Subtract short-distance subgraphs before the overall divergence

An ultraviolet-divergent subgraph arises when some loop momenta become large while the remainder is held fixed. Nested divergences have one such subgraph inside another; disjoint ones involve separate subgraphs; overlapping ones share lines without either containing the other. Each describes a different potentially singular region. A counterterm for a proper subgraph replaces that subgraph by its local ultraviolet approximation, while its external legs remain connected to the rest of the graph. Consequently the inserted contribution can depend nonlocally on the full graph’s external momenta.

This nonlocal dependence of an inserted graph is compatible with a local counterterm vertex. Locality constrains the replacement itself, not the complete surrounding propagation. A pole multiplied by ln(Q/μ) in an unsubtracted two-loop expression is therefore not permission to add a logarithmic momentum-dependent vertex to the Lagrangian. It is often evidence that a subdivergence has not yet been removed. One must also separate ultraviolet from infrared poles; dimensional regularization can display both as poles in the same ε, and a scaleless zero can hide their cancellation.

For disjoint subgraphs, subtracting either one alone is insufficient because their common ultraviolet region would be counted incorrectly. The forest expansion includes the product of their counterterms. Nested graphs require the inner subtraction before the outer remaining ultraviolet part is identified. Overlapping subgraphs cannot simply be put simultaneously into one forest, since their contractions are incompatible; the forest organization sums the permitted alternatives. This combinatorial structure is what replaces an informal instruction to subtract every visible loop independently.

After proper subdivergences are subtracted, the remaining overall ultraviolet divergence is a polynomial in external momenta and masses of bounded degree, under the assumptions of perturbative locality and power counting. It can be cancelled by an allowed local interaction. This need not mean there is only a simple 1/ε pole: higher poles can occur, with consistency relations tying them to lower-order data. Nonrenormalizable effective theories remain systematic when the required local operators are organized at a specified order in the low-energy expansion.

A lower-order counterterm is fixed once and then reused wherever that interaction appears. Its coefficient cannot be retuned separately for each larger diagram. This reuse is a strong check on signs, symmetry factors, and field-normalization conventions. Gauge identities further correlate counterterms in gauge theories. If a nonlocal ultraviolet pole survives after the complete set of insertions, likely causes include a missing graph, a mistaken combinatorial factor, a mixed convention, or infrared contamination; a new nonlocal subtraction is not the routine remedy.

The waves lab illustrates cancellation of a shared functional shape. Reset and set both controls to phase difference zero and amplitude ratio one, predicting summed amplitude two. Reset and set phase difference 180 degrees and ratio one, predicting cancellation of the entire profile. This resembles an insertion cancelling a nonlocal surrounding factor because both contributions contain it. These equal-frequency classical waves contain no subgraphs, momentum integrals, or ultraviolet regulator. Their exact cancellation illustrates an algebraic requirement, while locality and the forest rules supply the field-theory content.