九州大学学術情報リポジトリ

Kyushu University Institutional Repository

A Study on Smoothed Particle Hydrodynamics for

Liquid Metal Flow Simulation
Al Salami, Jabir
Interdisciplinary Graduate School of Engineering Sciences, Kyushu University

Hu, Changhong
Faculty of Research Institute of Applied Mechanics, Kyushu University : Professor

Hanada, Kazuaki
Faculty of Research Institute of Applied Mechanics, Kyushu University : Professor

https://doi.org/10.5109/2349294

出版情報：Evergreen. 6 (3), pp.190-199, 2019-09. 九州大学グリーンテクノロジー研究教育センター

バージョン：
権利関係：
 EVERGREEN Joint Journal of Novel Carbon Resource Sciences & Green Asia Strategy, Vol. 06, Issue 03, pp190-199, September, 2019

A Study on Smoothed Particle Hydrodynamics for Liquid

Metal Flow Simulation
Jabir Al Salami1*, Changhong Hu2, Kazuaki Hanada2
1
Interdisciplinary Graduate School of Engineering Sciences, Kyushu University, Japan
2
Faculty of Research Institute of Applied Mechanics, Kyushu University, Japan

*Author to whom correspondence should be addressed,

E-mail: [email protected]

(Received December 6, 2018; accepted July 12, 2019).

In order to address the problem of heat and particle removal in tokamak-type, magnetic
confinement nuclear fusion reactors, a divertor that utilizes liquid metal is suggested to replace solid
tungsten divertors due to concerns regarding their structural integrity at high energy fluxes. The
operation of such a device gives rise to phenomena spanning multiple disciplines of physics such as
fluid dynamics, electromagnetics, thermodynamics and plasma physics. Smoothed Particle
Hydrodynamics (SPH) is a Lagrangian, mesh-free numerical method that has been proven effective
in a variety of disciplines. In this work, the hydrodynamic aspect of the liquid metal divertor is
simulated using SPH, paving the way to implement additional physics in future work.

Keywords: Smoothed Particle Hydrodynamics, Liquid Metal.

1. Introduction Hydrodynamics (SPH) is better adapted and is easier to

implement in cases of highly non-linear phenomena, such
Nuclear fusion, the reaction that powers the stars in our as free-surface flows, splashing and fragmentation when
universe, is a potential source of zero-carbon-emissions, compared to its Eulerian counterparts due to its purely
highly safe and abundant energy here on the Earth. In a mesh-free, Lagrangian nature14,25). Furthermore, the SPH
fusion reactor, Deuterium and Tritium are fused at framework facilitates the integration of multi-physics
temperatures in excess of tens of millions of degrees to without the need for extensive and computationally
produce energy. Because no known material can demanding and complicated approaches aiming to resolve
withstand and contain temperatures anywhere near those free-surfaces and non-hydrodynamical phenomena. In this
needed to achieve nuclear fusion, the core plasma in a work, the hydrodynamical aspect of the operation of a
tokamak-type fusion reactor is contained by means of a liquid metal divertor tank is simulated using SPH, where
strong magnetic field to avoid direct contact with the solid liquid metal is circulated in a tank as it is being bombarded
walls of the reaction chamber. An essential component of by plasma particles from above. A key desired simulation
the tokamak reactor is the divertor, which is responsible outcome is the extent and speed at which new fluid is
for expelling reaction byproducts and unspent fuel that circulated, and whether there are spots where the fluid is
adversely affect the plasma performance in addition to still and not being circulated, in addition to identifying any
help extract energy from the reaction to produce usable potential pitfalls or limitations for applying of the SPH
energy. method to this problem.
One of the main challenges facing divertor design is the
potential damage to solid plasma facing components, 2. Smoothed Particle Hydrodynamics
primarily made from tungsten due to the extremely high
thermal fluxes (~ 10 MW/m2) anticipated in future designs In the SPH framework, the fluid is discretized into
of reactors planned to provide energy to the public material particles that in addition to carrying the fluid
electricity grid1,2). Liquid metals were suggested to replace properties, serve as interpolation nodes for the discretized
solid plasma facing components as viable alternatives for fields. Although invented originally to simulate
future fusion reactor designs20,21,22). astrophysical problems, SPH’s inherent ability to bridge
the gap between continuity and fragmentation with little
Smoothed Particle Hydrodynamics is a Computational
effort carried over well to hydrodynamical and solid-
Fluid Dynamics (CFD) method that meets the mechanical applications, in addition to problems
requirements to be able to model multi-physics involving both, such as fluid-structure interaction, melting
phenomena similar to the case with the liquid metal and solidification and industrial processing3,4).
divertor23,24,25). In particular, Smoothed Particle Computationally, despite the large number of calculations

- 190 -
 EVERGREEN Joint Journal of Novel Carbon Resource Sciences & Green Asia Strategy, Vol. 06, Issue 03, pp190-199, September, 2019

required when compared to mesh-based methods, the high As Monaghan pointed out, the gradient approximation in
locality of SPH gives rise to ease of parallelization, which Eq. 6 has the advantage of conserving linear and angular
in addition to the lack of need for complicated and momentum, and is therefore the form of choice in the
expensive interface-capturing techniques renders SPH
attractively competitive to Eulerian CFD techniques. present work14).
For the purpose of carrying out simulations featured in this
work, an in-house SPH code was written in the C++ 2.2 SPH Discretization of the Navier-Stokes Equation
programing language. The open-access library Aboria Using the SPH interpolation to discretize the Lagrangian
was used for the general data structure and particle search form of the momentum equation we have
algorithms26).

2.1 The SPH Interpolation ⟨ ⟩ ∑ 𝑚 𝜵𝑾𝒂𝒃 ⟨ 𝛻 ∙ 𝜈𝛻 𝑢 ⟩ (7)

A continuous field’s value at a point in space can be

obtained using the Dirac delta function as follows where the second term on the right of Eq. (7) is the SPH
approximation of viscosity for which there are several
𝐴 𝑥 𝐴 𝑥 𝛿 𝑥 𝑥 𝑑𝑥 (1) options in the literature. In this work we use the SPH
viscosity formulation by Morris5).
Then, if the delta function is replaced by symmetric
smoothing kernel, W, such that ∑ 𝑚
⟨ 𝛻 ∙ 𝜈𝛻 𝑢 ⟩ 𝛻𝑊 (8)
lim 𝑊 |𝑥 𝑥 |, ℎ 𝛿 𝑥 𝑥 , (2)
→ The density is calculated by means of solving the
conservation of mass, or continuity equation, discretized
where h is the smoothing length, we end up with the SPH
discretized approximation of a field value as a summation in SPH as
over neighboring particles
⟨ ⟩ ∑ 𝑚 𝑢 ∙ 𝛻𝑊 (9)
⟨𝐴 ⟩ ∑ 𝐴 𝑊 (3)
In this work, the weakly compressible formulation of SPH
Where Wab = W(xa-xb,h). There are several options for is adopted, where pressure is computed by means of an
smoothing kernels available in the literature but in this equation of state
work, the fifth order Wendland kernel is used, which is 𝑝 𝜌 1 (10)
based on a family of kernels by Wendland 4). The choice
of kernel is based on its low computational cost, in Where 𝛾 7, and the zero subscript denotes reference
addition to resulting is less oscillatory and a more stable initial values.
behavior with higher resolutions, when compared to other To reduce the order of density fluctuations, the density
kernels. The Wendland kernel is defined as field is reset every 40 iterations using the so-called
Shepard filter as follows6).
2 𝑞 1 2𝑞 𝑓𝑜𝑟 0 𝑞 2,
𝑊 𝑞 (4) 𝜌
∑
(11)
0 𝑓𝑜𝑟 𝑞 2 ∑

‖ ‖ The XSPH correction is used to reduce the divergence of

where 𝑞 , d is the number of dimensions, and
a particle’s velocity from the average velocity in its
𝛽 7/ 64𝜋 in two dimensions and 21/ 64𝜋 in three vicinity, where the XSPH term is added to the velocity of
dimensions. One of the strong suits of SPH is the ease by each particle
which the derivatives of fields can be calculated, as the
derivative of the field itself is not needed, but it is 𝑋𝑆𝑃𝐻 𝜀∑ 𝑢 𝑊 (12)
calculated by means of the derivative of the smoothing
kernel, which is predetermined by the choice of kernel. As
such, the gradient of a scalar field can be approximated as 𝜀 being a tunable parameter between 0 and 1.

2.3 Solid Boundary Treatment

⟨𝛻𝐴 ⟩ ∑ 𝑚 𝐴 𝜵𝑾𝒂𝒃 (5)
Boundary treatment is known to be one of the most
Where we adopt the following convention challenging aspects of the SPH method, since they do not
arise naturally from the formal SPH discretization9),
𝛻𝑊 , and 𝛻𝑾𝒂𝒃 𝑥̅ . making boundary treatment a major area of research in the
‖ ‖
The gradient can also be approximated by SPH community in the past three decades.
One of the earliest and most widely solid boundary
implementations in SPH is the method of virtual or ghost
⟨𝛻𝐴 ⟩ 𝜌 ∑ 𝑚 𝛻𝑾𝒂𝒃 (6) particles, where fictitious particles are placed outside of

- 191 -
 A Study on Smoothed Particle Hydrodynamics for Liquid Metal Flow Simulation

the solid boundary that mirror the velocity of any fluid undesirable fluctuations in pressure and free-surface
particles within the smoothing length of the boundary profile near the boundary; it was therefore found best to
particles, but possess the same density and pressure7). This prohibit inter-boundary-particle interactions in the
method helps mitigate the issue of “kernel truncation” that simulation, and alleviate the resulting discontinuity of
reduces the accuracy of the SPH approximation near pressure values by filtering their density in the same
boundaries of the fluid when compared to its interior, due manner as fluid particles. In order to minimize kernel
to the scarcity of particles within the support domain of truncation error near the boundary and prevent particle
the fluid particle in question. However, the constantly penetration at high fluid-solid wall impact velocities, 3
changing number of virtual boundary particles increases rows of dynamic particles are placed at the boundary in a
implementation complexity and computational cost staggered manner.
especially for complicated geometries and high
resolutions. Fig. 1 illustrates the manner in which boundary particles
The particle nature of SPH has also inspired the adoption are arranged, where Δp is the initial particle spacing.
of inter-particle repulsive forces, analogous to the The radius of the circle or sphere on which the SPH
repulsive forces that arise when neutral atoms or interpolation is carried out, is 2h, where
molecules interact, to enforce solid wall boundary
conditions. ℎ √2𝛥𝑝 (13)
The most popular way of enforcing solid boundaries by
means of such inter-particle repulsive forces is the
Lennard-Jones potential, where a fluid particle 2.4 Open Boundary Treatment
experiences a radially directed repulsive force inversely Fig. 2 below shows an illustration of the implementation
proportional to its distance from a boundary particle. of the inlet boundary treatment used in this work.
Repulsive force boundary formulations, although The structure of buffer zones adopted by Tafuni et. al is
relatively easy to implement, may lead to spurious particle used to prevent kernel truncation for fluid particles near
motion near the boundary as it fails to compensate for the boundary11).
kernel truncation8).
In this work, the so-called “dynamic boundary condition”
is used to model solid boundaries10).

In the dynamic boundary treatment, particles are placed

at the boundary and are subjected to the same treatment as
fluid particles, with the exception of having a pre-
determined positions and velocities. Since boundary
particles are governed by the same momentum and
continuity equations as the fluid, when fluid particles
approach the boundary, the pressure and density increases,
consequently repelling fluid particles and preventing
fluid-boundary penetration. Also, the viscous term in the Fig. 2: Inlet buffer particles (orange), whose pressure is
interpolated from ghost nodes (red diamonds)
momentum equation enforces the no-slip condition at the before crossing the inlet boundary and becoming
boundary10). regular fluid particles (blue).

A buffer layer of particles is created with a thickness of

2ℎ 𝜖, where 𝜖 is a small number to ensure the inclusion
of all buffer particles in the support domain of particles
near the boundary.
Inlet buffer particles are created in a regular lattice and
assigned with the inlet velocity, u0, in the direction normal
to the inlet boundary. The position of the buffer region
particles is simply marched in time towards the inlet
boundary, and when an inlet buffer particle crosses the
boundary, it is converted to a fluid particle and another
buffer particle is created in its place at the beginning of
Fig. 1: Staggered configuration of dynamic boundary the buffer.
particles (red) and fluid particles’ (blue)
support domain (green dashed circle) Simply assigning a fixed pressure to the buffer particles
was found to create unphysical fluctuations in velocity
and pressure near the boundary and eventually lead to a
In addition to being convenient, the ability to perform
crash in the simulation. In order to mitigate such spurious
inter-particle interaction computations of the boundary
behavior and better enforce the natural, or zero gradient,
particles in the same loop as the fluid particles has the
pressure boundary condition at the inlet boundary, the
advantage of improving computational efficiency and
density, and therefore pressure, of inlet buffer particles is
parallelization. This, however, allows pressure waves to
interpolated from within the fluid domain by means of
travel across the boundary, inducing spurious and

- 192 -
 EVERGREEN Joint Journal of Novel Carbon Resource Sciences & Green Asia Strategy, Vol. 06, Issue 03, pp190-199, September, 2019

interpolation ghost nodes as shown in Fig. 2. The position 𝜌 ⁄

𝜌 𝐷 𝑥 ⁄
,𝑢 ⁄
(16)
of the ghost nodes is found by mirroring the position of
the buffer particles about the inlet boundary, and the where F is the momentum equation and D is the continuity
density at the node is found by using the Shepard filter equation. Finally, the variables are corrected by advancing
given by Eq. 7, but without including other buffer particles one more half a time-step, using the continuity and
in the interpolation to minimize dissipation. The pressure momentum equations evaluated with the predicted values
is then calculated by means of Eq. 6. in the previous stage of the integration scheme
As regards the outlet boundary treatment, simply
deleting particles that cross the outlet results in pressure ⁄ ⁄ ⁄
and velocity fluctuations and pressure waves reflecting off 𝑢 𝑢 𝛥𝑡𝐹 𝑥 ,𝑢 ,𝜌 (17)
the boundary, resulting in unphysical behavior and 𝑥 𝑟 ⁄
𝑢 (18)
artifacts in the solution. To avoid such issues, the buffer
zone approach is also used to model the outlet boundary 𝜌 𝜌 ⁄
𝐷 𝑥 ,𝑢 (19)
condition as shown in Fig. 3. Once the fluid particles cross
the outlet boundary, they are converted into outlet buffer then the pressure’s equation of state is evaluated using the
zone particles, and their pressure is set to the prescribed corrected density at the end of the time step.
outlet pressure, while their density is frozen. The The size of the time-step used in the integration scheme is
particles’ velocity component that is orthogonal to the updated every time-step to ensure numerical stability
boundary normal is zeroed once particles cross the according to the Courant-Friedrichs-Lewy (CFL)
boundary. The position of the outlet buffer zone particles condition and criteria depending on the force per unit mass
are marched with time until they cross the end of the
of fluid the and viscous diffusion term, as per the work
buffer zone, where they are marked for deletion at the next
done by Monaghan and Kos15).
time step.
The minimum time step due to the force is computed as
follows

𝛥𝑡 𝑚𝑖𝑛 (20)
‖ ‖

where Fa is the force per unit mass evaluated for all

particles. The time-step according to the CFL criteria and
viscous diffusion is

𝛥𝑡 , min ⋅
(21)
Fig. 3: Fluid particles (blue) are converted to outlet buffer
particles (orange) when they cross the outlet
boundary, then they are marked for deletion (red) where cs is the speed of sound, set to be 10 times the
once they leave the buffer zone maximum velocity anticipated in the simulation, and ε =
0.01h2. The time-step is then found as follows
2.5 Time Integration 𝛥𝑡 𝛼𝑚𝑖𝑛 𝛥𝑡 , 𝛥𝑡 (22)
,
Geometric, or symplectic time integrators are often
used in particle-based physics simulations due to their α being a factor of safety set to 0.3 in this work.
ability to exactly preserve the Lagrangian, and therefore
the total energy of the system14). Although higher order 3. Dam-Break Validation
time integrators, such as the Runge-Kutta method, are
superior in terms of absolute accuracy, geometric Dam-break, or the collapse of a water column, is one of
integrators’ ability to preserve angular momentum yield the most widely studied phenomena in computational
time-reversible and more stable results14). fluid dynamics due to the presence of highly non-linear
In this work, the second-order symplectic time-stepping phenomena and violent fluid-boundary interactions, and
scheme is used. its similarity to flow regimes that occur in naval and
First, the position, velocity and density are predicted by coastal engineering. The numerical simulation of dam-
advancing half a time-step as follows break problems have been extensively studied using both
Lagrangian and Eulerian methods, and even statistical
𝑥 ⁄
𝑟 𝑢 (14) mechanical methods such as the Lattice-Boltzmann
method. Example of such works are the Finite-Volume
⁄ / simulations by Kamra et. al and the Lattice-Boltzmann
𝑢 𝑢 𝐹 𝑥 ,𝑢 ,𝜌 (15)
simulations by Nik et. al12,19). In order to validate the SPH

- 193 -
 A Study on Smoothed Particle Hydrodynamics for Liquid Metal Flow Simulation

implementation in this work, the dam break flow over a

dry bed experiment by Lebovsky et. al 16) is simulated in
two dimensions, and the water level, propagation of the In Fig. 6, adequate agreement is found between the water
wave front, pressure and free surface profile are compared elevation obtained from the simulation and the
with the experimental measurements. Fig. 4 below shows experimental measurements by Lobovsky et. al.
a sketch of the initial configuration and measurement Fig. 7 compares the evolution of the wave front’s non-
locations. dimensional location vs. time obtained from SPH
Fig. 5 shows satisfactory agreement between the pressure simulation and the experiments by Lobovsky et. al and Hu
measurement on the left wall from experiment and the and Sueyoshi 18). The deviation between numerical and
SPH simulation. In the SPH simulation, the effect of the experimental results is consistent with what is expected
gate motion, which is used to release the fluid and initiate when the gate effect is not accounted for, as found by
the flow, has not been accounted for, resulting in an early Suyoshi and Hu19), and is comparable to Finite Volume
time of impact in the simulation when compared to the simulations done by Kamra attempting to quantify extent
experiment as has been noted by Kamra17). to which the gate effect can alter the flow17).
A comparison between the evolution of the free surface in
the experiment and simulation is shown in Fig. 8. The
pictures of the free surface from the Lobovsky experiment
are snapshots from a video.

Fig. 4: Sketch of the initial configuration of the 2-

dimensional dam break simulation showing the
pressure measurement point (red dot) and the
water elevation (H) measurement location

Fig. 5: Pressure from SPH simulation vs. pressure measurement by Lobovsky et. al.

- 194 -
 EVERGREEN Joint Journal of Novel Carbon Resource Sciences & Green Asia Strategy, Vol. 06, Issue 03, pp190-199, September, 2019

Fig. 6: Water level from SPH vs. experiment by Lobovsky et. al

Fig. 7: Wave front propagation vs. dimensionless time from SPH compared to measurements from experiments by Hu
et. al and Lobovsky et. al

- 195 -
 A Study on Smoothed Particle Hydrodynamics for Liquid Metal Flow Simulation

Fig. 8: Free surface snapshots of the dam break SPH simulation (left) vs. experiment by Lobovsky et.
al (right)
Table 1: Some physical properties of GaInSn
4. Liquid Metal Divertor Simulation
Fig. 9 below shows a sketch of the front view of a tank Melting point(oC) 10.5
initially partially filled with liquid metal. The physical Ga (%vol) 67
parameters used for the liquid metal in the simulation are
In (%vol) 20.5
those of GaInSn, a candidate for use as liquid metal
divertor material. Table 1 shows some of the physical Sn (%vol) 12.5
parameters that were used in the simulation 16).
Density (Kg/m3) 6360

Viscosity (m2/s) 2.98 x 10-7

In the simulation of the liquid metal divertor, the tank

was initially partially filled with fluid, so that the free
surface was 40 mm high. The velocity of the liquid metal
entering the tank through the inlet was set to 0.25 m/s, or
0.0785 l/s and the pressure at the inlet was interpolated
from inside the domain, as per section 2.3 in this paper.
The velocity at the inlet was chosen to avoid completely
draining the tank or overtopping and the pressure at the
Fig. 9: Sketch of the numerical liquid metal divertor
tank outlet was set to zero. Fig.10 and Fig.11 show the free
surface evolution, distribution of the new fluid particles
introduced to the domain through the inlet and the velocity
vector field on a plane parallel to the x-y plane, at a height
of 20 mm. The free surface shown in the figures below
was captured by means of the SPH first moment, since it
is approximately equal to 1 in the interior of the domain
and its value drops near the free surface, as follows:

∑ 𝑊 0.80 (23)

- 196 -
- 197 -
EVERGREEN Joint Journal of Novel Carbon Resource Sciences & Green Asia Strategy, Vol. 06, Issue 03, pp190-199, September, 2019

Fig. 10: Free surface time-evolution (left column), distribution of new fluid introduced through inlet (middle column) and velocity distribution and vector field taken on
plane bisecting the tank parallel to the x-y plane (right column).
- 198 -
A Study on Smoothed Particle Hydrodynamics for Liquid Metal Flow Simulation

Fig. 11: Free surface time-evolution (left column), distribution of new fluid introduced through inlet (middle column) and velocity distribution and vector field taken on
plane bisecting the tank parallel to the x-y plane (right column).
 EVERGREEN Joint Journal of Novel Carbon Resource Sciences & Green Asia Strategy, Vol. 06, Issue 03, pp190-199, September, 2019

The simulation shows that in areas around the corners 5) J. Morris, P. Fox, Y. Zhu, J. of Comp. Phys., 136, p.
of the tank opposite to the inlet and outlet in addition to 21 (1997).
the area in the center of the tank fluid is not circulated 6) D. Shepard, ACM Nat. Conference, 23, p. 517, New
sufficiently which may result in reduced heat removal York (1968).
efficiency. Simply increasing the velocity at the inlet to 7) J. C. Crespo, M. Gomez-Gesteira, R. A. Dalrymple,
attempt to induce more highly turbulent flow patterns in Comput. Mater. Contin., 5, p. 173 (2007).
the tank may result in overtopping, therefore additional 8) P. W. Randles, L. D. Libersky, Comput. Meth- ods
solid structures inside the tank that aim to enhance mixing Appl. Mech. Eng., 138, 375 (1996).
and circulation may be needed. The flow pattern may be 9) M. Ferrand, D. R. Laurence, B. D. Rogers, D.
significantly altered by the presence of a magnetic field, Violeau, and C. Kassiotis, Int. J. Numer. Methods
which was not accounted for in this work. An accurate and Fluids, 71, p. 446 (2013).
comprehensive simulation of all physical aspects relevant 10) J. C. Crespo, M. Gomez-Gesteira, R. A. Dalrymple,
to the operation of the liquid metal divertor necessitates Comput. Mater. Contin., 5, p. 173 (2007).
the inclusion of magneto-hydrodynamic effects, which is 11) Tafuni, I. Sahin, and R. Vacondio, Spheric, Munich
part of planned future work. (2016).
12) M. Kamra, C. Hu, Evergreen, 4, p.25, (2017).
13) J. Monaghan, Reports Prog. Phys., 68, p. 1703
5. Conclusion and Future Work (2005).
14) J. Monaghan, A Kos, Coastal and Ocean
This paper aimed to lay the groundwork for running Engineering, 125, p. 145 (1999)
multi-physics simulations of liquid metal divertors using 15) N. Morely, J. Burris, L. Cadwallader, M. Norenberg,
Smoothed Particle Hydrodynamics by means of first Rev. of Sci. Instruments, 79, (2008).
instigating a hydrodynamically accurate implementation 16) L. Lobovský, E. Botia-Vera, F. Castellana, J. Mas-
of SPH that can then be used to simulate free-surface Soler, and A. Souto-Iglesias, J. Fluids Struct., 48, p
problems. In this paper, Smoothed Particle 407 ( 2014).
Hydrodynamics formalism was introduced and then it was 17) M. Kamra, Kyushu University, PhD thesis (2018).
shown how it could be used to discretize the Navier- 18) Hu and M. Sueyoshi, Jour. of Mar. Sci. and App. 9,
Stokes equations to simulate hydrodynamical problems. p. 109 ( 2010).
The details of the SPH implementation used in 19) N. Mohd, M. Kamra, M. Sueyoshi, C. Hu, Evergreen,
simulations shown in this work were detailed and then 4, (2017).
validated by simulating a dam break problem and 20) A. Y. Ying, M. A. Abdou, N. Morley, T. Sketchley,
comparing the results with experimental measurements R. Woolley, J. Burris, and M. Narula. Fusion Eng.
from the literature. The validation case yielded and Design, 72, p. 35-62, (2004).
satisfactory matching with the experimental results. A 21) M. Shimada and Y. Hirooka. Nuclear Fusion, 54,
simplified and scaled model of the liquid metal divertor p.122002 (2014).
tank was then simulated and some potential pitfalls of the 22) C. Liao and M. S. Kazimi. Fusion Technology, 21, p.
suggested design were identified. 1845-1851 (1992).
In order to capture the full dynamics of the divertor in a 23) M. S. Shadloo, G. Oger, and David Le
fully realistic and accurate manner, the addition of Touzé. Computers & Fluids, 136, p. 11-34 (2016).
magneto-hydrodynamics, surface tension and turbulence 24) C. Ulrich, M. Leonardi and T. Rung. Ocean
modes is necessary and planned as part of future work. Engineering, 64, 109-121 (2013).
Furthermore, to be able to qualify and evaluate the 25) J. Hall, T. C. S. Rendall, C. B. Allen, and H. Peel.
practicality and effectiveness of different liquid metal Journal of Fluids and Structures, 56, p. 11-32 (2015).
divertor designs for use in nuclear fusion reactors, 26) M. Robinson and M. Bruna. SoftwareX, 6, p. 172-178
simulating the magnetohydrodynmic and heat transfer (2017).
physics of the problem is essential, as such effects may
significantly alter the flow pattern.

References
1) S. V. Mirnov, V. N. Dem'yanenko, E.V. Murav'ev,
J. of Nuclear Material, 196-198, 105 (1992).
2) T. W. Morgan, P. Rindt, G. G. van Eden, V. Kvon,
M. A. Jaworksi, N J. Lopes Cardozo, Plasma Phys.
and Controlled Fusion, 60, (2017).
3) C. Alvarado-Rodriguez, J. Klapp and L. Sigalotti,
Computers and Fluids, 159, p. 177 (2017).
4) H. Wendland, Adv. in Comp. Math., 4, p. 389 (1995)

- 199 -