Renormalization of singular stochastic products

For a centered scalar Gaussian Z of variance C>0, define $:Z^2:=Z^2-C$ and $:Z^3:=Z^3-3CZ$. The quadratic expression is centered. The cubic expression is centered too, but that fact alone does not explain its correction, since an uncorrected centered Gaussian cube already has mean zero. Instead, the correction removes the projection of the cube onto the linear Gaussian component. Gaussian moment identities verify the orthogonality explicitly.

The resulting polynomials belong to distinct Gaussian chaos orders and have variances two C squared and six C cubed. These variances grow with C. Therefore Wick subtraction does not make a sequence of scalar variables with variance tending to infinity converge at each point. In a random field, the useful limit can exist only after considering spatial or space-time testing and the correlations between nearby points. This is a distributional statement that requires estimates beyond the one-variable polynomial identities.

For a stationary massive linear heat field on the two-dimensional torus, smooth the field at spatial scale ε and let $C_\varepsilon$ be its point variance. With a fixed Fourier and mollifier convention, this variance grows logarithmically as ε decreases. Wick powers subtract the corresponding covariance contractions. A cubic stochastic quantization equation is then approximated with a compensating linear drift, schematically $\Delta u_\varepsilon-u_\varepsilon^3+(3C_\varepsilon-m)u_\varepsilon+\xi_\varepsilon$ for positive reference mass m.

This schematic display records the sign and algebra of the cubic correction; it is not a theorem for arbitrary mollifiers, initial distributions or solution topologies. One must choose a consistent reference field and approximation, construct the required Wick distributions, and solve the remainder equation. In three-dimensional cubic quantization, further divergent contractions occur, so replacing the cube by its first Gaussian Wick correction alone does not complete the construction. Additional equation-dependent counterterms and enhanced distributions are needed.

Counterterms are derived from the covariance and singular contractions of the chosen approximation; they are not arbitrary empirical adjustments intended to make a plotted solution look stable. Divergent constants may depend on the mollifier, while an appropriately calibrated limiting equation can be independent of that choice. Finite changes in the counterterms can still alter effective coefficients such as a mass. An assertion of universality must specify which finite parameters are held fixed across approximation schemes.

The analytic solution map is usually made continuous in an enhanced noise topology containing the corrected products. Probability then establishes convergence of that enhanced input, and continuity transfers it to the solution, often locally in time before a possible blow-up criterion. Global existence, invariant measures and approximation rates are additional results. A finite-dimensional Gaussian calculation or a deterministic quadratic animation can clarify the algebra, but cannot certify the convergence or long-time behavior of a singular stochastic field.