Field Note: The Shell
The shell. This is where the cutting happens. Not in some abstract UV limit. Not in the infrared where everything is smooth and dilute. The shell is where you actually do the work. It is a shell in momentum space, centered at the cutoff $\Lambda$, with thickness $\Lambda/b - \Lambda$ — a thin annulus of momenta you are about to integrate out. You look at your field $\phi(x)$, you Fourier transform it, and you see that every mode is either inside the shell or outside it. The ones inside are about to vanish. The ones outside will remember them.
Here is what a shell integration looks like. You have your Wilsonian effective action $S_\Lambda[\phi_s]$, defined at cutoff $\Lambda$. You introduce a slower field $\phi_s$ (momenta $|p| < \Lambda/b$) and a faster field $\phi_f$ (momenta in the shell $\Lambda/b < |p| < \Lambda$). The full field is $\phi = \phi_s + \phi_f$. The action splits:
$S_\Lambda[\phi_s + \phi_f] = S_\Lambda[\phi_s] + \int \phi_f \frac{\delta S}{\delta \phi}\bigg|{\phi_s} + \frac{1}{2} \int \phi_f \frac{\delta^2 S}{\delta \phi^2}\bigg|{\phi_s} \phi_f + \dots$
Now you integrate over $\phi_f$. The result is an effective action for $\phi_s$ alone, but with modified couplings. This is the Wilsonian step: integrate out a shell, rescale, read off the flow. It is iterative. It is mechanical. It is how you turn a UV theory into an IR one without losing a single bit of information that matters.
The shell is thin because you want to compute the flow infinitesimally. Make it thin enough and the change in couplings is small: $g(\Lambda - d\Lambda) = g(\Lambda) + \beta(g) \frac{d\Lambda}{\Lambda}$. The beta function emerges from the shell integration. And the shell, being thin, means you can trust your perturbative expansion even in strongly coupled theories — because at any given scale, only a little bit of new physics has been integrated out. A small correction. A small adjustment. The theory breathes.
What lives in the shell? Modes that are correlated. Modes that, when you trace over them, generate new interactions. A $\phi^4$ vertex in the original action generates, through shell integration, corrections to the mass term, corrections to the $\phi^4$ coupling, and — if you are lucky — new terms like $(\partial \phi)^2 \phi^2$ that were not in the original action but are generated by the integrating out. These are the induced interactions. They are the fossil record of the modes you removed.
The shell is where Wilson's insight becomes computable. You do not integrate out all the UV at once — that is impossibly hard and conceptually messy. You do it shell by shell. A little bit at a time. And in the limit of infinitesimally thin shells, you get differential equations for the couplings. Flow equations. Renormalization group equations. The continuous version of a discrete process. You took a continuum of momentum modes and sliced them into infinitesimal shells, and each shell left its fingerprint on the effective action.
Field note: the shell is also where intuition lives. You can feel what happens. A heavy particle of mass $M \gg \Lambda$ — it lives in the shell. When you integrate it out, you generate contact interactions at scale $\Lambda$, suppressed by powers of $M$. A photon coupled to a charged particle — the shell generates vacuum polarization, a correction to the photon propagator. The shell is the mechanism by which heavy physics talks to light physics. It is a conversation, mediated by integration. The heavy modes speak, the light modes listen, and the coupling constants change.
This is the Wilsonian step. Not a limit. Not an approximation. A step. You lower the cutoff by a factor $b$, you integrate out the shell, you rescale, you repeat. Step by step, shell by shell, you flow from the UV to the IR. And at each step, you ask: what has changed? The answer is always the same: the couplings have shifted. The theory has adjusted. The world has been coarsened. And you are one step closer to understanding what the theory looks like at large distances.
The shell is small. But it contains everything.