Chemical Engineering Science 301 (2025) 120666

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Effects of the boundary conditions at the gas-liquid interface on single

hydrogen bubble growth in alkaline water electrolysis
F. Khalighi a , A.W. Vreman a,b,c,∗ , Y. Tang a,b , N.G. Deen a,b
a Power & Flow group, Department of Mechanical Engineering, Eindhoven University of Technology, PO Box 513, Eindhoven 5600 MB, the Netherlands
b Eindhoven Institute for Renewable Energy Systems (EIRES), Eindhoven University of Technology, PO Box 513, Eindhoven 5600 MB, the Netherlands
c Nobian Industrial Chemicals, P. O. Box 75730, Amersfoort 3811 LP, the Netherlands

A R T I C L E I N F O A B S T R A C T

Keywords: Alkaline water electrolysis is important for green hydrogen production. We simulate the growth of a single
Alkaline water electrolysis hydrogen bubble on a cathode in a 30 wt% KOH solution in a narrow channel. We develop and use a sharp
Hydrogen evolution reaction interface method to solve the Navier-Stokes equations, the species transport equations, and the potential equation
Growing hydrogen bubble
for a tertiary current distribution. To investigate the role of the mobility of the bubble interface, three different
Immersed boundary method
Marangoni flow
boundary conditions are used: the no-slip, the free-slip, and the Marangoni stress condition. The surface tension
No-slip and free-slip boundary conditions depends on the local electrolyte concentration. The simulation results show that different boundary conditions
lead to minor changes in electrochemical quantities but significantly affect the force on the bubble. The Marangoni
boundary condition leads to a relatively large force on the bubble, which is expected to accelerate bubble
detachment. This result makes plausible why the hydrogen bubbles in alkaline electrolysis are relatively small.

1. Introduction numerical simulations of bubble dynamics. Research has demonstrated

that boundary conditions can influence the bubbles rising in bulk, coa-
Hydrogen, known as a clean and sustainable energy carrier, can lescence (Liu et al., 2020; Vakarelski et al., 2024), and detachment of
be produced through water electrolysis. However, making this process bubbles (Park et al., 2023). The question thus arises: what is the effect
economically viable remains a challenge. During electrolysis, hydrogen of the boundary condition at H2 bubble interface in alkaline water elec-
is generated at the cathode, forming bubbles due to the diffusion of trolysis?
dissolved hydrogen in the surrounding supersaturated electrolyte. How- In clean water, the bubble surface cannot resist any tangential shear,
ever, these bubbles can have a negative impact on the electrochemical
which means that the tangential velocity at the interface should equal
reactions, i.e., leading to high overpotential, blocking the active surface
the velocity of liquid flow right next to it. This boundary condition is de-
area of the electrode, and consequently reducing electrolysis efficiency.
fined as the free-slip boundary condition (Liu et al., 2020). The presence
To develop efficient electrolysis systems for hydrogen production, it
of contaminants and surfactants results in zero tangential velocity at the
is crucial to understand bubble dynamics and their interactions with
interface. This boundary is called no-slip boundary condition (Liu et al.,
the flow field. Bubbles in electrolysis have been studied experimentally
and computationally, see for example references (de Groot and Vreman, 2020). Both the free-slip and no-slip boundary conditions are commonly
2021; Granados Mendoza, 2016; Haverkort and Rajaei, 2021; Hossain used in numerical simulations of bubbly flows. The Marangoni effect
et al., 2020; Janssen and Van Stralen, 1981; Raman et al., 2022; Sepahi occurs as stress at the gas-liquid interface due to gradients in surface
et al., 2022; Van Der Linde et al., 2017). tension, which induces convection in the adjacent phases. Depending on
Several studies have shown that flow fields can affect electrochemi- the origin of the surface tension gradient, temperature, concentration,
cal quantities (Ebadi et al., 2020; Khalighi et al., 2023; Li et al., 2018; or potential, it is referred to as thermocapillary convection, solutocap-
Nagai et al., 2006). On the other hand, the velocity field is influenced illary convection, or electrocapillary convection, respectively (Meulen-
by the boundary conditions at the gas-liquid interface. Properly address- broek et al., 2021). The Marangoni effect due to thermal effects has
ing the boundary condition at the gas-liquid interface is essential for been simulated for a bubble on a microelectrode (Massing et al., 2019;

* Corresponding author at: Power & Flow group, Department of Mechanical Engineering, Eindhoven University of Technology, PO Box 513, Eindhoven 5600 MB,
the Netherlands.
E-mail address: [email protected] (A.W. Vreman).

https://doi.org/10.1016/j.ces.2024.120666
Received 27 June 2024; Received in revised form 25 August 2024; Accepted 27 August 2024
Available online 30 August 2024
0009-2509/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Meulenbroek et al., 2021, 2024). In these cases, the bubble was larger Section 5 examines the effects of various boundary conditions in detail.
than the electrode, and the current density was very large (of the or- A summary and conclusion are provided in Section 6.
der of 105 A∕m2 ). In this paper, we consider a hydrogen bubble on a Since this work builds on our previous work (Khalighi et al., 2023),
macro-electrode (an electrode that is larger than the bubble) at a mod- many of the model assumptions in that paper are also used in the
est current density (approximately 1600 A∕m2 ), which is of the order present paper. One of these assumptions is equal mass diffusivities of
of current densities typically achieved at macro-electrodes. We simulate ions (𝐷K + = 𝐷OH− ), initially assumed because it simplifies the equations
the bubble in a narrow channel, representing the narrow space between to some extent and because the ion diffusivities for the electrolyte of in-
the electrode and the membrane in nearly zero-gap electrolysis. Further- terest (30 wt% KOH at 80 ◦ C) could not be found in literature. However,
more, we consider an electrolyte with high electrical conductivity (30 in reality the ion diffusivities are different (𝐷OH− > 𝐷K + ). Therefore, we
wt% KOH at 80 ◦ C, typically used in alkaline water electrolysis). Since include Appendix A, in which we extend the model and show the effect
the heat generated in the electrolyte is proportional to the square of of unequal ion diffusivities.
the current density and reversely proportional to the electrical conduc-
tivity, the thermal Marangoni effect is not expected to be relevant in 2. Mathematical and numerical modeling
our case. Furthermore, Haverkort (2024) introduced an estimate show-
ing that the solutal Marangoni force dominates the thermal Marangoni In this paper, the following assumptions are made: constant tem-
force up to 0.5 V overpotential. The electrocapillary effect might play perature, no water evaporation, electroneutrality (𝑐K + = 𝑐OH− = 𝑐KOH ).
a role for large bubble surface charge, but it is not clear how it should Furthermore, equal ion diffusivity coefficients are assumed, except in
be modeled (Meulenbroek, 2024). In this paper, we only simulate so- Appendix A.
lutocapillary convection, in which the surface tension depends on the
concentration of a solute. In this case, the solute is the electrolyte (KOH). 2.1. Governing equations
Dissolved hydrogen cannot cause the Marangoni convection because
gas-vapor thermodynamic equilibrium prevents concentration gradients The 3D time-dependent Navier-Stokes equations for a viscous, New-
tonian, laminar, and incompressible flow, with constant density and
across the bubble surface (Haverkort, 2024; Yang et al., 2018).
viscosity, are employed:
Overall, the Marangoni forces/effects have not been extensively in-
vestigated in open literature until recently. The effect of anions in acidic
∇ ⋅ 𝐮 = 𝟎, (1)
electrolytes on hydrogen gas bubble dynamics during the hydrogen evo-
𝜕𝐮 1
lution reaction on a platinum microelectrode was measured by Park et + (𝐮 ⋅ ∇) 𝐮 = − ∇𝑝 + 𝜈∇2 𝐮, (2)
𝜕𝑡 𝜌𝑓
al. (2023). They concluded that different surface tension of different
electrolytes can change the solutal Marangoni forces acting on the hy- where 𝐮 is the fluid velocity, 𝑝 is the pressure relative to the ambient
drogen gas bubbles. As a result, depending on the type of anion in the pressure, 𝜈 is kinematic viscosity, and 𝜌𝑓 is the fluid density. To simulate
electrolyte, the detachment radii and periods of hydrogen bubbles are the transport of hydrogen, water, and electrolyte, the following species
different. Meulenbroek et al. (2024) simulated the solutal Marangoni transport equation is used:
flow on a hydrogen bubble in various acidic electrolytes and simulated 𝜕𝑐𝑘
the growth of the oxygen bubble at the anode. They concluded that the + ∇ ⋅ (𝑐𝑘 𝐮) = 𝐷𝑘 ∇2 𝑐𝑘 , (3)
𝜕𝑡
Marangoni flow in the anodic compartment is opposite to the cathodic where 𝑐𝑘 is the concentration of species 𝑘, and 𝐷𝑘 is the constant Fickian
compartment in the same electrolyte. If the force delays the detachment diffusion coefficient of species 𝑘, while 𝑘 can be H2 , KOH and H2 O. As
at the cathode, it accelerates the detachment at the anode or vice versa. in our previous paper (Khalighi et al., 2023), we have chosen to solve a
Park et al. (2023) and Meulenbroek et al. (2024) showed that the detach- separate transport equation for H2 O with 𝐷H O = 𝐷KOH . An alternative
ment hydrogen bubble radius in different acidic electrolytes depends on 2
approach is to omit the transport equation for H2 O, to assume that the
Marangoni flow caused by the electrolyte concentration gradient. Park
total concentration 𝑐H + 𝑐KOH + 𝑐H O is constant, and to compute 𝑐H O
et al. (2023) and Meulenbroek et al. (2024) investigated a bubble on a 2 2 2
by subtracting 𝑐H and 𝑐KOH from the total concentration. If the total
microelectrode, in which the bubble diameter is typically larger than the 2
electrode diameter. Haverkort (2024) concluded that both electrostatic concentration is constant and the diffusivities of H2 , KOH and H2 O are
forces and solutal Marangoni forces can explain the differences between equal to each other then both approaches are formally equivalent. If
oxygen and hydrogen behavior in acidic and alkaline electrolytes. While we relax the latter condition to equal diffusivities of non-dilute species
both forces may play a role, solutal Marangoni flows are more effective (𝐷H O = 𝐷KOH ) then both approaches are expected to produce similar
2
in explaining various observed bubble behaviors, such as bubble jump- results, at least for the type of results shown in this paper.
off and return, floating on bubble carpets above a certain critical current Next, we introduce the current density and specify the equation for
density, and radial coalescence. In this work, we aim to understand how the electric potential and the boundary conditions for the tertiary den-
the velocity boundary conditions at the bubble surface affect the bub- sity system given by the Nernst-Planck equations for an electroneutral
ble dynamics. The growth of a single hydrogen bubble attached to an binary electrolyte (Fuller and Harb, 2018; Newman and Thomas-Alyea,
electrode in an alkaline electrolyte in a narrow channel is investigated. 2012). As in our previous paper (Khalighi et al., 2023), we simplify the
We analyze the effects of different velocity boundary conditions (no- equations by assuming 𝐷K + = 𝐷OH− = 𝐷KOH . This assumption implies
slip, free-slip, and Marangoni stress boundary) at the gas-liquid interface the following expressions for current density, electrical conductivity,
on bubble growth behavior, species concentration, and current density and charge balance:
distribution. To solve the Navier-Stokes equations, the mass transport 2𝐹 2 𝐷KOH
equations, and the potential equation for a tertiary current distribution, 𝐢 = −𝜅∇𝜙, 𝜅= 𝑐KOH , (4)
𝑅u 𝑇
we employ a sharp interface immersed boundary method implemented
in an in-house code.
∇ ⋅ 𝐢 = −∇ ⋅ (𝜅∇𝜙) = 0. (5)
This paper is organized as follows: Section 2 introduces the math- Here 𝐢 denotes the current density, 𝜅 the electrical conductivity, 𝜙 the
ematical and numerical modeling framework. Section 3 provides a de- electric potential, 𝐹 the Faraday constant, 𝑅u the universal gas constant,
tailed description of the immersed boundary method and its application and 𝑇 the absolute temperature. Equations and simulation results for
to different boundary conditions. In Section 4, we validate the method unequal ion diffusivities are shown in Appendix A.
for two steady-state flows past a sphere. Furthermore, a grid refinement Using Fick’s first law, the diffusive molar hydrogen flux at the bubble
study for the case of a growing bubble at the electrode is performed. surface (𝑆 ) can be calculated:

2
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

( ) ( )1∕2 ( ) ( )
⎡ 𝑐H O −𝛼𝑐 𝐹 𝜂 𝑐H 𝑐KOH 𝛼𝑎 𝐹 𝜂 ⎤⎥
𝑖 = −𝑖0 ⎢ 2 exp − 2
exp ,
⎢ 𝑐H O,0 𝑅𝑢 𝑇 𝑐H ,0 𝑐KOH,0 𝑅𝑢 𝑇 ⎥
⎣ 2 2 ⎦
(9c)
𝜕𝜙
𝑖 = −𝜅 , 𝜙 + 𝜂 = 𝜙1 . (9d)
𝜕𝑥
Equation (9c) presents the Butler-Volmer equation for a one-electron
reaction. The concentration factors are justified in the next paragraph.
This equation is used in combination with Equation (9d) to calculate cur-
rent density and overpotential. 𝛼𝑐 and 𝛼𝑎 are charge transfer coefficients
for the forward and reverse directions of the reaction, respectively, while
𝑖0 and 𝜂 are the exchange current density and overpotential, respec-
tively. 𝜙1 , a prescribed constant, is the potential of the electrode plus
a standard potential. The value of 𝜙 at 𝑥=0 represents the electric po-
tential in the electrolyte just outside the very thin electric double layer.
Fig. 1. 3D configuration of the case study.
Equation (9c) is used to compute the species flux at the wetted part of
the electrode surface.
𝑑𝑚 In fact, the chemical reaction at the cathode is a multi-step reaction.
= 𝐷 ∇𝑐 ⋅ 𝐧𝑑𝑆, (6) Various mechanisms can be constructed from the Volmer, Heyrovski,
𝑑𝑡 ∫ H2
𝑆 and Tafel reaction steps (Newman and Thomas-Alyea, 2012; Shinagawa
where 𝐧 represents the unit normal vector pointing outward from the et al., 2015). We assume a dominant reaction pathway consisting of a
bubble surface. The new bubble radius is obtained using the ideal gas rate-determining Volmer step, H2 O + e− ⇌ H𝑎𝑑 + OH− , and an equili-
law (Khalighi et al., 2023; van der Linde, 2019): brated Tafel step, 2 H𝑎𝑑 ⇌ H2 , where H𝑎𝑑 denotes a proton adsorbed to
the electrode. According to elementary reaction kinetics, the forward
4 3
𝜋𝑅 = 𝑚𝑅𝑢 𝑇 ∕𝑃g , (7) reaction rate in the Tafel step is proportional to 𝑐H 2 and the reverse
3 𝑎𝑑
reaction rate proportional to 𝑐H . Here 𝑐H denotes the surface concen-
where 𝑚 is the number of hydrogen moles in the bubble, 𝑅 is the bubble 2 𝑎𝑑
tration of adsorbed protons. Since the Tafel reaction is equilibrated in
radius, and 𝑃g = 𝑃liquid + 2𝜎∕𝑅 is the absolute pressure of the gas inside
this pathway, forward and reverse rates in the Tafel reaction are nearly
the bubble. 𝜎 denotes the liquid-gas surface tension. In the gas phase, equal, which implies that 𝑐H is proportional to (𝑐H )1∕2 . Likewise, the
the effects of inertia and viscosity are considered negligible. 𝑃liquid rep- 𝑎𝑑 2
forward reaction rate in the Volmer step is proportional to 𝑐H and
resents the sum of the local mean value of 𝑝 on the bubble surface and 2
O

the ambient pressure (𝑃 ). the reverse reaction rate proportional to 𝑐H 𝑐OH− . Due to the equi-
𝑎𝑑
libration of the Tafel step, the latter is proportional to (𝑐H )1∕2 𝑐OH− .
2
2.2. Configuration, initial conditions and boundary conditions Therefore, the first term in the Butler-volmer equation is taken propor-
tional to 𝑐H O and the second term is taken proportional to (𝑐H )1∕2 𝑐OH− .
2 2
It is remarked that Dukovic and Tobias (1987) also used a Butler-Volmer
Following a similar approach to Khalighi et al. (2023), our focus
equation with a reverse reaction rate proportional to (𝑐H )1∕2 . However,
is directed towards the hydrogen evolution process, consequently, only 2
one-half of the electrolyzer (the cathodic compartment) is simulated (see they did not provide the justification given above.
Fig. 1). The computational domain is displayed as a cube with dimen- The boundary conditions at the membrane (𝑥 = 𝐿) are:
sions 𝐿, with an electrode on the left, a membrane on the right, an inlet
𝑢𝑥 = 𝑢𝑦 = 𝑢𝑧 = 0, (10a)
at the bottom, and an outlet at the top. The bubble volume creates a
(time-dependent) hole in the flow domain, and the partial differential 𝑐k = 𝑐k,0 , (10b)
equations are not solved inside the bubble.
𝜙 = 0. (10c)
The initial conditions for flow, concentration species, and electric
field are as follows: The boundary conditions at the inlet (𝑦 = 0) are:
{ ( ) ( )
𝑥 𝑥
6 𝑢𝑦,inlet 𝐿𝑥 1 − 𝑥
if 𝑦 = 0 𝑢𝑥 = 𝑢𝑧 = 0, 𝑢𝑦 = 6 𝑢𝑦,inlet 1− , (11a)
𝑢𝑥 = 𝑢𝑧 = 0, 𝑢𝑦 = 𝐿 (8a) 𝐿 𝐿
0 otherwise, 𝑐k = 𝑐k,0 , (11b)
𝑐k = 𝑐k,0 , (8b) 𝜕𝜙
= 0. (11c)
𝜕𝑦
where 𝑢𝑦,inlet is the bulk velocity, and 𝑐k,0 is reference concentrations
for species 𝑘. As mentioned earlier, 𝑘 can be H2 , H2 O, and KOH. The The boundary conditions at the outlet (𝑦 = 𝐿) are:
reference concentration for hydrogen is equal to the saturation concen- 𝜕𝑢𝑥 𝜕𝑢𝑦 𝜕𝑢𝑧
tration of hydrogen at ambient pressure. = = = 0, 𝑝 = 0, (12a)
𝜕𝑦 𝜕𝑦 𝜕𝑦
The boundary conditions for flow, concentration species, and electric 𝜕𝑐k
field are specified in the following. = 0, (12b)
𝜕𝑦
The boundary conditions at the electrode (𝑥 = 0) are (Khalighi et al.,
𝜕𝜙
2023): = 0. (12c)
𝜕𝑦
𝑢𝑥 = 𝑢𝑦 = 𝑢𝑧 = 0, (9a) The boundary conditions at the bubble surface (𝑟 = 𝑅) are:

𝜕𝑐H 𝜕𝑐H O 𝜕𝑐KOH 𝜕𝑐H O

𝑖 𝜕𝑐 𝑖 𝑖 𝑐H = 𝑐H ,0 , = 0, 2
= 0, (13a)
−𝐷H 2
=− , −𝐷KOH KOH = − , −𝐷H O 2
= , 2 2 𝜕𝑟 𝜕𝑟
2 𝜕𝑥 2𝐹 𝜕𝑥 2𝐹 2 𝜕𝑥 𝐹
𝜕𝜙
(9b) = 0. (13b)
𝜕𝑟
3
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

For the case when the bubble surface is completely immobile due
to contaminations, a no-slip boundary condition is implemented at the
bubble surface.
𝑑𝑅
𝑢𝑟 = , 𝑢𝜑 = 0, 𝑢𝜃 = 0. (16)
𝑑𝑡
For the free-slip boundary condition, a Dirichlet boundary condition
is imposed for the normal velocity component only, while the tangential
components of the stress tensor are zero at the interface. With free-slip
boundaries, no fluid force is exerted from the fluid to the interface in
the tangential directions. Hence, the free-slip boundary condition is as
follows:
𝑑𝑅
𝑢𝑟 = , (17a)
𝑑𝑡
(𝜏 ⋅ 𝐞𝑟 ) ⋅ 𝐞𝜑 = 0, (17b)
Fig. 2. Spherical coordinates.
(𝜏 ⋅ 𝐞𝑟 ) ⋅ 𝐞𝜃 = 0, (17c)
where 𝜏 is the viscous stress tensor and 𝐞𝑟 , 𝐞𝜃 , and 𝐞𝜑 are standard base
Laplace pressure on the solubility at the surface is ignored, an approx-
vectors of the spherical coordinate system. The last two conditions can
imation that is expected to have negligible effects on the results in
be written as Kempe et al. (2015); Vreman (2016):
this paper. Equation (13a) indicates zero mass fluxes of KOH and H2 O
through the surface. Equation (13b) reveals that the normal component | 1 𝜕𝑢𝑟 𝜕𝑢𝜃 𝑢𝜃
𝜏𝑟𝜃 || = 𝜇( + − ) = 0, (18a)
of the current density vector at this surface is zero. The velocity bound- |𝑟=𝑅 𝑟 𝜕𝜃 𝜕𝑟 𝑟
ary conditions at the bubble interface are explained in more detail in the ( )
| 1 𝜕𝑢𝑟 𝜕𝑢 𝜑 𝑢𝜑
next section. Periodic boundary conditions are applied in the 𝑧-direction 𝜏𝑟𝜑 || =𝜇 + − = 0. (18b)
for all variables. As indicated above, we use inlet and outlet conditions |𝑟=𝑅 𝑟 sin 𝜃 𝜕𝜑 𝜕𝑟 𝑟
in 𝑦-direction. Since the bubble grows liquid should leave the domain As the normal velocity at the bubble interface does not depend on 𝜑 and
and therefore an outlet is needed. In realistic electrolyzers, upward liq- 𝜃 , Equation (18) reduces to:
uid flow is expected. The parabolic shape of the velocity profile is simply | 𝜕𝑢 𝑢
a convenient assumption; experimental conditions are probably differ- 𝜏𝑟𝜃 || = 𝜇( 𝜃 − 𝜃 ) = 0, (19a)
|𝑟=𝑅 𝜕𝑟 𝑟
ent. The effect of the magnitude of the inlet velocity was investigated in
our previous work Khalighi et al. (2023). | 𝜕𝑢𝜑 𝑢𝜑
𝜏𝑟𝜑 || = 𝜇( − ) = 0. (19b)
|𝑟=𝑅 𝜕𝑟 𝑟
2.2.1. Velocity boundary conditions at the bubble surface Marangoni flow has a similar boundary condition at the bubble inter-
In addition to the Cartesian velocity components, it is convenient face, with the distinction that the stress tensor in the tangential direction
to define spherical velocity components using the standard transforma- is not homogeneous:
tion from Cartesian to spherical coordinates. The origin of the spherical
frame of reference is at the center of the sphere. The velocity compo- 1 𝜕𝜎
(𝜏 ⋅ 𝐞𝑟 ) ⋅ 𝐞𝜑 = , (20a)
nents in the Cartesian and spherical coordinates in the vector form are 𝑟 𝜕𝜑
defined as 𝐮 = (𝑢1 , 𝑢2 , 𝑢3 ) = (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ) and 𝐮̃ = (𝑢̃ 1 , 𝑢̃ 2 , 𝑢̃ 3 ) = (𝑢̃ 𝑟 , 𝑢̃ 𝜃 , 𝑢̃ 𝜑 ), 1 𝜕𝜎
(𝜏 ⋅ 𝐞𝑟 ) ⋅ 𝐞𝜃 = . (20b)
respectively. The velocity transformation from Cartesian coordinates (𝐮) 𝑟 𝜕𝜃
to spherical coordinates (𝐮 ̃ ) and vice versa are as follows Vreman (2016): 𝜎 is surface tension at the bubble interface. We consider a surface ten-
sion that is a function of the electrolyte concentration (𝑐KOH ) only and
𝐮̃ 𝑖 = 𝐸𝑖,𝑗 (𝐮𝑗 − 𝐮b𝑗 ), 𝐮𝑗 = 𝐸𝑗,𝑖 𝐮̃ 𝑖 + 𝐮b𝑗 , (14a) assume that the derivative with respect to the electrolyte concentration
⎡ sin𝜃cos𝜑 sin𝜃sin𝜑 cos𝜃 ⎤ 𝜎𝑐 = 𝜕𝜎 /𝜕𝑐KOH is constant. Thus Equation (20) can be written as fol-
𝐸 = ⎢ cos𝜃cos𝜑 cos𝜃sin𝜑 −sin𝜃 ⎥ , (14b) lows:
⎢ ⎥
⎣ −sin𝜑 cos𝜑 0 ⎦ | 𝜕𝑢 𝑢 𝜎 𝜕𝑐
𝜏𝑟𝜃 || = 𝜇( 𝜃 − 𝜃 ) = − 𝑐 KOH , (21a)
where |𝑟=𝑅 𝜕𝑟 𝑟 𝑟 𝜕𝜃
𝑧 − 𝑧𝑐 𝑦 − 𝑦𝑐 | 𝜕𝑢𝜑 𝑢𝜑 𝜎 𝜕𝑐
𝜃 = arccos , 𝜑 = atan2 , 𝜏𝑟𝜑 || = 𝜇( − ) = − 𝑐 KOH . (21b)
𝑟 𝑥 − 𝑥𝑐 |𝑟=𝑅 𝜕𝑟 𝑟 𝑟 𝜕𝜑
√ (15)
𝑟 = (𝑥 − 𝑥𝑐 )2 + (𝑦 − 𝑦𝑐 )2 + (𝑧 − 𝑧𝑐 )2 , 3. Numerical method
where 𝑥𝑐 , 𝑦𝑐 , and 𝑧𝑐 are coordinates of the bubble center (see Fig. 2). 𝐮𝑏
is the velocity of the bubble center. Since the bubble has a small radius, The numerical method is an immersed boundary method (IBM), im-
the Bond number is very small (𝐵𝑜 = (𝜌𝑓 − 𝜌𝑔 )𝑔𝑅2 ∕𝜎 < 1, where 𝜌𝑔 is plemented as an in-house code. A staggered grid is used to discretize
the gas density, and 𝜌𝑔 ≪ 𝜌𝑓 ), meaning surface tension forces dominate the equations. The spatial derivatives are approximated by second-order
gravitational forces. As a result, the bubble is assumed to be spherical, central differences, except in the species equations, in which the con-
with surface area 𝑆 and radius 𝑅. The point of 𝑆 nearest to the electrode vective terms are approximated by the first-order upwind method in
remains fixed in time (in this study, the distance between the electrode advective form. The continuity equation is approximated by an artifi-
and the bubble surface is assumed to be 𝛿 = Δ𝑥, unless mentioned oth- cial compressibility method (Khalighi et al., 2023; Vreman, 2020). The
erwise). Thus the center of the bubble moves in the 𝑥 direction, with explicit Euler scheme is used for the temporal discretization. The pres-
velocity 𝑑𝑅∕𝑑𝑡, while 𝑦𝑐 = 𝑧𝑐 = 𝐿∕2. Note that “atan2” is similar to sure in the artificial compressibility method and the potential in the
“arctan” with the key difference that it can recognize the quadrant of iterative method used for the potential equation are both initialized by
the point and returns angles in the range (-𝜋 ,𝜋 ] radians. zero.

4
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Kempe et al. (2015)introduced the following procedure to impose

free-slip boundary conditions on the surface of a sphere in the context of
an immersed boundary method with a diffuse interface. For each veloc-
ity forcing point of the diffuse interface, a line perpendicular to the exact
interface was drawn, connecting the forcing point and a probe point out-
side the sphere. Because the IBM used a diffuse interface, a rather large
distance of two grid sizes between the probe point and the exact sur-
face was found to be most accurate and was therefore recommended. At
the probe point, trilinear interpolation was used to obtain the Cartesian
velocity vector, which was subsequently transformed into the spherical
velocity vector. Next, the spherical velocity vector at the forcing point
was computed from the discretization of the free-slip boundary condi-
tion, Equation (17). Finally, the spherical velocity vector at the forcing
point was transformed into the Cartesian velocity vector, which includes
Fig. 3. Immersed boundary method illustration.
the desired Cartesian velocity component at the forcing point. We use
the same procedure for the free-slip boundary condition. However, an
3.1. Immersed boundary method important difference is that we use this method to obtain the velocity at
IB points in a sharp interface IBM and not at forcing points in a diffuse in-
In the immersed boundary method, cells are classified based on the terface method. Therefore, we also use a smaller distance between probe
location of their centers. Cells situated in the fluid are called fluid cells. and surface, namely one grid size instead of two grid sizes, which likely
Cells within the bubble that are adjacent to at least one fluid cell are improves the accuracy of the method in a sharp-interface IBM. Further-
known as immersed boundary (IB) cells, also referred to as ghost cells, more, we extend the method to prescribe Marangoni stresses, while we
see for example Das et al. (2018). also use it to impose the no-slip velocity boundary condition.
Boundary faces are defined as the interfaces between IB cells and
fluid cells. A staggered grid is employed, where IB velocity points are 3.1.1. Dirichlet condition
located at the midpoint between two adjacent scalar points (pressure, To impose a Dirichlet boundary condition at the bubble surface on
concentration, or potential). Consequently, unlike pressure, concentra- flow variable Φ, the value at the IB point should be calculated using
tion, or potential IB points, some velocity IB points are not inside but adjacent points, ensuring that Φ = Φ𝑆 at the liquid-gas interface. One-
outside the bubble. The governing equations are not solved inside the dimensional linear interpolation is employed to compute the value at
bubble. For further explanation, readers are directed to Khalighi et al. the IB point using the values at the interface point and probe point. The
(2023); Vreman (2020). interpolation formula is given by:
In this paper, the Dirichlet, Neumann, and stress conditions at the
bubble surface are discretized along the normal direction to the surface, 𝑟S − 𝑟IB 𝑟 − 𝑟IB
ΦIB = − ΦS + P Φ , (22)
as illustrated in Fig. 3. The inspiration for this approach came from 𝑟P − 𝑟S 𝑟P − 𝑟S P
the distant image method, one of the sharp-interface IBM variants in- where 𝑟IB , 𝑟S = 𝑅, and 𝑟P are the distance of the IB, S, and P points from
troduced and tested in Vreman (2020). Fig. 3 shows that the method the bubble center.
relies on a probe point P (the image point of the IB point) placed at a For the velocity, under no-slip boundary condition, Φ in Equa-
distance 𝑑 outside the interface, such that the line connecting the IB tion (22) represents the three velocity components in spherical coor-
point and the probe point is normal to the surface. The value of 𝑑 is dinates. For free-slip boundary conditions and Marangoni flow, Φ in
the same for all probe points; we use 𝑑 = Δ𝑥 = Δ𝑦 = Δ𝑧, as in Vreman Equation (22) represents the radial velocity component only (𝑢𝑟 ). For
(2020). The variable values at the probe point are calculated using tri- the scalars, Φ in Equation (22) represents the hydrogen concentration
linear interpolation. The distant image method is similar to the mirror only. Of course, it is also possible to obtain the no-slip condition by ap-
image method proposed by Ghias et al. (2007); Mark and van Wachem plying Equation (22) directly to the Cartesian velocity components (after
(2008). In the latter, 𝑑 is not the same for each probe point, but equal transformation of the no-slip boundary condition Equation (16) to Carte-
to the distance between the surface and the IB point, which means that sian coordinates). Then the distant image method of Vreman (2020) is
𝑑 can become much smaller than the grid size or even zero. The advan- recovered. It should not change the results because Equation (14a) and
tage of the distant image method is that an explicit discretization can be Equation (22) are linear relations.
used for the viscous terms because the distance between the probe point
and interface is chosen equal to (or of the same order as) the grid size. 3.1.2. Neumann and stress conditions
Like the mirror image method has been extended to Neumann bound- The Neumann boundary condition defines the gradient of Φ at the
ary conditions (Ghias et al., 2007; Mark and van Wachem, 2008), the bubble surface. For a surface flux 𝐴, it is defined as 𝜕Φ∕𝜕𝑟 = 𝐴. The
distant image method can readily be extended to Neumann conditions value of Φ at the IB point (ΦIB ) can be obtained from a straightforward
for scalars. We do this in the present paper. first-order discretization of the Neumann condition,
Das et al. (2018) introduced another IBM discretization for the Neu-
mann condition, which inspired the discretization of the Neumann con- ΦIB = ΦP − (𝑟P − 𝑟IB )𝐴. (23)
dition that we used in our previous work Khalighi et al. (2023). In
Das et al. (2018), the Neumann condition is imposed via a Dirichlet The above equation is directly used to obtain the potential, electrolyte
condition. The Dirichlet condition is discretized using a third-order ac- concentration, and water concentration at IB points (𝐴 = 0 then). For the
curate scheme aligned along a Cartesian grid line. The scalar value at pressure, a simpler extrapolation technique is used (Vreman, 2020)).
the corresponding surface intersection point is obtained from the Neu- The stress conditions in the free-slip and Marangoni boundary can
mann condition, discretized by a second-order accurate scheme along symbolically be written as 𝜕Φ∕𝜕𝑟 − Φ∕𝑟 = 𝐴, where Φ represents the
a straight line normal to the surface. This line contains the surface in- tangential velocity components 𝑢𝜃 and 𝑢𝜑 and 𝐴 a stress component at
tersection point and two probe points outside the object. In order to the surface. The tangential velocity (ΦIB ) at the IB point is computed
cope with multiple boundaries possibly near to each other, Claassen et from:
al. (2024) modified this method into a hybrid first-order second-order ΦP − ΦIB ΦS
method. − = 𝐴. (24)
𝑟P − 𝑟IB 𝑟S

5
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

The velocity at the bubble interface can be expressed in terms of the Table 1
velocity at the probe point and the IB point using linear interpolation. Drag forces for different boundary conditions on the bubble surface at 𝑅𝑒 =
Therefore, Equation (24), using this interpolation term and under some 0.075 obtained using the IBM code.
manipulations, is modified as follows: Domain length No-slip Free-slip
( )
ΦP − ΦIB 1 𝑟P − 𝑟S IB 𝑟S − 𝑟IB P 4𝑅 1.25 nN 0.55 nN
− Φ + Φ = 𝐴, (25) 8𝑅 0.63 nN 0.36 nN
𝑟P − 𝑟IB 𝑅 𝑟P − 𝑟IB 𝑟P − 𝑟IB
16𝑅 0.49 nN 0.31 nN
which implies Analytical 0.40 nN 0.27 nN

𝑟IB P 𝑟
ΦIB = Φ + ( IB − 1)𝑅𝐴. (26)
𝑟P 𝑟P Table 2
The value of 𝐴 is zero for the free-slip boundary condition, and then Drag forces for different boundary conditions on the bubble surface obtained
the above equation becomes equivalent to the one used by Kempe using Fluent and the IBM code.
et al. (2015) for the tangential velocity at the forcing point in a dif- No-slip Free-slip
fuse interface IBM. However, for Marangoni flow, 𝐴 is nonzero and IBM code 1.00 nN 0.48 nN
equals −𝜎𝑐 𝜕𝑐KOH /𝜕𝜑 or −𝜎𝑐 𝜕𝑐KOH /𝜕𝜃 . Thus, to simulate Marangoni Fluent 0.97 nN 0.46 nN
flow, the electrolyte concentration gradient in the tangential directions
(𝜕𝑐KOH /𝜕𝜑 and 𝜕𝑐KOH /𝜕𝜃 ) should be known. Two probe points (de-
noted by P1 and P2 ) are selected along the tangential direction, at is replaced by a free-slip boundary condition. The bubble radius, the
distances 𝑑 and -𝑑 from the S point (see Fig. 3). The electrolyte con- uniform velocity at the inlet, and the dynamic viscosity of the fluid are
centrations at these probe locations are then determined using trilinear 𝑅 = 25 μm, 𝑈∞ = 1 mm∕s, 𝜌𝑓 = 1258 kg∕m3 and 𝜇 = 8.43 ×10−4 Pa s, re-
interpolation. Subsequently, the gradient of electrolyte concentration at spectively. This corresponds to a Reynolds number (𝑅𝑒 = 2 𝜌𝑓 𝑈∞ 𝑅∕𝜇 )
the bubble interface is computed using the following equation: of 0.075. The analytical solutions for the drag force at zero Reynolds
P P
number for both no-slip and free-slip boundary conditions yield 0.40 nN
𝜕𝑐KOH 𝑐KOH − 𝑐KOH
1 2
and 0.27 nN, respectively.
= . (27)
𝜕𝜑 2𝑑 Table 1 shows the drag force acting on the sphere for various domain
lengths, as computed using the IBM code (in-house code). As the domain
Similarly, the electrolyte gradient is calculated in the 𝜃 direction.
length in all three directions increases, the drag force approaches closer
Once the velocities in spherical coordinates are determined at the
to the analytical solution. It is reasonable to assume that this difference
IB point, they are converted from spherical to Cartesian coordinates
becomes even smaller if the domain size is further increased. This was
using the relation between Cartesian and spherical coordinates (Equa-
however not done because of the large computational demand (the IBM
tion (14a)).
code uses a uniform grid with grid size Δ𝑥 = 𝑅∕12).
4. Validation and verification
4.2. A stagnant sphere close to a wall
The computational simulations in our earlier publication (Khalighi
et al., 2023) for no-slip boundary conditions were validated. For fur- The domain remains unchanged with the description provided ear-
ther validation of the numerical method, we simulate steady-state flows lier in the configuration section, i.e. the bubble is again a sphere very
around a stagnant sphere for which analytical solutions are available, close to the electrode, which is a wall. However, in this case, the bubble
for no-slip and free-slip boundary conditions (section 4.1). Next, we sim- does not grow. The bubble radius is 𝑅 = 25 μm. The physical proper-
ulate a stagnant sphere near an electrode in the geometry and compare ties of this validation case are similar to the studied case of the growing
the results with results obtained using ANSYS-Fluent software, also for bubble and are presented in Table 5. In the IBM simulation, a uniform
no-slip and free-slip boundary conditions (section 4.2). Furthermore, a grid with grid size Δ𝑥 = 𝑅∕24 is used. Table 2 shows the drag force as
mesh refinement study of the growing bubble case is presented for all computed by the IBM code and by ANSYS-Fluent. The drag force in Flu-
three types of boundary conditions (section 4.3). ent was obtained from the momentum balance in the y-direction. The
difference between the obtained values is small, another validation of
4.1. Uniform flow past a sphere the IBM code.

Stokes (1851) introduced the basic formula for calculating the drag 4.3. Mesh refinement study for the growing bubble case
force on a stagnant sphere in a uniform flow. The assumptions of this
formula are an infinite domain, an infinitely small Reynolds number, A mesh refinement study is performed to evaluate the dependence
and no-slip boundary conditions at the sphere surface. The drag force of the numerical solution to the mesh resolution. The domain configu-
equation is as follows: ration and physical properties are similar to the model in the previous
section 4.2. This study investigates the lift and drag forces on a grow-
𝐹𝑑 = 6𝜋𝜇𝑅𝑈∞ , (28) ing hydrogen bubble with final radius 𝑅 = 15 μm. The mesh refinement
process started with a coarse mesh 𝑁 = 48 cells per direction, with sub-
where 𝜇 is the dynamic viscosity, 𝑅 is the radius of the sphere, and 𝑈∞ sequent refinements 𝑁 = 96 and 𝑁 = 192 cells in each direction. For
is the velocity of the uniform flow infinitely far away from the sphere. the different meshes, the initial bubble radius (𝑅0 ) is respectively 3Δ𝑥,
Hadamard (1911) and Rybczynski (1911) introduced an equation to cal- 6Δ𝑥, and 12Δ𝑥. The distance between the bubble and the electrode (𝛿 )
culate the drag force for free-slip boundary conditions at the bubble is respectively Δ𝑥, 2Δ𝑥, and 4Δ𝑥. This approach ensures that the geom-
surface: etry is consistent across all mesh resolutions.
Fig. 4 shows the bubble radius as a function of time for various
𝐹𝑑 = 4𝜋𝜇𝑅𝑈∞ . (29)
boundary conditions with different mesh resolutions. For each boundary
As a first validation case, we consider a domain similar to the previously condition, the profiles of the bubble radius are close across the different
introduced configuration, with the difference that the bubble is located mesh resolutions. The mesh dependence is shown negligible for the no-
in the center of the domain, a uniform velocity is applied at the inlet, and slip and free-slip boundary conditions, and acceptable for the Marangoni
the no-slip velocity boundary condition at the electrode and membrane stress boundary condition. Table 3 shows the results of the lift (𝐹𝑥 ) and

6
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Table 5
Physical parameters of the base case.

𝐿 (μm) 100 𝜈 (m2 ∕s) 6.7 × 10−7

𝑖0 (A∕m2 ) 1 𝑇 (K) 353
𝑃 (Pa) 105 𝑣inlet (mm∕s) 0.67
𝜎 (N∕m) 7.7 × 10−2 KOH (wt%) 30
𝜙1 (V) −0.45 𝜌𝑓 (kg∕m3 ) 1258
𝛼𝑐 0.5 𝛼𝑎 0.5
𝜎𝑐 (Nm2 ∕mol) 2 × 10−6

𝑘 H2 KOH H2 O
𝐷𝑘 (m2 ∕s) 5.8 × 10−9 3.2 × 10−9 3.2 × 10−9
𝑐𝑘,0 (mol∕m3 ) 0.16 6700 49000

Fig. 4. Bubble radius 𝑅(𝑡) versus time for various boundary conditions at differ- Table 6
ent mesh resolutions. Lift (𝐹𝑥 ) and drag (𝐹𝑦 ) forces for different boundary conditions on the bubble
surface obtained using the IBM code.

Table 3 𝐹𝑥 (nN) 𝐹𝑦 (nN)

Mesh refinement study of lift and drag forces on a growing bubble with radius No-slip −0.18 1.1
𝑅 = 15 μm for various boundary conditions with the same geometry. Free-slip 0.004 0.62
Marangoni 19.5 1.5
Boundary condition 𝑁 𝐹𝑥 (nN) 𝐹𝑦 (nN)
Buoyancy - 0.8
48 −0.13 0.28
No-slip 96 −0.14 0.28
192 −0.15 0.28 plained by the fact that the Marangoni flow mainly affects the region
48 −0.04 0.17 near the bubble foot. As the distance between the bubble and the elec-
Free-slip 96 −0.04 0.18 trode increases, the effect of the Marangoni flow decreases and leads
192 −0.03 0.19 to a decrease in the lift force. This finding highlights the sensitivity of
48 11.10 0.43 the lift force in the Marangoni flow case to the distance between the
Marangoni 96 5.90 0.46 electrode and the bubble.
192 4.73 0.42
5. Results and discussion
Table 4
Mesh refinement study of lift and drag forces on a growing bubble with radius The physical properties of the study case are detailed in Table 5.
𝑅 = 15 μm for various boundary conditions, using 𝑁 = 96. These features correspond to a 30 wt% KOH solution at 80 ◦ C, which is
commonly used in industrial alkaline water electrolysis. More specifi-
Boundary condition 𝛿∕Δ𝑥 𝑅0 ∕Δ𝑥 𝐹𝑥 (nN) 𝐹𝑦 (nN)
cally, we choose 𝐷KOH such that the initial 𝜅 equals 136𝑆∕𝑚, which
1 3 −0.18 0.27 is a good approximation of the electric conductivity of a 30 wt%
No-slip 1 6 −0.17 0.27
KOH solution at 80 ◦ C (de Groot and Vreman, 2021; Gilliam et al.,
2 3 −0.15 0.28
2007). The derivative of the KOH concentration (𝜎𝑐 = 𝜕𝜎 /𝜕𝑐KOH ) is
1 3 −0.03 0.17 2 × 10−6 N m2 /mol (Weissenborn and Pugh, 1996).
Free-slip 1 6 −0.03 0.17
The number of grid cells is 963 in the simulations, the grid is uniform,
2 3 −0.04 0.18
and the time step corresponding to this grid is Δ𝑡 = 2.02 × 10−7 s.
1 3 10.38 0.62 Fig. 5 compares the velocity fields for the case with no-slip and
Marangoni 1 6 10.37 0.58 free-slip boundary conditions at the 𝑥-𝑦 plane in the center of the compu-
2 3 5.9 0.48
tational domain when the bubble radius reaches 𝑅 = 25 μm. Simulations
with no-slip, free-slip, and Marangoni stress conditions are conducted to
drag force (𝐹𝑦 ) for all velocity boundary conditions at the bubble surface study the boundary condition’s effects on bubble growth and surround-
for various mesh resolutions. The results for both no-slip and free-slip ing flow/current field. As seen, the tangential velocity at the bubble
boundary conditions are almost consistent at all mesh resolutions. How- surface is zero for the no-slip condition, while it is non-zero for the
ever, the Marangoni flow case is significantly more sensitive to mesh free-slip condition. Fig. 6 shows the velocity vector and contour plots
refinement. Specifically, when refining from 𝑁 = 48 to 𝑁 = 96, the lat- for the Marangoni flow. The velocity range is significantly larger in the
eral Marangoni force changes by approximately 88%, while the change Marangoni case, approximately 33 times greater than in the free-slip
from 𝑁 = 96 to 𝑁 = 192 is less than 25%. This is considerable, but case. In the Marangoni case, the velocity is highest at the half of the
sufficiently small to trust at least the order of magnitude of the force bubble surface facing the electrode. Furthermore, the velocity at the
computed for the Marangoni stress boundary condition from simula- bubble surface in the Marangoni case is directed toward the electrode.
tions with a resolution of 𝑁 =96. The azimuthal velocity (𝑢𝜑 ) at the bubble surface for a growing bub-
Table 4 shows the drag and lift forces for all boundary conditions ble with 𝑅 = 25 μm, for different boundary conditions, is shown in Fig. 7.
when 𝑁 = 96, considering different initial bubble radius (𝑅0 ) and dis- The Marangoni effect strongly increases at the bubble surface.
tances between the electrode and the bubble (𝛿 ). For both no-slip and For a growing bubble with a radius of 𝑅 = 25 μm, the values of the
free-slip boundary conditions, the results are almost the same for all lift force and drag force on the bubble in the 𝑥 and 𝑦 direction (exclud-
geometries. In the case of Marangoni flow, changing the initial bubble ing the buoyancy force) are presented in Table 6. The buoyancy force
radius does not significantly change the acting forces. However, dou- is computed by (𝜌𝑓 − 𝜌𝑔 )𝑔𝑉bubble , where 𝜌𝑔 is the gas density, 𝑔 is the
bling the distance between the electrode and the bubble can result in gravitational acceleration, and 𝑉bubble is the bubble volume. As for the
a reduction of the horizontal force (𝐹𝑥 ) of approximately 43%, while forces exerted on the bubble, shown in Table 6, the free slip boundary
the effect on the vertical force (𝐹𝑦 ) remains very small. This can be ex- condition leads to the lowest drag and lift forces compared to the other

7
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Fig. 5. Velocity (vector, 𝑥 component, 𝑦 component) for no-slip (a, b, c) and free-slip (d, e, f) boundary conditions at the bubble surface at the 𝑥-𝑦 plane in the center
of the computational domain (𝑧∕𝐿 = 0.5).

Fig. 6. Velocity (vector, 𝑥 component, 𝑦 component) for Marangoni flow (a, b, c) at the 𝑥-𝑦 plane in the center of the computational domain (𝑧∕𝐿 = 0.5).

two boundary conditions, because viscous forces do not play a role in no-slip, free-slip, and Marangoni flow, respectively. The mean concen-
the free slip boundary condition. These forces indicate the critical in- tration in the Marangoni case is slightly lower because the outlet velocity
fluence of boundary conditions at the bubble interface on the bubble is larger near the electrode (see Figs. 5, and 6), so more hydrogen is lost
detachment process. Considering that the lift force for Marangoni flow through the outlet. Consequently, the mean hydrogen concentration in
is 114 times larger than for no-slip conditions and 487 times larger than the domain decreases. Please note that the hydrogen supersaturation
for free-slip conditions, Marangoni flow significantly increases the prob- concentration is very high, which makes our assumption of only one
ability of electrolytic hydrogen bubble detachment in a KOH solution. bubble nucleation unrealistic, as multiple bubbles typically form under
Accelerating the bubble detachment process may increase the efficiency these conditions. However, to simplify the analysis and focus solely on
of electrolysis. the effect of the boundary conditions, we stick to the single bubble case.
Fig. 8 compares the effects of boundary conditions on the hydrogen Fig. 9 shows the radial hydrogen flux at the bubble interface versus
concentration profile during the growth of a single hydrogen bubble. the angle for all boundary conditions studied. For the Marangoni case,
Despite the clear differences in flow fields led by different boundary the velocity at the bubble surface increases, and this velocity is towards
conditions, the hydrogen concentration profiles resemble each other. the electrode. Therefore, the bubble surface boundary layer is faster re-
The mean values of the hydrogen concentration in the plane shown, placed by a liquid of low hydrogen concentration. This causes the lower
the 𝑥-𝑦 plane at 𝑧∕𝐿 = 0.5, are 𝑐H ,mean = 2.5, 2.4 and 2.1 mol∕m3 for radial hydrogen flux observed in the Marangoni profile. As shown in
2

8
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Fig. 7. Azimuthal velocity (𝑢𝜑 ) at the bubble surface in the 𝑥-𝑦 plane at 𝑧∕𝐿 = 0.5 (𝜃 = 90◦ ) for (a) no-slip and free-slip boundary conditions at the bubble surface
and (b) Marangoni flow.

Fig. 8. The dissolved hydrogen concentration normalized by 𝑐H ,0 for a bubble growing at 𝑅 = 25 μm in the 𝑥-𝑦 plane at 𝑧∕𝐿 = 0.5 for different boundary conditions.
2

Fig. 9. The radial hydrogen flux across the bubble interface versus angle 𝜑 for
a single bubble in the 𝑥-𝑦 plane at 𝑧∕𝐿 = 0.5 (𝜃 = 90◦ ) with 𝑅 = 25 μm for Fig. 10. The growth of radius 𝑅(𝑡) versus time for a single bubble for various
different boundary conditions at the bubble surface. 𝜑 = 180◦ corresponds to boundary conditions. The dotted line refers to the power laws 𝑅 ∝ 𝑡0.8 .
the bubble foot.
at the bubble foot. This causes a strong gradient of KOH concentra-
Fig. 8, the hydrogen concentration on the side of the inlet is lower than tion along the bubble surface near the bubble foot. This gradient causes
on the side of the outlet, which explains why in Fig. 9 the radial hydro- the Marangoni flow when the Marangoni stress boundary condition is
gen flux for 𝜑 > 180 is smaller than for 𝜑 < 180, for all three boundary switched on. The Marangoni flow is so strong that it significantly mod-
conditions. ified the KOH concentration profile at the bubble surface, see Fig. 11.
A logarithmic plot of the bubble radius is shown in Fig. 10. As men- This figure shows a lower concentration in the Marangoni case, with the
tioned earlier, the hydrogen concentration is lowest in the Marangoni most significant drop occurring at the bubble foot (𝜑 = 180◦ ). Fig. 12
flow case, which leads to slower bubble growth. Furthermore, the dif- shows the KOH concentration at the bubble surface. The narrower shape
ference in the growth profiles between the no-slip and free-slip bound- of the Marangoni profile indicates that the KOH concentration bound-
ary conditions is negligible. Fig. 11 shows the electrolyte concentration ary layer at the electrode is thinner compared to the no-slip and free-slip
(KOH) in the midplane, where the maximum concentration is observed boundary conditions. Similar to the hydrogen concentration boundary

9
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Fig. 11. The electrolyte concentration normalized by 𝑐KOH,0 for a bubble growing at 𝑅 = 25 μm in the 𝑥-𝑦 plane at 𝑧∕𝐿 = 0.5 for different boundary conditions.

are checked. The validated code is used for numerical simulations of a

growing H2 bubble under different boundary conditions. The boundary
conditions effects are investigated on bubble growing behavior as well
as the surrounding flow/current dynamics. The following conclusions
can be drawn.
The velocity magnitude in the Marangoni case is approximately 20 times
larger than in the no-slip and free-slip cases. In the Marangoni case, the
highest velocity occurs on the half of the bubble surface facing the elec-
trode, and the velocity is directed toward the electrode. This results in
a significant increase in lift force, 114 times larger than in the no-slip
and the free-slip conditions. Consequently, this might promote bubble
detachment. Furthermore, this results in a smaller detachment hydro-
gen bubble radius.
The free-slip boundary condition has the lowest drag and lift forces com-
Fig. 12. KOH concentration at the bubble surface for 𝑅 = 25 μm for different
pared to the other two boundary conditions because viscous forces do
boundary conditions.
not play a role.
In the Marangoni case, the average concentration of hydrogen in the
layer at the electrode, the KOH concentration at the electrode is com- fluid domain decreases. This happens because the outlet velocity is
pressed by the Marangoni flow that is directed toward the electrode. larger near the electrode, so more hydrogen is lost through the outlet.
The minimum KOH concentration is not significantly lower than the However, its effect is very small.
mean KOH concentration in the domain, so its effect on conductivity Since the velocity at the bubble surface increases in the Marangoni case
and current density is small. The difference in KOH concentration be- and the velocity on the interface is directed towards the electrode, the
tween no-slip and free-slip boundary conditions is relatively small. boundary layer around the bubble is more rapidly replaced by liquid
Fig. 13 shows the current density at the electrode at instantaneous with a low hydrogen concentration. This results in the lower radial hy-
times when the bubble radius is 25 μm for various boundary conditions drogen flux for the Marangoni profile.
at the bubble surface. The average magnitudes of the current density for Marangoni flow significantly modifies the KOH concentration profile
Fig. 13(a-c) are −𝑖 = 1592.9, 1592.9, 1593.3 A∕m2 , respectively. These at the bubble surface. The lowest concentration of KOH occurs in the
values reveal that the different boundary conditions at the bubble sur- Marangoni case, and its largest drop occurs at the bubble foot.
face have a negligible effect on the average current density in this case. Similar to the hydrogen concentration boundary layer at the electrode,
However, the current density pattern in the Marangoni case differs from the KOH concentration boundary layer at the electrode is the thinnest
the other two cases: the current density is relatively less reduced near in the Marangoni case. The difference in KOH concentration between
the bubble foot (blue region) in Fig. 13c compared to Figs. 13a and 13b. no-slip and free-slip boundary conditions is relatively small.
As the value of the overpotential is very high (𝜂 itself is negative), only Bubble growth is the slowest in the Marangoni flow case due to the
the water part of the Butler-Volmer equation (Equation (9c)) is impor- lowest hydrogen concentration. Additionally, the difference in growth
tant, i.e. the current density is proportional to the water concentration. profiles between the no-slip and free-slip boundary conditions is negli-
As the velocity is higher at the foot of the bubble in the Marangoni case, gible.
the water is refreshed faster in that region, so the water concentration The pattern of water concentration in Marangoni flow is different com-
remains higher there (see Fig. 14). pared to other boundary conditions. In particular, in the Marangoni
flow, the water concentration is higher at the foot of the bubble because
6. Conclusions the Marangoni velocity causes a faster refreshment of water in that re-
gion. However, this variation in water concentration does not affect the
The effects of velocity boundary conditions on a growing hydrogen mean value of the current density at the electrode.
bubble are studied numerically using an immersed boundary method Unfortunately, we cannot provide guidance on improving the effi-
implemented in an in-house code. To verify the numerical accuracy of ciency of alkaline water electrolysis based on the current study solely, as
the method, the drag force obtained from the in-house code is compared our model does not include multiple bubbles, bubble coalescence, bub-
with that derived from the analytical solutions for the Stokes problem ble detachment, or the anodic compartment. However, the developed
(no-slip) and Hadamard-Rybczynski problem (free-slip) as well as re- simulation tool is promising and can be used to explore fundamental
sults from ANSYS-Fluent simulations. Moreover, a grid refinement study questions about the hydrogen mobility of the bubble surface and its ef-
is performed, and the global momentum and hydrogen mass balances fect on water electrolysis processes. It represents an important building

10
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Fig. 13. The current density of a growing bubble at the electrode in the 𝑦-𝑧 plane at 𝑥∕𝐿 = 0 with 𝑅 = 25 μm for different boundary conditions.

Fig. 14. The water concentration normalized by 𝑐H O,0 for a bubble growing at the electrode in the 𝑦-𝑧 plane at 𝑥∕𝐿 = 0 with 𝑅=25 μm for different boundary
2
conditions.

block for a more realistic/comprehensive model for enhancing the effi- Declaration of generative AI and AI-assisted technologies in the
ciency of water electrolysis. writing process

CRediT authorship contribution statement During the preparation of this work, the author(s) used ChatGPT in
order to improve the language and readability. After using this tool/ser-
F. Khalighi: Writing – original draft, Visualization, Validation, Soft- vice, the author(s) reviewed and edited the content as needed and
ware, Methodology, Investigation. A.W. Vreman: Writing – review & take(s) full responsibility for the content of the publication.
editing, Supervision, Software, Methodology, Funding acquisition. Y.
Tang: Writing – review & editing, Supervision, Funding acquisition.
N.G. Deen: Writing – review & editing, Supervision, Project adminis- Acknowledgements
tration, Funding acquisition.
This research received funding from the Dutch Research Council
Declaration of competing interest (NWO) in the framework of the ENW PPP Fund for the top sectors [and
from the Ministry of Economic Affairs in the framework of the PPS-
The authors declare the following financial interests/personal rela- toeslagregeling, grant number 741.019.201. Additionally, the research
tionships which may be considered as potential competing interests: is funded by Shell, Nobian, and Nouryon. We thank Björn van der Lande
N.G. Deen reports financial support was provided by Dutch Research for starting the implementation of the free-slip boundary condition in
Council. N.G. Deen reports financial support was provided by Top Con- the immersed boundary method during his master’s graduation project.
sortia for Knowledge and Innovation. N.G. Deen reports financial sup-
port was provided by Shell. N.G. Deen reports financial support was
Appendix A. The effect of unequal ion diffusivities
provided by Nobian. N.G. Deen reports financial support was provided
by Nouryon. A.W. Vreman reports a relationship with Nobian that in-
cludes: employment. N.G. Deen was previously an editor of this journal. This appendix presents a simulation for unequal ion diffusion coeffi-
If there are other authors, they declare that they have no known compet- cients for the case of stress Marangoni flow, as this boundary condition
ing financial interests or personal relationships that could have appeared is sensitive to variations in KOH concentration due to changes in ion dif-
to influence the work reported in this paper. fusion coefficients. Below, we will present the necessary modifications
to the model to accommodate for unequal ion diffusion coefficients.
Data availability For general (unequal) ion diffusion coefficients 𝐷K + and 𝐷OH− and
assuming electroneutrality, Faraday’s law describing the current density
The authors do not have permission to share data. reads (Newman and Thomas-Alyea, 2012):

11
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Fig. 15. The current density of a growing bubble at the electrode in the 𝑦-𝑧 plane at 𝑥∕𝐿 = 0 with 𝑅 = 25 μm for stress Marangoni boundary condition for (a) equal
diffusion coefficient of ions (𝐷K + = 𝐷OH− ), and (b) unequal diffusion coefficient of ions (𝐷K + ≠ 𝐷OH− ).

𝐹 2 (𝐷K + + 𝐷OH− ) Table 7

𝐢 = −𝜅∇𝜙 − 𝐹 (𝐷K + − 𝐷OH− )∇𝑐KOH , 𝜅= 𝑐KOH .
𝑅u 𝑇 Electrolyte diffusion coefficients for equal and un-
(30) equal ion diffusivities (m2 /s). In both cases the initial
electrical conductivity 𝜅 is 136 S/m.
The charge balance in the electrolyte is satisfied by the following equa-
tion (Newman and Thomas-Alyea, 2012): 𝐶𝑎𝑠𝑒 KOH K+ OH−

( ) equal 3.2 × 10−9 3.2 × 10−9 3.2 × 10−9

∇ ⋅ 𝐢 = −∇ ⋅ 𝜅∇𝜙 + 𝐹 (𝐷K + − 𝐷OH− )∇𝑐KOH = 0. (31) unequal 2.53 × 10−9 1.74 × 10−9 4.66 × 10−9

The equivalent diffusion coefficient of KOH is Fuller and Harb (2018);

Newman and Thomas-Alyea (2012):
2𝐷K + 𝐷OH−
𝐷KOH = . (32)
𝐷K + + 𝐷OH−
The boundary conditions for the ion concentrations at the electrode
are not used in the numerical model, but we need them to derive the
boundary condition for 𝑐KOH , which is used in the numerical model. The
boundary conditions for the ion concentrations are as follows:
𝐷K + 𝜕𝜙 𝜕𝑐 −
𝑁K + ,𝑥 = − 𝐹𝑐 + − 𝐷K + OH = 0, (33a)
𝑅u 𝑇 K 𝜕𝑥 𝜕𝑥
𝐷OH− 𝜕𝜙 𝜕𝑐 − 𝑖
𝑁OH− ,𝑥 = 𝐹 𝑐KOH − 𝐷OH− OH = − , (33b)
𝑅u 𝑇 𝜕𝑥 𝜕𝑥 𝐹
where 𝑖 is the 𝑥-component of the current density vector at the electrode,
and 𝑁K + ,𝑥 and 𝑁OH− ,𝑥 are the fluxes in 𝑥-direction for K + and OH− , Fig. 16. KOH concentration at the bubble surface for 𝑅 = 25 μm for the cases
respectively. Potassium ions do not participate in the reaction at the of equal and unequal ion diffusion coefficients.
electrode, so 𝑁K + ,𝑥 is zero at the electrode. Since due to electroneutral-
ity 𝑐K + = 𝑐OH− = 𝑐KOH , equation (33) implies the boundary condition in Section 5 (𝐷K + = 𝐷OH− ). The geometry and all physical properties
for 𝑐KOH ,
are the same in both cases, except for the diffusion coefficients listed in
𝜕𝑐KOH 𝑖 Table 7 and 𝐷H O , which equals 𝐷KOH in each case.
−𝐷OH− =− . (34) 2
𝜕𝑥 2𝐹 Fig. 15 shows the current density for two cases at the electrode in
If the second term in Equation (33b) is replaced with Equation (34), it the 𝑦-𝑧 plane at 𝑥∕𝐿 = 0. The mean current density values are −𝑖 =
leads to: 1593.3 A∕m2 for the equal diffusivity case and −𝑖 = 1597.2 A∕m2 for
the unequal diffusivity case, indicating a very small difference. Suppose
2𝐷OH− 𝐹 2 𝑐KOH 𝜕𝜙
𝑖=− , (35) we ignore the negligible increase of the current density. In that case,
𝑅𝑢 𝑇 𝜕𝑥 the right-hand side of the boundary conditions of the electrolyte at the
which replaces the first expression in Equation (9d). electrode is the same for both cases (see Equations (9b) and (34)). As a
According to literature, 𝐷K + = 1.96 × 10−9 m2 ∕s and 𝐷OH− = 5.26 × result, we can express (𝐷KOH 𝜕𝑐KOH ∕𝜕𝑥) equal ≈ (𝐷OH− 𝜕𝑐KOH ∕𝜕𝑥)unequal .
10−9 m2 ∕s at infinite dilution at 25 ◦ C (Newman and Thomas-Alyea, Since the diffusion coefficient of hydroxide (𝐷OH− ) is larger than the
2012). Unfortunately, we could not find the values for the conditions of equivalent diffusion coefficient (𝐷KOH ), the resulting KOH gradient
interest, 30 wt% KOH at 80 ◦ C. Therefore, we choose the ion diffusivities (𝜕𝑐KOH ∕𝜕𝑥) at the electrode is smaller in the unequal case. This results
such that the electrical conductivity 𝜅 keeps the physical value 136 S/m in a weaker Marangoni flow and a lower KOH concentration at the elec-
at 𝑐KOH = 6700 mol/m3 at 80 ◦ C, while we choose the ratio 𝐷OH− ∕𝐷K + trode surface in the unequal case.
equal to 2.68, the ratio at infinite dilution at 25 ◦ C. Fig. 16 shows the KOH concentration across the bubble surface for
Next, we compare the case with unequal ion diffusion coefficients both cases, which confirms our expectations. The KOH concentration at
(𝐷K + ≠ 𝐷OH− ) to the case with equal ion diffusion coefficients solved the bubble surface is lower in the unequal case. Table 8 shows the lift

12
F. Khalighi, A.W. Vreman, Y. Tang et al. Chemical Engineering Science 301 (2025) 120666

Table 8 Haverkort, J.W., 2024. A general mass transfer equation for gas-evolving electrodes. Int.
Marangoni forces on the bubble surface J. Hydrog. Energy 74, 283–296.
with 𝑅 = 25 μm obtained for equal and Haverkort, J.W., Rajaei, H., 2021. Voltage losses in zero-gap alkaline water electrolysis.
J. Power Sources 497, 229864.
unequal ion diffusion coefficients.
Hossain, S.S., Mutschke, G., Bhaskatov, A., Eckert, K., 2020. Thermocapillary effect on gas
Force 𝐷K + = 𝐷OH− 𝐷K + ≠ 𝐷OH− bubbles growing on electrodes of different sizes. Electrochim. Acta 353, 136461.
Janssen, L.J., Van Stralen, S., 1981. Bubble behaviour on and mass transfer to an oxygen-
𝐹𝑥 (nN) 19.5 14.8 evolving transparent nickel electrode in alkaline solution. Electrochim. Acta 26,
𝐹𝑦 (nN) 1.5 1.4 1011–1022.
Kempe, T., Lennartz, M., Schwarz, S., Fröhlich, J., 2015. Imposing the free-slip condi-
tion with a continuous forcing immersed boundary method. J. Comput. Phys. 282,
183–209.
Khalighi, F., Deen, N.G., Tang, Y., Vreman, A.W., 2023. Hydrogen bubble growth in alka-
line water electrolysis: an immersed boundary simulation study. Chem. Eng. Sci. 267,
118280.
Li, Y., Kang, Z., Mo, J., Yang, G., Yu, S., Talley, D.A., Han, B., Zhang, F.-Y., 2018. In-situ
investigation of bubble dynamics and two-phase flow in proton exchange membrane
electrolyzer cells. Int. J. Hydrog. Energy 43, 11223–11233.
Liu, B., Manica, R., Xu, Z., Liu, Q., 2020. The boundary condition at the air–liquid interface
and its effect on film drainage between colliding bubbles. Curr. Opin. Colloid Interface
Sci. 50, 101374.
Mark, A., van Wachem, G.M., 2008. Derivation and validation of a novel implicit second-
order accurate immersed boundary method. J. Comput. Phys. 227, 6660–6680.
Massing, J., Mutschke, G., Baczyzmalski, D., Hossain, S.S., Yang, X., Eckert, K., Cierpka,
C., 2019. Thermocapillary convection during hydrogen evolution at microelectrodes.
Electrochim. Acta 297, 929–940.
Meulenbroek, A.M., 2024. Marangoni effects on hydrogen bubbles on a microelectrode.
PhD thesis. Eindhoven University of Technology.
Meulenbroek, A.M., Vreman, A.W., Deen, N.G., 2021. Competing Marangoni effects form
Fig. 17. The bubble radius 𝑅(𝑡) versus time for equal and unequal ion diffusion a stagnant cap on the interface of a hydrogen bubble attached to a microelectrode.
coefficients. Electrochim. Acta 385, 138298.
Meulenbroek, A.M., Deen, N.G., Vreman, A.W., 2024. Marangoni forces on electrolytic
bubbles on microelectrodes. Electrochim. Acta, 144510.
and drag forces for both cases. Weaker Marangoni flow in the unequal Nagai, N., Takeuchi, M., Furuta, T., 2006. Effects of bubbles between electrodes on alkaline
diffusion coefficient for ions leads to reduced velocity, resulting in lower water electrolysis efficiency under forced convection of electrolyte. In: Proceedings
of 16th World Hydrogen Energy Conference. Lyon, pp. 1–10.
forces acting on the bubble surface.
Newman, J., Thomas-Alyea, K.E., 2012. Electrochemical Systems. John Wiley & Sons.
Fig. 17 also shows the bubble radius over time for both cases, show- Park, S., Liu, L., Demirkır, Ç., van der Heijden, O., Lohse, D., Krug, D., Koper, M.T., 2023.
ing the faster bubble growth in the unequal case. This is due to the Solutal Marangoni effect determines bubble dynamics during electrocatalytic hydro-
weaker Marangoni flow, which results in a lower velocity in the domain gen evolution. Nat. Chem. 15, 1532–1540.
Raman, A., Peñas, P., van der Meer, D., Lohse, D., Gardeniers, H., Fernández Rivas, D.,
and less hydrogen leaving the domain through the outlet.
2022. Potential response of single successive constant-current-driven electrolytic hy-
drogen bubbles spatially separated from the electrode. Electrochim. Acta.
References Rybczynski, W., 1911. On the progressive motion of a liquid sphere in a viscous medium.
Bull. Acad. Sci. Crac., Ser. A 1, 40–46.
Sepahi, F., Pande, N., Chong, K.L., Mul, G., Verzicco, R., Lohse, D., Mei, B.T., Krug, D.,
Claassen, C., Baltussen, M.W., Peters, E.A.J., Kuipers, J.A.M., 2024. An improved ghost cell
2022. The effect of buoyancy driven convection on the growth and dissolution of
immersed boundary method for conjugate mass and heat transport in fluid-particle
bubbles on electrodes. Electrochim. Acta 403, 139616.
systems. Chem. Eng. Sci. 291, 119936.
Shinagawa, T., Garcia-Esparza, A.T., Takanabe, K., 2015. Insight on Tafel slopes from a
Das, S., Panda, A., Deen, N.G., Kuipers, J.A.M., 2018. A sharp-interface immersed bound-
microkinetic analysis of aqueous electrocatalysis for energy conversion. Sci. Rep. 5,
ary method to simulate convective and conjugate heat transfer through highly com-
13801.
plex periodic porous structures. Chem. Eng. Sci. 191, 1–18.
Stokes, G.G., 1851. On the Effect of the Internal Friction of Fluids on the Motion of Pen-
de Groot, M.T., Vreman, A.W., 2021. Ohmic resistance in zero gap alkaline electrolysis
dulums, vol. 9, p. 8.
with a zirfon diaphragm. Electrochim. Acta 369, 137684.
Vakarelski, I.U., Kamoliddinov, F., Thoroddsen, S.T., 2024. Why bubbles coalesce faster
Dukovic, J., Tobias, C.W., 1987. The influence of attached bubbles on potential drop and
than droplets: the effects of interface mobility and surface charge. Langmuir.
current distribution at gas-evolving electrodes. J. Electrochem. Soc. 134, 331–343. van der Linde, P., 2019. On electrolytic bubbles. PhD thesis. University of Twente.
Ebadi, B., Behbahani-Nejad, M., Changizian, M., Pop, I., 2020. Fluid flow effects on diffu- Van Der Linde, P., Moreno Soto, Á., Peñas-Lóp ez, P., Rodríguez, J., Lohse, D., Gar-
sion layer and current density for electrochemical systems. Korean J. Chem. Eng. 37, deniers, H., Van Der Meer, D., Fernández Rivas, D., 2017. Electrolysis-driven and
1453–1465. pressure-controlled diffusive growth of successive bubbles on microstructured sur-
Fuller, T.F., Harb, J.N., 2018. Electrochemical Engineering. John Wiley & Sons. faces. Langmuir 33, 12873–12886.
Ghias, R., Mittal, R., Dong, H., 2007. A sharp interface immersed boundary method for Vreman, A.W., 2016. Particle-resolved direct numerical simulation of homogeneous
compressible viscous flows. J. Comput. Phys. 225, 528–533. isotropic turbulence modified by small fixed spheres. J. Fluid Mech. 796, 40–85.
Gilliam, R.J., Graydon, J.W., Kirk, D., Thorpe, S., 2007. A review of specific conductivities Vreman, A.W., 2020. Immersed boundary and overset grid methods assessed for Stokes
of potassium hydroxide solutions for various concentrations and temperatures. Int. J. flow due to an oscillating sphere. J. Comput. Phys. 423, 109783.
Hydrog. Energy 32, 359–365. Weissenborn, P.K., Pugh, R.J., 1996. Surface tension of aqueous solutions of electrolytes:
Granados Mendoza, P., 2016. Intensification of the chlor-alkali process using a rotor-stator relationship with ion hydration, oxygen solubility, and bubble coalescence. J. Colloid
spinning disc membrane electrochemical reactor. PhD thesis. Eindhoven University of Interface Sci. 184, 550–563.
Technology. Yang, X., Baczyzmalski, D., Cierpka, C., Mutschke, G., Eckert, K., 2018. Marangoni
Hadamard, J., 1911. Mouvement permanent lent d’une sphere liquid et visqueuse dans un convection at electrogenerated hydrogen bubbles. Phys. Chem. Chem. Phys. 20,
liquid visqueux. Acad. Sci. 152, 1735–1752. 11542–11548.

13