Momentum-shell renormalization for scalar models

Use a real scalar field with a reversal-symmetric local action $\mathcal S=F/(k_BT)$, so the statistical weight is $e^{-\mathcal S}$. The coefficient of the squared gradient is normalized to one half. A finite ultraviolet momentum cutoff represents the microscopic coarse-graining scale. Split the field into slow modes below $\Lambda/b$ and fast modes in the shell between that value and $\Lambda$, with $b>1$.

Define the effective slow-field action by integrating the fast fields in the partition function while holding the slow field fixed. Taking the negative logarithm generates connected cumulants. The fast Gaussian variance shifts the quadratic coupling at first order in the quartic interaction; connected second-order contractions also change the quartic coupling. Just setting fast amplitudes to zero would miss those terms and is therefore not the Wilson integration step.

After the shell integral, the largest retained momentum is smaller. Rescale coordinates by $x'=x/b$ and momenta by $q'=bq$ to restore the numerical cutoff. Rescale the field as $\phi'(x')=b^{(d-2)/2}\phi_<(x)$ at Gaussian order so the gradient term again has coefficient one half. This step gives tree-level transformations $r'=b^2r$ and $u'=b^{4-d}u$, before fluctuation corrections.

Introduce dimensionless mass $\bar r=r/\Lambda^2$ and coupling $g=K_d u\Lambda^{d-4}$, with $K_d$ the area of the unit sphere divided by $(2\pi)^d$. Let $\ell=\ln b$ increase toward longer physical length scales. In a local small-coupling truncation, a thin shell then produces explicit differential flows. Another quartic normalization or reversed flow direction changes the displayed coefficients and signs, so comparing formulas without translating conventions is unreliable.

At zero bare quadratic coefficient the shell correction is positive when u is positive. Thus the critical surface is not generally the bare equation r=0. One tunes a temperature-like scaling field relative to the fluctuation-shifted critical surface. The numerical critical bare mass depends on cutoff and scheme, whereas the properly extracted long-distance exponents can be universal. A negative fixed-point coordinate for the bare mass does not mean the physical critical susceptibility is negative.

Shell integration also generates higher powers and derivative interactions allowed by symmetry. Keeping only r and u is a truncation whose justification improves near the appropriate upper critical dimension and small coupling. Marginal or relevant omitted operators cannot be discarded merely because they were absent microscopically. The result is an approximate statistical transformation with an error hierarchy, not a general solver for every critical material or a numerical fit to a preferred curve.