Gradient-Flow Regularized Local Operators

• Gradient-flow regularized local operators are defined by evolving gauge fields via a diffusion process, yielding UV-finite, renormalization-independent observables.
• They enable precise matching between continuum renormalization and lattice computations, allowing accurate extraction of conserved currents like the energy-momentum tensor.
• The method uses small flow-time expansions and RG equations to compute operator coefficients, linking flowed observables to renormalized physical matrix elements.

Gradient-flow regularized local operators are composite or pointlike quantum field theory observables defined by evolving the fundamental field variables through a diffusion-type partial differential equation—termed the "gradient flow"—in a fictitious flow-time dimension before constructing local operator products. This procedure smooths ultraviolet (UV) fluctuations, rendering local products UV finite for any positive flow time and independent of the regularization scheme. The method provides a powerful bridge between continuum renormalization and non-perturbative regularization on the lattice, enabling precise definitions and computations of conserved currents such as the energy-momentum tensor, as well as a formalism for matching flowed operators to renormalized operators in the ultraviolet limit. The foundational analysis, which introduced many of the core features of this approach, appears in the context of Yang-Mills theory (Suzuki, 2013).

1. Gradient Flow Evolution and Fundamental Equations

The gradient flow is an auxiliary smoothing process applied to the gauge field $A_\mu(x)$, producing a "flowed" field $B_\mu(t,x)$ that evolves according to the parabolic PDE

$$\partial_t B_\mu(t,x) = D_\nu G_{\nu\mu}(t,x) + \alpha_0 D_\mu \partial_\nu B_\nu(t,x),\quad B_\mu(0,x) = A_\mu(x),$$

where $D_\nu$ is the covariant derivative with respect to $B_\mu(t,x)$, and $G_{\mu\nu}(t,x)$ is the associated field strength

$$G_{\mu\nu}(t,x) = \partial_\mu B_\nu(t,x) - \partial_\nu B_\mu(t,x) + [B_\mu(t,x), B_\nu(t,x)].$$

This equation is of diffusion type: in momentum space, UV components are exponentially suppressed by $\sim e^{-t k^2}$, so the flow acts as a gauge-covariant smearing with physical radius $\sqrt{8t}$.

Flowed composite operators, such as

$$U_{\mu\nu}(t,x) = G_{\mu\rho}^a(t,x) G_{\nu\rho}^a(t,x) - \tfrac{1}{4}\delta_{\mu\nu}G_{\rho\sigma}^a(t,x) G_{\rho\sigma}^a(t,x),$$

and the energy density

$$E(t,x) = \tfrac{1}{4}G_{\mu\nu}^a(t,x)G_{\mu\nu}^a(t,x),$$

are then constructed from these smooth fields.

2. UV Finiteness and Regularization Independence

For any positive flow time $t>0$, local products of flowed fields are free from the coincidence-point singularities that afflict conventional local products at vanishing separation. The exponential damping in the momentum representation ensures that correlation functions of the form $\langle U_{\mu\nu}(t,x)\cdots\rangle$ are ultraviolet finite, regardless of the regularization (lattice, dimensional regularization, etc.) used for the underlying theory. The UV finiteness is nonperturbative in nature, with the scale $\sqrt{8t}$ playing the role of a physical UV cutoff, which can be safely taken to zero after other limits. This property holds for any gauge-invariant local product constructed from the flow-evolved fields, as shown both formally and by explicit perturbative evaluation.

3. Small Flow-Time Expansion and Operator Matching

While flowed local operators at finite $t$ are well-defined and finite, physical observables correspond to $t\to 0$ (the original theory). The small flow-time expansion expresses flowed operators as a series in $t$ whose leading coefficients are renormalized local operators at $t=0$, with calculable, finite coefficient functions: $$U_{\mu\nu}(t,x) = c_T(t) \{T_{\mu\nu}(x)\}_R + c_S(t)\delta_{\mu\nu} \left\{ \tfrac{1}{4}F_{\rho\sigma}^aF_{\rho\sigma}^a(x) \right\}_R + \mathcal{O}(t),$$

$$E(t,x) = \langle E(t,x) \rangle + c_E(t)\left\{\tfrac{1}{4}F_{\rho\sigma}^aF_{\rho\sigma}^a(x)\right\}_R + \mathcal{O}(t).$$

The coefficients $c_T(t)$, $c_S(t)$, and $c_E(t)$ depend on both the flow time $t$ and the renormalized coupling $g$, and obey specific renormalization group equations: $$(\mu\partial_\mu + \beta\partial_g) c_T(t) = 0,\quad (\mu\partial_\mu + \beta\partial_g + \gamma_S)c_S(t) = 0.$$ The trace anomaly implements a relation among the coefficients:

$c_S(t) = \frac{\beta}{2g^3}c_T(t),$

encoding the breaking of classical scale invariance by quantum effects.

4. Construction of the Renormalized Energy-Momentum Tensor

A principal illustration is the explicit construction of the correctly normalized, conserved energy-momentum tensor in Yang-Mills theory purely in terms of flowed composite operators: $$\{T_{\mu\nu}(x)\}_R = \frac{1}{c_T(t)} U_{\mu\nu}(t,x) - \frac{c_S(t)}{c_T(t)c_E(t)}\delta_{\mu\nu}[E(t,x) - \langle E(t,x)\rangle] + \mathcal{O}(t).$$ As $t\rightarrow0$, the right-hand side becomes increasingly dominated by the underlying local operators, with the coefficients determined by a combination of RG flow and perturbative matching. An explicit one-loop computation yields the asymptotic relation

$$\{T_{\mu\nu}(x)\}_R \sim -2b_0[\ln(\sqrt{8t}\Lambda) + c_1]U_{\mu\nu}(t,x) + \frac{b_0}{2}[1 - \frac{c_1-c_2}{-\ln(\sqrt{8t}\Lambda)}]\delta_{\mu\nu}[E(t,x) - \langle E(t,x)\rangle],$$

with $b_0$ the first coefficient of the $\beta$-function and numerical constants $c_1,c_2$.

5. Computation and Extraction from Lattice Simulations

On the lattice, the gradient flow allows practical, nonperturbative computation of correlators of physically normalized composite operators. The procedure is:

• Simulate on a lattice with spacing $a$ much smaller than the flow radius $\sqrt{8t}$ to ensure continuum-like behavior at the scales probed by the flow.
• Implement the discretized analog of the gradient flow (Wilson or Zeuthen flow) to generate flowed gauge configurations.
• Measure regularized operators such as $U_{\mu\nu}(t,x)$, $E(t,x)$ at positive $t$, guaranteeing finite expectation values.
• Fit the $t\to 0$ asymptotics of measured quantities according to the small flow-time expansion to extract continuum, renormalized operator matrix elements. Monte Carlo sampling over gauge fields provides the necessary statistical ensemble.

The method's independence of regularization is evidenced by the fact that UV divergences are eliminated at the flow stage, prior to any continuum limit $a\to0$.

6. Generalizations and Broader Implications

The framework developed for the Yang-Mills case extends to a wide class of local operators in gauge theories (e.g., higher-dimensional operators, composite fermion bilinears via associated fermionic gradient flows). In each instance, the principle is to construct UV-finite flowed correlators whose small flow-time expansion admits an operator product expansion with universal, RG-determined coefficients, enabling matching to a standard renormalization scheme.

This regularization mechanism underlies modern approaches to computing energy-momentum tensor correlators, topological susceptibilities, and other key quantities on the lattice, for which traditional methods are rendered ambiguous or computationally fragile by operator mixing and power divergences. The construction remains closely related to continuum quantum field theory renormalization, preserving symmetries and enabling the use of RG-improved perturbation theory in interpreting data.

7. Summary Table: Key Properties and Functional Relations

Property	Flowed Operator Construction	Renormalization/Matching
Evolved field	$B_\mu(t,x)$, diffusion PDE	Initial value $B_\mu(0,x) = A_\mu(x)$
UV finiteness	Exponential momentum damping	No point-splitting or further regularization required
Small-t expansion	$U_{\mu\nu}(t,x) \sim c_T(t) T_{\mu\nu}(x) + ...$	RG eqs determine $c_T(t)$, $c_S(t)$, and $c_E(t)$
Energy-momentum tensor	Extracted from flowed observables	Formula for $\{T_{\mu\nu}(x)\}_R$ as function of $U_{\mu\nu}(t,x)$, $E(t,x)$
Lattice implementation	Standard flow on discretized fields	Use of finite $a \ll \sqrt{8t}$, extraction via Monte Carlo

The introduction of gradient-flow regularized local operators provides a robust, symmetry-preserving route to defining and computing local observables in quantum field theory, both in continuum analysis and large-scale numerical computations. The approach's universality and deep connection to RG concepts continue to inform research across quantum field theory and lattice gauge theory.