A new computational fluid dynamics framework combining large eddy simulation with a finite difference method has significantly outperformed traditional Reynolds-averaged Navier-Stokes approaches in predicting local gas volume fractions within a benchmark bubble column reactor. Developed by researchers Matheus Furtado Grings, Ligia Damasceno Ferreira Marczak, and Álvaro Luiz de Bortoli, the numerical model addresses persistent hydrodynamic modeling difficulties found in multiphase gas-liquid systems. Bubble columns are widely utilized across biochemical operations, notably wastewater treatment and microalgae photobioreactors, because their simple vertical cylindrical geometry provides effective mass and heat transfer without requiring mechanical impellers that exert damaging shear stress on living cells. However, accurately forecasting internal flow dynamics, phase interactions, and gas holdup distributions has historically proven mathematically challenging. By explicitly resolving large-scale unsteady flow structures rather than relying on statistical time averaging, the newly formulated Eulerian-Eulerian framework delivers superior fidelity against physical experimental data across challenging transitional and lower column zones.
The modeling methodology was validated against the experimental and computational benchmark established by Krepper and colleagues for a flat rectangular bubble column. The physical system features a width of 0.1 meters, a depth of 0.02 meters, and an initial water height of 1.0 meter, into which air is injected through a centralized bottom porous airstone measuring 0.02 meters wide, 0.01 meters deep, and 0.01 meters high. Operating at superficial gas velocities of 6, 8, and 10 millimeters per second, the benchmark reactor functions within a dispersed bubbly regime characterized by the absence of bubble breakup and coalescence, maintaining a nominal fixed bubble diameter of 3 millimeters. High-frequency wire mesh sensors positioned at elevations of 0.08 meters and 0.63 meters above the sparger recorded cross-sectional gas fraction distributions at 2500 hertz over sixty seconds. These twin elevations captured two hydrodynamically distinct environments: a turbulent, highly oscillatory entrance zone near the sparger and a stabilized, fully developed flow zone higher in the column.
To resolve transport equations within this rectangular geometry, the authors developed an in-house solver written in Fortran 90 utilizing the finite difference method in Cartesian coordinates. Within their Eulerian-Eulerian continuum framework, continuous liquid water and dispersed air phases are represented as interpenetrating media governed by phase-weighted mass and momentum balances. Spatial derivatives were discretized using second-order central finite difference approximations, while temporal integration advanced via a simplified third-order Runge-Kutta scheme. To overcome numerical stability challenges with steep phase gradients, the authors incorporated a liquid-phase turbulent diffusion term into the mass conservation equation with a turbulent Schmidt number of 1.0. Physical properties were set at standard room conditions of 25 degrees Celsius, with water density and viscosity at 997 kilograms per cubic meter and 8.899 times 10 to the negative fourth kilogram per meter-second, while air density and viscosity were 1.185 kilograms per cubic meter and 1.831 times 10 to the negative fifth kilogram per meter-second.
In contrast to conventional Reynolds-averaged formulations that statistically time-average every scale of fluid motion, the large eddy simulation methodology splits the velocity field into resolved large-scale eddies and modeled unresolved subgrid scales. The authors employed the Smagorinsky subgrid-scale model to approximate unresolved eddy momentum transfer, where filter width was defined by the cube root of cell volume. Subgrid turbulent viscosity for the dispersed gas phase was closed through the dispersed phase zero-equation model, while the turbulent Prandtl number was assigned a standard value of unity. Because rising bubbles actively shed vortices that agitate the surrounding liquid medium, bubble-induced turbulence was explicitly integrated into the effective liquid viscosity using the empirical model proposed by Sato and Sekoguchi. Simulation boundary conditions enforced strict no-slip constraints along vertical walls alongside Neumann zero-gradient conditions for air volume fraction and pressure, while a degassing condition at the upper boundary allowed air to vent without liquid loss or numerical backflow.
Establishing rigorous spatial and temporal discretization independence formed an essential stage of the numerical verification process. The researchers evaluated three uniform Cartesian grids comprising 9,999 coarse points, 37,989 reference points, and 147,969 refined points, ensuring each mesh maintained nine discrete nodes across the narrow depth coordinate to prevent cross-sectional under-resolution. Grid convergence was formally verified through the Grid Convergence Index method, which yielded a maximum discretization uncertainty of only 2.1 percent between the reference and refined meshes. Given its balance between scale resolution and computational demand, the 37,989-point reference grid was adopted for comprehensive validation, corresponding to an effective filter-to-bubble diameter ratio of approximately 1.67 relative to the Milelli criteria. Temporal stability was maintained by selecting a small time step of 2.5 times 10 to the negative fifth seconds, keeping maximum dispersed Courant-Friedrichs-Lewy numbers below 0.05. Simulations ran for 200 physical seconds, with statistical averaging initiated after 50 seconds to exclude initial startup transients.
The team conducted an extensive comparative investigation into interphase momentum closures, focusing specifically on three classic drag coefficient correlations: the Schiller and Naumann spherical model, the Grace ellipsoidal model, and the Ishii and Zuber three-regime model. While bubble drag dictates vertical momentum transfer and slip velocity, lateral dispersion across the column cross section is governed by transverse interfacial forces. To remain consistent with the Krepper benchmark, the authors incorporated the Tomiyama distorted bubble lift force correlation alongside the Antal wall lubrication force, which prevents unphysical bubble accumulation at rigid boundaries through empirical wall proximity terms. Because the thin rectangular vessel enforces strong geometric confinement along the depth coordinate, lateral z-direction velocity components and z-dependent force contributions were set to zero, following established flat-column assumptions. Crucially, the researchers excluded artificial turbulent dispersion force models from their formulation, examining whether resolving transient flow structures via large eddy simulation could naturally disperse bubbles without empirical calibration.
Evaluating overall time-averaged gas holdup across superficial gas velocities of 6, 8, and 10 millimeters per second revealed distinct predictive capabilities among the numerical implementations. All three drag formulations produced a strong linear correlation between gas holdup and superficial gas velocity, each achieving determination coefficients of 0.999 that mirrored experimental trends. However, their absolute magnitudes diverged noticeably from physical measurements: the spherical Schiller-Naumann model and the Grace formulation consistently underestimated total volumetric gas fraction, whereas the original Reynolds-averaged numerical simulation developed by Krepper and colleagues systematically overestimated holdup. The Ishii and Zuber model generated results closest to the experimental benchmark curve, exhibiting only minor overestimation at the highest velocity of 10 millimeters per second. The authors emphasized that global gas holdup predictions do not depend on the drag correlation alone, but rather reflect an intricate dynamic equilibrium established between drag, lift, wall lubrication, subgrid dissipation, and the underlying finite difference discretization.
The most pronounced advantage of the finite difference large eddy simulation approach emerged in local air volume fraction profiles across different column heights. In the lower oscillation zone at 0.08 meters above the distributor, the rising bubble plume is dominated by large-scale coherent vortex shedding, lateral flapping instabilities, and pronounced unsteady gradients. Traditional Reynolds-averaged models fail in this lower region because statistical temporal averaging suppresses unsteady meandering, generating unphysically sharp peaks at the plume core and underestimating gas spreading toward the walls. Conversely, all three large eddy simulation runs accurately replicated the broader, bell-shaped experimental profiles, successfully capturing natural bubble redistribution driven by transient plume motion. In the upper, non-oscillatory region at 0.63 meters, the flow field stabilizes into a flatter profile with elevated void fractions near boundaries due to wall lubrication forces, where the Ishii-Zuber correlation yielded superior agreement at 10 millimeters per second while the Schiller-Naumann model excelled at lower velocities.
To benchmark their code against the wider computational fluid dynamics literature, the authors compared their results with published Reynolds-averaged studies that utilized the identical experimental data set. These included shear stress transport simulations by Besagni and colleagues, polydisperse formulations by Charton and colleagues, and OpenFOAM investigations conducted by Godino and collaborators. Although previous Reynolds-averaged studies introduced empirical turbulent dispersion forces to artificially spread bubbles, they consistently struggled to match physical observations across both elevations simultaneously. For instance, Godino and colleagues achieved close overall gas holdup predictions but substantially overpredicted central gas fractions while underpredicting near-wall concentrations in the lower plume zone. The simulations of Besagni and colleagues similarly underestimated gas fractions within the oscillatory region. By contrast, because the large eddy simulation framework directly resolves unsteady macroscopic structures in space and time, the resulting time-averaged fields naturally exhibit realistic physical diffusion without requiring empirical tuning parameters.
Finally, the validated hydrodynamic framework was applied to predict local gas-liquid mass transfer efficiency by estimating the overall volumetric mass transfer coefficient for carbon dioxide in water. Utilizing Higbie penetration theory coupled with the Boussinesq hypothesis, the volumetric coefficient was calculated from local dispersed gas fraction, bubble diameter, and turbulent dissipation rate. The resulting contours revealed that mass transfer is strongly concentrated within the central plume in the lower oscillatory region, mirroring the local gas concentration peak. Higher up the column, the distribution becomes smoother across the cross section, with secondary peaks appearing near outer walls as a direct consequence of elevated turbulent dissipation generated by no-slip boundary conditions. By accurately capturing both turbulent dissipation and local phase distributions, the authors demonstrated that their computationally efficient finite difference large eddy simulation code provides an effective tool for optimizing aeration, mixing, and carbon dioxide dissolution in industrial photobioreactors and microalgae cultivation facilities.
Subject of Research: Large eddy simulation modeling of two-phase hydrodynamics in bubble column reactors
Article Title: LES-based simulations for bubble columns: a comparative study based on turbulence models and interfacial forces
Article References: Grings, M. F., Marczak, L. D. F., & de Bortoli, Á. L. (2026). LES-based simulations for bubble columns: a comparative study based on turbulence models and interfacial forces. Multiscale and Multidisciplinary Modeling, Experiments and Design, 9(1), Article 180. https://doi.org/10.1007/s41939-026-01260-4
Image Credits: AI Generated
DOI: 10.1007/s41939-026-01260-4
Keywords: Bubble columns, Large eddy simulation, Computational fluid dynamics, Finite difference method, Eulerian-Eulerian model, Gas holdup, Gas volume fraction, Drag coefficient models, Ishii-Zuber model, Grace drag model, Schiller-Naumann model, Bubble-induced turbulence, Smagorinsky model, Mass transfer coefficient, Photobioreactors, Microalgae cultivation, Multiphase flow, Grid Convergence Index
Cite Scienmag News
Audrey Campbell. (November 18, 2026). Large Eddy Simulation Accurately Predicts Bubble Column Gas Distributions. Scienmag. https://scienmag.com/large-eddy-simulation-accurately-predicts-bubble-column-gas-distributions/
Audrey Campbell. "Large Eddy Simulation Accurately Predicts Bubble Column Gas Distributions." Scienmag, 18 November 2026, https://scienmag.com/large-eddy-simulation-accurately-predicts-bubble-column-gas-distributions/. Accessed 18 November 2026.
Audrey Campbell. "Large Eddy Simulation Accurately Predicts Bubble Column Gas Distributions." Scienmag. November 18, 2026. https://scienmag.com/large-eddy-simulation-accurately-predicts-bubble-column-gas-distributions/








