Stout Smearing in Lattice Gauge Theory

• Stout smearing is an analytic, gauge-covariant smoothing technique that replaces thin link variables with locally averaged SU(N) matrices to suppress UV fluctuations.
• It enables controlled continuum extrapolations by reducing discretization artifacts and improving the properties of Dirac operators in QCD and beyond.
• Its differentiability and computational efficiency make it a powerful alternative to Wilson flow in large-scale lattice simulations and operator renormalization.

Stout smearing is an analytic, gauge-covariant, and differentiable procedure for replacing thin link variables on the lattice by locally averaged, smoother SU(N) matrices. Developed to suppress ultraviolet (UV) fluctuations while maintaining unitarity and locality, it is a key tool in modern lattice gauge theory, facilitating controlled continuum extrapolations, suppressing discretization artifacts, and improving the properties of Dirac operators used for both QCD and beyond-the-Standard-Model theories (Risch, 2023).

1. Formal Definition and Algorithmic Structure

Stout smearing is defined recursively as follows. Starting from an initial set of SU(N) link variables $U_\mu^{(0)}(x) = U_\mu(x)$, each subsequent stout step $k \to k+1$ is given by

$$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$

where $\rho$ is the smearing parameter and $Q_\mu^{(k)}(x)$ is an appropriately constructed traceless, anti-Hermitian staple matrix: $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$ Here, the staple sum $\Omega_\mu(x)$ is

$$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$

The update preserves exact unitarity (no re-projection required) since the exponential map is within SU(N). This link replacement can be iterated $n$ times, producing the $n$-step stout-smeared link $k \to k+1$0 (Risch, 2024, Risch, 2023).

Iteratively, the effective smearing radius achieved is related to the parameters by $k \to k+1$1, where $k \to k+1$2 is the equivalent Wilson flow time and $k \to k+1$3 the lattice spacing. For practical precision and computational efficiency, $k \to k+1$4–$k \to k+1$5 with $k \to k+1$6 in the range $k \to k+1$7–$k \to k+1$8 are commonly used (Risch, 2024, Risch, 2023).

The projection to su(N) is built into the construction of $k \to k+1$9: no further step is necessary. This analyticity and differentiability in the gauge links make stout smearing suitable for molecular dynamics-based simulation algorithms such as HMC (Constantinou et al., 2013).

2. Analytical Properties and Equivalence to the Wilson Flow

Stout smearing, in the limit of $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$0, $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$1 with $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$2 fixed, is analytically equivalent to the Wilson or Yang–Mills gradient flow. Under these conditions, stout smearing acts as a first-order (Euler) integrator of the Wilson flow equation: $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$3 with $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$4 the Wilson plaquette action. The correspondence is set by matching $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$5, so $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$6 (Nagatsuka et al., 2023, Nagatsuka et al., 2023, Risch, 2024).

The leading discretization artifacts, for finite $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$7, arise at $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$8—i.e., the difference between stout smearing (at finite $$U_\mu^{(k+1)}(x) = \exp\left[ \rho\, Q_\mu^{(k)}(x) \right] U_\mu^{(k)}(x)$$9, $\rho$0) and Wilson flow vanishes in the continuum limit $\rho$1. At fixed $\rho$2, to keep these artifacts below sub-percent levels, it is necessary to choose $\rho$3, with $\rho$4 the reference flow scale (Nagatsuka et al., 2023, Risch, 2024). Numerically, for $\rho$5 fm, one finds that $\rho$6 (i.e., $\rho$7) ensures that smearing does not spoil controlled continuum extrapolation for observables at distances $\rho$8 fm (Risch, 2023).

In practical large-scale lattice simulations, stout smearing is computationally more efficient than explicit Wilson flow integration (typically $\rho$9 faster per iteration) and is natively supported in HMC and related algorithms thanks to its differentiability (Nagatsuka et al., 2023, Risch, 2024).

3. Effects on Discretization Errors and Continuum Extrapolation

Stout smearing acts as a UV filter on the gauge fields entering the Dirac operator, efficiently suppressing high-momentum modes responsible for discretization artifacts such as taste violation in staggered fermions and large $Q_\mu^{(k)}(x)$0 matching corrections. For Wilson, clover, or staggered quarks, the procedure:

• Dramatically reduces the near-zero mode density and the condition number of the Dirac operator.
• Leads to smaller one-loop corrections to operator renormalization constants, improving the perturbative convergence and reducing the need for large matching corrections (Constantinou et al., 2013, Bali et al., 2013, Gupta et al., 2013).
• Reduces taste-breaking effects in the pion and kaon sectors by a factor $Q_\mu^{(k)}(x)$1 after the first stout step, with further iterations asymptoting after the second or third step (Bali et al., 2013, Risch, 2024).

The impact is quantitatively characterized via observables such as the diagonal Creutz ratio,

$Q_\mu^{(k)}(x)$2

where $Q_\mu^{(k)}(x)$3 is the rectangular Wilson loop. At too large smearing radii ($Q_\mu^{(k)}(x)$4 at $Q_\mu^{(k)}(x)$5 fm) the approach of $Q_\mu^{(k)}(x)$6 to the continuum limit in $Q_\mu^{(k)}(x)$7 becomes non-monotonic, signaling loss of control over discretization systematics at small distances. Empirically, mild smearing ($Q_\mu^{(k)}(x)$8) ensures monotonicity and controlled extrapolation for $Q_\mu^{(k)}(x)$9 fm (Risch, 2023).

Tables of optimal stout parameters for different setups are compiled in (Risch, 2023), and the monotonicity criterion for the continuum approach is emphasized as a diagnostic for safe smearing.

4. Applications: Renormalization, Improvement, and Algorithmics

Renormalization and Operator Matching

Perturbative calculations show that one-step and two-step stout smearing, with tunable weights, lead to renormalization constants for fermion bilinears $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$0, $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$1, $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$2, etc., that are explicit polynomials in the smearing parameters. This flexible structure enables:

• Tuning smearing parameters to achieve minimal operator matching corrections,
• Exact conservation of the vector current at one loop if the same smearing is used in the Dirac operator and in the operator definition,
• Substantial suppression of additive mass renormalization and $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$3 taste-exchange terms in the staggered pion spectrum (Constantinou et al., 2013, Bali et al., 2013, Costa et al., 16 Jan 2026).

One-Loop Improvement Coefficient $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$4

Stout smearing directly enters the determination of the $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$5 improvement coefficient $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$6 for clover/Wilson/Brillouin fermions. A single stout iteration with moderate $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$7 reduces $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$8 by a factor of 2–5; after two or three steps, $$Q_\mu(x) = \frac{1}{2i}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right] - \frac{1}{2iN} \operatorname{Tr}\left[ \Omega_\mu(x) - \Omega_\mu^\dagger(x) \right]$$9 is typically $\Omega_\mu(x)$0 for $\Omega_\mu(x)$1 (bare QCD coupling), allowing simulations with $\Omega_\mu(x)$2 close to its tree-level value (Ammer et al., 18 Jan 2026, Ammer et al., 2024).

A summary of $\Omega_\mu(x)$3 as a function of $\Omega_\mu(x)$4 and $\Omega_\mu(x)$5 (Wilson glue, $\Omega_\mu(x)$6):

$\Omega_\mu(x)$7, $\Omega_\mu(x)$8	$\Omega_\mu(x)$9 (Wilson)	$$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$0 (Symanzik)
1, 0.05	0.125	0.096
2, 0.12	0.023	0.012
3, 0.12	0.014	0.004

These values indicate robust suppression of $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$1 artifacts in on-shell lattice observables (Ammer et al., 18 Jan 2026, Ammer et al., 2024).

Algorithmic and Scaling Implications

Stout smearing, being local and analytic, enables efficient stochastic simulation (e.g., HMC) and easy implementation in perturbative calculations. In scaling studies, mapping the stout-smeared flow radius to Wilson flow time ensures direct comparison of smeared and flowed observables, with careful matching requirements detailed in (Risch, 2023, Risch, 2024).

5. Comparative Analysis: Stout Smearing vs. Wilson Gradient Flow

The equivalence of stout smearing (in the infinitesimal step and infinite iteration limit) and Wilson flow is now analytically and numerically established. Analytically, the mapping $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$2 ensures that both methods implement equivalent smoothing up to $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$3 artifacts for finite $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$4 and $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$5 (Nagatsuka et al., 2023, Nagatsuka et al., 2023).

In numerical applications:

• For matched flow radii, stout smearing with $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$6, $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$7, or $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$8, $$\Omega_\mu(x) = \sum_{\nu\neq\mu} \left[ U_\nu(x) U_\mu(x+\hat{\nu}) U_\nu^\dagger(x+\hat{\mu}) + U_\nu^\dagger(x-\hat{\nu}) U_\mu(x-\hat{\nu}) U_\nu(x-\hat{\nu}+\hat{\mu}) \right]$$9 already reproduces Wilson flow results for action densities and scale setting, with fractional differences $n$0 even at $n$1 fm (Nagatsuka et al., 2023, Risch, 2023).
• Variance reduction in noisy observables is essentially saturated at the same radius for both methods, i.e., further smearing or flow beyond $n$2 brings little additional reduction (Risch, 2023).
• Computational cost is strongly in favor of stout smearing: a single sweep is at least $n$3 cheaper than one Wilson flow step with standard ODE integrators (Nagatsuka et al., 2023).

A direct summary of matching prescription:

Scheme	Flow/smearing parameter	Matched smoothing radius
Wilson flow	$n$4	$n$5
Stout smearing	$n$6	$n$7

6. Parameter Selection and Practical Guidelines

Rigorous continuum scaling and preservation of physical short-distance behavior require judicious selection of stout parameters. Guidelines from (Risch, 2023, Risch, 2024):

• For distances $n$8 fm and $n$9 fm, enforce $n$0.
• One stout iteration at $n$1–$n$2 or $n$3–$n$4 with $n$5–$n$6 yields safe smoothing.
• Always check for monotonic approach of continuum-extrapolated observables in $n$7 as a function of the smearing strength.
• For longer-distance observables, the smearing radius may be increased within the constraint that short-distance physics is not oversmoothed.

These criteria are universally applicable for pure gauge, dynamical-fermion, and improved fermion actions.

7. Physical Impact and Empirical Results

Extensive nonperturbative studies corroborate the impact of stout smearing:

• In staggered fermion QCD, stout smearing reduces splitting among pion tastes, resulting in greatly improved restoration of taste symmetry and smaller corrections in high-$n$8 screening masses. The suppression of UV power in gauge observables is up to $n$9–$k \to k+1$00 for optimally tuned smearing parameters (Gupta et al., 2013).
• In the study of topological observables, matching the stout radius to Wilson flow time ensures nearly perfect agreement in average action, topological susceptibility, and correlation functions (Alexandrou et al., 2017).
• Excessive stout smearing, however, weakens physical crossovers or sharpens overlap problems in finite-density studies, constraining its use in reweighting-based phase-diagram investigations (Giordano et al., 2020).
• In 2D models, stout smearing shifts taste violation in the Dirac eigenvalue spectrum from $k \to k+1$01 to $k \to k+1$02 scaling, but logarithmic corrections can become relevant at coarse $k \to k+1$03 (Ammer et al., 2024).

In summary, stout smearing is an analytically well-controlled, cost-effective alternative to continuous gradient/Wilson flow for UV filtering on the lattice, robustly reducing discretization artifacts and enabling controlled continuum limits in both fermionic and gluonic sectors, provided smearing parameters are chosen within the empirically established safe range (Risch, 2023, Nagatsuka et al., 2023, Risch, 2024).