The Operator Product
When two local operators sit close together in space — closer than the typical correlation length of the theory, but farther apart than the microscopic cutoff — something remarkable happens. They multiply. Not in the elementary sense of arithmetic, but in the structural sense that defines what a quantum field theory is. The operator product expansion, discovered by Kenneth Wilson, states that the product of two operators at nearby points can be expanded as a sum of single operators at one of those points, with coefficient functions that depend on the separation.
$$\mathcal{O}_i(x) \mathcal{O}j(0) = \sum_k C{ij}^k(x) \mathcal{O}_k(0)$$
This is not an approximation. In a conformal field theory, it is an exact operator identity. In a general QFT, it is an asymptotic expansion valid as $x \to 0$. The coefficient functions $C_{ij}^k(x)$ are c-numbers — they carry no operator indices, no quantum fluctuations. All the operator content is in the $\mathcal{O}_k$. The coefficient functions are determined by the dynamics and by the symmetry of the theory. They encode the short-distance physics. The operators encode the long-distance physics. This separation is the power of the OPE.
Consider two scalar fields $\phi(x)\phi(0)$. As $x \to 0$, the product can be expanded in terms of the identity operator, the operator $\phi^2$, the stress tensor, and an infinite tower of higher-dimension operators. The leading term is the identity, with coefficient function proportional to $|x|^{-2\Delta_\phi}$. This is the singular part — the vacuum expectation value $\langle \phi(x)\phi(0) \rangle$ that diverges as the points approach. The next term involves $\phi^2(0)$, with coefficient function $|x|^{\Delta_{\phi^2} - 2\Delta_\phi}$, which is less singular because the dimension of $\phi^2$ is greater than $2\Delta_\phi$ (at least in any interacting theory). And so on, to an infinite tower.
The OPE is not just a formal device. It is the organizing principle of QFT. It tells you that at short distances, the theory becomes local in a very precise sense: any complicated product of operators can be replaced by a sum of simpler ones. This is why renormalization works. The divergences that appear in perturbation theory are the singularities of the OPE coefficient functions, and the renormalization program is the process of subtracting these singularities order by order.
In perturbation theory, the OPE can be computed diagram by diagram. You calculate the correlation functions $\langle \mathcal{O}i(x) \mathcal{O}j(0) \mathcal{O}k(0) \mathcal{O}{m_1}(y_1) \cdots \mathcal{O}{m_N}(y_N) \rangle$, and the OPE coefficients $C{ij}^k(x)$ are the coefficients that reproduce these correlation functions when you insert the OPE into the correlation function. By the completeness of the operator basis, if the OPE reproduces all correlation functions, it is correct.
The OPE has a radius of convergence. Inside this radius — the "OPE window" — the expansion converges. Outside it, the expansion diverges (at best, it is asymptotic). The radius is determined by the nearest singularity in the complexified position space, which in practice is set by the distance to the nearest other operator insertion or to the boundary of the spacetime. In Euclidean QFT, the OPE converges absolutely when $|x| < \min(|y_i|)$, i.e., when the separation of the two operators is smaller than the distance to any other insertion.
The OPE is associative. This is a deep property. It means that the result of expanding three operators $\mathcal{O}_i(x_1)\mathcal{O}_j(x_2)\mathcal{O}_k(x_3)$ does not depend on whether you first expand the pair $(i,j)$ and then expand the result with $k$, or first expand $(j,k)$ and then expand $i$ with the result. The associativity of the OPE translates into the conformal bootstrap equations in a CFT — the crossing symmetry constraints that we will discuss elsewhere. In a general QFT, associativity is the foundation of the renormalization group: the RG flow of the OPE coefficients is consistent because the OPE is associative at every scale.
The OPE is also the key to understanding operator mixing. When you compute correlation functions at short distances, operators of the same quantum numbers mix under renormalization. The OPE makes this mixing manifest: the coefficient functions of mixed operators are coupled, and diagonalizing the mixing matrix is equivalent to finding the eigen-operators of the dilatation operator — the operators with definite scaling dimension.
In AdS/CFT, the OPE has a beautiful holographic interpretation. A bulk field $\Phi(z,x)$, where $z$ is the radial AdS coordinate, has a boundary limit that is a CFT operator. As two boundary operators approach each other, the dual bulk process is two strings coming together, interacting in the bulk, and producing a single string that propagates back. The OPE coefficients are bulk interaction vertices. The tower of operators in the OPE is the tower of Kaluza-Klein modes or string excitations in the bulk. The OPE is the boundary manifestation of locality in the bulk.
The OPE is the structure that makes QFT calculable. Without it, we would have to deal with products of distributions at coincident points, which is mathematically meaningless. The OPE says: don't multiply distributions. Expand them. Replace the product with a sum of well-defined operators with c-number coefficients. This is the mathematical content of Wilson's insight, and it is the reason the OPE is one of the deepest structures in quantum field theory.
In the language of the renormalization group, the OPE is the statement that integrating out short-distance degrees of freedom generates an infinite series of local operators, each suppressed by powers of the short-distance scale. The coefficient functions are the Wilson coefficients, and they are universal — they depend only on the short-distance physics, not on the long-distance state or boundary conditions. This is the essence of effective field theory: the short-distance physics is encoded in a finite (or infinite, but systematically organized) set of Wilson coefficients multiplying a complete basis of local operators.