Gas Holdup, Bubble Size Distribution, and Impeller Flooding in Aerated Bioreactors

Every aerated stirred tank lives or dies by what its gas does after it leaves the sparger. Bioreactor gas holdup measurement, bubble size, and the risk of impeller flooding in a bioreactor are three views of the same problem: how much gas the vessel holds, how finely it is divided, and whether the impeller can still disperse it. Get them right and oxygen transfer scales predictably. Get them wrong and a process that ran cleanly at 10 L stalls at 2,000 L with the dissolved oxygen sliding to zero while the agitator and gas controller both look healthy.

This article connects the three ideas into one working framework for gas dispersion scale-up in bioreactors. We define gas holdup, walk through the four dispersion regimes of a Rushton turbine, compare measurement methods and published correlations, relate bubble size distribution to interfacial area, and then derive a flooding check you can run on a spreadsheet. A worked example scales a 10 L vessel to 2,000 L and shows exactly where the flooding boundary sits.

What Is Gas Holdup and Why Does It Matter for Oxygen Transfer?

Gas holdup (ε_g) is the volume fraction of the gassed dispersion that is gas rather than liquid. In a sparged stirred tank you can define it from volumes or from liquid height:

ε_g = V_g / (V_g + V_L) = (H_gassed − H_ungassed) / H_gassed

Typical gas holdup in stirred-tank bioreactors for bioprocess work sits between 2% and 10%, depending on power input, gas flow and media. Bubble columns and airlift reactors, which rely on the gas itself for mixing, can reach 30%. A mammalian cell culture at 0.01–0.1 VVM may run at well under 1% holdup, while a vigorously aerated microbial fermentation at 1 VVM and 2–3 kW/m³ can approach 10%.

Gas holdup matters because it is half of the interfacial area equation. The gas–liquid interfacial area per unit dispersion volume is:

a = 6ε_g / d₃₂

and the volumetric mass transfer coefficient is the product of that area and the liquid-side coefficient, kLa = k_L × a. With k_L usually in the range 2–5 × 10⁻⁴ m/s for bioreactor conditions, almost all of the variation in kLa comes from a, which means from gas holdup and bubble size. If you want the full measurement picture, see our guide to how to calculate kLa. Here we focus on the physical quantities underneath it.

Holdup also has a second, less obvious role. Gas that is held in the vessel displaces liquid, so higher gas holdup raises the gassed liquid level and changes the working volume you can fill to. It also governs residence time of bubbles: with a rise velocity of 0.2–0.3 m/s, small bubbles in a 1.5 m tall vessel are in contact with the liquid for only 5–8 seconds unless the impeller recirculates them. That recirculation is exactly what separates a good dispersion regime from a flooded one. The relationship between gas flow and tank geometry is covered in more detail in our guide to bioreactor aeration scale-up.

The Four Gas Dispersion Regimes in Stirred Bioreactors

For a given gas flow rate, a Rushton turbine passes through four distinct dispersion states as the impeller speed increases. These are visible in a transparent vessel and, indirectly, in the power draw and the gas holdup. Understanding them is the key to gas holdup interpretation, because holdup and kLa change sharply at each transition.

1. Flooding. At low speed relative to the gas flow, the impeller cannot pump the gas outward. Gas rises as a channel up the shaft and the impeller sits in a bubble of its own making. Radial dispersion is essentially absent, large coalescing bubbles reach the surface, and the liquid is poorly mixed. The gassed power draw falls by 60–70% from the ungassed value. Flooding is the worst case for gas holdup and kLa.

2. Loading. As speed rises, the impeller begins to capture gas. Large cavities form behind the blades in the characteristic 3-3-3 or 3-3 arrangement, and gas is dispersed only in the zone above the impeller. The lower part of the vessel stays gas-free. Power draw is reduced by 30–50%. Loading is an unstable and inefficient state, and a stirred tank held here will show noisy power and drifting kLa.

3. Complete dispersion. At the minimum speed for complete dispersion (N_CD), gas reaches the whole vessel above the impeller as small, uniform bubbles. The cavities stabilise into a symmetric six-cavity pattern. This is the minimum acceptable operating point for aerated bioprocesses. Below N_CD you are in loading, and the oxygen transfer you calculated from a correlation is not what you will get.

4. Recirculation. At higher speed again, fine bubbles are pulled back down and recirculate through every zone, including below the impeller. Gas holdup and kLa reach their maximum. The cost is power and, for cell culture, bubble-burst and shear exposure, which is why many mammalian processes deliberately operate near complete dispersion rather than full recirculation.

The practical reading: your gas holdup target is not a number, it is a regime. If you operate below N_CD, no tuning of VVM will restore predictable oxygen transfer. For a deeper comparison of what different impellers do at each regime, see our guide to bioreactor impeller selection.

How to Measure Gas Holdup in Bioreactors

Bioreactor gas holdup measurement is simpler than most characterisation work, but the method has to match the vessel and the media. Four approaches cover almost every case.

Liquid height (bed expansion). Record the liquid level before and after gassing and apply ε_g = (H_gassed − H_ungassed)/H_gassed. It needs a sight glass or a level probe and works well in glass lab vessels. It is inaccurate when foam or a rippling surface blurs the reading, and error grows when the rise is only a few millimetres, as with low-VVM mammalian cultures.

Hydrostatic pressure difference. Two pressure transducers at different heights give the mean density of the dispersion, and ε_g = 1 − ΔP/(ρ_L g Δh). This works in opaque broths and large vessels, but the transducers must be matched and the result is a local average between the two ports.

Gas disengagement. Stop the gas and the agitator at the same instant and watch the liquid level fall. The step change gives total gas holdup and the decay shape separates small and large bubble populations. It is a quick pilot-scale diagnostic but it disturbs the process, so it is a characterisation run, not an in-process measurement.

Local probes and imaging. Electrical conductivity or resistivity probes, optical fibre probes, and capillary suction systems give local holdup and, with imaging, bubble size. Endoscope probes with image analysis report a bubble size distribution directly. These are the tools for validating CFD or population balance models, but they need transparent or sterilisable port access.

Whichever method you choose, measure at the gas flow and impeller speed you plan to run, in the actual medium. Antifoam, surfactants such as Pluronic F-68, and proteins all suppress coalescence and raise gas holdup relative to water, sometimes by a factor of two. A holdup measurement in water is a starting point, not the answer. For the sparger side of this problem, see bioreactor sparger design and selection.

Gas Holdup Correlations for Scale-Up Prediction

The most widely used gas holdup correlations for stirred bioreactors follow a power-law form in gassed power per volume and superficial gas velocity. The general form from van’t Riet (1979) is:

ε_g ∝ (P_g/V)^0.2–0.4 × v_s^0.5–0.7

For coalescent (water-like) media, a common working form is:

ε_g = 0.026 × (P_g/V)^0.4 × v_s^0.5

with P_g/V in W/m³, v_s in m/s, and ε_g dimensionless (multiply by 100 for %). At P_g/V = 1,000 W/m³ and v_s = 1 cm/s, this gives ε_g = 0.026 × 15.8 × 0.1 = 4.1%. For non-coalescent media, such as salt-containing or protein-containing broths, Smith et al. (1977) proposed:

ε_g = 1.44 × [(P_g² N D_i³)/(σ_L² V_L)]^0.34 × (v_s/√(σ_L/ρ_L))^0.68

The critical detail is that every one of these correlations depends on the gassed power P_g, not the ungassed power P₀. For a Rushton turbine P_g/P₀ is typically 0.3–0.5 under production aeration, so plugging in P₀ overpredicts gas holdup and kLa. Calculate ungassed power from the power number and impeller speed, then apply a gassed-to-ungassed ratio that matches your gas flow number.

Figure 2 shows the shape of the response. Both exponents are below one, so gas holdup rises with diminishing returns: doubling gas velocity raises holdup by 2^0.5 − 1 = 41%, and doubling P_g/V raises it by 2^0.4 − 1 = 32%. This is why pushing agitation harder to rescue a gas-limited run gives so little oxygen transfer per watt, and why the cheaper levers are usually oxygen enrichment or headspace pressure.

Use the correlations to rank scale-up options and to detect when a proposed condition is implausible. Do not treat them as a substitute for measuring gas holdup at the target scale, especially in cell culture media containing surfactants. Our OTR and kLa Estimator applies the van’t Riet form so you can see how P/V and v_s trade off.

Bubble Size Distribution and Interfacial Area

Bubble size distribution in bioprocess vessels is broad. A single stirred tank holds bubbles from under 1 mm, formed near the impeller where turbulent shear is highest, to more than 10 mm in the bulk where coalescence dominates. A distribution that wide cannot be summarised by one diameter, so engineers use the Sauter mean diameter d₃₂, the diameter of a bubble with the same volume-to-surface ratio as the whole population. It is the right average for mass transfer because it weights interfacial area correctly.

In stirred bioreactors, d₃₂ typically falls between 1 and 6 mm. The classic Calderbank (1958) relationship, derived from the balance between turbulent breakup and surface tension, scales as:

d₃₂ ∝ (P_g/V)^−0.4 × σ^0.6

Two consequences follow. First, doubling gassed power per volume reduces d₃₂ by 2^−0.4, about 24%, so bubbles shrink slowly with agitation. Second, lowering surface tension is powerful: σ^0.6 means a drop from 72 to 45 mN/m cuts d₃₂ by about 25%. This is one reason surfactants and antifoams change oxygen transfer, though antifoams also add a film resistance that lowers k_L.

Combining the two quantities gives a feel for what you are buying. A bioreactor at ε_g = 4% with d₃₂ = 3 mm has an interfacial area of 6 × 0.04 / 0.003 = 80 m²/m³. With k_L = 3 × 10⁻⁴ m/s that gives kLa = 0.024 s⁻¹, or 86 h⁻¹. Halve the bubble diameter at constant holdup and area doubles to 160 m²/m³. That is the argument for microspargers, although fine bubbles carry two costs: they raise foaming and they strip CO₂ less effectively, as Sieblist et al. (2011) showed for large-scale cell-culture vessels.

When you need to measure the distribution rather than infer it, photographic image analysis in a transparent vessel gives the cleanest histogram, capillary suction probes handle opaque media, and laser diffraction or endoscope probes fit pilot and research reactors. All of them sample a small region, so repeat at several radial and axial positions before quoting a vessel-average d₃₂. For the mixing side of the picture, see our guide to mixing time in bioreactors, since bubbles that are not circulated spend most of their life where the oxygen is already depleted.

Impeller Flooding: When Gas Overwhelms the Impeller

Impeller flooding in a bioreactor is the loss of gas control by the impeller. It happens when the gas flow is high relative to the pumping capacity of the impeller, which is set by speed and diameter. The dimensionless groups that describe it are the gas flow number and the Froude number:

Fl_g = Q_g / (N D³)     Fr = N² D / g

Here Q_g is the volumetric gas flow, N the impeller speed in rev/s, D the impeller diameter and T the tank diameter. Fl_g measures how much gas the impeller is asked to handle per revolution, and Fr measures how strong its inertial pumping is against gravity.

The flooding–loading transition for a Rushton turbine was correlated by Nienow and co-workers as:

Fl_g,flood = 30 × Fr × (D/T)^3.5

If your operating Fl_g is below this value, the impeller is not flooded. If it is above, the impeller floods. For D/T = 0.33, (D/T)^3.5 = 0.0207, so Fl_g,flood = 0.62 × Fr. For complete dispersion the minimum speed correlates with Fl_g^0.5 × (D/T)^0.25, which means that once you are safely clear of flooding, the next target is N_CD, which sits at a higher speed than the flooding boundary.

Not every impeller floods at the same point. A pitched-blade turbine tolerates roughly 1.5 times more gas than a Rushton at the same speed, and axial hydrofoil impellers tolerate 2–3 times more, because their blades do not shed the large trailing gas cavities that choke a radial turbine. That higher flooding tolerance is why modern large-scale bioreactors often use a Rushton or a disc turbine at the sparger plus hydrofoils above. Figure 3 maps the boundaries.

Reading the map, the four marked points are the worked example in the next section. The 10 L vessel sits far below the Rushton line. The 2,000 L vessel at the same VVM sits closer, and when aeration is raised to 0.3 VVM it falls within about 20% of the flooding boundary. In practice, any operating point closer than about a factor of two to the boundary is unreliable, because feed and antifoam additions change bubble coalescence and the foam height shifts the effective sparging depth. The signature of approaching flooding is a falling power draw and a rising DO deficit that agitation no longer corrects.

Also remember that this correlation uses ungassed conditions. As the impeller approaches flooding the power draw falls, so real pumping is weaker than the nominal speed suggests. Monitor motor power or torque on the drive: a sustained drop of more than about 30% from the ungassed baseline at the same speed means loading or flooding. Linking this to impeller Reynolds number in the turbulent regime is simple, as covered in Reynolds number in bioreactors, since flooding limits are normally quoted for fully turbulent flow, Re > 10⁴.

How to Prevent Impeller Flooding During Bioreactor Scale-Up

The root cause of gas dispersion problems at scale is geometric. Superficial gas velocity equals VVM multiplied by liquid height (v_s = VVM × H), and height grows with vessel size. If you hold VVM constant, v_s rises in proportion to H, so a 2,000 L vessel at the same VVM sees roughly 6–7 times the superficial gas velocity of a 10 L vessel. Meanwhile, scale-up at constant P/V lowers impeller speed (N ∝ D^−2/3), which reduces the impeller’s ability to disperse gas. Fl_g increases and Fr falls, so the operating point drifts toward the flooding boundary in two ways at once.

Constant P/V also keeps gas holdup roughly similar, since holdup depends on P_g/V and v_s, but mixing time lengthens. For a 1,000-fold volume increase, mixing time grows about 5 times at constant P/V. Above roughly 2,000 L a single impeller cannot do both jobs, so large vessels use multiple impellers: a bottom impeller near the sparger disperses gas, and upper impellers distribute liquid and gas through the vessel. The bottom impeller is the one at risk of flooding, and the upper impellers inherit whatever gas it passes.

A practical prevention checklist

• Check Fl_g against the flooding limit at every scale. Aim for at least a 2-fold margin at the highest planned gas flow, including peak OUR in late culture.
• Scale on v_s and kLa, not VVM alone. Our guide to aeration scale-up explains why VVM is a lab-scale convenience.
• Use a higher-tolerance impeller where gas flow is high. Hydrofoils and PBTs raise the flooding ceiling by 1.5–3 times.
• Add oxygen, not air. Enrichment raises the driving force without raising Q_g. The Gas Mixing Calculator shows the blend needed.
• Match the sparger to the impeller. Placing it centrally under the impeller with a diameter smaller than D keeps gas in the swept zone.
• Watch power draw. Trend the drive power against the ungassed baseline as an online flooding alarm.

The example shows why a scale-up that looks correct on P/V and kLa can fail on dispersion: the correlations assume a dispersed impeller, and the flooding check tells you whether that assumption holds. Run the arithmetic for your own geometry with the Scale-Up Calculator, which compares P/V, tip speed and kLa side by side.

Frequently Asked Questions

Related Tools

• OTR & kLa Estimator — Estimate kLa for stirred tanks using van’t Riet and Büchs correlations
• Scale-Up Calculator — Compare scale-up criteria side by side for any vessel geometry
• Gas Mixing Calculator — Calculate O₂/air/N₂/CO₂ blends for oxygen enrichment
• DOE Generator — Design experiments for agitation and aeration characterization

References

1. van’t Riet K. Review of measuring methods and results in nonviscous gas-liquid mass transfer in stirred vessels. Ind Eng Chem Process Des Dev. 1979;18(3):357–364. doi:10.1021/i260071a001
2. Nienow AW, Wisdom DJ, Middleton JC. The effect of scale and geometry on flooding, recirculation, and power in gassed stirred vessels. Proceedings of the 2nd European Conference on Mixing. 1977;F1-1. (Conference proceedings; no DOI assigned.)
3. Calderbank PH. Physical rate processes in industrial fermentation, Part I: The interfacial area in gas-liquid contacting with mechanical agitation. Trans Inst Chem Eng. 1958;36:443–463. (No DOI assigned.)
4. Nienow AW. Hydrodynamics of stirred bioreactors. Appl Mech Rev. 1998;51(1):3–32. doi:10.1115/1.3098990
5. Sieblist C, Hägeholz O, Aehle M, Jenzsch M, Pohlscheidt M, Lübbert A. Insights into large-scale cell-culture reactors: II. Gas-phase mixing and CO₂ stripping. Biotechnol J. 2011;6(12):1547–1556. doi:10.1002/biot.201100153