lattice is the only known framework in which all three are simultaneously under control.

Full proof set Master

Proposition (area law from the leading character expansion). For $G = SU (N)$ with the Wilson action, the leading strong-coupling contribution to a planar rectangular Wilson loop $C$ of area $A (C)$ plaquettes is $W (C) = (β /2 N^{2})^{A (C)} (1 + O (β))$ , yielding the area law with string tension $σ = - lo g (β /2 N^{2}) + O (β) > 0$ for $β$ small.

Proof. Expand each plaquette weight in characters of $G$ : $$ \exp!\Big(\tfrac{\beta}{N}\operatorname{Re}\operatorname{tr} U_\square\Big) = \sum_r d_r, c_r(\beta), \chi_r(U_\square), \qquad c_r(\beta) = \frac{1}{d_r}\int_G \chi_r(U),\exp!\Big(\tfrac{\beta}{N}\operatorname{Re}\operatorname{tr} U\Big)\mathrm{d}U. $$ Expanding the exponential to first order, $\int_{G} χ_{r} (U) \frac{β}{N} Re tr U d U$ is nonzero only for $r$ the fundamental (or antifundamental) by Schur orthogonality, since $tr U = χ_{F} (U)$ . This gives $c_{0} = 1 + O (β^{2})$ , $c_{F} = \frac{β}{2 N ^{2}} + O (β^{3})$ , and $c_{r} = O (β^{∣ r ∣})$ with $∣ r ∣ \geq 2$ for higher representations.

The fundamental Wilson loop is $W (C) = Z^{- 1} \int \prod_{ℓ} d U_{ℓ} χ_{F} (\prod_{ℓ \in C} U_{ℓ}) \prod_{□} (\sum_{r} d_{r} c_{r} χ_{r} (U_{□}))$ . Integrate link by link using $$ \int_G U^{r}{ij},(U^{-1})^{s}{kl},\mathrm{d}U = \frac{\delta_{rs}}{d_r},\delta_{il}\delta_{jk}, $$ the Peter-Weyl orthogonality of matrix elements. A link integral is nonzero only when the representations on the two plaquettes meeting at that link, together with any loop line through it, match and contract. For an interior link the only matched assignment that connects to the fundamental line of $C$ is the fundamental representation on the adjacent plaquettes; the identity representation elsewhere integrates each free link to unity and cancels against $Z$ .

The contractions force the fundamental-representation plaquettes to form a connected surface $Σ$ with $\partial Σ = C$ . Each plaquette of $Σ$ carries one factor $c_{F} = \frac{β}{2 N ^{2}}$ , and the matrix-element contractions along interior links and the loop produce one overall factor of $1$ after normalisation (the $d_{r}$ and $1/ d_{r}$ factors telescope on a tiling). Summing over surfaces, $$ W(C) = \sum_{\Sigma:,\partial\Sigma = C} \Big(\tfrac{\beta}{2N^2}\Big)^{|\Sigma|}(1 + O(\beta)) = \Big(\tfrac{\beta}{2N^2}\Big)^{A(C)}\big(1 + O(\beta)\big), $$ where the last step keeps the minimal surface $∣Σ∣ = A (C)$ , all others being suppressed by extra powers of $β$ . Taking logarithms, $- lo g W (C) = A (C) (- lo g \frac{β}{2 N ^{2}}) + O (β A (C))$ , so $σ = - lo g (β /2 N^{2}) + O (β)$ , which is positive once $β /2 N^{2} < 1$ . The static potential $V (R) = σ R / a$ is linear. $□$

Proposition (string tension is gauge invariant and reflection-positive). The string tension $σ$ extracted from $W (R \times T) \sim e^{- σ R T}$ equals the energy gap of the transfer matrix in the sector of a static quark-antiquark pair at separation $R$ , divided by $R$ in the linear regime, and this gap is real and non-negative.

Proof. By Osterwalder-Seiler reflection positivity there is a self-adjoint transfer matrix $T = e^{- a H}$ with $H \geq 0$ on a physical Hilbert space. Insert a complete set of energy eigenstates into the temporal evolution of the Wilson loop: a rectangular loop of temporal extent $T = N_{t} a$ is the trace of $T^{N_{t}}$ restricted to the sector with static sources at separation $R$ , so $W (R \times T) = \sum_{n} ∣ ⟨ n ∣ ψ_{R} ⟩ ∣^{2} e^{- E_{n} (R) T}$ . For large $T$ the lowest energy $E_{0} (R)$ dominates, giving $W \sim e^{- E_{0} (R) T}$ , hence $V (R) = E_{0} (R)$ . Comparison with the area law $W \sim e^{- σ R T}$ identifies $E_{0} (R) = σ R$ in the linear regime, so $σ = E_{0} (R) / R$ . Reflection positivity makes each $E_{n} (R)$ real with $E_{n} (R) \geq 0$ , so $σ \geq 0$ , and the strong-coupling computation shows $σ > 0$ . Gauge invariance is automatic because $W (C)$ is built from a trace of an ordered link product and the Haar measure is gauge invariant. $□$

Theorem (absence of a bulk transition; stated without full proof — see Creutz 1980 ^{[source pending]}). The statement that $SU (2)$ and $SU (3)$ have no bulk phase transition between strong and weak coupling is established numerically by Monte Carlo measurement of the average plaquette and of the string tension across the coupling range; no analytic proof exists, and the analytic continuity of the confining phase to the continuum limit remains, in the precise sense of the Yang-Mills mass-gap problem, open. The lattice provides the constructive control; the existence of the continuum limit with a mass gap is a Clay Millennium Problem.

Connections Master

Yang-Mills action 03.07.05. The Wilson plaquette action is the lattice discretisation of the continuum Yang-Mills action: expanding the plaquette $U_{□} = exp (i a^{2} F_{μν} + \dots)$ recovers $\frac{1}{2 g ^{2}} \int tr F^{2}$ as the spacing goes to zero. The lattice is the only known regulator that keeps exact gauge invariance at finite cutoff, which is why it is the definition of choice for non-perturbative gauge theory and the setting in which the Yang-Mills mass-gap problem is posed.
Asymptotic freedom and the running gauge coupling 12.18.03. Confinement and asymptotic freedom are the two ends of one renormalisation-group flow. Asymptotic freedom makes the coupling small in the ultraviolet, justifying the continuum limit at large $β$ ; the lattice strong-coupling area law lives at small $β$ . Creutz's demonstration that the lattice string tension scales as $a σ \propto exp (- 1/2 b_{0} g^{2})$ with the two-loop beta-function coefficient $b_{0}$ is the quantitative bridge: the same flow that is free in the ultraviolet confines in the infrared.
Wilson's lattice gauge theory 08.08.01. The statistical-mechanics framing of the same construction reads the plaquette weight as a Boltzmann factor of a generalised spin model and studies its thermodynamic limit, transfer matrix, and phase structure. This QFT-pointer unit imports that machinery and re-reads it as the confinement narrative of a Lorentzian gauge theory: the same partition function, two complementary readings, with the stat-mech unit owning the thermodynamic-limit theory and this unit owning the quark-confinement interpretation.
Wilson action 08.08.02. The character expansion that proves the area law is the same expansion used on the stat-mech side to compute the lattice free energy; the per-plaquette character coefficient $c_{F} (β)$ that becomes the string-tension factor here is the high-temperature expansion coefficient there. The two units share one computation and split its interpretation.
BRST cohomology and Faddeev-Popov ghost quantisation 03.07.31. The lattice needs no ghosts because the compact-group Haar measure makes the gauge volume finite and cancelable, in direct contrast to the continuum, where the divergent gauge volume forces the Faddeev-Popov determinant and the BRST formalism. The two approaches are complementary regularisations of the same gauge redundancy: the lattice handles it geometrically, the continuum handles it cohomologically.

Historical & philosophical context Master

Kenneth Wilson introduced lattice gauge theory in his 1974 paper Confinement of Quarks (Phys. Rev. D 10, 2445) ^{[source pending]}, motivated by the failure of weak-coupling perturbation theory to address confinement and by his own renormalisation-group programme. The construction grew directly out of Wilson's work on the renormalisation group in statistical mechanics: he recognised that a gauge theory on a Euclidean lattice is a classical statistical-mechanical system, that the strong-coupling regime is its high-temperature expansion, and that the Wilson loop is the gauge-invariant analogue of a spin correlation. The area-law criterion for confinement and its proof at strong coupling appeared in that single paper. John Kogut and Leonard Susskind gave the Hamiltonian (continuous-time) formulation in 1975 (Phys. Rev. D 11, 395) ^{[source pending]}, making the electric-flux-string picture of confinement explicit, and Kogut's 1979 review (Rev. Mod. Phys. 51, 659) ^{[source pending]} became the standard pedagogical account.

The rigorous side developed in parallel. Konrad Osterwalder and Erhard Seiler established reflection positivity and the transfer-matrix construction for lattice gauge theories in 1978 (Ann. Phys. 110, 440) ^{[source pending]}, placing the strong-coupling confinement result on a constructive-field-theory footing and giving the cluster-expansion proof of the area law's convergence. Shmuel Elitzur's 1975 theorem (Phys. Rev. D 12, 3978) ^{[source pending]} clarified that local gauge symmetry cannot break spontaneously and so the order parameter must be the nonlocal Wilson loop, settling a conceptual confusion inherited from the Higgs literature. Michael Creutz's 1980 Monte Carlo study (Phys. Rev. D 21, 2308) ^{[source pending]} turned the lattice into a computational instrument, showing that $SU (2)$ and $SU (3)$ have no bulk transition and that the measured string tension scales with the asymptotic-freedom beta function, joining the strong-coupling and continuum descriptions quantitatively. Steven Weinberg's treatment in The Quantum Theory of Fields, Vol. II §15.9 (Cambridge, 1996) presents the lattice as the gauge-invariant non-perturbative definition that the Lorentzian-Lagrangian development of the rest of the volume points toward. The existence of the continuum limit with a mass gap remains a Clay Millennium Problem, and lattice gauge theory is the framework in which that problem is precisely formulated.

Bibliography Master

@article{Wilson1974Confinement,
  author  = {Wilson, Kenneth G.},
  title   = {Confinement of Quarks},
  journal = {Phys. Rev. D},
  volume  = {10},
  year    = {1974},
  pages   = {2445--2459}
}

@article{KogutSusskind1975,
  author  = {Kogut, John B. and Susskind, Leonard},
  title   = {Hamiltonian Formulation of Wilson's Lattice Gauge Theories},
  journal = {Phys. Rev. D},
  volume  = {11},
  year    = {1975},
  pages   = {395--408}
}

@article{Kogut1979Review,
  author  = {Kogut, John B.},
  title   = {An Introduction to Lattice Gauge Theory and Spin Systems},
  journal = {Rev. Mod. Phys.},
  volume  = {51},
  year    = {1979},
  pages   = {659--713}
}

@article{OsterwalderSeiler1978,
  author  = {Osterwalder, Konrad and Seiler, Erhard},
  title   = {Gauge Field Theories on a Lattice},
  journal = {Ann. Phys. (NY)},
  volume  = {110},
  year    = {1978},
  pages   = {440--471}
}

@article{Elitzur1975,
  author  = {Elitzur, Shmuel},
  title   = {Impossibility of Spontaneously Breaking Local Symmetries},
  journal = {Phys. Rev. D},
  volume  = {12},
  year    = {1975},
  pages   = {3978--3982}
}

@article{Creutz1980,
  author  = {Creutz, Michael},
  title   = {Monte Carlo Study of Quantized SU(2) Gauge Theory},
  journal = {Phys. Rev. D},
  volume  = {21},
  year    = {1980},
  pages   = {2308--2315}
}

@article{BanksMyersonKogut1977,
  author  = {Banks, Tom and Myerson, Robert and Kogut, John},
  title   = {Phase Transitions in Abelian Lattice Gauge Theories},
  journal = {Nucl. Phys. B},
  volume  = {129},
  year    = {1977},
  pages   = {493--510}
}

@book{Weinberg1996QTF2,
  author    = {Weinberg, Steven},
  title     = {The Quantum Theory of Fields, Vol. 2: Modern Applications},
  publisher = {Cambridge University Press},
  year      = {1996}
}

@book{MontvayMunster1994,
  author    = {Montvay, Istvan and M{\"u}nster, Gernot},
  title     = {Quantum Fields on a Lattice},
  publisher = {Cambridge University Press},
  year      = {1994}
}

@book{Smit2002,
  author    = {Smit, Jan},
  title     = {Introduction to Quantum Fields on a Lattice},
  publisher = {Cambridge University Press},
  year      = {2002}
}

Prerequisites

12.12.01
03.07.05
12.18.03
08.08.01
08.08.02

Tier anchors

beginner: Wilczek, The Lightness of Being (2008), Chs. 3-5 (confinement and the QCD vacuum); Wilson, Scientific American 'The Origins of Lattice Gauge Theory' reflections
intermediate: Peskin & Schroeder, An Introduction to Quantum Field Theory (1995), §15 (Wilson lines and loops) with Smit, Introduction to Quantum Fields on a Lattice (2002), Chs. 4-6
master: Weinberg, The Quantum Theory of Fields, Vol. II (1996), §15.9; Wilson, Phys. Rev. D 10, 2445 (1974); Kogut, Rev. Mod. Phys. 51, 659 (1979); Coleman, Aspects of Symmetry (1985), Ch. 5 §3 (1/N and confinement); Montvay-Munster, Quantum Fields on a Lattice (1994)

References

Wilson, K. G. — Confinement of Quarks · Phys. Rev. D 10, 2445 (1974); the plaquette action, the Wilson loop, the strong-coupling area law
Kogut, J. B. — An Introduction to Lattice Gauge Theory and Spin Systems · Rev. Mod. Phys. 51, 659 (1979); Hamiltonian lattice gauge theory, strong-coupling expansion, the confinement criterion
Kogut, J. B. & Susskind, L. — Hamiltonian Formulation of Wilson's Lattice Gauge Theories · Phys. Rev. D 11, 395 (1975); the electric-field Hamiltonian and the strong-coupling vacuum
Osterwalder, K. & Seiler, E. — Gauge Field Theories on a Lattice · Ann. Phys. (NY) 110, 440 (1978); reflection positivity, transfer matrix, and the rigorous strong-coupling confinement proof
Weinberg, S. — The Quantum Theory of Fields, Vol. II: Modern Applications · Cambridge University Press (1996), §15.9 (the lattice formulation of gauge theories; the Wilson loop and the confinement criterion)
Creutz, M. — Monte Carlo Study of Quantized SU(2) Gauge Theory · Phys. Rev. D 21, 2308 (1980); numerical evidence for a smooth crossover from strong to weak coupling with no bulk phase transition for SU(2) and SU(3)
Elitzur, S. — Impossibility of Spontaneously Breaking Local Symmetries · Phys. Rev. D 12, 3978 (1975); gauge-invariant order parameters are forced, motivating the Wilson loop

Estimated time

beginner: 18m
intermediate: 45m
master: 85m