Stout: A Multifaceted Term in Science

• Stout is a multifaceted term describing distinct phenomena, including the sinking-bubble dynamics in Guinness beer and stout-link smearing in lattice QCD.
• In computational astrophysics, Stout serves as the atomic and molecular database for Cloudy, ensuring accurate spectral synthesis with data traced to original literature.
• In probability, Stout refers to extensions of the law of the iterated logarithm for martingales, with both classical and noncommutative formulations establishing optimal fluctuation bounds.

Searching arXiv for the relevant senses of “Stout” to ground the article in the cited literature. arXiv search 1: stout smearing / Wilson flow / lattice QCD. arXiv search 2: Guinness stout / fluid dynamics. arXiv search 3: Cloudy Stout database / astrophysical spectroscopy. arXiv search 4: Stout in probability / law of the iterated logarithm. “Stout” appears in several unrelated technical settings in the scientific literature. In fluid mechanics it denotes stout beer, especially Guinness, whose settling dynamics exhibit the counter-intuitive appearance of sinking bubbles. In lattice gauge theory it usually denotes stout-link smearing, an analytic gauge-link smoothing transformation that is closely related to Wilson flow. In computational astrophysics it denotes Stout, the atomic and molecular database developed for the spectral-synthesis code Cloudy. In probability, “Stout” refers to the classical martingale law of the iterated logarithm associated with Stout’s extension of Kolmogorov-type results, and to later noncommutative analogues framed explicitly as Stout-type theorems (Benilov et al., 2012, Nagatsuka et al., 2023, Lykins et al., 2015, Zeng, 2012).

1. Principal scientific referents

Domain	Referent	Representative source
Fluid mechanics	Stout beer and the sinking-bubble effect	(Benilov et al., 2012)
Lattice gauge theory	Stout-link smearing and related improvement schemes	(Nagatsuka et al., 2023)
Computational astrophysics	Stout, Cloudy’s atomic and molecular database	(Lykins et al., 2015)
Probability theory	Stout-type martingale law of the iterated logarithm	(Panja et al., 26 Sep 2025)

These usages are historically and conceptually independent. Two of them are proper names attached to prior work or a software component, while the beer-related usage is literal. The lattice-gauge usage is the most extensive in the cited arXiv corpus: “stout” there functions both as a specific link-smearing map and as a descriptor for actions, operators, renormalization calculations, and improvement programs built from that map (Borsanyi et al., 2010).

A common source of confusion is that the same word therefore names both a physical system and a computational method. The beer literature concerns buoyancy-driven bubbly flow in a settling pint, whereas the lattice literature concerns UV smoothing of gauge fields; the connection is lexical rather than scientific (Benilov et al., 2012, Ammer et al., 2024).

2. Stout beer and the sinking-bubble effect

In stout beers such as Guinness, bubbles can be seen moving downward along the wall of the glass while the beer is settling after the pour. The effect is real rather than an optical illusion, and it occurs during the settling stage between pouring and final formation of the creamy head. It is not typical of ordinary beers, whose foam is driven mainly by carbon dioxide and whose bubbles are generally larger. In Guinness and other stouts, the dissolved-gas mixture includes nitrogen as well as carbon dioxide; the nitrogen leads to smaller bubbles and a long-lasting creamy head, and those small bubbles are central to the mechanism (Benilov et al., 2012).

The essential point is that the bubbles remain buoyant relative to the surrounding liquid, but the liquid is not stationary. A recirculating flow develops, with downward liquid motion near the wall and upward motion in the interior. If the downward liquid velocity near the wall exceeds the bubble’s intrinsic upward rise speed relative to the liquid, then the bubbles move downward in the laboratory frame and appear to sink. In the model used for Guinness at $6^\circ \mathrm{C}$, the liquid density and viscosity are taken as

$$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$

the gas density and viscosity as

$$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$

and the characteristic bubble diameter as

$d_b = 122\,\mu\mathrm{m}.$

The Bond number

$\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$

is estimated as $\mathrm{Bo}\approx 0.002$, so the bubbles are essentially spherical. Treating them as rigid spheres because stout contains many surfactants, the Stokes rise velocity is estimated as

$$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$

with Reynolds number

$$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$

This places the problem in a creeping-flow regime in which slow bubble drift can be overwhelmed by a modest recirculating current (Benilov et al., 2012).

The paper’s main contribution is the identification of glass shape as the control parameter for the sign of the circulation. In the traditional Guinness pint, which narrows downward, bubbles rising approximately vertically move away from the sloping wall, leaving a near-wall bubble-poor layer. Since bubbles drag liquid upward where they are present, stronger bubble drag in the interior than near the wall drives upward flow in the middle and downward flow near the wall. In an “anti-pint,” the same glass inverted so that it widens downward, the effect reverses: rising bubbles approach the wall, the near-wall void fraction increases, and the circulation becomes upward near the wall and downward in the center. The mechanism is explicitly compared with the Boycott effect known from sedimentation in inclined containers (Benilov et al., 2012).

The numerical study uses finite-element simulations in COMSOL Multiphysics with a two-phase bubbly-flow model based on Sokolichin, Eigenberger, and Lapin. The formulation is axisymmetric and assumes monodisperse spherical bubbles with an initially uniform bubble distribution. For a served glass the void fraction is taken as $f\approx 0.02$, and settling time is defined by

$f(T_s)=10^{-6}.$

For the pint with $$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$0, the computed settling time is about $$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$1, whereas an experimental estimate in a real Guinness pint is approximately $$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$2. The paper attributes much of this discrepancy to the monodisperse approximation, since smaller-than-median bubbles in the real size distribution extend the tail of the settling process. It also reports a simple qualitative test: in a tilted cylindrical vessel, bubbles rise near the upper side and sink near the lower side, as predicted by the geometry-driven depletion mechanism (Benilov et al., 2012).

An alternative wall-lift explanation is discussed and largely dismissed. At the relevant low Reynolds numbers, wall-induced lift is estimated to be too weak to generate the observed near-wall depletion, and it does not naturally explain the opposite behavior of pint and anti-pint geometries. The favored interpretation is therefore geometry-induced redistribution of bubble concentration, not a loss of buoyancy and not a wall-lift effect (Benilov et al., 2012).

3. Stout-link smearing in lattice gauge theory

In lattice gauge theory, stout-link smearing is a recursive, analytic smoothing transformation on gauge links $$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$3. In the isotropic case,

$$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$4

and one stout step is written as

$$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$5

The Lie-algebra-valued generator is

$$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$6

and can be written explicitly in terms of staple sums. Because the update is expressed through an exponential map, the smeared link remains in $$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$7, which is one of the reasons stout smearing is useful in both dynamical simulations and perturbation theory (Nagatsuka et al., 2023).

The Wilson or Yang–Mills gradient flow evolves links continuously in a fictitious time $$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$8,

$$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$9

and the cited literature proves that stout smearing becomes the Wilson flow in the limit $$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$0 at finite lattice spacing. Introducing a continuous smearing-time variable $$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$1, one obtains the continuous stout equation

$$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$2

which is exactly the Wilson-flow equation after the identification

$$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$3

The formal mismatch at finite $$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$4 begins at $$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$5 and is interpreted as inducing $$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$6 effects (Nagatsuka et al., 2023, Nagatsuka et al., 2023).

The numerical comparison is based on the action density

$$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$7

and the relative difference

$$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$8

Across several lattice spacings, the results show that discrepancy decreases with both decreasing $$\rho_g = 1.223\,\mathrm{kg\,m^{-3}}, \qquad \mu_g = 0.017\times 10^{-3}\,\mathrm{Pa\,s},$$9 and decreasing $d_b = 122\,\mu\mathrm{m}.$0, and practical equivalence for $d_b = 122\,\mu\mathrm{m}.$1 and $d_b = 122\,\mu\mathrm{m}.$2 is achieved within statistical precision when $d_b = 122\,\mu\mathrm{m}.$3 is chosen small enough for the lattice spacing under study. The same literature also states that stout smearing is roughly an order of magnitude cheaper than Wilson flow as a smoothing procedure in the tested setups (Nagatsuka et al., 2023, Nagatsuka et al., 2023).

A further development is the perturbative expansion of stout smearing and Wilson flow through $d_b = 122\,\mu\mathrm{m}.$4, sufficient for one-loop calculations. In momentum space, the leading stout kernel is

$d_b = 122\,\mu\mathrm{m}.$5

while in the Wilson-flow limit it becomes

$d_b = 122\,\mu\mathrm{m}.$6

This perturbative machinery is used to dress generic fermion-action Feynman rules with stout or flow kernels rather than recomputing them from scratch for each smoothed action (Ammer et al., 2024).

The stout idea has also been adapted to digital quantum simulation. In that setting the paper “Stout Smearing on a Quantum Computer” formulates a reversible, unitary, representation-agnostic quantum analogue of stout-link smearing for Hamiltonian lattice gauge theory and shows that it reduces coupling to high-energy modes in the discrete nonabelian gauge theory $d_b = 122\,\mu\mathrm{m}.$7 (Gustafson, 2022). This suggests that, beyond its classical lattice role, stout can function as a structured filtering primitive in quantum algorithms.

4. Stout smearing as an improvement and renormalization tool

In thermodynamic lattice QCD, stout improvement is used primarily to reduce taste-symmetry violation in staggered fermions. In the Budapest–Wuppertal equation-of-state program, the calculation employs a tree-level Symanzik-improved gauge action together with a stout-link improved staggered fermion action for $d_b = 122\,\mu\mathrm{m}.$8 dynamical flavors, with physical quark masses fixed through $d_b = 122\,\mu\mathrm{m}.$9 and $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$0 on lattices with $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$1 and checks at $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$2. The logic is that reducing taste splitting yields a more physical pseudo-Goldstone sector at finite lattice spacing and supports a controlled continuum extrapolation of the pressure, interaction measure, energy density, entropy density, and speed of sound. This stout-based strategy is also used to interpret discrepancies with earlier hotQCD results obtained from p4 and asqtad actions (Borsanyi et al., 2010).

The same basic construction appears in other fermion formulations with different practical goals. In SU(2) $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$3 SYM, one-step stout smearing with $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$4 is applied only to the links entering the Wilson–Dirac operator for adjoint gluinos, while the gauge action remains tree-level Symanzik improved. The stated purpose is removal of short-range topological defects and the corresponding small eigenvalues of the fermion matrix, and the stout-link ensembles are reported to give better signal quality in difficult scalar channels. In the Schrödinger-functional study of two-color six-flavor Wilson fermions, one level of stout smearing with $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$5 is used in the fermion action to avoid coupling the fermions to unphysical fluctuations of the gauge field on the scale of the lattice spacing; the work emphasizes that near-boundary smearing conventions matter for SF observables, even though tested alternatives do not significantly shift the critical mass or phase boundary (Demmouche et al., 2010, Voronov, 2012).

A distinct set of papers treats stout parameters as perturbative control variables. For staggered fermions with up to two stout iterations applied separately in the action and in bilinear operators, one-loop renormalization functions are given explicitly in terms of four stout parameters $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$6, and the tensor, axial, pseudoscalar, vector, scalar, quark-field, and mass renormalizations are made available in RI′ and $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$7 forms. One notable structural result is that $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$8 when the same stout parameters are used in the action and vector operator. In the two-loop singlet–nonsinglet study of staggered bilinears with twice stout-smeared links, the singlet–nonsinglet difference vanishes for scalar, pseudoscalar, vector, and tensor operators, while the axial-vector difference is a nontrivial polynomial in the action and operator stout parameters (Constantinou et al., 2013, Bali et al., 2013, Panagopoulos et al., 2017).

For Wilson-like fermions, stout smearing is tightly connected with clover improvement. One proceedings paper develops the perturbative machinery required to determine $\mathrm{Bo} = \frac{\rho_l g d_b^2}{\sigma}$9 for overall stout-smeared Wilson fermions and makes explicit the relation to gradient flow. A later one-loop calculation for Wilson and Brillouin fermions with stout smearing or Wilson flow reports that even a small amount of smoothing rapidly decreases the one-loop coefficient, with minima near $\mathrm{Bo}\approx 0.002$0 for several action combinations. The same paper argues that once $\mathrm{Bo}\approx 0.002$1 is moderately large, the one-loop correction becomes small enough that the full $\mathrm{Bo}\approx 0.002$2 may remain close to its tree-level value over a wide coupling range (Ammer et al., 2021, Ammer et al., 18 Jan 2026).

The parameter $\mathrm{Bo}\approx 0.002$3 can also be tuned directly against Ward identities or operator normalization conditions. In lattice SYM with clover fermions, a Symanzik-improved gauge action, and stout-smeared links in the covariant derivatives, one-loop expressions are derived for the gluino field, scalar, pseudoscalar, and axial-vector bilinears, the critical gluino mass, and other renormalization factors. The paper identifies a value of the smearing parameter that ensures axial-current conservation at one loop and, for the Wilson gauge action, finds a simultaneous solution of

$\mathrm{Bo}\approx 0.002$4

at

$\mathrm{Bo}\approx 0.002$5

This suggests that stout is not merely a generic UV smoother but a tunable ingredient in fine-tuning programs (Costa et al., 16 Jan 2026).

Not all consequences are uniformly beneficial. In finite-density lattice QCD with multiparameter reweighting on $\mathrm{Bo}\approx 0.002$6 lattices, adding one step of stout smearing to the unimproved staggered fermion action weakens the $\mathrm{Bo}\approx 0.002$7 crossover, pushes the leading Fisher zero farther from the real axis, and makes the overlap problem substantially worse, to the point that the critical endpoint becomes inaccessible with the smeared action in the studied setup. In the quenched Schwinger model, by contrast, stout smoothing of 0, 1, or 3 steps with $\mathrm{Bo}\approx 0.002$8 improves pairing of low Dirac eigenvalues and changes the asymptotic taste-splitting scaling of would-be zero modes from $\mathrm{Bo}\approx 0.002$9 without smearing to $$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$0 with smearing, while non-topological modes remain action dependent (Giordano et al., 2020, Ammer et al., 2024).

5. Stout as Cloudy’s atomic and molecular database

In computational astrophysics, Stout is the atomic and molecular database developed for the spectral-synthesis code Cloudy. It is explicitly not intended as a review database; rather, it is a practical, maintainable, physics-driven data system built around Cloudy’s requirements. Cloudy models molecular, atomic, and ionized gas over

$$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$1

with radiation fields from

$$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$2

to

$$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$3

and aims to treat all atoms and ions of the lightest 30 elements together with approximately $$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$4 molecules (Lykins et al., 2015).

A central design principle is that data are stored as close as possible to the original literature sources. This is intended to simplify maintenance, provenance tracking, and later updates. Within Cloudy, Stout is one of three principal external spectral databases, alongside Chianti and LAMDA, while the H-like and He-like sequences are handled by dedicated internal models. After those internal sequence models, the preference order is Stout, then Chianti, then LAMDA. Species data are organized in a file structure similar to Chianti, typically with separate .nrg, .tp, and .coll files for level energies, transition probabilities, and collision data (Lykins et al., 2015).

The database must ingest heterogeneous radiative and collisional inputs. For radiative transitions it accepts line strengths $$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$5, oscillator strengths $$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$6 or $$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$7, and Einstein $$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$8-values. One of the central formulas is

$$u_b = \frac{(\rho_l-\rho_b) g d_b^2}{18\mu_l} \approx 3.96\,\mathrm{mm\,s^{-1}},$$9

or equivalently in terms of transition energy,

$$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$0

If theoretical transition data are combined with experimental level energies, the transition probabilities are corrected by

$$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$1

Experimental energies from NIST are preferred whenever available (Lykins et al., 2015).

For collisional excitation, the preferred quantity is the effective collision strength $$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$2, with rate coefficients

$$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$3

Stout is unusual among spectral databases in that it must cover a broad range of colliders, not only electrons but also $$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$4, $$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$5, He, and $$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$6, because Cloudy is used in photodissociation regions and molecular gas as well as coronal or photoionized plasmas (Lykins et al., 2015).

The paper is equally explicit about incompleteness handling. Interpolation uses the Fritsch–Butland local, piecewise-cubic, monotonicity-preserving method to avoid overshoot and unphysical negative collision rates. Extrapolation rules depend on species and process. For molecular de-excitation rates, the low-temperature rule is

$$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$7

and the high-temperature rule is

$$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$8

If collision data are entirely missing, the $$\mathrm{Re} = \frac{\rho_l u_b d_b}{\mu_l} \approx 0.24.$$9 approximation may be used; if the level-population matrix would otherwise become ill-conditioned, an electron effective collision strength of $f\approx 0.02$0 is assigned as a numerical floor (Lykins et al., 2015).

The significance of Stout within Cloudy is therefore infrastructural. It broadens line coverage, especially for complex ions and molecules, while tying the database version directly to the Cloudy release. The paper emphasizes that this richer line list matters even when the dominant cooling budget changes little, because synthetic spectra depend on many weak and subordinate lines, rare species, and line blends (Lykins et al., 2015).

6. Stout in martingale laws of the iterated logarithm

In probability theory, “Stout” denotes the classical martingale extension of Kolmogorov’s law of the iterated logarithm. The cited noncommutative literature recalls the commutative theorem in the following form. For a martingale $f\approx 0.02$1 with differences $f\approx 0.02$2, define

$f\approx 0.02$3

If $f\approx 0.02$4 and

$f\approx 0.02$5

for some positive sequence $f\approx 0.02$6, then Stout’s martingale theorem gives

$f\approx 0.02$7

This is the specific classical result to which later noncommutative papers refer when they speak of a Stout-type LIL (Zeng, 2012, Panja et al., 26 Sep 2025).

The first cited noncommutative extension, due to Zeng, replaces scalar random variables by self-adjoint martingales in a finite von Neumann algebra $f\approx 0.02$8, with martingale differences $f\approx 0.02$9 and operator-valued predictable square function. The normalization becomes

$f(T_s)=10^{-6}.$0

and the conclusion is only an almost-uniform upper bound: $f(T_s)=10^{-6}.$1 The paper is explicit that the constant $f(T_s)=10^{-6}.$2 is weaker than the classical $f(T_s)=10^{-6}.$3, and also that in the noncommutative setting one can in general expect only an upper bound, not a full equality, because free semicircular examples preclude a classical-type lower limsup statement (Zeng, 2012).

A later paper sharpens this result to the optimal constant $f(T_s)=10^{-6}.$4. For a self-adjoint martingale with the same normalization and increment condition

$f(T_s)=10^{-6}.$5

it proves

$f(T_s)=10^{-6}.$6

It then extends the result to non-self-adjoint martingales using

$f(T_s)=10^{-6}.$7

yielding

$f(T_s)=10^{-6}.$8

The proof relies on an improved exponential inequality essentially due to Randrianantoanina, replacing the weaker inequality used in the earlier Junge–Zeng framework. The same paper also derives a noncommutative Hartman–Wintner type result for independent sequences (Panja et al., 26 Sep 2025).

The probability-theoretic use of “Stout” is therefore not a named object analogous to stout smearing or the Cloudy database. It denotes a theorem lineage: Stout’s martingale LIL in the commutative setting, Zeng’s noncommutative analogue with constant $f(T_s)=10^{-6}.$9, and the later optimal noncommutative Stout-type LIL with constant $$\rho_l = 1007 \,\mathrm{kg\,m^{-3}}, \qquad \mu_l = 2.06\times 10^{-3}\,\mathrm{Pa\,s},$$00. A common misconception is that the noncommutative result should also be an equality; the cited literature states that this is not expected in general, because the free CLT changes fluctuation behavior fundamentally (Zeng, 2012, Panja et al., 26 Sep 2025).