Functional-RG Applications and Cross-Checks

The functional renormalization group provides an exact one-parameter equation for a scale-dependent effective action. Exactness belongs to the unprojected functional flow. An application replaces that functional by a finite operator or momentum ansatz, chooses projection conditions, and integrates a regulator-dependent finite system. Physical results should become insensitive to those choices only along a demonstrated convergence sequence.

Required background. Functional-RG truncations and projection methods supplies the exact flow and theory-space language. Helpful background. Closure, symmetry constraints, and branch selection supplies the finite-ansatz and residual record.

Exact flow and its boundary data

Add an infrared regulator

$$\Delta S_k[\varphi] =\frac12\int_q \varphi(-q)R_k(q)\varphi(q)$$

and define the effective average action $\Gamma_k$ by a modified Legendre transform. For bosonic fields,

$$\partial_k\Gamma_k[\phi] =\frac12\operatorname{Tr} \left[ \left( \Gamma_k^{(2)}[\phi]+R_k \right)^{-1} \partial_kR_k \right].$$

The equation is exact when the complete functional $\Gamma_k$ is retained. It is a one-loop trace with the full field-dependent inverse propagator, not a one-loop approximation. Wetterich derived the form and its interpolation between microscopic and full effective actions Wetterich 1993, Eqs. (1)–(7).

A regulator family must satisfy, in the intended sense,

$$R_k(q)\gtrsim k^2 \quad(q^2\ll k^2), \qquad R_k(q)\to0 \quad(q^2\gg k^2), \qquad R_{k\to0}(q)\to0.$$

The ultraviolet condition $\Gamma_{k=\Lambda}$ also requires a microscopic action and regulator matching. At finite $\Lambda$, irrelevant operators generated above the starting scale can matter unless they are included or bounded.

O(N) effective potential projection

For $\rho=\phi^a\phi^a/2$, choose the derivative-expansion ansatz

$$\Gamma_k[\phi] =\int\mathrm d^d x \left[ U_k(\rho) +\frac12Z_k(\rho) \partial_\mu\phi^a\partial_\mu\phi^a +O(\partial^4) \right].$$

For a constant background, the transverse and radial inverse propagators are

$$\begin{aligned} P_T(q;\rho) &=Z_kq^2+R_k(q)+U_k'(\rho),\\ P_L(q;\rho) &=Z_kq^2+R_k(q)+U_k'(\rho)+2\rho U_k''(\rho). \end{aligned}$$

Projecting the exact flow on constant fields gives

$$\partial_kU_k(\rho) =\frac12\int_q\partial_kR_k(q) \left[ \frac{N-1}{P_T(q;\rho)} +\frac{1}{P_L(q;\rho)} \right].$$

This equation is exact only as the constant-field projection of the exact $\Gamma_k^{(2)}$. Substituting a field-independent $Z_k$, truncating $U_k$ at finite polynomial order, or dropping $O(\partial^4)$ terms makes it a finite approximation.

Dimensionless variables expose a fixed point. Critical exponents are eigenvalues of the linearized flow about that fixed point. Their digits depend on field-expansion order, derivative order, projection momenta, and regulator until convergence is demonstrated. The $O(N)$ potential flow, derivative expansion, and associated approximation limits are reviewed in Berges, Tetradis, and Wetterich 2002, §§ 2–3.

Regulator and projection variation

At infinite theory-space resolution, admissible regulator choices give the same physical effective action at $k=0$. At finite truncation, residual regulator dependence is a diagnostic. A useful comparison varies:

• regulator shape inside an admissible family;
• the normalization convention for $Z_k$;
• polynomial versus grid representations of $U_k$;
• expansion about $\rho=0$ versus the running minimum;
• derivative order and momentum-dependent vertices;
• projection momentum and field point.

The principle of minimal sensitivity can identify a locally stationary regulator parameter, but stationarity in one family is not an error bound. Agreement with a second truncation basis or an independent method is stronger.

Morris analyzes derivative expansions and regulator dependence at finite order Morris 1994, §§ 2–4.

Symmetry and convexity checks

An $O(N)$-invariant ansatz preserves the global symmetry manifestly, but a field or momentum truncation can still violate relations among vertices. In gauge theories, $R_k$ breaks standard BRST symmetry at intermediate $k$ and produces modified Ward or Slavnov–Taylor identities. A projected flow must track their residuals or include symmetry-restoring counterterms.

As $k\to0$, the exact effective potential is convex. A finite polynomial expansion around one minimum can hide the flattening of a coexistence region. Failure to reach convexity may indicate insufficient infrared resolution, a stopped flow, or a basis unable to represent the solution; it is not automatically a new metastable phase.

Cross-checks and justified digits

For an $O(N)$ critical exponent or amplitude ratio, a defensible report includes:

1. regulator-family and parameter variation;
2. field and derivative-order sequences;
3. grid, domain, and integrator refinement;
4. perturbative $\epsilon$ expansion, large-$N$, or exactly known limiting cases;
5. comparison with Monte Carlo or conformal-bootstrap data using matched definitions.

Correlated inputs must stay correlated. Fitting several microscopic couplings to the same observables and then varying each independently generally overstates or understates the uncertainty.

The functional-equation closure and validation map locates ansatz and projection. The functional-method validation comparison states the regulator, convergence, benchmark, covariance, and digit requirements.

Common pitfalls

Calling the projected flow exact. The Wetterich equation is exact; a finite derivative or vertex expansion is not.

Using one optimized regulator as convergence. Optimization can improve one truncation. It does not replace regulator-family, basis, and order variation.

Reporting a fixed-point digit from solver precision. An integrator can solve a truncated flow to ten digits while the derivative expansion controls only three.

Exercises

1. Derive the transverse and longitudinal curvature masses from $U_k(\rho)$.

Solution

Since

$$\frac{\partial^2U}{\partial\phi^a\partial\phi^b} =U'(\rho)\delta_{ab} +U''(\rho)\phi_a\phi_b,$$

vectors orthogonal to $\phi^a$ have eigenvalue $U'$, with multiplicity $N-1$. The vector parallel to $\phi^a$ has eigenvalue $U'+2\rho U''$. Adding the kinetic and regulator terms gives $P_T$ and $P_L$.

2. A critical exponent changes by $10^{-5}$ under tighter ODE tolerances but by $3\times10^{-2}$ between derivative orders. Which variation controls the quoted digits?

Solution

The numerical integration error is small, but the truncation sequence changes the result by $3\times10^{-2}$. Unless higher orders show a controlled convergence pattern, digits below that scale are not justified. Solver precision cannot be used as the scientific uncertainty.

Continue

Coupled Propagator, Vertex, and Bound-State Systems treats dependencies shared with other functional equations. Functional-Method Validation and Error Control turns variation sequences into a bounded uncertainty statement.

References

• Berges, Jürgen, Nikolaos Tetradis, and Christof Wetterich. “Non-Perturbative Renormalization Flow in Quantum Field Theory and Statistical Physics.” Physics Reports 363 (2002): 223–386. DOI.
• Morris, Tim R. “Derivative Expansion of the Exact Renormalization Group.” Physics Letters B 329 (1994): 241–248. DOI.
• Wetterich, Christof. “Exact Evolution Equation for the Effective Potential.” Physics Letters B 301 (1993): 90–94. DOI.