Epstein–Glaser Induction

Epstein–Glaser induction constructs the $n$th time-ordered product first on configuration space away from the total diagonal, where causal factorization reduces it to already known lower orders. A causal cover and partition of unity glue these definitions consistently. Renormalization is the final extension across the diagonal; the induction proves neither uniqueness of that extension nor convergence of the perturbation series.

Required background. Causal factorization supplies the induction equation, and microcausal functionals and Peierls brackets supply the admissible functional domain. Helpful background. Wightman fields and common domains provide the free Wick fields, while isotony, additivity, and primitive causality explain the local-net target.

Induction off the total diagonal

Assume symmetric maps $T_k$ have been constructed for $k<n$ and satisfy causal factorization. Let

$$D_n=\{(x_1,\ldots,x_n)\in M^n:x_1=\cdots=x_n\}$$

be the total diagonal. For every nonempty proper subset $I\subset\{1,\ldots,n\}$, let $C_I$ be the open set on which the cluster $x_I$ is not earlier than $x_{I^c}$. On $C_I$ define

$$T_n^{I}=T_{|I|}(x_I)\star T_{|I^c|}(x_{I^c}).$$

The sets $C_I$ cover $M^n\setminus D_n$ on a globally hyperbolic spacetime: away from total coincidence one can find a Cauchy-time separation of a nontrivial cluster. On overlaps, repeated causal factorization and the induction hypothesis show $T_n^I=T_n^J$. A smooth partition of unity subordinate to the cover therefore glues them to a distribution $T_n^0$ on $M^n\setminus D_n$ independent of the chosen partition.

Partial diagonals cause no new independent problem at this stage. Near a configuration with several coincident clusters but not total coincidence, the lower-order $T_{|I|}$ already include their prescribed extensions. The only missing data are supported on $D_n$. This causal-cover construction is the core of Epstein and Glaser 1973, §§3–4, pp. 223–253, and its functional formulation is summarized in Brunetti, Dütsch, and Fredenhagen 2009, §4.1, pp. 1557–1563.

At $n=3$, for example, the region where $x_1$ is later than $\{x_2,x_3\}$ gives $T_1(x_1)\star T_2(x_2,x_3)$. Where $\{x_1,x_2\}$ is later than $x_3$, it gives $T_2(x_1,x_2)\star T_1(x_3)$. On their overlap, either $x_1$ or $x_2$ can be separated once more, and associativity reduces both expressions to the same product of three $T_1$ factors in the corresponding order. If $x_2=x_3\ne x_1$, the already-renormalized $T_2(x_2,x_3)$ is used; only $x_1=x_2=x_3$ remains missing. This concrete case displays why the induction resolves subdivergences before the new total-diagonal extension.

Second order for φ⁴

For $\mathcal L(x)=\lambda:\!\phi(x)^4\!:/4!$ in four-dimensional Minkowski space, Wick expansion gives the off-diagonal product

$$T_2^0(\mathcal L(x),\mathcal L(y)) =\frac{\lambda^2}{(4!)^2} \sum_{k=0}^{4}\binom4k^2k!\, H_F(x,y)^k :\!\phi(x)^{4-k}\phi(y)^{4-k}\!:$$

for $x\neq y$, where $H_F$ is the chosen Feynman two-point distribution and its convention absorbs the contraction factors. On the region $x$ later than $y$, this equals $\mathcal L(x)\star\mathcal L(y)$; on the reverse region it equals the reversed product. Their difference is the causal distribution whose retarded/advanced splitting produces the same off-diagonal $T_2^0$.

The numerical distributions $H_F^k$ are the quantities to extend to $x=y$. Translation covariance reduces the problem to the relative coordinate $z=x-y$. Each extension may differ by derivatives of $\delta(z)$ allowed by scaling degree, Lorentz covariance, field parity, and the declared normalization conditions. The next step does not choose these coefficients; it merely records them as local data. This is the construction underlying renormalized contact products of composite operators.

At $k=2$, for example, $H_F^2$ has scaling degree four in four relative dimensions, so a scalar delta term is possible. At $k=4$, the larger degree permits more derivatives, but multiplication by the remaining Wick monomial and dimensional constraints organize the result into local composite counterterms. The exact enumeration belongs to the scaling-degree theorem.

An independent consistency check restricts the glued $T_2^0$ back to either causal region. Only one partition function is then needed, and the result reduces to the required ordered product. Thus the gluing has introduced no extra off-diagonal freedom.

Symmetrizing the glued distribution does not add freedom either: permutation covariance of the cover and symmetry of all lower $T_k$ make the symmetrized distribution agree with $T_n^0$ off $D_n$. Any difference after extension is therefore diagonal and belongs to the separately constrained renormalization freedom.

Consistency across orders

Adversarial test. Suppose $T_3$ is extended independently without requiring its restriction to each $C_I$ to equal the product of the already fixed $T_1$ and $T_2$. On an overlap where $x_1$ is later than $x_2,x_3$, causal factorization demands

$$T_3(x_1,x_2,x_3)=T_1(x_1)\star T_2(x_2,x_3).$$

An arbitrary added term not supported on the total diagonal changes this restriction and therefore changes the third derivative of $S(F+H)-S(F)S(H)$ on causally ordered supports. The local $S$-matrix ceases to factorize. Terms supported only on the total diagonal may remain as renormalization freedom, but they must still satisfy symmetry, covariance, and lower-order normalization identities.

Exercises

1. Cover at second order. Identify the two causal regions covering $M^2\setminus D_2$ and show their definitions agree on spacelike separation.

Solution

The regions are $x\notin J^-(y)$ and $y\notin J^-(x)$. Their overlap is spacelike separation. Locality makes $\mathcal L(x)\star\mathcal L(y)=\mathcal L(y)\star\mathcal L(x)$ there, so the two definitions glue.

2. Count contractions. Verify the coefficient of the $H_F^2:\!\phi^2(x)\phi^2(y)\!:$ term.

Solution

Choose two of four fields at $x$ and two of four at $y$, then pair them in $2!$ ways. The coefficient is $\binom42^2 2!=6^2\cdot2=72$, before the overall $(4!)^{-2}$ factor.

References

• Brunetti, Romeo, Michael Dütsch, and Klaus Fredenhagen. “Perturbative Algebraic Quantum Field Theory and the Renormalization Groups.” Advances in Theoretical and Mathematical Physics 13 (2009): 1541–1599. DOI; Open PDF.
• Epstein, Henri, and Vladimir Glaser. “The Role of Locality in Perturbation Theory.” Annales de l’Institut Henri Poincaré A 19 (1973): 211–295. EuDML record and PDF.