12.18.06 · quantum / gauge-and-symmetry

Operator product expansion and short-distance behaviour

shipped3 tiersLean: none

Anchor (Master): Weinberg, S., *The Quantum Theory of Fields, Vol. 2: Modern Applications* (Cambridge, 1996), Ch. 20 (the operator product expansion; Wilson coefficients and their renormalisation-group running; deep-inelastic sum rules); Collins, J. C., *Renormalization* (Cambridge, 1984), Ch. 10 (the OPE as an operator statement and its proof in perturbation theory); Zimmermann, W., 'Local Operator Products and Renormalization in Quantum Field Theory', in *Lectures on Elementary Particles and Quantum Field Theory* (MIT Press, 1970)

Intuition Beginner

When you measure two things at almost the same place at almost the same instant, the answer often blows up. In a field theory, the basic measurable objects are local operators: the electric current at a point, the energy density at a point, the value of a field at a point. Multiply two of them together and slide the points toward each other, and the product grows without bound. The operator product expansion tames this blow-up. It says the messy product of two nearby operators can be rewritten as a clean sum: each term is an ordinary number that carries all the singular dependence on the separation, times a single well-behaved operator sitting at the midpoint.

Think of two magnets brought close together. Far apart they barely interact; up close the force grows like one over the distance squared. The force is a known function of the separation, and the magnets themselves are unchanged objects. The operator product expansion does the same factorisation for quantum fields: the strength of the short-distance singularity is captured by a number, the coefficient function, while the physical content that survives at the midpoint is a local operator.

Why bother? Because short distances mean high energies, and high-energy experiments probe exactly this regime. The expansion sorts the many possible operators by how badly each one diverges as the points merge. The most singular terms dominate, so a handful of operators controls the entire short-distance behaviour. That sorting is what makes deep-inelastic scattering off a proton computable.

Visual Beginner

Picture two points, labelled $x$ and $y$ , drawn close together inside a cloud that represents a quantum field. As $x$ slides toward $y$ , the product of the two operators is redrawn on the right as a stack of terms: each term is a coefficient (a number that depends only on the separation) times a single operator pinned at the midpoint between the two points. The coefficients are sorted top to bottom by how strongly they grow as the separation shrinks, the most singular one on top.

The picture captures the central move: a two-operator product at short separation becomes a sum of one-operator terms, with all the separation dependence pushed into the c-number coefficients. The leading term in the stack is the one whose coefficient is most singular as the points come together.

Worked example Beginner

Take the simplest quantum field, a single free scalar field $ϕ$ , and look at the product of its value at two nearby points in the lowest two-point average. The two-point average of the field with itself is the propagator. In four spacetime dimensions, at short separation $r$ between the two points, this average behaves like a number divided by $r$ squared: $$ \langle \phi(x),\phi(y) \rangle = \frac{1}{4\pi^2},\frac{1}{r^2}, \qquad r = |x - y|. $$

Step 1. Read off the singular coefficient. The whole separation dependence is the factor $1/ r^{2}$ . The number $1/ (4 π^{2})$ is a fixed constant. So the coefficient function multiplying the simplest operator (the identity, the operator that just gives back $1$ ) is $C (r) = 1/ (4 π^{2} r^{2})$ .

Step 2. Identify the operator at the midpoint. In this lowest average, the operator left behind is the identity: the average of the product is a pure number, with no field operator surviving. So the leading short-distance term of $ϕ (x) ϕ (y)$ is the coefficient $1/ (4 π^{2} r^{2})$ times the identity operator.

Step 3. Find the next term. Subtract the singular piece. What remains, as the points merge, is the average of $ϕ$ squared at the midpoint, a finite operator. Its coefficient is a constant times $r^{0}$ , meaning it survives smoothly as $r$ goes to zero. So the next term is a finite coefficient times the operator $ϕ^{2}$ at the midpoint.

Step 4. Order the terms by singularity. The identity term grows like $1/ r^{2}$ . The $ϕ^{2}$ term stays finite. The identity dominates at the shortest distances; the $ϕ^{2}$ term is the leading surviving operator.

What this tells us: even for a free field, the product of two operators at short separation splits cleanly into a singular number times the identity, plus a finite number times a single operator at the midpoint. The whole singularity lives in the coefficient. That clean split is the operator product expansion in its simplest possible instance.

Check your understanding Beginner

Exercise (easy, multiple choice).

In the operator product expansion, where does the singular dependence on the separation between the two points live?

A. In the operators sitting at the midpoint B. In the c-number coefficient functions C. It cancels and there is no singularity D. Split evenly between coefficients and operators

Hint

The whole point of the expansion is to factor the product into a part that knows about the separation and a part that does not.

Answer

B. In the c-number coefficient functions.

The coefficient functions carry all the singular dependence on the separation. The operators sitting at the midpoint are finite, well-defined local operators that do not depend on the separation. This factorisation is the central feature of the expansion.

Correct: You have the central split right. If you picked another option: the operators at the midpoint are evaluated at a single point and carry no separation dependence; the singularity has to go into the numbers multiplying them.

Formal definition Intermediate+

Let $A (x)$ and $B (y)$ be local operators of a relativistic quantum field theory, each carrying a definite scaling dimension $d_{A}$ , $d_{B}$ (the engineering dimension corrected by the operator's anomalous dimension). The operator product expansion (OPE) is the asymptotic statement, due to Wilson ^{[Wilson 1969]}, that as $x \to y$ the product factorises into a sum over a complete set of local operators $O_{n}$ , $$ A(x),B(y) ;\underset{x\to y}{\sim}; \sum_n C^n_{AB}(x-y);\mathcal{O}n!\left(\tfrac{x+y}{2}\right), $$ where the $C^n{AB} $a r ec - n u mb er * * coe f f i c i e n t f u n c t i o n s * * (W i l so n coe f f i c i e n t s) d e p e n d in g o n l y o n t h ese p a r a t i o n$ x-y $an d o n t h eco u pl in g, an d t h eo p er a t or s$ \mathcal{O}_n $a r er e n or ma l i se d l oc a l o p er a t or so f d e f ini t esc a l in g d im e n s i o n$ d_n $. T h es t a t e m e n t i s a sy m pt o t i c ina p r ec i sese n se : in ser t e d in t o an y G r ee n f u n c t i o n w i t h t h er e mainin g f i e l d s k e pt a t f i x e d se p a r a t i o n f r o m$ x $,$ y $, t h e d i f f er e n ce b e tw ee n t h e l e f t s i d e an d t h es u m t r u n c a t e d a f t er t h eo p er a t or so f d im e n s i o n$ \le D $i s b o u n d e d b y a p o w er$ |x-y|^{D - d_A - d_B + \epsilon} $f or e v er y$ \epsilon > 0$ ^{[Wilson-Zimmermann 1972]}.

By Lorentz invariance and dimensional analysis the leading short-distance behaviour of each coefficient is fixed: $$ C^n_{AB}(x-y) ;\sim; |x-y|^{,d_n - d_A - d_B}\times(\text{logarithms}), $$ up to tensor structures carrying the Lorentz indices. The exponent $d_{n} - d_{A} - d_{B}$ is negative — the coefficient is singular — exactly when $d_{n} < d_{A} + d_{B}$ . Operators of low dimension therefore dominate the short-distance product; operators of high dimension are suppressed by extra powers of the separation. This is the operative content of the expansion: it sorts the contributing operators by dimension, and a finite set controls the leading behaviour.

The identity operator $1$ is always present with the most singular coefficient, $C_{A B}^{1} \sim ∣ x - y ∣^{- d_{A} - d_{B}}$ ; it reproduces the leading singularity of the vacuum expectation value $⟨ A (x) B (y)⟩$ . Conserved currents, the stress tensor, and other low-dimension composites appear as subleading operators.

The logarithms in $C_{A B}^{n}$ are not optional: in an interacting theory the dimensions $d_{n}$ shift away from their free values by anomalous dimensions $γ_{n} (g)$ , and the resummation of logarithms is governed by the renormalisation group. The expansion as written holds in Euclidean position space or, after Fourier transform, in deep-Euclidean momentum space where all invariants are large and spacelike; the light-cone version, governing $x^{2} \to 0$ with $x$ not necessarily small, is the form relevant to deep-inelastic scattering and is organised by twist rather than by dimension.

Key theorem with proof Intermediate+

The cleanest complete statement is the free-field OPE, where the coefficient functions are computed exactly by Wick contraction and the expansion terminates at each order. It is the rigorous skeleton on which the interacting expansion is built.

Theorem (free-scalar OPE). Let $ϕ$ be a free real scalar field of mass $m$ in four-dimensional Minkowski space, with Feynman propagator $Δ_{F} (x - y) = ⟨ 0∣ T ϕ (x) ϕ (y) ∣0 ⟩$ satisfying $Δ_{F} (z) \to 1/ (4 π^{2} z^{2})$ as $z^{2} \to 0$ (with $z^{2} = z_{μ} z^{μ}$ in the Euclidean continuation). Then the time-ordered product admits the exact expansion $$ T,\phi(x)\phi(y) ;=; \Delta_F(x-y),\mathbf{1} ;+; {:}\phi(x)\phi(y){:}, $$ and expanding the normal-ordered term about the midpoint $z = \frac{x + y}{2}$ in powers of $ξ = x - y$ produces the operator product expansion $$ T,\phi(x)\phi(y) ;=; \Delta_F(\xi),\mathbf{1} ;+; {:}\phi^2{:}(z) ;+; \tfrac{1}{4},\xi^\mu\xi^\nu,{:}\partial_\mu\phi,\partial_\nu\phi{:}(z) ;-; \tfrac{1}{4},\xi^\mu\xi^\nu,{:}\phi,\partial_\mu\partial_\nu\phi{:}(z);+;\cdots, $$ with coefficient functions $C^{1} = Δ_{F} (ξ) \sim 1/ (4 π^{2} ξ^{2})$ , $C^{ϕ^{2}} = 1$ , and $C_{μν}^{\partial\partial} = O (ξ^{2})$ .

Proof. Apply Wick's theorem ^{[Itzykson-Zuber §8-4]} to the time-ordered product of two free fields. Wick's theorem states that a time-ordered product equals the sum of its normal-ordered terms with all possible contractions, a contraction of two fields being the Feynman propagator. For two fields there is exactly one contraction and one uncontracted term: $$ T,\phi(x)\phi(y) ;=; \underbrace{\Delta_F(x-y)}{\text{contraction}},\mathbf{1} ;+; \underbrace{{:}\phi(x)\phi(y){:}}{\text{no contraction}}. $$ The contraction is a c-number and multiplies the identity; the normal-ordered product is a finite operator (its vacuum expectation vanishes by construction). This is already an exact OPE with two terms, the first singular and the second regular as $x \to y$ .

It remains to expand the regular operator about the midpoint to expose the higher operators and their coefficients. Write $x = z + ξ /2$ , $y = z - ξ /2$ with $z = (x + y) /2$ . Taylor-expand each field in $ξ$ : $$ \phi(x) = \phi(z) + \tfrac12 \xi^\mu \partial_\mu \phi(z) + \tfrac18 \xi^\mu\xi^\nu \partial_\mu\partial_\nu\phi(z) + \cdots, $$ $$ \phi(y) = \phi(z) - \tfrac12 \xi^\mu \partial_\mu \phi(z) + \tfrac18 \xi^\mu\xi^\nu \partial_\mu\partial_\nu\phi(z) + \cdots. $$ Multiply inside the normal ordering (normal ordering is linear, so it commutes with the Taylor expansion). The zeroth order gives $: ϕ^{2} : (z)$ with coefficient $1$ . The terms linear in $ξ$ cancel between the two expansions in the symmetric combination, as required since the product is symmetric under $ξ \to - ξ$ together with $x \leftrightarrow y$ . The order- $ξ^{2}$ terms assemble into $$ {:}\Big(\tfrac18 \xi^\mu\xi^\nu \phi,\partial_\mu\partial_\nu\phi - \tfrac14 \xi^\mu\xi^\nu \partial_\mu\phi,\partial_\nu\phi + \tfrac18 \xi^\mu\xi^\nu \partial_\mu\partial_\nu\phi,\phi\Big){:}(z), $$ which regroups into the displayed dimension-six operators with $O (ξ^{2})$ coefficients. Each coefficient function carries exactly the power of $ξ$ predicted by dimensions: the identity coefficient $Δ_{F} (ξ)$ scales as $ξ^{- 2}$ (dimension $d_{1} - 2 d_{ϕ} = 0 - 2 = - 2$ ), the $ϕ^{2}$ coefficient as $ξ^{0}$ (dimension $2 - 2 = 0$ ), the dimension-four operators as $ξ^{2}$ . The expansion is exact at each finite order in $ξ$ because the free field has no interactions to generate further contractions. $□$

Bridge. This free-field computation builds toward the interacting OPE, and the bridge is the recognition that the same dimensional bookkeeping survives renormalisation with one modification: the dimensions acquire anomalous corrections. The foundational reason the expansion exists at all is exactly the factorisation just proved — the singular separation dependence lives entirely in c-number coefficients, the surviving operators sit at one point — and this is exactly what generalises to the interacting theory, where the coefficient functions $C_{A B}^{n}$ are no longer pure powers but pick up logarithms resummed by the renormalisation group. The central insight is that the coefficient functions satisfy their own Callan-Symanzik equation, with the anomalous dimensions of $A$ , $B$ , and $O_{n}$ as inhomogeneous terms; this pattern appears again in the running of the gauge coupling, where the same beta function that governs 12.18.03 (asymptotic freedom) controls the logarithmic scaling of the Wilson coefficients. Putting these together, the OPE identifies the short-distance physics of a product of currents with a renormalisation-group-improved sum over local operators, and the bridge to experiment is that the moments of deep-inelastic structure functions are precisely these coefficient functions, weighted by the matrix elements of the operators in the target.

Exercises Intermediate+

Exercise 4 (medium, short-answer).

Explain why the identity operator always appears in the OPE of two operators with the most singular coefficient, and why its coefficient reproduces the vacuum two-point function.

Hint

The identity has dimension zero, the smallest possible. Take the vacuum expectation value of both sides of the expansion.

Answer

The identity $1$ has scaling dimension $d_{1} = 0$ , the smallest dimension of any operator, so its coefficient scales as $∣ ξ ∣^{0 - d_{A} - d_{B}} = ∣ ξ ∣^{- d_{A} - d_{B}}$ , the most singular power available. Every other operator has positive dimension and a less singular (or finite) coefficient. Taking the vacuum expectation value of $A (x) B (y) \sim \sum_{n} C_{A B}^{n} O_{n} (z)$ , every $O_{n}$ with $n \neq = 1$ has vanishing vacuum expectation value by definition of the renormalised composites, so $⟨ A (x) B (y)⟩ = C_{A B}^{1} (ξ) ⟨ 1 ⟩ = C_{A B}^{1} (ξ)$ . The identity coefficient is exactly the vacuum two-point function. Rubric: full credit for the dimension argument plus the vacuum-expectation argument.

Exercise 5 (medium, symbolic).

Write the Callan-Symanzik (renormalisation-group) equation satisfied by a Wilson coefficient $C_{A B}^{n} (ξ; g, μ)$ , where $A$ , $B$ , $O_{n}$ have anomalous dimensions $γ_{A}$ , $γ_{B}$ , $γ_{n}$ and $β (g)$ is the beta function. Explain the sign of each anomalous-dimension term.

Hint

The product $A (x) B (y)$ on the left and the operator $O_{n}$ on the right must each be renormalisation-group invariant when multiplied by their own renormalisation factors. The coefficient absorbs the mismatch.

Answer

[μ \frac{\partial}{\partial μ} + β (g) \frac{\partial}{\partial g} - γ_{A} (g) - γ_{B} (g) + γ_{n} (g)] C_{A B}^{n} (ξ; g, μ) = 0.

The operators $A$ and $B$ on the left carry their renormalisation factors $Z_{A}^{- 1}, Z_{B}^{- 1}$ , contributing $- γ_{A} - γ_{B}$ (the coefficient must compensate as the products are made finite); the operator $O_{n}$ on the right contributes $+ γ_{n}$ (the coefficient compensates the renormalisation of the operator it multiplies). The solution exponentiates the anomalous dimensions, replacing the naive power $∣ ξ ∣^{d_{n} - d_{A} - d_{B}}$ by a running power with logarithmic corrections. Rubric: full credit for the equation with correct signs and the explanation.

Exercise 6 (medium, short-answer).

Define the twist of a local operator and explain why, in the light-cone OPE relevant to deep-inelastic scattering, operators are organised by twist rather than by dimension.

Hint

Twist is dimension minus spin. On the light cone $x^{2} \to 0$ the relevant suppression is by powers of $x^{2}$ times the large invariant, which counts twist, not dimension.

Answer

The twist of an operator is $τ = d - s$ , its scaling dimension minus its Lorentz spin (the number of symmetric traceless indices). In deep-inelastic scattering the relevant short-distance limit is the light-cone limit $x^{2} \to 0$ with the momentum transfer $Q^{2}$ large, not the limit $x \to 0$ . On the light cone the suppression of an operator's contribution to a structure-function moment is by powers of $1/ Q^{2}$ controlled by $τ - 2$ , not by the full dimension: spin- $s$ operators contribute to the $s$ -th moment, and among operators of given spin the lowest twist dominates at large $Q^{2}$ . Leading twist is $τ = 2$ (the quark and gluon bilinears with covariant derivatives), giving the parton-model scaling; higher-twist operators give power corrections in $1/ Q^{2}$ . Rubric: full credit for the definition $τ = d - s$ plus the $1/ Q^{2}$ -power argument.

Exercise 7 (hard, short-answer).

Derive the leading-order solution of the Callan-Symanzik equation for a Wilson coefficient in an asymptotically free theory, and show that the coefficient runs as a power of the logarithm of the scale.

Hint

Solve along characteristics with the running coupling $\overset{g}{ˉ} (ξ)$ . In an asymptotically free theory $\overset{g}{ˉ}^{2} \sim 1/ (b_{0} lo g (1/ ξ^{2} μ^{2}))$ , and $γ (g) = γ_{0} g^{2} + \dots$ .

Answer

Solve $[μ \partial_{μ} + β \partial_{g} + Γ (g)] C = 0$ with $Γ = - γ_{A} - γ_{B} + γ_{n}$ by the method of characteristics: $C (ξ; g, μ) = C (ξ; \overset{g}{ˉ}, \overset{μ}{ˉ}) exp (- \int_{g}^{\overset{g}{ˉ}} \frac{Γ ( g ^{'} )}{β ( g ^{'} )} d g^{'})$ , choosing the scale so $\overset{μ}{ˉ} \sim 1/∣ ξ ∣$ removes the large logarithm from the fixed-order coefficient. With $β = - b_{0} g^{3} / (16 π^{2})$ and $Γ = Γ_{0} g^{2} / (16 π^{2})$ at leading order, $\int \frac{Γ}{β} d g = - \frac{Γ _{0}}{b _{0}} \int \frac{d g}{g} = - \frac{Γ _{0}}{b _{0}} lo g \overset{g}{ˉ} + const$ , so the exponential becomes $(\overset{g}{ˉ} / g)^{- Γ_{0} / b_{0}}$ . Using the asymptotically free running $\overset{g}{ˉ}^{2} (ξ) ≃ 16 π^{2} / (b_{0} lo g (1/ ξ^{2} μ^{2}))$ , the coefficient acquires the factor $$ C^n_{AB}(\xi) ;\propto; |\xi|^{,d_n - d_A - d_B},\big[\log(1/\xi^2\mu^2)\big]^{-\Gamma_0/(2b_0)}, $$ a power-of-logarithm modification of the naive dimensional scaling. For the leading-twist operators of deep-inelastic scattering this reproduces the logarithmic scaling violations of the structure-function moments. Rubric: full credit for the characteristic-method solution plus the power-of-logarithm result with the exponent $- Γ_{0} / (2 b_{0})$ .

Exercise 8 (hard, short-answer).

State the OPE for the time-ordered product of two electromagnetic currents $T j^{μ} (x) j^{ν} (0)$ near the light cone and identify which operators control the moments of the deep-inelastic structure functions.

Hint

The leading-twist operators are the symmetric traceless spin- $n$ quark and gluon operators $O^{μ_{1} \dots μ_{n}}$ built from the field with $n - 1$ covariant derivatives. The $n$ -th moment of a structure function picks out the spin- $n$ operator.

Answer

Near the light cone $x^{2} \to 0$ , $$ T,j^\mu(x)j^\nu(0) \sim \sum_{n,i} C^{(n)}i(x^2),x{\mu_1}\cdots x_{\mu_n},\mathcal{O}_i^{\mu\nu\mu_1\cdots\mu_n}(0), $$ where $O_{i}^{μ_{1} \dots μ_{n}}$ are the leading-twist (twist-two) symmetric traceless spin- $n$ operators: the quark operators $\overset{ˉ}{ψ} γ^{{μ_{1}} i D^{μ_{2}} \dots i D^{μ_{n}}} ψ$ and the analogous gluon operators, with traces removed. Taking the forward matrix element in a nucleon state of momentum $p$ and matching to the Bjorken- $x$ moments, the $n$ -th Mellin moment $\int_{0}^{1} d x x^{n - 1} F_{2} (x, Q^{2})$ is the product of the Wilson coefficient $C^{(n)} (Q^{2})$ and the reduced matrix element $⟨ p ∣ O^{(n)} ∣ p ⟩$ . The $Q^{2}$ dependence of each moment is governed by the anomalous dimension of the corresponding spin- $n$ operator, giving the logarithmic scaling violations of the (later DGLAP) evolution. The leading twist $τ = 2$ dominates at large $Q^{2}$ ; higher twist gives $1/ Q^{2}$ power corrections. Rubric: full credit for the light-cone expansion, the identification of the leading-twist spin- $n$ operators, and the moment-by-moment matching.

Advanced results Master

Theorem (existence of the OPE in renormalised perturbation theory; Wilson-Zimmermann). In a renormalisable field theory with renormalised composite operators defined by Zimmermann normal products (the BPHZ subtraction), the operator product of two such composites admits, to every order in perturbation theory, an asymptotic expansion $$ A(x)B(0) ;=; \sum_{n : d_n \le D} C^n_{AB}(x),\mathcal{O}n(0) ;+; R_D(x), $$ *where the remainder, inserted into any Green function with spectator fields at finite separation, satisfies $R_{D} (x) = O (∣ x ∣^{D - d_{A} - d_{B} + 1 - ϵ})$ for every $ϵ > 0$ as $x \to 0$ . The coefficient functions $C^n{AB} $a r eco m p u t e d b y t h e 1 P I am p u t a t e d G r ee n f u n c t i o n s w i t h o n e in ser t i o n o f t h e p r o d u c t$ A B $an d o n e in ser t i o n o f$ \mathcal{O}_n$, projected onto the appropriate tensor and dimension.* ^{[Wilson-Zimmermann 1972]}

The proof is an induction on the dimension cut-off $D$ together with the BPHZ forest formula. At each order one subtracts the contributions of all operators of dimension $\leq D$ ; Weinberg's power-counting theorem then bounds the remainder by the stated power. The expansion is not convergent but asymptotic — adding more terms improves the short-distance approximation but the series need not sum. Collins gives the modern renormalisation-group proof, in which the coefficient functions are extracted as the solution of an inhomogeneous Callan-Symanzik equation ^{[Collins Ch. 10]}.

Theorem (renormalisation-group equation for the coefficients). The Wilson coefficients satisfy $$ \Big[\mu\frac{\partial}{\partial\mu} + \beta(g)\frac{\partial}{\partial g} - \gamma_A(g) - \gamma_B(g)\Big]C^n_{AB}(x;g,\mu) = -\sum_m \gamma_{nm}(g),C^m_{AB}(x;g,\mu), $$ where $γ_{nm}$ is the anomalous-dimension matrix mixing the operators $O_{n}$ of equal dimension and quantum numbers under renormalisation. Diagonalising $γ_{nm}$ produces multiplicatively renormalised operators with definite anomalous dimensions, and the coefficients of these eigen-operators run as powers of the logarithm of the scale in an asymptotically free theory. ^{[Weinberg Ch. 20]}

Operator mixing is generic: operators of the same dimension and symmetry that are not forbidden to mix do mix under renormalisation, so $γ$ is a matrix, not a number. In deep-inelastic scattering the spin- $n$ quark and gluon operators mix, and the eigenvalues of the $2 \times 2$ mixing matrix are the leading-order DGLAP anomalous dimensions.

Theorem (application to deep-inelastic scattering and sum rules). The forward Compton amplitude $T^{μν} (p, q)$ for virtual-photon-nucleon scattering, analytically continued in the Bjorken variable, has moments controlled by the light-cone OPE: the $n$ -th moment of the structure function $F_{2}$ , $$ M_n(Q^2) = \int_0^1 dx;x^{n-2},F_2(x,Q^2) = \sum_i C^{(n)}_i(Q^2),\langle\langle\mathcal{O}^{(n)}_i\rangle\rangle, $$ factorises into a perturbatively computable Wilson coefficient $C_{i}^{(n)} (Q^{2})$ and a non-perturbative reduced matrix element $⟨⟨ O_{i}^{(n)} ⟩⟩$ of the spin- $n$ twist-two operator in the nucleon. The $Q^{2}$ evolution of each moment is fixed by the anomalous dimension of the operator, giving the logarithmic scaling violations. ^{[Peskin-Schroeder §18.5]}

This factorisation is the rigorous version of the parton model: the Wilson coefficients are the hard partonic cross-sections, the operator matrix elements are the parton distributions' moments, and the OPE proves that the two factorise at leading twist. The Bjorken and Gross-Llewellyn-Smith sum rules are the lowest moments evaluated with current-algebra matrix elements.

Theorem (free-field termination and the radius of the leading singularity). For free fields the OPE coefficient of every operator is a finite polynomial in the propagator and its derivatives, and the expansion of a product of $k$ fields is exact at each order in the separation. The most singular coefficient is the fully contracted one, scaling as the $⌊ k /2 ⌋$ -th power of the propagator singularity.

This is the statement that the free-field OPE is purely combinatorial (Wick's theorem) and carries no logarithms; logarithms are the signature of interactions and anomalous dimensions, and their absence in the free theory is the boundary condition for the renormalisation-group running.

Synthesis. The operator product expansion is the foundational reason short-distance physics in a quantum field theory is computable: it factorises the singular product of two local operators into c-number coefficient functions, which carry the entire short-distance singularity and are perturbatively calculable, times finite local operators whose matrix elements encode the long-distance, non-perturbative content. The central insight is that the coefficient functions obey their own renormalisation-group equation, so the same beta function and anomalous dimensions that govern the running coupling control the scaling of the coefficients; this is exactly the bridge between the free-field expansion, where Wick's theorem gives pure-power coefficients with no logarithms, and the interacting expansion, where the logarithms resum into anomalous-dimension powers. Putting these together, the light-cone OPE for two electromagnetic currents identifies the moments of deep-inelastic structure functions with products of Wilson coefficients and twist-two operator matrix elements, and this factorisation generalises the parton model into a controlled expansion in inverse powers of the momentum transfer. The expansion is dual to the Wilsonian effective-action picture: integrating out short distances and expanding a product of operators are two views of the same short-distance organisation, and the bridge is that the irrelevant operators of the effective field theory are exactly the high-dimension operators suppressed in the OPE. What began as Wilson's heuristic for current algebra is the operative tool of every modern short-distance calculation, from QCD sum rules to lattice matching to the conformal bootstrap, where the OPE coefficients become the dynamical data of the theory.

Full proof set Master

Proposition (free-scalar OPE coefficients are fixed by dimensions). In the free-scalar OPE of $ϕ (x) ϕ (y)$ , the coefficient function multiplying a normal-ordered operator $O_{n}$ of dimension $d_{n}$ scales as $∣ ξ ∣^{d_{n} - 2}$ as $ξ \to 0$ , where $ξ = x - y$ and $d_{ϕ} = 1$ .

Proof. From the Wick expansion proved in the Key theorem section, $T ϕ (x) ϕ (y) = Δ_{F} (ξ) 1 + : ϕ (x) ϕ (y) :$ , and the Taylor expansion of the normal-ordered factor about the midpoint $z = (x + y) /2$ produces operators of the form $$ \mathcal{O}n^{(a,b)} = {:}(\partial^a\phi)(\partial^b\phi){:}(z),\qquad d_n = 2 + a + b, $$ with $a + b$ derivatives distributed by the Taylor coefficients. The coefficient of $O_{n}^{(a, b)}$ is the corresponding monomial $ξ^{a + b}$ times a numerical factor from the multinomial expansion. Hence the coefficient scales as $∣ ξ ∣^{a + b} = ∣ ξ ∣^{(2 + a + b) - 2} = ∣ ξ ∣^{d_{n} - 2}$ . For the identity, $d_{1} = 0$ and the coefficient $\Delta_F(\xi) \sim 1/(4\pi^2\xi^2) \sim |\xi|^{-2} = |\xi|^{d{\mathbf 1} - 2} $, ma t c hin g t h es am e f or m u l a . T h e g e n er a l d im e n s i o na l r u l e$ C \sim |\xi|^{d_n - d_A - d_B} $s p ec ia l i ses h er e t o$ d_A = d_B = 1 $, g i v in g$ |\xi|^{d_n - 2} $.$ \square$

Proposition (renormalisation-group invariance of the physical product). The combination $C_{A B}^{n} (x; g, μ) O_{n} (0)$ summed over $n$ is renormalisation-group invariant: $μ d / d μ [A (x) B (0)] = 0$ for the bare product, and the Callan-Symanzik equation for $C_{A B}^{n}$ follows from the renormalisation of $A$ , $B$ , and $O_{n}$ .

Proof. The bare operators $A_{0} = Z_{A} A$ , $B_{0} = Z_{B} B$ , $O_{0, n} = Z_{nm} O_{m}$ are independent of the renormalisation scale $μ$ , so $μ d / d μ$ annihilates the bare product $A_{0} (x) B_{0} (0)$ and the bare operators $O_{0, n}$ . Write the renormalised OPE $A (x) B (0) = \sum_{n} C_{A B}^{n} (x; g, μ) O_{n} (0)$ . Substituting the renormalised operators in terms of bare ones, $$ Z_A^{-1}Z_B^{-1}A_0(x)B_0(0) = \sum_n C^n_{AB},(Z^{-1}){nm},\mathcal{O}{0,m}(0). $$ Apply $μ d / d μ = μ \partial_{μ} + β \partial_{g}$ . The left side gives $- (γ_{A} + γ_{B})$ times the product, with $γ_{A} = μ d lo g Z_{A} / d μ$ ; the bare operators are inert. The right side gives $(μ d / d μ C_{A B}^{n}) (Z^{- 1})_{nm} O_{0, m} + C_{A B}^{n} (μ d / d μ (Z^{- 1})_{nm}) O_{0, m}$ , and $μ d / d μ (Z^{- 1})_{nm} = (Z^{- 1} γ)_{nm}$ with $γ_{nm} = (μ d Z / d μ Z^{- 1})_{nm}$ the anomalous-dimension matrix. Equating coefficients of each independent bare operator $O_{0, m}$ yields $$ \Big[\mu\partial_\mu + \beta\partial_g - \gamma_A - \gamma_B\Big]C^n_{AB} = -\sum_m C^m_{AB},\gamma_{mn}, $$ the matrix Callan-Symanzik equation for the coefficients. $□$

Proposition (leading-log running of a multiplicatively renormalised coefficient). If $O$ is multiplicatively renormalised with anomalous dimension $γ (g) = γ_{0} g^{2} / (16 π^{2}) + \dots$ in an asymptotically free theory with $β (g) = - b_{0} g^{3} / (16 π^{2}) + \dots$ , the associated Wilson coefficient runs as $C (x) \propto ∣ x ∣^{d_{n} - d_{A} - d_{B}} [lo g (1/ x^{2} μ^{2})]^{- Γ_{0} / (2 b_{0})}$ at short distance, with $Γ_{0} = - γ_{A, 0} - γ_{B, 0} + γ_{0}$ .

Proof. With a single operator the matrix equation reduces to $[μ \partial_{μ} + β \partial_{g} + Γ (g)] C = 0$ , $Γ = - γ_{A} - γ_{B} + γ$ . The general solution by characteristics is $C (x; g, μ) = \hat{C} (\overset{g}{ˉ}) exp (- \int_{g}^{\overset{g}{ˉ}} \frac{Γ ( g ^{'} )}{β ( g ^{'} )} d g^{'})$ where $\overset{g}{ˉ} = \overset{g}{ˉ} (xμ)$ is the running coupling evaluated at the scale $1/∣ x ∣$ . To leading order $Γ/ β = (Γ_{0} g^{2}) / (- b_{0} g^{3}) = - Γ_{0} / (b_{0} g)$ , so $\int_{g}^{\overset{g}{ˉ}} \frac{Γ}{β} d g^{'} = - (Γ_{0} / b_{0}) lo g (\overset{g}{ˉ} / g)$ and the exponential is $(\overset{g}{ˉ} / g)^{Γ_{0} / b_{0}}$ . The one-loop running coupling satisfies $\overset{g}{ˉ}^{2} (xμ) = 16 π^{2} / [b_{0} lo g (1/ x^{2} μ^{2})]$ , so $\overset{g}{ˉ} \propto [lo g (1/ x^{2} μ^{2})]^{- 1/2}$ and $$ C(x) \propto \big[\log(1/x^2\mu^2)\big]^{-\Gamma_0/(2b_0)}, $$ multiplying the naive dimensional power $∣ x ∣^{d_{n} - d_{A} - d_{B}}$ . $□$

Proposition (free-field termination), proof. The OPE of a product of two free fields has only finitely many operators at each fixed order in the separation $ξ$ , each with a polynomial coefficient. By Wick's theorem the product is the finite sum over contractions; for two fields there is one contraction and one uncontracted term, and the Taylor expansion of the uncontracted normal product about the midpoint produces, at order $ξ^{k}$ , exactly the operators $: (\partial^{a} ϕ) (\partial^{b} ϕ) :$ with $a + b = k$ , a finite set. No contraction generates a logarithm because the free propagator is a pure power. Logarithms enter only through loop integrals in the interacting theory, where they signal a non-zero anomalous dimension. $□$

Connections Master

Canonical quantum field theory 12.12.01. The OPE is a statement about products of the local field operators of a canonically quantised theory. The free-field expansion is computed entirely by the contractions of Wick's theorem applied to the operator-valued fields, and the Feynman propagator that appears as the leading coefficient is the basic two-point function of the canonical formalism. Without the operator structure of the fields and their propagators the expansion has no objects to expand.
Vacuum polarization at one loop and the Uehling potential 12.16.03. The short-distance logarithms that the OPE coefficients resum are the same logarithms that appear in the one-loop vacuum polarisation: both are the position-space and momentum-space faces of the running of couplings and operator dimensions. The Uehling correction is the leading short-distance modification of the photon propagator, and the OPE organises exactly this class of short-distance corrections for arbitrary operator products.
Beta function (renormalisation group) 08.04.03. The Callan-Symanzik equation for the Wilson coefficients is driven by the same beta function that governs the running coupling. The anomalous-dimension matrix of the composite operators sits alongside $β (g)$ in the renormalisation-group flow, and the eigenvalues of that matrix are the powers of the logarithm by which the coefficient functions scale. The OPE is the place where the beta function and the operator anomalous dimensions act together.
Effective field theory 08.08.03. The OPE is dual to the Wilsonian effective-action picture: integrating out short-distance modes generates a tower of local operators ordered by dimension, exactly the operators that appear in the expansion of a short-distance product, with the high-dimension (irrelevant) operators suppressed by powers of the separation. The Wilson coefficients of the OPE are the matching coefficients of the effective theory; the two constructions are one short-distance organisation viewed from operator-product and effective-Lagrangian sides.
Asymptotic freedom and the running gauge coupling 12.18.03. In an asymptotically free gauge theory the coupling weakens at short distances, so the OPE coefficient functions are computable in perturbation theory there, and their logarithmic running is the power-of-logarithm scaling derived from the negative beta function. The leading-twist anomalous dimensions of the deep-inelastic operators are the gauge-theory analogue of the running coupling, and the scaling violations they predict are the experimental signature that confirmed both asymptotic freedom and the OPE.

Historical & philosophical context Master

Kenneth Wilson introduced the operator product expansion in his 1969 paper Non-Lagrangian Models of Current Algebra (Phys. Rev. 179, 1499) ^{[Wilson 1969]}, where he postulated that the product of two local operators at short separation expands in a series of local operators whose c-number coefficient functions carry the singular separation dependence, with the power of the singularity fixed by operator dimensions that need not equal the free-field values. The motivation was current algebra: Wilson sought a framework for the short-distance behaviour of products of currents that did not depend on a specific Lagrangian, and the expansion let him define operator dimensions as dynamical quantities. The proposal that dimensions could differ from their canonical values — anomalous dimensions — was the conceptual core, and it connected directly to Wilson's contemporaneous work on the renormalisation group and critical phenomena.

The perturbative construction was completed by Wilson and Wolfhart Zimmermann in Operator product expansions and composite field operators in the general framework of quantum field theory (Comm. Math. Phys. 24, 87, 1972) ^{[Wilson-Zimmermann 1972]}, building on Zimmermann's BPHZ definition of normal products of composite operators (the 1970 Brandeis lectures). They proved the expansion as an asymptotic statement to all orders, with the precise remainder bound, and matched the coefficient functions to renormalised Green functions. The renormalisation-group equation for the coefficients, with the anomalous-dimension matrix of the mixing operators, followed from Callan's and Symanzik's 1970 equations applied to the coefficient functions; the asymptotic-freedom discovery of Gross, Wilczek, and Politzer in 1973 made the coefficients computable at short distance in QCD.

The decisive application was deep-inelastic scattering. The light-cone OPE for two electromagnetic currents, organised by twist rather than dimension, identified the moments of the structure functions with products of Wilson coefficients and twist-two operator matrix elements, turning Bjorken scaling and its logarithmic violations into a controlled prediction. Christ, Hasslacher, and Mueller (Phys. Rev. D 6, 3543, 1972) and Georgi and Politzer (Phys. Rev. D 9, 416, 1974) developed the moment analysis that connected the OPE to the parton model and the evolution equations. The expansion has since become the operative tool of QCD sum rules (Shifman, Vainshtein, Zakharov 1979) and, with the operator coefficients reinterpreted as dynamical data, of the conformal bootstrap.

Bibliography Master

@article{Wilson1969OPE,
  author  = {Wilson, Kenneth G.},
  title   = {Non-Lagrangian Models of Current Algebra},
  journal = {Phys. Rev.},
  volume  = {179},
  year    = {1969},
  pages   = {1499--1512}
}

@article{WilsonZimmermann1972,
  author  = {Wilson, Kenneth G. and Zimmermann, Wolfhart},
  title   = {Operator product expansions and composite field operators in the general framework of quantum field theory},
  journal = {Comm. Math. Phys.},
  volume  = {24},
  year    = {1972},
  pages   = {87--106}
}

@incollection{Zimmermann1970,
  author    = {Zimmermann, Wolfhart},
  title     = {Local Operator Products and Renormalization in Quantum Field Theory},
  booktitle = {Lectures on Elementary Particles and Quantum Field Theory (Brandeis 1970)},
  publisher = {MIT Press},
  year      = {1970}
}

@article{ChristHasslacherMueller1972,
  author  = {Christ, Norman and Hasslacher, Brosl and Mueller, Alfred H.},
  title   = {Light-Cone Behavior of Perturbation Theory},
  journal = {Phys. Rev. D},
  volume  = {6},
  year    = {1972},
  pages   = {3543--3562}
}

@article{GeorgiPolitzer1974,
  author  = {Georgi, Howard and Politzer, H. David},
  title   = {Freedom at Moderate Energies: Masses in Color Dynamics},
  journal = {Phys. Rev. D},
  volume  = {9},
  year    = {1974},
  pages   = {416--420}
}

@article{ShifmanVainshteinZakharov1979,
  author  = {Shifman, M. A. and Vainshtein, A. I. and Zakharov, V. I.},
  title   = {QCD and Resonance Physics. Theoretical Foundations},
  journal = {Nucl. Phys. B},
  volume  = {147},
  year    = {1979},
  pages   = {385--447}
}

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

@book{Collins1984Renormalization,
  author    = {Collins, John C.},
  title     = {Renormalization},
  publisher = {Cambridge University Press},
  series    = {Cambridge Monographs on Mathematical Physics},
  year      = {1984}
}

@book{PeskinSchroeder1995,
  author    = {Peskin, Michael E. and Schroeder, Daniel V.},
  title     = {An Introduction to Quantum Field Theory},
  publisher = {Addison-Wesley},
  year      = {1995}
}

@book{ItzyksonZuber1980,
  author    = {Itzykson, Claude and Zuber, Jean-Bernard},
  title     = {Quantum Field Theory},
  publisher = {McGraw-Hill},
  year      = {1980}
}

Prerequisites

12.12.01
12.16.03
08.04.03
08.08.03
08.10.04

Tier anchors

beginner: Wilson, K. G., 'Non-Lagrangian Models of Current Algebra', *Phys. Rev.* 179, 1499 (1969) — conceptual opening on the short-distance product of two local operators; Coleman, S., *Aspects of Symmetry* (Cambridge, 1985), Ch. 3 (dilatations and scaling, popular-to-graduate framing)
intermediate: Peskin, M. E. & Schroeder, D. V., *An Introduction to Quantum Field Theory* (Addison-Wesley, 1995), §18.3–18.5 (the OPE and deep-inelastic scattering); Itzykson, C. & Zuber, J.-B., *Quantum Field Theory* (McGraw-Hill, 1980), §8-4, §13-2 (short-distance expansion; Wilson coefficients)
master: Weinberg, S., *The Quantum Theory of Fields, Vol. 2: Modern Applications* (Cambridge, 1996), Ch. 20 (the operator product expansion; Wilson coefficients and their renormalisation-group running; deep-inelastic sum rules); Collins, J. C., *Renormalization* (Cambridge, 1984), Ch. 10 (the OPE as an operator statement and its proof in perturbation theory); Zimmermann, W., 'Local Operator Products and Renormalization in Quantum Field Theory', in *Lectures on Elementary Particles and Quantum Field Theory* (MIT Press, 1970)

References

Weinberg, S. — The Quantum Theory of Fields, Vol. 2: Modern Applications (Cambridge, 1996) · Ch. 20 §20.1 (the operator product expansion $A(x)B(y) \to \sum_n C^n_{AB}(x-y)\mathcal{O}_n(\tfrac{x+y}2)$ and the scaling of $C_n$ by dimension counting); §20.2 (momentum-space OPE and the renormalisation-group equation for the Wilson coefficients $[\mu\partial_\mu + \beta\partial_g - \gamma_A - \gamma_B + \gamma_n]C_n = 0$); §20.3 (application to deep-inelastic scattering, moments of structure functions, and the leading-twist dominance)
Wilson, K. G. — Non-Lagrangian Models of Current Algebra · Phys. Rev. 179, 1499 (1969) — the postulate that the product of two local field operators at short separation expands in a series of local operators with c-number coefficient functions singular as a power of the separation fixed by dimensions; the application to current algebra and the definition of operator dimensions away from the free-field values
Wilson, K. G. & Zimmermann, W. — Operator product expansions and composite field operators in the general framework of quantum field theory · Comm. Math. Phys. 24, 87 (1972) — the perturbative construction of the expansion to all orders, the precise asymptotic statement (the remainder after $N$ terms is suppressed by extra powers of the separation), and the matching to Zimmermann's normal-product (BPHZ) definition of composite operators
Peskin, M. E. & Schroeder, D. V. — An Introduction to Quantum Field Theory (Addison-Wesley, 1995) · §18.3 (the OPE for the product of two electromagnetic currents); §18.4–18.5 (moments of deep-inelastic structure functions, the twist expansion, the leading-twist operators and their anomalous dimensions, scaling violations)
Collins, J. C. — Renormalization (Cambridge, 1984) · Ch. 10 (the operator product expansion as an operator-valued asymptotic statement, the proof by the renormalisation-group analysis of the relevant 1PI functions, and the role of the anomalous dimensions of the composite operators in the running of the coefficient functions)
Itzykson, C. & Zuber, J.-B. — Quantum Field Theory (McGraw-Hill, 1980; Dover reprint, 2005) · §8-4 (the short-distance expansion of operator products and the singular coefficient functions); §13-2 (Wilson coefficients, the OPE applied to deep-inelastic scattering and the parton model)

Estimated time

beginner: 16m
intermediate: 38m
master: 80m