PHD Thesis Andre Calado Coroado

Self-consistent simulation of multi-component
plasma turbulence and neutral dynamics in the

tokamak boundary
THIS IS A TEMPORARY TITLE PAGE
It will be replaced for the final print by a version
provided by the service academique.
Thèse présenté le 9 septembre 2021

à la Faculté des Sciences de Base
laboratoire Swiss Plasma Center
programme doctoral en Physique
École Polytechnique Fédérale de Lausanne
pour l’obtention du grade de Docteur ès Sciences
par
André Calado Coroado
acceptée sur proposition du jury:

Prof. Alfredo Pasquarello, président du jury
Prof. Paolo Ricci, directeur de thèse
Prof. Christian Theiler, rapporteur
Prof. Jens Juul Rasmussen, rapporteur
Prof. Fulvio Militello, rapporteur
Lausanne, EPFL, 2021

Abstract
The interaction between neutral particles and the plasma plays a key role in determining the
dynamics of the tokamak boundary that, in turn, significantly impact the overall performance
of the device. Leveraging the work in Wersal and Ricci [Nucl. Fusion 55, 123014 (2015)],
the present thesis describes the development and implementation in the GBS code of a
mass-conserving multi-component self-consistent model to simulate the interplay between
neutrals and plasma in the tokamak boundary. The simulation results are shown to be a useful
tool to disentangle the physics at play in the tokamak boundary, and in particular for the
interpretation of gas puff imaging (GPI).
Developed in the last decade, the GBS code [Ricci et al, Plasma Phys. Control. Fusion 54,
124047 (2012)] allows for self-consistent three-dimensional numerical simulations of the
turbulent plasma and neutral dynamics in the tokamak boundary. In GBS, a set of Braginskii
equations in the drift limit describes the plasma time evolution, and the neutrals are mod-
elled by solving a kinetic equation using the method of characteristics. While GBS enables
simulations in arbitrary magnetic configurations, here we focus on limited plasmas.
We first describe the geometrical operators and proper boundary conditions to ensure that
mass conservation is satisfied. In comparison to the previous non-mass-conserving model of
GBS, the mass-conserving simulations capture more accurately the sharp transition of the
plasma and neutral quantities between the edge and scrape-off layer regions. In addition, we
show that mass conserving simulations allow for reliable quantitative studies of particle fluxes
in the tokamak boundary.
A multi-component model of the neutral-plasma interaction is then developed by extending
the single-component model to the description of a deuterium plasma that includes electrons,
D+ ions, D atoms, D2 molecules and D+ 2 ions. The molecular dynamics is introduced through a
set of drift-reduced Braginskii equations for the D+
2 species and considering a kinetic equation
for D2 molecules, in addition to the kinetic equation for D atoms, thus resulting in a coupled
system of kinetic equations for the atomic and molecular neutral distribution functions. The
first multi-component GBS simulations show that, in the relatively long mean-free-path,
sheath limited conditions under consideration, most of the D2 molecules cross the last closed
flux surface (LCFS) and are dissociated or ionized in the edge region, thus giving rise to sources
of D atoms inside the LCFS. This leads to an inward radial shift of the peak of the plasma
source due to ionization of D atoms with respect to the single-component simulations.
The multi-component model is applied to the simulation of GPI diagnostics, where the pres-
ence of a molecular gas puff is simulated self-consistently with the plasma and neutral dy-
i
namics of the tokamak boundary. The injected molecules interact with the boundary plasma,
resulting in the emission of light in the Dα wavelength that can be measured to infer the
turbulent properties of the plasma. The simulated mechanisms underlying the light emission,
which include the excitation of D atoms and dissociation of both D2 and D+ 2 , provide a reli-
able tool for the interpretation of GPI experimental measurements. The impact of neutral
fluctuations on the Dα emission rate is investigated, as well as the correlation between the Dα
emission and the plasma and neutral quantities.
Key words:
plasma physics, controlled fusion, tokamak boundary, scrape-off layer, limited configu-
ration, plasma turbulence, fluid simulations, drift-reduced Braginskii model, kinetic neu-
tral dynamics, neutral-plasma interaction, mass conservation, molecular dynamics, multi-
component plasma, gas puff imaging, molecular gas puff, neutral fluctuations
ii
Riassunto
L’interazione tra particelle neutre e il plasma svolge un ruolo fondamentale nel determinare
la dinamica al bordo di un tokamak. A sua volta, questa regione ha un impatto significativo
sulle prestazioni della macchina. Partendo dal lavoro di Wersal e Ricci [Nucl. Fusion 55,
123014 (2015)], la presente tesi descrive lo sviluppo e l’implementazione nel codice GBS di un
modello a più specie, auto-consistente, che verifica la conservazione della massa, utilizzato
per simulare l’interazione tra i neutri e il plasma al bordo di un tokamak. I risultati delle
simulazioni si dimostrano uno strumento utile per districare i meccanismi fisici che hanno
luogo al bordo di un tokamak, particolarmente per l’interpretazione del GPI, una diagnostica
basata sull’iniezione di gas per visualizzare la dinamica del plasma.
Sviluppato nell’ultimo decennio, il codice GBS [Ricci et al, Plasma Phys. Cont. Fus. 54,
124047 (2012)] permette simulazioni numeriche tridimensionali auto-consistenti del plasma
e dei neutri. In GBS, un insieme di equazioni fluide nel limite di deriva descrive l’evoluzione
temporale della dinamica torbolenta del plasma e i neutri sono modellati risolvendo la loro
equazione cinetica. Anche se il codice GBS permette simulazioni in configurazioni magnetiche
diverse, nel presente lavoro ci concentriamo su plasmi con un limiter.
Questa tesi comincia con la descrizione degli operatori geometrici e delle condizioni al bordo
adeguati per garantire la conservazione della massa. In confronto con il modello precedente
di GBS che non conservava la massa, le nuove simulazioni catturano più precisamente la
transizione subita delle quantità del plasma e delle particelle neutre attraverso la LCFS (l’ultima
superficie di flusso chiusa).
Successivamente, un modello a più specie è sviluppato, estendendo il modello di un plasma
ad una sola specie ad uno che è in grado di descrivere un plasma di deuterio composto da elet-
troni, ioni D+ , atomi D, molecole D2 e ioni D+2 . La dinamica molecolare è introdotta attraverso
le equazioni di Braginskii nel limite di deriva per la specie D+ 2 e considerando un’equazione
cinetica per le molecole D2 , oltre all’equazione per gli atomi. Le prime simulazioni a più specie
con il codice GBS dimostrano che, nel regime di trasporto convettivo considerato in questo
lavoro, la maggior parte delle molecole D2 attraversano la LCFS e sono dissociate o ionizzate
nella regione del bordo, originando quindi atomi D dentro la LCFS. Questo conduce ad uno
spostamento radiale del picco della sorgente di plasma dovuta all’ionizzazione di atomi D
verso l’interno del tokamak, rispetto delle simulazioni con una sola specie.
Il modello a più specie è applicato alla simulazione della diagnostica GPI. In questa simulazio-
ne, l’iniezione di molecole è simulata in modo auto-consistente con la dinamica del plasma
e dei neutri. Le molecole iniettate interagiscono con il plasma e l’interazione risulta nell’e-
iii
missione di luce nella lunghezza d’onda Dα , misurata per inferire le proprietà turbolente del
plasma. I meccanismi alla base dell’emissione di luce simulati in questo studio, che includono
l’eccitazione di atomi D e la dissociazione di D2 e D+
2 , forniscono uno strumento affidabile per
l’interpretazione di misure sperimentali prodotte dal GPI. In questo lavoro studiamo l’impatto
delle fluttuazioni dei neutri nell’emissione Dα , così come la correlazione tra l’emissione Dα e
varie grandezze fisiche del plasma e dei neutri.
Parole chiave:
fisica dei plasmi, fusione controllata, periferia del tokamak, scrape-off layer, limiter, pla-
smi turbolenti, simulazioni di fluido, modello di Braginskii nel limite di deriva, dinamica
cinetica di neutri, interazione neutri-plasma, conservazione di massa, dinamica moleco-
lare, plasma a più specie, gas puff imaging, iniezione di gas molecolare, fluttuazioni dei
neutri
iv
Contents
Abstract (English/Italiano) i
1 Introduction 1
1.1 Controlled nuclear fusion as an energy source . . . . . . . . . . . . . . . . . . . . 1
1.2 The tokamak boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Simulation models for tokamak boundary . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Scope and outline of the present Thesis . . . . . . . . . . . . . . . . . . . . . . . . 9
2 The mass-conserving GBS model of plasma turbulence and kinetic neutrals 13

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 The mass-conserving model for the plasma and neutral particles . . . . . . . . . 15
2.2.1 A conserving model for the plasma . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 A mass-conserving neutral model . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Mass-conserving boundary conditions . . . . . . . . . . . . . . . . . . . . 26
2.3 Verification of Mass Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.1 Differences between conserving and non-conserving models . . . . . . . 33
2.4.2 A conserving model for the plasma . . . . . . . . . . . . . . . . . . . . . . 36
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3 A multi-component model of plasma turbulence and kinetic neutral dynamics 41

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Collisional processes in multi-component deuterium plasmas . . . . . . . . . . 44
3.3 The three-fluid drift-reduced Braginskii equations . . . . . . . . . . . . . . . . . 49
3.4 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.5 The kinetic model for the neutral species and its formal solution . . . . . . . . . 63
3.6 First simulation of a multi-component plasma with the GBS code . . . . . . . . 74
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4 Numerical simulations of Gas Puff Imaging 87

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.2 The GPI diagnostics and Dα emission rate . . . . . . . . . . . . . . . . . . . . . . 91
4.3 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
v
Contents
4.4 Analysis of the correlation between the D α emission and the plasma and neutral
quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5 Impact of neutral fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5 Summary and conclusions 115
A Proof of mass conservation in the formal solution of the neutral kinetic equation 119
B Evaluation of the average electron energy loss and reaction product energies in col-
lisional processes 123
C Zhdanov collisional closure 127
D List of kernel functions 133
E Numerical solution of the neutral equations 141
Bibliography 155
Curriculum Vitae 157
vi
1 Introduction
1.1 Controlled nuclear fusion as an energy source

Since the rising of civilizations, people join their efforts to grant each other access to basic
needs in a safe and comfortable environment. This has always relied on the ability of human
communities to exploit the natural resources in order to meet these needs. The industrial
revolution of the XVIII and XIX centuries brought a significant increase into this demand,
requiring energetic resources for large scale production of goods and transportation. Later
in the XIX century, the rising of the first power plants for electric energy generation opened
the door for a new electricity-based society. This led to the technological advances of the
XX century revolutionizing the means of production and life standards of an ever-increasing
number of people. The demand for energy sources grew exponentially in the last century,
leading to an unprecedented increase in fossil fuel consumption (coal, oil, gas) and the quest
for alternative energy sources. Even today, population growth and the continuous rise of the
energy consumption per capita make the overall energy consumption rise year after year,
despite technological progress allowing for a more efficient energy use [1, 2] (see Fig. 1.1).
Fossil fuel exploitation certainly do not provide the solution for such ever-increasing global
energy demand, for two main reasons. First, there is a finite amount of these resources (see
Fig. 1.2). Second, fossil fuel burning has led to an increase of the concentration of CO2 and
other pollutants in the atmosphere, leading to severe environmental consequences, of which
the increase of greenhouse effect and subsequent man-induced changes in the Earth’s climate
are the most worrisome [3] (see Fig. 1.3).
Current alternatives to fossil fuels are based on nuclear fission and renewable energy sources.
The later include hydro-power, the most relevant contribution on a global scale to date (see
Fig. 1.1), and also solar, wind, geothermal and waves. These resources are not exhausted
when exploited for electricity generation and, in general, have no associated greenhouse gas
emissions. However, the availability of these sources is highly dependent on the geographical,
geological and climatological situation of a given country. Moreover, due to the inherent
intermittent nature of most of these sources, major developments on energy storage and
1
Chapter 1. Introduction
Figure 1.1: Global primary energy consumption by source, according to Vaclav Smil’s Energy
Transitions: Global and National Perspectives (Second expanded and updated edition) (2017)
[1] and BP Statistical Review of World Energy [2]. Primary energy is calculated based on the
substitution method which takes account of the inefficiencies in fossil fuel production by
converting non-fossil fuel energy into the energy inputs if they had the same conversion losses
as fossil fuels. Image Source: ourworldindata.org
management systems are required to take full advantage of their potential [4]. Finally, although
recent progress has been reported on making energy generation from renewable sources more
efficient, their net power outcome is still largely insufficient to meet global needs.
Nuclear fission has been exploited for the last half century with successful results. The energy
density of uranium and other nuclear fission fuels is 5 orders of magnitude larger than that of
fossil fuels and the operation of nuclear power plants does not involve pollutant gas emissions.
Nuclear fission nowadays contributes with a significant fraction to the overall world electricity
generation (see Fig. 1.1). However, being based on a chain reaction and leading to radioactive
waste, this energy source suffers from important drawbacks. Release of radioactive isotopes
into the environment represents a major threat for the environment and public health. This
may happen due to operating an obsolete nuclear power plant in disregard of proper safety
procedures (Chernobyl, 1986) or leakage originated by earthquakes or other natural disasters
(Fukushima, 2011). Although nuclear accidents of this kind are rare and only responsible
for a small number of casualties in comparison with fossil fuel exploitation [5], the public
acceptance of these events undermines the possible use of this resource. In addition, storage
of radioactive waste generated as a side product of nuclear fission poses a major challenge.
Finally, we remark that even fission resources are not renewable on the long term (uranium
reserves are expected to last approximately a century if exploited in traditional nuclear power
2
1.1. Controlled nuclear fusion as an energy source
Figure 1.2: Years of fossil fuel reserves (coal, oil and natural gas) left, reported as the reserves-
to-product (R/P) ratio which measures the number of years of production left based on known
reserves and annual production levels in 2015. These values can change with time, depending
on the discovery of new reserves and changes in annual production. Based on data from the
BP Statistical Review of World Energy 2016 [7]. Image Source: ourworldindata.org
plants at the current consumption rate [6]).
In this context, finding a reliable long-lasting clean energy source that answers mankind’s
urgent need for energy is one of the great scientific and technological challenges of today. For
this reason, controlled fusion has been on the spotlight of scientific research during the last
half century. The aim of fusion research is to reproduce the fusion reactions taking place in
the stars and to make use of the energy they release to generate electricity. Indeed, the sun, as
the other stars with a similar mass, continuously converts hydrogen into helium, thanks to a
chain of fusion reactions, generating the energy responsible for most phenomena taking place
on Earth.
In the most promising among the set of fusion reactions, a deuterium and tritium nuclei fuse,
originating an α particle (helium-4 nucleus) and a neutron,
2 3 4
1 D +1 T →2 He(3.5MeV) +10 n(14.1MeV). (1.1)
In this reaction, most of the released energy is carried by the neutron (14.1 MeV), while the
energy of the helium nucleus constitutes the remaining 3.5 MeV. In future fusion power plants
the energy of the fast neutrons will be deposited in an external blanket and used to generate
electricity. On the other hand, the α particles are meant to remain confined and collide with
the fresh D,T fuel to deposit their energy and sustain the fusion reaction reducing the need of
3
Figure 1.3: Time evolution of the average temperature on the Earth’s surface compared to
solar irradiance since 1880, considering averages over 11-year periods to eliminate statistical
noise associated with annual fluctuations. While the two plots are clearly correlated until
1980, the subsequent divergence of the two strongly suggests an important role of enhanced
greenhouse effect on the Earth’s climate (Source: NASA/JPL-Caltech).
external heating.
The reaction in Eq. (1.1) is promising if compared to nuclear fission reactions, as the amount
of energy it releases is about four times larger (per mass unity), thus resulting in a higher
energy density [8, 9]. The fuel for the reaction in Eq. (1.1) is in general easily accessible in
nature, since deuterium is abundant in Earth’s water (about 35g of deuterium can be found in
each cubic metre of sea water [8]). As for tritium, although it does not occur naturally on Earth,
it can be bred from lithium, which is present in the Earth’s crust. There are some drawbacks
of the D-T fusion, such as the non-renewable nature of lithium reserves and the possibility
that fast 14.1MeV neutrons activate the wall materials. For this reason, other reactions are
also being studied for electricity generation, such as D-D fusion, which releases neutrons with
2.5MeV. Notwithstanding those alternatives, D-T is still the main focus of fusion research, as
its energy outcome is larger and the peak cross section is reached at lower temperature [10].
The high temperature necessary for the fusion reaction to occur is due to the small range of
4
1.1. Controlled nuclear fusion as an energy source
the strong nuclear force and the repulsive nature of the electromagnetic interaction between
positively charged nuclei. This makes it necessary an energy of the order of 10keV to bring the
fusion isotopes sufficiently close to each other so that the attractive strong nuclear force can
overcome the electrostatic repulsion. This energy corresponds to a temperature of the order of
108 K. At such high temperatures, matter is in the plasma state, atoms are ionized and matter
consists of charged particles, ions and electrons, that interact through electromagnetic fields
that they contribute to generate. At the same time, to self-sustain fusion reactions, the plasma
has to be confined in a region of space at sufficiently high densities and with sufficiently
high confinement times. To achieve self-sustained fusion reactions, experiments must aim at
increasing the triple product nT τE , with n the plasma density, T the plasma temperature and
τE the energy confinement time. In fact, the Lawson criterion [11] stipulates that self-sustained
energy generation in a nuclear fusion reactor is only possible for nT τE & 3 × 1021 m−3 keVs.
Achieving this triple product is therefore the main goal of current research, and improving the
plasma confinement proves to be a real challenge.
While confinement arises in the Sun as a natural consequence of the gravitational force
generated by the mass of the star itself (of the order of 1030 kg), man-made devices must follow
another strategy to confine the plasma (the mass of the whole Earth itself is about 1024 kg, one
million times smaller than the sun). The Lorentz force, which the charged particles of the
plasma are subject to, F = q(E + v × B), with q and v the particle charge and velocity and
E and B the electric and magnetic field, can be used. In the presence of a magnetic field,
because of the Lorenz force, charged particles rotate around the magnetic field lines, while
their motion along the direction parallel to the field lines is not constrained. In magnetic
confinement fusion devices, of which the stellarator and the tokamak are the most widely
exploited examples, a field is created inside a toroidal magnetic chamber, with magnetic
field lines closed within the plasma volume ensuring that the plasma remains confined. The
magnetic field inside the device cannot be merely toroidal: if that was the case, charged
particles would not remain confined as their motion would be affected by drifts arising from
the magnetic field curvature and gradient, leading to a loss of confinement. Therefore, a
poloidal component is also created to confine the plasma particles.
In a stellarator, both the toroidal and poloidal components of the magnetic field are generated
by means of external coils, whose complex design is optimized in order to allow for the
magnetic field topologies that best confine the plasma. In contrast, the tokamak (Russian
acronym for toroidal magnetic chamber) uses a different approach, as the toroidal magnetic
field is created by a series of coils positioned around an axially symmetric torus, while the
poloidal field is obtained by exploiting an electromagnetic induction effect. More precisely,
the tokamak works as the secondary circuit of a transformer that crosses the center of the
torus, with the primary circuit relying on an external magnet. This leads to a toroidal current
in the plasma that, in turn, generates a poloidal magnetic field. Additional coils around the
torus are used to adjust the shape of the magnetic field. The resulting helical magnetic-field
lines lie over nested magnetic flux surfaces, that is geometrical surfaces along which the flux
of the magnetic field is constant. The present thesis focuses on the tokamak configuration.
5
Figure 1.4: The plasma is confined inside a tokamak by means of a helical magnetic field
(Source: EUROfusion).
Even though there are some drawbacks of inducing an electric current in the plasma, namely
providing a drive for plasma disruptions, the tokamak has been widely studied over the last
decades, emerging as a strong candidate for a high performance fusion device. The quality of
fusion performances is measured by the gain factor Q, which is the ratio between the energy
output and the required power input. The world record for the value of Q was set by JET (Joint
European Torus), in the United Kingdom, in 1998, when Q = 0.67 was achieved.
Nowadays, the scientific community is joining forces on the construction and future operation
of ITER, a large-scale tokamak which is intended to show the feasibility of fusion as an energy
source by reaching a gain factor Q = 10 in a steady state operation. The first plasma is currently
scheduled for December 2025 and ITER is expected to reach the operational target conditions
by 2035.
1.2 The tokamak boundary

The plasma volume inside the tokamak can be divided in two regions according to the mag-
netic field topology: a closed-field line region where the magnetic field lines lie on nested
closed flux surfaces and an open-field line region where the magnetic field lines intersect the
vessel walls. The two regions are separated by the last closed flux surface (LCFS), or separatrix.
6
1.2. The tokamak boundary
In the closed-field line region, the high density and high temperature plasma is confined and
nuclear fusion reactions are set to take place in burning plasma conditions. Temperature
and density drop significantly in the outermost layer of the closed-field region, which sets
the sharp transition between the core and the open-field line region. We henceforth denote
this region as the edge, which extends from the outer core to the LCFS. We remark that the
core-edge interface is set based on the change of the plasma pressure scale length and thus it
is not strictly defined.
On the other hand, the open-field line region is filled with plasma outflowing from the closed-
field line region due to cross-field transport and local ionization events of neutral atoms.
Transport is the result of the collisions among charged particles (classical transport), further
enhanced by the device toroidal geometry (neoclassical transport), and also the turbulent
collective dynamics generated in the plasma as a result of pressure and temperature gradients
(turbulent transport). In the open field line region, by flowing along the magnetic field-lines,
the plasma particles hit the vessel walls of the device, being recycled and coming back into the
plasma as neutrals. For this reason, the open field line region is commonly referred to as the
Scrape-Off Layer (SOL).
Successful operation of fusion devices must make sure that the heat load on the vessel wall
does not exceed the maximum threshold allowed by the material. The magnetic field lines are
made to end on metal plates specifically designed to withstand large heat fluxes (e.g. by active
cooling of the material). These can be metal plates, called limiters, which extend radially into
the plasma, thus defining the LCFS, or divertor plates, in case the SOL magnetic field lines
are diverted away from the core region by creating one or more X-points where the poloidal
component of the magnetic field vanishes. In addition to regulating the particle and heat
exhaust, the SOL of a magnetic confinement device influences the overall performance of the
machine by setting the boundary conditions for plasma and energy confinement. The SOL
also plays an important role on fueling, impurity level control and fusion ashes removal [12].
When plasma particles heat the walls, ions and electrons recombine and are recycled into
the plasma as neutrals. A fraction of the ions are simply reflected back into the plasma as
neutral atoms, preserving their energy, while others are absorbed and then reemitted at the
temperature of the wall, in particular when this is saturated. A fraction of the remitted particles
are neutral hydrogen atoms, while others associate to form hydrogen molecules. While neutral
atoms are ionized into the plasma, hydrogen molecules are also ionized, but undergo a number
of dissociation processes as well. An important part of the plasma fueling thus results from
this recycling process [13, 14].
Due to the recycling taking place at the boundary, physical phenomena occurring in the SOL
involve the complex interaction between plasma and neutral particles. Indeed, neutrals are
present in the tokamak boundary as they are formed also by recombination events within
the plasma volume. In addition, they can be injected in the tokamak (for instance, via gas
puffing) with the purpose of fuelling, controlling the heat exhaust, or diagnosing the plasma
7
dynamics. Neutral particles interact with the low-temperature boundary plasma through a
number of collisional processes, thus playing a crucial role in determining the dynamics of
plasma turbulence in the SOL.
If the plasma densities and temperatures in the SOL are sufficiently low, the neutrals interact
weakly with the plasma and ionization takes place mostly inside the LCFS. In this case, there are
no large temperature gradients along the SOL and there is a strong flow of plasma particles all
along the magnetic field lines towards the limiter/divertor. As a consequence, heat convection
dominates heat transport in the SOL. This is referred to as the sheath-limited or convection-
limited regime [12].
On the other hand, if the plasma density is increased, the neutral mean free path decreases,
which results in a large amount of neutrals being ionized in the open-field line region. As a
result, the parallel particle flux in the SOL is significantly reduced, while strong temperature
gradients rise because of the cooling effect due to the ionization of neutrals close to the
limiter/divertor plates. In this condition, heat transport is mainly conductive. The tokamak is
then operated in the so-called conduction-limited or high-recycling regime.
In a divertor configuration, it is possible to extend the high-recycling regime by further de-

creasing the neutral mean free path, thus leading to a dense cloud of neutral atoms near the
divertor plates. As a result, temperature becomes sufficiently low so that volumetric electron-
ion recombination becomes important, thus greatly reducing the flux of particles and heat
to the divertor. In these conditions, known as the detached regime, a large fraction of heat is
radiated, thus making detachment a good candidate for future operation of nuclear fusion
reactors [15].
This regime can also be attained by injecting impurities into the system to mimic the role of
the neutral cloud and, following this approach, a detached regime can also be obtained in
limiter configurations [16, 12].
1.3 Simulation models for tokamak boundary

While the low collisionality of the plasma in the hot dense core of the device demands a kinetic
approach for its description, the considerably lower temperature observed in the tokamak
boundary, which makes the plasma significantly more collisional, opens the possibility to a
fluid description. In the SOL, where most often k k λmfp ¿ 1, with λmfp the electron mean free
path and k k the typical plasma perturbation wavenumber, simulations of plasma turbulence
can be performed by solving a set of fluid equations, derived following the approach developed
by Braginskii [17]. On the other hand, the edge is characterized by a large temperature gradient,
setting the transition from low to high collisionality, and therefore the fluid description is
valid only to the extent that k k λmfp ¿ 1. In addition, with respect to the tokamak core, the
simulations of the boundary region must address the neutral-plasma interaction in order to
take into account one of the key aspects that determine the plasma dynamics in this region.
8
1.4. Scope and outline of the present Thesis
Historically, the simulations of the plasma in the tokamak boundary have been based on
fluid models that describe cross-field plasma transport by means of a convective-diffusive
approach, implemented in codes such as B2 [18, 19], EDGE2D [20], SOLEDGE-2D [21] and
TECXY [22]. The neutral dynamics is approached either by using a diffusive fluid model, as in
the UEDGE [23] code, or by means of a kinetic description, usually based on a Monte Carlo
method, implemented e.g. in the DEGAS2 [24], EIRENE [25], GTNEUT [26] and NEUT2D [27]
codes. These simulation models remain the standard of reference, and the study and design
of the heat exhaust in magnetically confined fusion devices, such as ITER, strongly relies on
simulations carried out by codes that couple these plasma and neutral models, such as the
EMC3-EIRENE [28] and SOLPS [29] codes.
During the last decade, first-principles numerical simulations of plasma turbulence in the
tokamak boundary have emerged in an attempt to overcome the limitations of diffusive models.
Progress has been made possible by the development of several fluid and gyrofluid codes,
namely BOUT++ [30] and its derivation Hermes [31], FELTOR [32], GBS [33, 34, 35, 36], GDB
[37], GRILLIX [38], HESEL [39] and TOKAM3X [40]. These codes are based on the Braginskii
fluid equations taken in the drift limit. Also kinetic codes based on the gyrokinetic model have
been considered, e.g. Gkeyll [41] and XGC1 [42, 43]. Notwithstanding the important progress
in the development of plasma turbulence codes, their coupling to the simulation of the neutral
dynamics is still a challenging open issue.
The coupling of fluid turbulence codes with neutral models is achieved in the BOUT++,
nHESEL, TOKAM3X and GBS codes. BOUT++ is a general framework for fluid model-based
simulation of turbulence in the tokamak boundary, having allowed for a number of derivations.
BOUT++ simulated the neutral-plasma interaction by coupling the gyro-fluid plasma model
with a fluid-diffusive description of the neutrals in a linear device geometry [44]. The nHESEL
[39] code, consists of the coupling of the HESEL code, which evolves density, electron and ion
pressure, and vorticity in a two-dimensional slab geometry, with a one-dimensional diffusive
model for neutral atoms generated by dissociation of molecules injected by gas puffs. nHESEL
simulations allowed for important insights on the neutral-plasma interaction, such as the
influence of blobs on the neutral particle dynamics [45, 39, 46]. In TOKAM3X, the plasma
fluid model is coupled with the Monte Carlo code EIRENE [25] that describes the dynamics of
neutral species. TOKAM3X-EIRENE simulations of a limited configuration showed that, in a
low-density plasma, neutral particle density and flows are weakly affected by the turbulent
fluctuations of the plasma [47]. The present thesis focuses on the GBS code.
1.4 Scope and outline of the present Thesis

The GBS code is a three-dimensional global turbulence code that simulates the plasma dy-
namics in the boundary of magnetic confinement devices. Following a flux-driven approach,
GBS describes the plasma dynamics based on a set of Braginskii equations [17] in the drift
limit [48], with no separation between fluctuations and equilibrium quantities.
9
The GBS code started as a two-dimensional fluid turbulence code and it first addressed the
plasma dynamics in simple magnetized plasma (SMT) configurations such as TORPEX [49]
by evolving the drift-reduced Braginskii equations in the plane perpendicular to the device
magnetic field [50, 51]. The parallel dynamics was then included in the model, by using a
straight field-line coordinate system. The first three-dimensional version of GBS allowed
for global simulations of plasma turbulence in linear devices [52], like LAPD [53], or SMT
configurations [54]. GBS was later extended to simulate the tokamak SOL, considering limited
magnetic configurations [33]. As described in Ref. [55], the first SOL simulations considered
the cold ion limit, neglected electromagnetic effects and made use of the Boussinesq approxi-
mation when evaluating the divergence of the polarization flux. Finite ion temperature effects
were included later [56], as well as electromagnetic effects [34]. The GBS simulation domain
was also extended to give the user the possibility to simulate the edge region. In addition,
the Boussinesq approximation was removed [34] in the open field-line region. Moreover,
an approach based upon the method of manufactured solutions (MMS) allowed for the full
verification of the numerical implementation of the GBS model [57]. In parallel, GBS was
adapted to enable the simulation of diverted magnetic configurations, which required the
development of a non-straight field line coordinate system and its implementation to treat
the presence of X-points [35]. The diverted geometry model of GBS has been exploited to
understand the mechanisms underlying intermittent transport in single-null, double-null and
snowflake configurations [58, 59, 60].
Since 2015, the limited configuration version of GBS code also addresses the neutral dynamics
by means of a kinetic model, thus allowing for self-consistent simulations of plasma turbulence
and neutral dynamics [61]. The neutrals and the plasma are coupled via a number of collisional
reactions taking place in the plasma volume and recycling processes at the domain boundary.
The model described in Ref. [61] considers a single neutral species. A kinetic advection
equation for the neutral species is formally solved by applying the method of characteristics
(in a similar way to the nSOLT code [62]). This formal solution, which describes the neutral
distribution function, is then integrated in the velocity space, thus leading to an integral
equation for the neutral density. This equation is then simplified using two assumptions: it
is assumed that the time scale of plasma turbulence is larger than the neutral time of flight,
which turns the neutral calculation into a static problem, and the neutral parallel motion
is also neglected with respect to the characteristic spatial scales of turbulence along the
direction parallel to the magnetic field, which reduces the neutral equation to a set of two-
dimensional independent equations on the plane perpendicular to the magnetic field. The
integral equation for the neutral density obtained within these simplifications is discretized
and solved in the whole plasma volume and the domain boundary. The neutral density and
neutral outflow to the boundary are then obtained. Once the neutral density in the plasma
volume is known, the higher order moments of the neutral distribution function, namely the
neutral flux and neutral temperature, are obtained in a straightforward way.
The neutral model implemented in GBS provides a description valid for all ranges of neutral
density and mean free path [61]. In fact, it has the advantage of avoiding the assumptions on
10
1.4. Scope and outline of the present Thesis
the neutral temperature and diffusion coefficients typically considered in fluid-diffusive mod-
els. On the other hand, we highlight that the approach to the neutral dynamics implemented
in GBS is deterministic, thus avoiding the statistical noise of the Monte Carlo method.
The self-consistent model of the neutral plasma interaction in boundary plasmas has been
exploited in Ref. [63] to derive a refined two-point model to link the properties of the plasma
in the upstream and divertor regions. This two point-model matches accurately the results
from GBS simulations. In fact, it accurately predicts the ratio between the electron temper-
atures upstream and at the limiter plates in the open field-line region, considering as input
parameters the particle and heat sources related to cross-field turbulent transport and the
ionization sources along the magnetic field-lines.
GBS simulations were also used to investigate the role of neutral density fluctuations on
the Dα emission, as reported in Ref. [64]. For this purpose, a simulation of a diagnostic gas
puff located on the low-field side equatorial midplane was performed. The results showed
a strong anti-correlation between the neutral density fluctuations and the perturbations of
the plasma quantities at distances from the gas puff smaller than the neutral mean free path,
λmfp,n . However, the neutral density fluctuations were observed to have a small impact on
the statistical moments and turbulence properties of the D α emission, which suggests that
neglecting the neutral density fluctuation is a valid approximation when integrating the results
of gas puff imaging (GPI) at distances from the gas puff smaller than λmfp,n . On the other
hand, the GBS simulation also showed that the anti-correlation is weaker when distances from
the diagnostic gas puff larger than λmfp,n are considered as the result of non-local shadowing
effects due to the fact that the neutrals cross several plasma structures along their path. As a
result, the neutral density was found to have a non-negligible impact on the Dα emission at
large distances from the gas puff source.
Despite enabling self-consistent simulations of plasma turbulence and neutral dynamics in

the tokamak boundary, the GBS model described in Ref. [61] is not mass-conserving. In fact,
several approximations are considered in the description of both the plasma and the neutrals,
that prevent mass conservation. These assumptions include the simplifications of the device
geometry both in the plasma and the neutral models, the use of the Boussinesq approximation
and a set of boundary conditions based on assumptions that prevent recycling at the vessel
walls from being mass-conserving. In order to account for a fully consistent description of the
neutral-plasma interaction and enable mass conservation, the GBS model is improved in the
present thesis. The modifications introduced in GBS in order to achieve mass conservation
are described in Ch. 2, where the changes implemented in the plasma model, neutral solver
and boundary conditions are discussed in detail. The results of the first mass-conserving GBS
simulations are also presented and analyzed in Ch.2.
The model developed in Ref. [61] considers one neutral species (D atoms) and the corre-
sponding ion species (D+ ), preventing the description of the processes relevant in realistic
multi-component plasmas. The multispecies description of the neutral and plasma dynamics
11
is of particular importance, since the molecular dynamics is expected to play an important

role on plasma recycling at the boundary and fuelling. This requires the description of D+2
ion dynamics, as well as of both D atoms and D2 molecules. The multi-component model
implemented in GBS to account for different ion and neutral species is the focus of Ch. 3,
where the model is presented in detail. The results obtained from the first multi-component
GBS simulations are also presented and discussed, being compared with the results from the
single-component model.
Finally, we remark that the GPI studies reported in Ref. [64] are based on a single-component
plasma framework. The improvement from a single-component to a multi-component de-
scription is expected to affect the interpretation of GPI results, as predicted in the studies
reported in Ref. [46], based upon fluid simulations of two-dimensional plasmas with multi-
component neutrals described by a fluid-diffusive approach. Ch. 4 focuses on the results of
multi-component GPI simulations. These results are presented and compared with the ones
observed in previous works [46, 64], shedding light on the mechanisms underlying the Dα
emission in realistic multi-component plasmas.
The conclusion of the present thesis follows, in Ch. 5. Following the conclusion, a few
appendices to the main text are included. App. A presents the proof of mass conservation
for the formal solution of the neutral kinetic equation. App. B carries out the derivation
of the average energy of the reaction products and the average electron energy loss for the
dissociative processes considered in the multi-species model presented in Ch. 3. In App. C,
the derivation of the friction and thermal force terms featuring in the velocity and temperature
equations in the multi-component plasma description is presented, following the Zhdanov
closure [65] and considering the approach described in Ref. [66]. App. D lists the kernel
functions used to express the system of equations solved for the neutral species in the multi-
component model presented in Ch. 3. Finally, App. E develops the system of equations
for the multi-component neutral model in the matrix form which is used for its numerical
implementation in GBS.
12
2 The mass-conserving GBS model of
plasma turbulence and kinetic neu-
trals
In this chapter, the single-component GBS model for the simulation of the tokamak boundary
developed in the present thesis is described. This model is mass-conserving to leading order
in ρ s0 /R 0 , where ρ s0 is the ion sound Larmor radius and R 0 is the tokamak major radius
at the magnetic axis. A three-dimensional description of plasma turbulence is provided by
the two-fluid drift-reduced Braginskii equations, while the neutral atom dynamics model is
based on the discretization of the formal solution of the Boltzmann equation, with proper
mass-conserving boundary conditions being applied at the vessel walls. The present chapter
describes the implementation of this model in GBS, and demonstrate mass conservation. The
results of the mass-conserving simulation are then shown and discussed, highlighting the
influence that mass conservation has on the profiles of the plasma and neutral quantities. The
formation of the electron density profile is also addressed by relating it to the observed radial
fluxes of plasma and neutral particles. The present chapter is based on the results published
in Ref. [67].
13
Chapter 2. The mass-conserving GBS model of plasma turbulence and kinetic neutrals
2.1 Introduction
Developed in the last decade, the GBS code [33, 34] has been used to study plasma turbulence
in basic plasma physics experiments [54, 52], in tokamak limited configurations [68] and it is
now used to perform simulation in diverted configurations [35, 59, 60]. GBS is based on the
drift-reduced Braginskii equations for the plasma. In the limited version of GBS, the plasma
equations are coupled to a kinetic model for the neutrals [61]. The neutral atom density
and higher order moments of the distribution function are computed by solving the kinetic
equation through the method of characteristics and discretizing the resulting integral along
the neutral path (similarly to the nSOLT code [62]). This method provides a description of all
ranges of neutral mean free path [61] while avoiding the statistical noise of the Monte Carlo
methods.
By building upon the self-consistent model for the neutral and plasma dynamics implemented
in the GBS code and described in Ref. [61], the present chapter describes an improved model
aiming at achieving mass conservation in a limited tokamak configuration. The improvements
presented here ensure that mass conservation is satisfied to leading order in ρ s0 /R 0 (with ρ s0
the ion sound Larmor radius and R 0 the tokamak major radius at the magnetic axis), thus
allowing for accurate quantitative studies in the tokamak boundary. In fact, the results of the
first numerical simulations highlight the crucial role of mass conservation on determining
the particle flux in the tokamak boundary, with physical implications, e.g., on the study of the
tokamak fueling and the flux of particles and heat flux to the vessel walls.
Mass conservation among the plasma species has already been addressed by other boundary
turbulence codes. In GRILLIX [38], plasma particles and energy are conserved by means of a
particle source that compensates on average the particle sink at the outermost flux surface. In
turn, the Hermes module [31], based on the BOUT++ [30] framework, verifies electron density
conservation by implementing the equations in divergence form and relying on finite volume
schemes, both in the limiter and divertor configurations. On the other hand, TOKAM3X
features conservation of the plasma density, parallel ion momentum, electron and ion energy
and charge by implementing the fluid equations in the divergence form [69].
In GBS, the discretization of the solution of the neutral kinetic equation prevents the statistical
noise of the Monte Carlo approach from affecting mass conservation in a coupled neutral-
plasma model. In fact, by improving the model previously implemented in GBS, presented
in Ref. [34], mass conservation is now achieved in GBS to leading order in ρ s0 /R 0 . This is
made possible thanks to the implementation of i) operators in the plasma equations that
consistently take the toroidal geometry into account, avoiding the Boussinesq approximation
and geometrical simplifications; ii) solution of the neutral kinetic equation written within
proper toroidal geometry; iii) accurate boundary conditions to ensure mass conservation in
the description of wall recycling. The resulting model enables self-consistent mass-conserving
simulations of the neutral-plasma interaction in the tokamak boundary to leading order in
ρ s0 /R 0 . These simulations ultimately allow for quantitative studies of particle fluxes. The first
14
2.2. The mass-conserving model for the plasma and neutral particles
results of the improved model, reported in this chapter, are compared with non-conserving
results from previous GBS simulations, highlighting a significant impact of mass conservation.
The role of the plasma and neutral fluxes in the formation of the plasma profile is also assessed
by varying the plasma density considered in the simulations.
This chapter is structured as follows. After the Introduction, the improved mass-conserving
GBS model is presented in Sec. 2.2, focusing separately on the plasma model, the kinetic
neutral model, and the boundary conditions. The verification of mass conservation in GBS is
the focus of Sec. 2.3. Sec. 2.4 reports on the results of the first mass-conserving simulations
carried out with the GBS model presented in this chapter. The differences between the
mass-conserving and the non-mass-conserving models are discussed and the role of plasma
and neutral particle fluxes on the formation of the plasma density profile is addressed, by
performing a density scan. The conclusion of the chapter follows, in Sec. 2.5.
2.2 The mass-conserving model for the plasma and neutral parti-
cles
In this chapter, a mass-conserving model to leading order in ρ s0 /R 0 is derived for the limiter
configuration, where we consider an infinitely thin wedge located on the HFS equatorial
mid-plane acting as a toroidal limiter, similarly to Ref. [70]. Also similarly to Ref. [70], we
consider a toroidal annulus as the simulation domain, which includes both the open field-line
region (SOL) and a fraction of the region inside the LCFS (edge). The mass-conserving plasma
equations are presented in Sec. 2.2.1, while the conserving neutral model, consistent with
the plasma model and the three-dimensional toroidal geometry, is presented in Sec. 2.2.2. In
order to ensure mass conservation in the context of the recycling processes taking place at the
domain boundaries, proper boundary conditions are implemented, being the subject under
discussion in Sec. 2.2.3.
2.2.1 A conserving model for the plasma
GBS describes plasma turbulence by solving the drift-reduced Braginskii equations [17, 48],
as it is justified by the typical low-temperature and high-collisionality plasma conditions
in the tokamak boundary. Due to the large amplitude fluctuations, no separation is made
between fluctuations and background quantities. The plasma density n, vorticity Ω, the
electron and ion temperatures Te and Ti , and their parallel velocities v ke and v ki , are evolved
in the electrostatic limit, according to the following system of equations,
∂n ρ −1 2£
= − ∗ [φ, n] +
¤
C (p e ) − nC (φ) − ∇ · (nv ke b)
∂t B B
(2.1)
+D ⊥n ∇2⊥ n + D kn ∇2k n + S n +n n νiz − nνrec
15
∂Ω ρ
· −1 ¸ hv i
ki
= −∇ · ∗2 [φ, B ω ] − ∇ ·
¡ ¢
∇k (B ω ) + ∇ · j k b
∂t B B
· µ
b×k
¶ µ
b×k
¶ ¸
2 2
+ C (p e + p i ) + ∇G i · +G i ∇ · − C (G i )
B B B 3B (2.2)
hn i
2 2 n
+η 0Ω ∇k Ω + D ⊥Ω ∇⊥ Ω − ∇ · (νcx + νiz )ω
n
∂v ke ρ −1
∗
=− [φ, v ke ] − v ke ∇k v ke
∂t B ·
mi ν j k ∇k p e
¸
2
+ + ∇k φ − − 0.71∇k Te − ∇kG e
me n n 3n (2.3)
nn
+ D ⊥v ke ∇2⊥ v ke + D kv ke ∇2k v ke + (νen + 2νiz )(v kn − v ke )
n
∂v ki ρ −1
∗ ∇k p 2
=− [φ, v ki ] − v ki ∇k v ki − − ∇kG i
∂t B n 3n
nn (2.4)
+ D ⊥v ki ∇2⊥ v ki + D kv ki ∇2k v ki + (νiz + νcx )(v kn − v ki )
n
∂Te ρ −1
= − ∗ [φ, Te ] − v ke ∇k Te
∂t B ·
4Te C (p e ) 5
¸
+ + C (Te ) −C (φ)
3B n 2
2Te £ ¡ ¢ ¡ ¢¤
+ 0.71∇ · j k b − n∇ · v ke b
3n
+ χ⊥e ∇2⊥ Te + ∇k χke ∇k Te + S Te (2.5)
nn me
· µ ¶¸
2 4
+ νiz − E iz − Te + v ke v ke − v kn
n 3 mi 3
nn me 2
− νen v ke (v kn − v ke )
n mi 3
∂Ti ρ −1
= − ∗ [φ, Ti ] − v ki ∇k Ti +
∂t B·
4Ti C (p e ) 5
¸
− C (Ti ) −C (φ)
3B n 2
2Ti £ ¡ ¢ ¡ ¢¤ (2.6)
+ ∇ · j k b − n∇ · v ki b
3n
+ χ⊥i ∇2⊥ Ti + ∇k χki ∇k Ti + S Ti
nn
· ¸
1 2
+ (νiz + νcx ) Tn − Ti + (v kn − v ki ) ,
n 3
16
with k the magnetic field curvature vector given by k = (b · ∇) b and the vector ω defined as
ω = 1/B 2 n∇⊥ φ + ∇⊥ p i . Since ω is related to the plasma vorticity by Ω = ∇ · ω , the Poisson
¡ ¢¡ ¢
equation for the electrostatic potential φ required to close the system of equations yields
³n ´ µ
1
¶
∇· ∇⊥ φ = Ω − ∇ · ∇ p
⊥ i . (2.7)
B2 B2
We remark that, in Eqs. (2.1-2.7), B denotes the local value of the magnetic field modulus for
a given magnetic equilibrium, as required to ensure conservation. We remark that S ne , S Te ,
S Ti in Eqs. (3.19), (2.5) and (2.6) denote respectively the plasma density, electron temperature
and ion temperature sources at the edge-core interface, which mimic the plasma and energy
outflow from the core into the simulation domain.
In Eqs. (2.1-2.7) and in the rest of the chapter, we use dimensionless quantities. Therefore,
the density, n, and the temperatures, Te and Ti , are normalized to the reference values, n 0
and Te0 , while lengths parallel to the magnetic field are normalized to the tokamak major
radius, R 0 , lengths perpendicular to the magnetic field are normalized to the ion sound Lar-
mor radius at the magnetic axis, ρ s0 = c s0 /Ωci0 , where c s0 = Te0 /m i is the sound speed and
Ωci0 = eB 0 /m i the ion cyclotron frequency at the magnetic axis (m i stands for the ion mass
and e is the elementary charge). In turn, time is normalized to R 0 /c s0 . The normalization of
all other GBS quantities follows from these normalizations. Precisely, the parallel velocities
v ke and v ki are normalized to the ion sound speed c s0 , the plasma vorticity Ω is normalized to
n 0 Te0 /(ρ 2s0 B 02 ), perpendicular diffusion coefficients D ⊥ and conductivities χ⊥ are normalized
to c s0 ρ 2s0 /R 0 , while the parallel diffusion coefficients D k (added for numerical stability pur-
poses) and conductivities χk are normalized to c s0 R 0 . We also highlight that, in Eqs. (2.1-2.7),
ρ ? = ρ s0 /R 0 is the normalized ion sound Larmor radius, b = B/B is the magnetic field unit
vector, p = n(Te + Ti ) is the total pressure and ν = en e0 R 0 /(m i c s0 σk ) is the dimensionless
Spitzer resistivity. It is remarked that the Spitzer parallel conductivity is considered. In fact, we
5/2
assume χke,i = χk0 Te,i , following the procedure used in previous GBS models (Refs. [33, 34]),
but the weaker spatial and temporal variation of the 2/(3n) factor is neglected. As a matter
of fact, based on previous GBS simulations, we expect that neglecting the 2/(3n) factor does
not affect our simulation results since conductivity terms are relatively small in the plasma
conditions considered, which lie on the border between the sheath-limited and conduction
limited regimes. The electron and ion gyroviscous terms (Refs. [33, 34]) in Eqs. (2.3) and (2.4)
are given by
C (φ) C (p i )
· ¸
G i = −η 0i 2∇k v ki + + , (2.8)
B nB
C (φ) C (p e )
· ¸
G e = −η 0e 2∇k v ke + − . (2.9)
B nB
17
The Poisson brackets, curvature, parallel and perpendicular gradient operators are introduced
as
[ f , g ] = b · (∇ f × ∇g ), (2.10)
B
· µ ¶¸
b
C(f ) = ∇× ·∇f (2.11)
2 B
∇k f = b · ∇ f . (2.12)
∇⊥ f = [∇ − b(b · ∇)] f . (2.13)
We finally note that νiz , νcx , νrec and νen are respectively the ionization, charge exchange,
recombination and electron-neutral elastic collision rates. Similarly to Ref. [61], these collision
rates are evaluated as
νiz = n 〈v e σiz (v e )〉 , (2.14)
νcx = n 〈v i σcx (v i )〉 , (2.15)
νrec = n 〈v e σrec (v e )〉 , (2.16)
νen = n 〈v e σen (v e )〉 , (2.17)
where σiz , σcx , σrec and σen are the ionization, charge exchange, recombination and electron-
neutral elastic collision cross-sections, and v e and v i denote the electron and ion velocities,
respectively. We remark that, while νiz , νrec and νen are evaluated by averaging the colli-
sion rates over the electron distribution function, νcx is obtained by averaging over the ion
distribution function [61].
It is remarked that the GBS model presented here does not consider the Boussinesq approx-
imation when evaluating the polarization current density. This constitutes an extension of
the work reported in Ref. [34], since the variation of the magnetic field is properly taken into
account in the polarization current in both the open and closed field-line regions. This enables
the vorticity equation to be written in a mass-conserving form, needed to ensure charge and
18
mass conservation.
Contrarily to the procedure followed in Hermes [31] and TOKAM3X [40], Eqs. (2.1-2.6) are not
expressed in divergence form. This choice is justified by the robust numerical implementation
of the Poisson brackets made possible by using the Arakawa scheme [71], similarly to BOUT++
[30]. Moreover, the GBS approach enables a clear separation of the contributions of different
physical mechanisms. For example, it allows for the effect of curvature in the E × B convection
terms to be isolated in a straightforward way. On the other hand, a careful evaluation of
the geometric operators featuring in Eqs. (2.1-2.7) is required for mass conservation to be
satisfied. These operators are derived in Ref. [72] taking into account finite aspect ratio effects
by assuming a constant aspect ratio ² = a 0 /R 0 , with a 0 the minor radius at the LCFS, and
considering the orderings k k /k ⊥ ¿ 1 and k ⊥ R 0 À 1. This derivation is generalized here to
allow for a consistent variation of the inverse aspect ratio ²(r ) = r /R 0 , with r the local minor
radius.
We perform the derivation of the geometric operators by assuming an axisymmetric magnetic

field,
B = F (ψ)∇ϕ + ∇ψ × ∇ϕ, (2.18)
where ψ denotes the poloidal magnetic flux function obtained as the solution of the Grad-
Shafranov equation, ϕ represents the toroidal angle, and F (ψ) is the current function that
yields the magnetic field toroidal component. In order to express the geometric operators, we
introduce the right-handed coordinates set (ξ1 , ξ2 , ξ3 ) = (θ ∗ , f (ψ), ϕ), where f (ψ) is a magnetic
flux coordinate and θ ∗ is a poloidal angle chosen such that B ϕ = q(ψ)B θ , with B ϕ = B · ∇ϕ
∗
and B θ = B · ∇θ ∗ the toroidal and poloidal contravariant components of the magnetic field
∗
respectively (in general B i = B · ∇ξi , with i = 1, 2, 3), and q(ψ) the safety factor at the magnetic
flux ψ, defined as
1 2π B ϕ (θ, ψ)
Z
q(ψ) = dθ , (2.19)
2π 0 B θ (θ, ψ)
with θ the poloidal angle. The reason behind this definition of θ ∗ is the fact that the magnetic
field lines are straight in the (θ ∗ , ϕ) plane, which simplifies the numerical implementation of
derivatives along the magnetic field. In fact, the (θ ∗ , f (ψ), ϕ) coordinate system is advanta-
geous for the numerical implementation of Eqs. (2.1-2.7) because it facilitates the decoupling
of the derivatives along the directions parallel and perpendicular to the magnetic field. We
then introduce the contravariant metric tensor for this coordinate set, g ij = ∇ξi ·∇ξj , the covari-
ant metric tensor, g ij = Inv(g ij ) (Inv denotes the inverse matrix), and the coordinate Jacobian
q
J = 1/ d et (g ij ) = 1/[∇θ ∗ · (∇ f (ψ) × ∇ϕ)]. The magnetic field modulus can be computed from
p
these definitions as B = B i B i , with the covariant components of the field being given by
19
B i = g ij B j (Einstein summation convention is used). The magnetic field unit vector is defined
through b i = B i /B , while c i = b i /B are the components of the magnetic field unit vector di-
vided by the magnetic field modulus. The Poisson brackets, curvature and parallel gradient
operators are computed from the definitions in Eqs. (2.10-2.12) as
1 ∂φ ∂A
[φ, A] = ²ijk b i i k , (2.20)
J ∂ξ ∂ξ
B ∂c m ∂A
C (A) = ²kjm , (2.21)
2J ∂ξj ∂ξk
∂A
∇k A = b · ∇A = b j , (2.22)
∂ξj
with ²kjm the Levi-Civita symbol. Similarly, the second derivative along the direction parallel
to the magnetic field is obtained by applying Eq. (2.12) twice,
∂2 A ∂b j ∂A
∇2k A = b · ∇(b · ∇A) = b i b j + bi , (2.23)
∂ξi ∂ξj ∂ξi ∂ξj
and the perpendicular derivative is defined from Eq. (2.13) as
[∇⊥ A]i = g ij − b i b j ∂j A.
£ ¤
(2.24)
For simplicity, and following Ref. [72], in the present chapter we consider an equilibrium
with circular magnetic flux surfaces and constant current function F (ψ) = B 0 R 0 . We highlight
that considering a more complex magnetic equilibrium would not change the conservation
properties of the model, since shaping effects do not impact mass conservation as long as they
are implemented consistently. The flux coordinate f (ψ) is chosen as the radial distance from
the magnetic axis to a given magnetic flux surface, that is r . This leads to the Jacobian for the
(θ ∗ , r, ϕ) coordinates being expressed as
¤3/2
1 − ²(r )2
£
∗
J (r, θ ) = r R 0 . (2.25)
[1 − ²(r ) cos(θ ∗ )]2
The magnetic field components, b i and c i , can be obtained by using Eqs. (2.18) and (2.25).
Having defined all quantities concerning the magnetic equilibrium, the Poisson bracket
operator is derived as
20
[φ, A] = P θ∗ r [φ, A]θ∗ r + P r ϕ [φ, A]r ϕ + P ϕθ∗ [φ, A]ϕθ∗ , (2.26)
with its components and respective coefficients defined as
∂φ ∂A ∂φ ∂A
[φ, A]θ∗ r = − , (2.27)
d θ ∗ ∂r ∂r ∂θ ∗
∂φ ∂A ∂φ ∂A
[φ, A]r ϕ = − , (2.28)
∂r ∂ϕ ∂ϕ ∂r
∂φ ∂A ∂φ ∂A
[φ, A]ϕθ∗ = − , (2.29)
∂ϕ ∂θ ∗ ∂θ ∗ ∂ϕ
bϕ
P θ∗ r = , (2.30)
J
bθ∗
Pr ϕ = , (2.31)
J
br
P ϕθ∗ = . (2.32)
J
The curvature operator is expressed as
∂A ∗ ∂A
C (A) = C r +C θ +C k ∇k A, (2.33)
∂r ∂θ ∗
where the geometric coefficients are given by
B ∂c ϕ
Cr = − , (2.34)
2J ∂θ ∗
B ∂c ϕ 1 ∂c θ∗ ∂c r
· µ ¶¸
θ∗
C = + − ∗ , (2.35)
2J ∂r q ∂r ∂θ
B ∂c r ∂c θ∗
µ ¶
Ck = ϕ
− . (2.36)
2J b ∂θ ∗ ∂r
We highlight that, while the Poisson bracket operator in Eq. (2.26) is written in (θ ∗ , r, ϕ)
coordinates, the curvature operator in Eq. (2.33) features derivatives along θ ∗ , r and the
direction parallel to the magnetic field. In fact, parallel derivatives can be directly computed
in GBS by taking advantage of the straight-field line coordinate set, (θ ∗ , r, ϕ), as discussed in
Ref. [33]. Otherwise, the derivative along the magnetic field would have to be projected into
the ϕ and θ ∗ directions. This can be derived from Eq. (2.22), yielding
∂ 1 ∂
· ¸
∇k = b ϕ + . (2.37)
∂ϕ q ∂θ ∗
21
It is then possible to reverse the relation in Eq. (2.37) to write the toroidal derivative in terms
of the parallel and poloidal derivatives,
∂ 1 1 ∂
= ϕ ∇k − . (2.38)
∂ϕ b q ∂θ ∗
We can thus use Eq. (2.38) to write the curvature operator in Eq. (2.33) in terms of derivatives
along θ ∗ and the parallel direction instead of derivatives along ϕ. Regarding the expressions
in Eqs. (2.27-2.29) for the Poisson bracket operator, the Arakawa scheme [71] is used to
implement them in (θ ∗ , r, ϕ) coordinates, taking advantage of the robust stability properties of
this numerical scheme.
In comparison with Ref. [72], we do not order k ⊥ R 0 À 1 and the divergence of the magnetic
field unit vector ∇ · b is consistently taken into account in the present chapter. Therefore,
applying the definition of the divergence of a vector,
1 ∂
∇·b = (J b i ), (2.39)
J ∂ξi
one obtains, for a circular axisymmetric magnetic equilibrium,
R 0 b θ ∂B
∗
∇·b = − . (2.40)
B ∂θ ∗
In turn, the ∇⊥ operator in Eq. (2.24) entering the plasma equations via Eq. (2.7) can be
expressed by means of Eqs. (2.24) and (2.39) as
i 1 ∂ · n
i k ∂A
hn ¸
ij
∇⊥ · 2 ∇⊥ A = J (g − b b ) k . (2.41)
B J ∂ξi B 2 ∂ξ
Eq. (2.41) is then developed consistently with the geometry under consideration and keeping
only leading order terms in ρ s0 /R 0 and k k /k ⊥ , which yields
rr ∂ n ∂A θ∗ θ∗ ∂ n ∂A
hn i µ ¶ µ ¶
∇⊥ · 2 ∇⊥ A =N +N
B ∂r B 2 ∂r ∂θ ∗ B 2 ∂θ ∗
(2.42)
1 r θ∗ ∂ n ∂A 1 r θ∗ ∂ n ∂A
µ ¶ µ ¶
+ N + N ,
2 ∂r B 2 ∂θ ∗ 2 ∂θ ∗ B 2 ∂r
where the geometric coefficients are given by
N r r = g r r = 1, (2.43)
22
g ϕϕ 1 1 − ² cos θ ∗ 1 sin2 θ ∗ + (1 − ² cos θ ∗ )2 /q 2

Nθ θ
= gθ θ
∗ ∗ ∗ ∗
+ = + , (2.44)
q2 r 2 1 − ²2 R 02 (1 − ²2 )2
2 sin θ ∗
N r θ = 2g r θ = −
∗ ∗
. (2.45)
R 0 1 − ²2
We remark that the k k /k ⊥ ¿ 1 ordering used in the Poisson equation in Eq. (2.7) is not adopted
when Eqs. (2.1-2.6) are developed. As a matter of fact, we include derivatives along the
magnetic field in the geometrical operators present in those equations. Therefore, while mass
is exactly conserved to all orders in k k /k ⊥ , the energy is only conserved to leading order in
k k /k ⊥ [48].
We highlight that the stencil used within the implementation of the operator in Eq. (2.42)
considers ²(r ) = r /R 0 consistently, therefore taking into account the radial variations of the
geometric coefficients. The stencil is implemented by using a second-order centered finite
difference scheme, which yields
" rr rr rr θ∗ θ∗ θ∗ θ∗ θ∗ θ∗ #
hn i A i,j M i-1,j + 2M i,j + M i+1,j M i,j-1 + 2M i,j + M i,j+1
∇ · 2 ∇⊥ A = − +
B 2 ∆r 2 ∆θ ∗ 2
A i+1,j ³ r r rr
´ A i-1,j ³
rr rr
´
+ M i+1,j + M i,j + M i-1,j + M i,j
2∆r 2 2∆r 2
A i,j+1 ³ θ∗ θ∗ θ θ
∗ ∗
´ A i,j-1 ³ θ∗ θ∗ θ∗ θ∗
´
+ M i,j+1 + M i,j + M i,j + M i,j-1
2∆θ ∗ 2 2∆θ ∗ 2
(2.46)
1 h ³
r θ∗ r θ∗
´
+ A i+1,j+1 M i+1,j + M i,j+1
8∆r ∆θ³ ∗
´
rθ ∗
rθ ∗
+ A i-1,j-1 M i-1,j + M i,j-1
³ ´
r θ∗ r θ∗
− A i-1,j+1 M i-1,j + M i,j+1
³ í
r θ∗ r θ∗
−A i+1,j-1 M i+1,j + M i,j-1 ,
with the geometric coefficients redefined as M r r = N r r (n/B 2 ), M r θ = N r θ (n/B 2 ) and M θ θ

∗ ∗ ∗ ∗
=
N θ θ (n/B 2 ).
∗ ∗
Finally, we underline that the model equations presented herein reduce to the ones in Ref. [72]
when the inverse aspect ratio is approximated by the constant value ² = a 0 /R 0 , thus neglecting
the radial variation of ² across the poloidal plane. In addition, one would have to consider
the limits k k /k ⊥ ¿ 1 and k ⊥ R 0 À 1, as well as the Boussinesq approximation, to retrieve the
model reported in Ref. [72].
23
2.2.2 A mass-conserving neutral model
Similarly to the model presented in Ref. [61], the neutral-plasma interaction is addressed in
GBS by solving the kinetic advection equation for the distribution function of neutral atoms
f n , that is
∂ fn ∂ fn fn fi
µ ¶
+v· = −νiz f n − νcx n n − + νrec f i , (2.47)
∂t ∂x nn ni
where n n and n i denote the neutral and ion density, respectively, and f i the ion distribution
function. We remark that we neglect collisions between neutral particles and elastic collisions
between neutrals and ions. Assuming known plasma properties, we obtain the solution of Eq.
(2.47) by using the method of characteristics, writing it as
Z r b0 · S(x0 , v, t 0 ) ¸
0
f n (x , v , t ) = + δ(r − r b0 ) f n (xb0 , v, t b0 )
0 v
Z 0 (2.48)
1 r J (x 0 ) 0
· ¸
× exp − νeff (x00 , t 00 )d r 00 dr ,
v 0 J (x )
since the neutrals that contribute to the distribution function at position x, velocity v and
time t are generated at position x0 = x − r 0 Ω and time t 0 = t − r 0 /v, with Ω = v/v designating
the unit vector along the velocity direction and r 0 the distance between x and x0 . We remark
that the subscript "b" refers to the intersection of the characteristic passing through x and
direction Ω with the domain boundary. The neutrals are solved in an (R, Z , ϕ) coordinate
set, which is related to the (r, θ) coordinates by R = R 0 [1 − ²(r ) cos θ] and Z = R 0 ²(r ) sin θ. The
Jacobian of this coordinate set if then given by J (x) = R(x) = R 0 [1 − ²(r ) cos θ]. The volumetric
source of neutrals due to charge-exchange interactions and recombination events occurring
within the plasma is represented by the term
S(x0 , v, t 0 ) = νcx (x0 , t 0 )n n (x0 , t 0 )φi (x0 , v, t 0 ) + νrec (x0 , t 0 ) f i (x0 , v, t 0 ), (2.49)
with φi (x0 , v, t 0 ) = m i /(2πTi )exp[−m i v 2 /(2Ti )] the velocity distribution of the ions undergoing
charge-exchange interactions, which is assumed to be a Maxwellian of temperature Ti . We
remark that the volumetric source S(x, v, t ) is proportional to the neutral density n n (x, t ) =
R
f n (x, v, t )d v, which makes Eq. (2.48) an integral equation for f n .
The effective cross-section for neutral depletion, νeff (x00 , t 00 ) = νiz (x00 , t 00 ) + νcx (x00 , t 00 ), appear-
ing in Eq. (2.48), takes into account the neutrals removed by ionization or charge-exchange
interactions along the path between the source x0 and target x locations. We note that the
neutrals removed by charge-exchange with ions appear as fast neutrals, as described by the
source term in Eq. (2.49). The location x00 is along the integration path, while t 00 denotes the
time at which a neutral is at position x00 . The neutral source at the domain boundary is given
24
by
f n (xb0 , v, t b0 ) = (1 − αrefl )(Γout,n + Γout,i )χin (xb , v, T w ) + αrefl [ f n (xb , v − 2v⊥ ) + Γout,i χin (xb , v, Ti )],
(2.50)
where Γout,n = v ⊥ <0 |v ⊥ |cosθ f n (xb , v)d v denotes the flux of neutrals outflowing to the bound-
R
ary, Γout,i is the flux of ions outflowing from the main plasma, which is calculated from the
quantities evolved by Eqs. (2.1-2.7), v⊥ = v ⊥ n̂ = (v · n̂)n̂ is the projection of the particle
velocity along the unit vector normal to the boundary pointing towards the plasma n̂, and
χin (xb , v, T w ) = 3m 2 /(4πTw2 )cos(θ)exp −mv 2 /(2Tw ) is the Knudsen cosine velocity distribu-
£ ¤
tion assumed for the reemitted neutrals [61, 73], with θ = arccos (Ω · n̂) and Tw the temperature
of neutral particles reemitted at the wall. Eq. (2.50) takes into account that a fraction, αrefl , of
the neutral particles reaching the wall is reflected, while the remaining fraction is absorbed
and then reemitted. Reflected neutrals have a reflected distribution function with respect
to the incoming particles, f n (xb , v − 2v⊥ ). Analogously, the ions outflowing to the boundary
recombine with electrons in the wall and are recycled as neutrals, with a fraction αrefl of them
being reflected and the rest reemitted. For reflected ions we assume a Knudsen cosine velocity
distribution χin (xb , v, Ti ) and ensure that kinetic energy is conserved by considering the ion
temperature Ti instead of wall temperature T w .
We then solve Eq. (2.48) by considering two approximations that hold within the typical
plasma parameters in the tokamak boundary, as explained in detail in Ref. [61]. We assume
that the plasma quantities vary over characteristic turbulent time scales, τturb , which are larger
than the characteristic neutral time of flight, τn . We also take advantage of the fact that the
p
typical neutral mean free path, λmfp,n ∼ v th,n /νeff (with v th,n = Tn /m n the neutral thermal
velocity, being Tn the neutral temperature and m n the neutral mass, and νeff = νiz + νcx the
effective collision frequency for neutral depletion), is considerably smaller than the typical
elongation of turbulent structures along the direction parallel to the magnetic field, 1/k k ∼ R 0 .
In fact, assuming a neutral temperature Tn ∼ 2eV and an effective frequency for neutral
depletion νeff ∼ 105 s−1 , it results λmfp,n ∼ 0.1m, which is considerably smaller than R 0 ∼ 1m.
This allows us to consider all quantities appearing in Eq. (2.48) independent of the parallel
¯ ¯
coordinate, approximately coincident with the toroidal direction in the limit ¯∇ψ × ∇ϕ¯ ¿
¯ ¯
¯F (ψ)∇ϕ¯ considered in this chapter. As a result of these approximations, Eq. (2.48) reduces
to a set of time-independent two-dimensional equations in the poloidal plane. Eq. (2.48)
is integrated in the velocity space (v p , ϑ, v t ), where v t denotes the projection of the neutral
velocity along the toroidal direction, v p is the modulus of the projection of the velocity on
the poloidal plane and ϑ is the angle between Ω and the horizontal plane. This approach
ultimately leads to a linear integral equation for the neutral density n n in the poloidal plane,
25
Z rb Z ∞ Z 2π Z ∞
n n (xp ) = dr 0 d vp vp dϑ d vt
0 0 0 −∞
R(xp0 ) S(xp0 , v)
( " #
0 0 0
× + δ(r − r b ) f n (xp,b , v) (2.51)
R(xp ) vp
Z 0
1 r
· ¸¾
×exp − νeff (xp00 )d r p00 ,
vp 0
where xp and xp0 designate the locations of the target and source of neutral particles within
the poloidal cross section and r 0 denotes the distance between these two points.
The numerical discretization of Eq. (2.51) leads to a matrix equation solved for the neutral
particle density n n (xp ) and the flux of neutral atoms at the boundary Γout (xp,b ),
" # " #" # " #" #

nn νcx K p→p K b→p nn νrec K p→p K b→p ni
= + , (2.52)
Γout,n νcx K p→b K b→b Γout,n νrec K p→b K b→b Γout,i
where the kernel functions K are velocity integrals, their expressions being detailed in Ref.
[61]. The kernels link the neutral density n n at a given location within the plasma volume
("p") and the flux of outflowing neutrals Γout,n at a given location at the domain boundary
("b") with the density and flux of neutrals or ions at all locations. Eq. (2.52) represents a
non-homogeneous matrix system, where the homogeneous part describes the contribution
of neutrals generated by charge-exchange processes in the plasma volume and the neutrals
outflowing to the boundary, while the non-homogeneous part accounts for the neutrals
generated by volumetric recombination and ion recycling at the wall.
We highlight that the present thesis takes toroidal geometry into account consistently, by
introducing the proper R(xp0 )/R(xp ) geometric factor in Eq. (2.51) with respect to the expres-
sions used in Ref. [61]. The geometric factor R(xp0 )/R(xp ) consistently takes toroidicity into
account, instead of the cylindrical geometry considered in Ref. [61], and is crucial for mass
conservation.
2.2.3 Mass-conserving boundary conditions
To ensure mass conservation, proper boundary conditions must be applied to both the plasma
and neutrals. We highlight that the domain boundary includes the interface with the limiter
plates, the edge-core interface and the vessel outer wall. We focus on the boundary conditions
implemented for the plasma fields, as well as the evaluation of the plasma quantities appearing
in Eq. (2.50) at the domain boundary, since these quantities provide the boundary conditions
for the computation of the neutrals.
We start by considering the boundary conditions for the plasma fields at the limiter plates,
where most of the plasma particles end by following the magnetic field lines, being recy-
26
cled back to the plasma as neutrals. These boundary conditions, imposed at the magnetic-
presheath entrance where the ion drift approximation breaks, extend the Bohm-Chodura
boundary conditions. Their derivation was first reported in Ref. [74] considering the cold ion
limit and extended for the Ti 6= 0 case in Ref. [75]. Neglecting plasma gradients along the wall,
the plasma boundary conditions at the limiter are given by
p
v ki = ± Te F T (2.53)
φ
µ ¶
p
v ke = ± Te exp Λ − (2.54)
Te
n
∂θ∗ n = ∓ p ∂θ∗ v ki (2.55)
Te F T
p
Te
∂θ∗ φ = ∓ p ∂θ∗ v ki (2.56)
FT
∂θ∗ Te = γe ∂θ∗ φ (2.57)
∂θ∗ Ti = γi ∂θ∗ φ (2.58)
p
Te 2
µ ¶
1 2
Ω=− (∂θ∗ v ki ) ± p ∂θ∗ v ki (2.59)
FT FT
where the ± signs accounts for magnetic field lines entering or leaving the vessel, F T = 1+Ti /Te ,
£p
Λ = log (1/2π)(m i /m e ) ' 3. The parameters γe = γi = 0.1 are numerical coefficients relating
¤
the poloidal derivatives of Te and Ti with the poloidal derivative of the potential at the limiter
[74].
On the other hand, ad hoc boundary conditions are used at the radial boundaries, where
magnetic field lines are parallel to the wall, since no first-principles boundary conditions have
been developed so far for the outer vessel wall and the edge-core interface. As a matter of
fact, the simulation domain is extended along the radial direction both towards the core and
outwards, so that the boundary conditions for the plasma density at the outer vessel walls do
not affect significantly the mass conservation properties of the simulation results. For n, Te ,
Ti , v ke and v ki , homogeneous Neumann boundary conditions are applied and Ω = 0 is set at
both the wall and the edge-core interface. Regarding φ, we follow an approach similar to the
one reported in Ref. [35], using φ = ΛTe at the wall, with ΛTe evaluated at the beginning of the
27
simulation, and φ = φ0 at the edge-core interface, with φ0 a properly chosen constant to avoid
strong potential gradients.
We now evaluate the flux of ions outflowing to the boundary, Γout,i , featuring in Eq. (2.50).
This constitutes the primary neutral source in the system and its proper evaluation plays an
important role on for mass conservation. We observe that the total ion flux to the boundary is
Γi = nv ki b + Γdi + ΓE×B + Γpol,i , (2.60)
where nv ki b denotes the parallel ion flux, Γdi = −(∇p i × B)/B 2 is the ion diamagnetic drift flux,
ΓE×B = (n B × ∇φ)/B 2 the E × B drift flux and Γpol,i = n vpol,i is the ion polarization drift flux,
with the ion polarization velocity written as
1 di n
µ ¶ · ¸
1 1 ∇G i
vpol,i = − ∇⊥ φ + ∇⊥ p i − b × Giκ − , (2.61)
nΩci0 d t B B m i nΩci0 3
where d i /d t = ∂t + v ki b · ∇ + vE×B · ∇. In this chapter, the polarization drift contributions vpol,i

are neglected in the expression of the ion flux to the boundary, since they are smaller than the
E × B contribution by a factor of the order of ρ s0 /R 0 . Therefore, the expression for the ion flux
to the limiter walls considered in the present chapter is given by the projection of Eq. (2.60)
along the poloidal direction θ ∗ ,
[Γi ]θlim = nv ki b θ + Γθdi + ΓθE×B ,

∗ ∗ ∗ ∗
(2.62)
with the diamagnetic and E × B flux poloidal components expressed as
1 ∂p i ∂p i
· ¸
θ∗
[Γdi ] = − 2 Bφ − Br , (2.63)
JB ∂r ∂ϕ
n ∂φ ∂φ
· ¸
θ∗
[ΓE×B ] = − 2 Bφ − Br . (2.64)
J B ∂r ∂ϕ
We remark that Eq. (2.62) is evaluated locally. In turn, at the vessel wall we consider the radial
component of the flux in Eq. (2.60),
[Γi ]rwall = Γrdi + ΓrE×B , (2.65)
with
28
∂p i ∂p i
· ¸
r 1
[Γdi ] = − 2 − ∗ B φ + B θ∗ , (2.66)
JB ∂θ ∂ϕ
n ∂φ ∂φ
· ¸
r
[ΓE×B ] = − 2 − ∗ B φ + B θ∗ . (2.67)
JB ∂θ ∂ϕ
The flux of ions perpendicular to the outer wall in Eq. (2.65), being considerably smaller than
the flux of ions outflowing towards the limiter in Eq. (2.62), fluctuates and may occasionally
reverse direction. This is also due to the ad hoc plasma boundary conditions used at the wall.
We hence solve this numerical drawback by performing a poloidal and toroidal average of the
flux when evaluating Γout,i , therefore uniformly redistributing it at the wall.
While the boundary conditions imposed at the limiter and outer wall ensure mass-conservation,
the edge-core interface is an open boundary of the simulation domain, which can be crossed
by both ions and neutrals, with the plasma outflow from the core compensating the neutral in-
flow to the core in a steady state situation. In our simulations, the plasma outflowing from the
core to the simulation domain is mimicked by a density source near the edge-core interface,
S n . For global mass conservation purposes, we consider a dynamical density source matching
the time variation of the volume-integrated plasma density at each time step. As a result, the
system evolves to a steady state where the ion flux at the edge-core interface compensates the
neutral flux. The mass of the system composed by the ions and the neutrals inside the domain
is thus conserved globally and locally, with the density profiles oscillating around a constant
steady-state value. Since energy conservation is not the scope of the present chapter, we do
not impose energy conservation when modelling the heat outflow from the core into the simu-
lation domain. Instead, the energy outflow from the core is mimicked by temperature sources
located at the edge-core interface, S Te and S Ti , which are described by a gaussian profile of
constant amplitude, S Te = A Te exp −(r − r c )2 /w 2 and S Ti = A Ti exp −(r − r c )2 /w 2 , with r the
radial coordinate, r c = 0ρ s0 the radial location of the edge-core boundary, w = 5ρ s0 the width
of the gaussian source and S Te and S Ti the electron and ion temperature source amplitudes,
respectively.
We remark that the boundary conditions described in the present chapter extend the ones
used in previous GBS models (see Refs. [61] and [34]) thanks to the introduction of boundary
conditions for the neutral model that account for the leading order drift contributions to the
ion flux on top of the parallel motion, while the previous model only considered the parallel
ion flow along the magnetic field lines. Therefore, the corrections implemented in the version
of GBS reported in this chapter ensure mass conservation to leading order in ρ s0 /R 0 .
The model considered here does not include gas puffs for fuelling purposes or other fuelling
mechanisms. In fact, in a nuclear fusion reactor, the plasma fuel is expected to be depleted
because of the fusion reactions taking place at the core, which has to be compensated by
the fuelling from the boundary in order to conserve the total mass of the plasma in the
device during operation. We remark that the plasma fuelling can be simulated by using the
29
model discussed in this chapter, while global and local mass conservation are ensured by also
including the modelling of the products of the fusion reactors in a multi-species framework,
as well as the presence of a plasma pump for the removal of fusion products. We also note
that in order to establish energy conservation in our simulation, the heat exchange between
the plasma and the walls has to be described, as well as a more comprehensive description of
the collisional heat exchange between different species provided, taking into account the role
played by atomic excitation and different energy levels of the hydrogen atom.
2.3 Verification of Mass Conservation

Aiming at verifying the implementation in GBS of the mass-conserving model described in
Section 2.2, and composed of the plasma model, Eqs. (2.1-2.7), the boundary conditions,
Eqs. (2.53-2.59), and the neutral model, Eq. (2.51), we start by deriving the proper mass
conservation law. We remark that the electron continuity equation, Eq. (2.1), can be expressed
in divergence form as
∂n
= −∇ · Γe + n n νiz + D ⊥n ∇2⊥ n + D kn ∇2k n, (2.68)
∂t
where the electron flux Γe is expressed as
Γe = nv ke b + Γde + ΓE×B , (2.69)
with ΓE×B the E × B drift contribution to the electron flux, and Γde = (∇p e × B)/B 2 the electron
diamagnetic drift flux. We note that in Eq. (2.68), as well as in the rest of the present chapter,
we neglect the recombination sink term, since the recombination rate is negligible compared
to the ones of ionization and charge-exchange at the typical SOL and edge temperatures
considered in our simulations.
The ion continuity equation can be derived from Eq. (2.68) by using the vorticity equation, Eq.
(2.2), which can also be written in divergence form, ∇ · j = 0, as
∇ · (Γi − Γe ) + η 0Ω ∇2k Ω + D ⊥Ω ∇2⊥ Ω = 0. (2.70)
Using Eqs. (2.70) and (2.68), one obtains the ion continuity equation,
∂n
= −∇ · Γi + n n νiz + D ⊥n ∇2⊥ n + D kn ∇2k n − η 0Ω ∇2k Ω − D ⊥Ω ∇2⊥ Ω. (2.71)
∂t
In what follows, we use D(n, Ω) = D ⊥n ∇2⊥ n + D kn ∇2k n − η 0Ω ∇2k Ω − D ⊥Ω ∇2⊥ Ω as a simplified

notation. In addition, we note that the solution of the neutral kinetic equation presented in
30
2.3. Verification of Mass Conservation
Eq. (2.51) obeys the conservation relation n n νiz = −∇ · Γn , the detailed demonstration being
presented in the Appendix A to this thesis. Thus, the ion continuity equation in Eq. (2.71) can
be written in terms of the divergence of the neutral flux, yielding
∂n
= −∇ · Γi − ∇ · Γn + D(n, Ω), (2.72)
∂t
which highlights that the variation of the plasma density in time matches the sum of the
divergence of the ion and neutral fluxes, apart from numerical diffusion terms featuring in
the continuity and vorticity equations, Eqs. (2.1-2.2). We note that Eq. (2.72) states the mass
conservation law of the model developed in Section 2.2.
In order to verify Eq. (2.72), we integrate Eq. (2.72) in time and over the poloidal and toroidal
directions, thus obtaining a one-dimensional conservation law,
Z Z Z
J d θ ∗ d φ∆n = dt J d θ ∗ d φ [−∇ · (Γi + Γn ) + D(n, Ω)] , (2.73)
∆t
where ∆n = ∆t (∂n/∂t ) d t denotes the local density variation over time. The relative error for
R
mass conservation, normalized to the volume-integrated density variation, is expressed as
J d θ ∗ d φ ∆n + J d θ ∗ d φ [∇ · (Γi + Γn ) − D(n, Ω)]

R R R
∆t dt
E (r ) = , (2.74)
J d θ ∗ d φd r |∆n| /L r
R
with L r the radial length of the simulation domain.
The error E (r ) is evaluated from a set of GBS simulations and presented in Fig. 2.1. We remark
that the results shown in Fig. 2.1 are independent of the particular time chosen to evaluate the
instantaneous error of mass-conservation.
Our convergence tests show that grid resolution in the radial direction plays a particularly
important role in mass conservation. Three different grid resolutions are considered in Fig. 2.1:
a coarser grid with n x,p = 64 plasma grid points and n x,n = 10 neutral grid points in the radial
direction, an intermediate grid with n x,p = 128 and n x,n = 18, and a finer grid with n x,p = 256
and n x,n = 36. The poloidal and toroidal resolution of the plasma and neutral grids is kept
constant in all simulations, with n y,p = 512, n z,p = 64, n y,n = 360, and n z,n = 64. Convergence
is found in the SOL region, since the error decreases as the grid is refined. In the closed flux
surface region, the error is very small for all the simulations. We remark that the peak of E (r )
at the LCFS is a consequence of the sharp gradients that are observed between the open and
closed field-line regions, ultimately due to the topological transition between the edge and
the SOL. This effect is more important if a higher radial resolution is considered, which results
from the fact that the grid approaches the topological boundary between the two regions.
31
Figure 2.1: Instantaneous error in particle conservation, defined according to Eq. (2.74),
evaluated over an interval of two time steps (∆t = 7.5 × 10−5 ) for three combinations of plasma
(n x,p ×n y,p ×n z,p ) and neutral (n x,n ×n y,n ×n z,n ) grids: i) (64×512×64) and (10×360×64), blue
line; ii) (128 × 512 × 64) and (18 × 360 × 64), red line; iii) (256 × 512 × 64) and (36 × 360 × 64),
green line. Simulation parameters: R 0 /ρ s0 = 500, q = 3.867, n 0 = 2 × 1013 cm−3 , T0 = 20.0eV,
Ωci = 5.0 × 107 s−1 , T w = 3.0eV, ν = 0.1, η 0e = η 0i = 1.0, η 0Ω = 4.0, S Te = S Ti = 0.3, χk0 = 0.5,
D kn = D kv ke = D kv ki = D ⊥n = D ⊥Ω = D ⊥v ke = D ⊥v ki = D ⊥Tke = D ⊥Tki = 7.0.
Aiming at a further demonstration of convergence, we obtain a global error measurement

by computing the l 2 norm corresponding to the relative error E (r ) in Eq. (2.74) for the same
pairs of plasma and neutral grid resolution values considered in Fig. (2.1). We evaluate the l 2
norm for the whole domain and then excluding the area around the LCFS where the peaking
is observed, which is given by [x LC F S − 1.5ρ s0 , x LC F S + 1.5ρ s0 ]. These results are presented in
Tab. 2.1, which shows clear convergence for the restricted domain. Therefore, provided the
numerical peaking effect is removed, the l 2 norm highlights the convergence, thus reinforcing
the conclusions driven from Fig. (2.1).
2.4 Simulation Results

In this section, we present and discuss the results of the first simulations obtained from the
mass-conserving GBS model described in Section 2.2. We start by underlining the differences
with respect to the non-mass-conserving model previously implemented in GBS. A parameter
scan on the plasma density is the performed to understand the influence on the plasma
density on the mechanisms underlying the neutral-plasma interaction. All results presented in
this section refer to time averages over an interval of ∆t = 10.0R 0 /c s0 , during the quasi-steady
state period of the simulations.
32
2.4. Simulation Results
Plasma resolution (nx p ) Neutral resolution (nx n ) l 2 (WD) l 2 (RD)

−2
64 10 6.7 ×10 6.0 × 10−2
128 18 4.0 × 10−2 9.5 × 10−3
256 36 3.8 × 10−2 4.0 × 10−3
Table 2.1: Values of the l 2 norm for the relative error E (r ) in Eq. (2.74) for different plasma and
neutral grid radial resolution (same pairs of values used in Fig. (2.1), considering the whole
simulation domain (WD) and for a restricted domain (RD), which excludes the points located
in the region around the LCFS where numerical peaking is observed, [x LC F S − 1.5ρ s0 , x LC F S +
1.5ρ s0 ].
2.4.1 Differences between conserving and non-conserving models
To evaluate the impact of mass conservation on the simulations, we compare the mass-
conserving model described in the present chapter and the model previously implemented
in GBS (described in Ref. [34]). Aiming at disentangling the contributions from each of the
modifications implemented in the plasma and the neutral models, we perform a comparison
with a hybrid model, which considers a mass-conserving model for the neutrals, but makes
use of the previously implemented model for the plasma, featuring inconsistent geometry, the
Boussinesq approximation and the k k /k ⊥ ¿ 1 and k ⊥ R 0 À 1 orderings. For the analysis, we
integrate the plasma and neutral quantities over the flux surfaces, which enables the study of
the plasma and neutral radial profiles. The plasma and neutral densities are presented in Fig.
2.2, for all three models.
Mass conservation is found to have a significant impact on the simulation results. The mass-
conserving simulation shows a considerably higher SOL neutral density that drops significantly
in the edge, an effect which is also observed in the hybrid simulations, where the neutrals are
described by a mass-conserving model. Therefore, the changes in the neutral density profile
arise essentially from the corrections within the neutral model, more precisely from the proper
geometric factor, R(xp0 )/R(xp ), included in Eq. (2.51). In addition, in the mass-conserving
model, the plasma density drop from the closed to the open field-line region is more significant
with respect to the one observed in the non-conserving model, while the hybrid model leads
to a result that lies somewhere in between. We thus conclude that the changes in the density
profile are partly a result of the geometrical corrections within the improved plasma model,
mostly thanks to the fact that the radial variation of the inverse aspect ratio ²(r ) = r /R 0 is now
consistently taken into account, but are also partly related to the neutral model corrections.
In fact, the modifications in the neutral dynamics also impact the ionization source profiles,
which in turn has a decisive influence on the plasma density and ion flux profiles.
We now turn to the analysis of the radial particle fluxes, so as to investigate the mechanisms
underlying the changes in the plasma and neutral profiles observed in the mass-conserving
model with respect to the one previously implemented in GBS. The radial profiles of the
33
Figure 2.2: Comparison of the radial profiles of plasma density for a non-mass-conserving
model, a hybrid version of GBS with conservation properties for the neutrals but not for
the plasma and a mass-conserving model, averaged over an interval ∆t = 10.0R 0 /c s0 from
the quasi-steady state period of the simulations. The same parameters considered for the
simulations presented in Fig. 2.1 are used, with the plasma and neutral grid sizes (n x,p × n y,p ×
n z,p ) = (255 × 511 × 64) and (n x,n × n y,n × n z,n ) = (24 × 360 × 64).
radial ion and neutral fluxes are presented in Fig. 2.3, where the contributions from E ×
B, diamagnetic, and polarization drift fluxes contributing to the overall ion flux are also
discriminated. In the mass-conserving simulations, the ion and neutral fluxes are opposite
to each other to a good approximation. Hence, the divergence of the ion and neutral fluxes
approximately balance each other, apart from small diffusive terms (see Eq. (2.72)). However,
this is not the case for the non-conserving model, where the inward-pointing neutral flux
remains about constant in the edge and decreases smoothly across the SOL to vanish at
the wall, while the outward-pointing ion flux exhibits a rather steep profile, growing sharply
from the core to the LCFS and decreasing towards the wall. Regarding the hybrid model,
which considers the improvements of the neutral model but not those relative to the plasma,
the neutral flux profile is quite similar to the one found for the improved mass-conserving
model, since there are no differences in the neutral computation. On the other hand, the
combination of the previously implemented non-conserving model for the plasma with the
mass-conserving neutral model produces unreliable results for the ion fluxes, especially when
close to the edge-core interface.
We further note that the non-conserving model exhibits a relatively large inward-pointing
diamagnetic drift flux in the edge and near SOL, which reverses sign in the far SOL. On the
other hand, in the mass-conserving model, the diamagnetic flux in Fig. 2.3 is considerably
less important and points radially outwards, except in the vicinity of the LCFS. While the
polarization drift flux plays a certain role in the near SOL in the non-conserving simulation, it
becomes negligible in the whole domain in the mass-conserving simulation. We also highlight
34
Figure 2.3: Comparison of the radial profiles of radial neutral and ion fluxes for a non-mass-
conserving model (top panel), a hybrid version of GBS with conservation properties for the
neutrals but not for the plasma (middle panel) and a mass-conserving model (bottom panel),
discriminating the contributions of E × B, diamagnetic and polarization drifts, averaged over
an interval ∆t = 10.0R 0 /c s0 from the quasi-steady state period, using the same grid sizes and
parameters considered for the simulations presented in Fig. 2.2.
that our results differ from the conclusions reported in Ref. [37], where the absence of neutrals
imposes ∇ · Γi = 0 in a steady state, thus forcing a balance between the E × B drift flux and
the diamagnetic component of the flux. In the simulation results reported in this chapter,
the E × B drift flux is instead balanced by the neutral particle flux, while the diamagnetic
contribution is negligible with respect to the E × B flux component.
As a final remark on the impact of the corrections in the boundary conditions for the neutral
calculation, we discuss the relevance of the matching between the flux of ions outflowing
to the limiter, [Γi ]θlim , described in Eq. (2.62), and the flux of recycled neutrals flowing back
∗
to the plasma, [Γn ]θlim . Aiming at this, we consider the same interval ∆t = 10.0R 0 /c s0 from
∗
the quasi-steady state period used in Figs. 2.2 and 2.3 and we compute the averaged ion
and neutral fluxes by integrating along the limiter plates. We then evaluate the relative error,
normalized to the total ion flux, i.e.
35
[Γi ]θlim,tot − [Γn ]θlim,tot

∗ ∗
E lim (r ) = , (2.75)
[Γi ]θlim,tot
∗
with
Z
[Γi ]θlim,tot J d ϕd r |[Γi ]θlim |,
∗ ∗
= (2.76)
lim
Z
[Γn ]θlim,tot J d ϕd r |[Γn ]θlim |.
∗ ∗
= (2.77)
lim
The procedure is applied to both sides of the limiter, with E lim (r ) = 28.3% at the lower plate for
the non-conserving boundary conditions and E lim (r ) = 9.6% for the mas-conserving model,
while the upper plate yields E lim (r ) = 7.2% for the non-conserving model and E lim (r ) = 2.5%
for the simulation considering improved boundary conditions. These values highlight that
mass-conserving recycling at the limiter becomes more accurate by a factor of three when the
E × B and diamagnetic drift flux components are included in the expression of the ion flux in
the boundary conditions for the neutral model.
2.4.2 A conserving model for the plasma
In this subsection we analyse the results obtained from simulations that consider different val-
ues of plasma density, which are obtained by varying the normalization density n 0 . Increasing
n 0 makes the reaction rates for ionization, charge-exchange and electron-neutral collisions
larger, which affects the plasma and neutral profiles, as shown in Fig. 2.4. In order to isolate
the role of the neutrals, all simulations consider the same plasma resistivity ν, as well as the
other simulation parameters.
We first consider the lowest density n 0 = 5.0 × 1012 cm−3 case (see Fig. 2.4). The neutral density
n n peaks in the SOL close to the limiter plates, where the plasma is recycled, and remains
almost constant in the closed field-line region, given the weak interaction of the neutrals with
the plasma (an important fraction of neutrals penetrate into the core with no interaction with
the plasma in the edge and SOL regions). In fact, because of the higher plasma temperature
in the closed field-line region, the ionization source n n νiz peaks in the vicinity of the limiter,
keeping a fairly constant value in the closed-flux surface region on the HFS. The analysis of
the cross section profile of charge-exchange, n n νcx , leads to similar conclusions.
Regarding the intermediate density case, n 0 = 2.0 × 1013 cm−3 , it is remarked that ionization is
predominantly localized near the LCFS, the neutrals penetrating considerably less into the
core with respect to the n 0 = 5.0 × 1012 cm−3 case. The spatial distribution of charge-exchange
is also characterized by a pronounced peak at the LCFS and a sharp decrease towards the
36
Figure 2.4: Cross sections of plasma density n (first row), neutral density n n (second row), ion-
ization source n n νiz (third row) and charge-exchange n n νcx (fourth row) for a mass-conserving
model, considering a time range of ∆t = 10.0R 0 /c s0 during the simulation quasi-steady
state period, for three different normalization densities: n 0 = 5.0 × 1012 cm−3 (left column),
n 0 = 2.0 × 1013 cm−3 (center column), and n 0 = 4.0 × 1013 cm−3 (right column). Grid sizes and
all other simulation parameters are the same as the ones used for the simulations presented
in Fig. 2.2.
edge-core interface. Due to the weaker penetration of the neutrals into the core and the strong
ionization in the closed field-line region, a significant drop in the neutral density between the
SOL and the edge-core interface is found. Analogously, since most of the ionizations occur
near the LCFS, a significant plasma density source is present in the edge near the limiter,
resulting in a smooth variation of n in the closed field-line region on the HFS.
Regarding the largest density considered in our scan, n 0 = 4.0 × 1013 cm−3 , the simulation
results exhibit a strong concentration of the ionization and charge-exchange interactions in
37
Figure 2.5: Radial plot of the time, toroidally and poloidally averaged density of plasma (top
panel) and neutrals (bottom panel) within a mass-conserving model, considering a time
range of ∆t = 10.0R 0 /c s0 during the quasi-steady state period, for three different values of the
normalization density: n 0 = 5.0 × 1012 cm−3 (green line), n 0 = 2.0 × 1013 cm−3 (red line) and
n 0 = 4.0 × 1013 cm−3 (blue line). Grid sizes and all other simulation parameters are the same as
the ones used for the simulations presented in Fig. 2.2.
the proximity of the LCFS, with an extremely weak penetration of neutrals across the edge.
Thus, neutral density drop between the SOL and edge regions is further enhanced, while the
plasma density peaks in the closed field-line region close to the limiter.
A more quantitative comparison of the radial profiles of n and n n is presented in Fig. 2.5. We
remark that, while small values of n 0 flatten the n n profile in the edge and SOL regions, the
variation of n 0 has a weak influence on the radial profile of n (keeping all other parameters
constant. In fact, the variation of ν, which is ultimately related to density, strongly influences
the plasma profile, as discussed in Ref. [76]).
On the other hand, density has a pronounced impact on the radial particle fluxes. As presented
in Fig. 2.6, at low density, the flux of neutrals increases radially in the SOL region, from the
wall to the LCFS, due to the increasing number of neutrals recycled into the plasma. Then, the
neutral flux Γn remains roughly constant in the edge region, since few neutrals are ionized on
their way to the core. As expected, the ion flux is approximately opposite to the neutral flux,
with the dominant contribution arising from the E × B drift, both in the edge and SOL regions,
and considerably smaller contributions from the diamagnetic and polarization drift fluxes. In
addition, a strong radial decrease of the neutral flux is observed from the LCFS to the edge-core
interface at high density, as most neutrals are ionized within the edge close to the LCFS. As
imposed by mass conservation, the ion flux is opposite to the neutral flux also in this case.
In fact, at high density, the ion flux increases radially, from the core to the LCFS. The reason
behind the radial increase of the ion flux is the fact that, at high densities, a large number of
38
2.5. Conclusion
neutrals are ionized in the edge region, thus generating the ions that contribute to increase the
flux from the core to the LCFS. Beyond the LCFS, the ion flux decreases gradually, as the ions
flow to the limiter plates where they are recycled. Nonetheless, the relative contributions to the
ion flux do not change, as the flux profile is mostly determined by the E × B drift contribution
in the whole simulation domain.
Figure 2.6: Time, toroidally and poloidally averaged radial fluxes of neutrals and ions, discrimi-
nating the contributions of E × B, diamagnetic and polarization drifts, for a mass-conserving
model, considering an average over a time interval ∆t = 10.0R 0 /c s0 during the quasi-steady
state period, for three different values of the normalization density: n 0 = 5.0 × 1012 cm−3 (top
panel), n 0 = 2.0 × 1013 cm−3 (middle panel), and n 0 = 4.0 × 1013 cm−3 (bottom panel). Grid
sizes and all other simulation parameters are the same as the ones used for the simulations
presented in Fig. 2.2.
2.5 Conclusion
In this chapter, a mass-conserving model to leading order in ρ s0 /R 0 for the study of the neutral-
plasma interaction in the tokamak boundary is presented. The numerical implementation of
this model in the GBS code is described in detail and the first simulation results are shown
and discussed.
The model relies on the drift-reduced two-fluid Braginskii equations for the description of
plasma turbulence and addresses the dynamics of neutral atoms by solving the neutral kinetic
equation, using the method of characteristics and discretizing the resulting formal solution. In
order to ensure mass conservation, ions and neutrals considered as a whole, proper boundary
conditions are implemented at the walls, thus accounting for mass-conserving recycling,
and toroidal geometry is consistently accounted for. More precisely, we consider the radial
39
variation of the inverse aspect ratio ² = a 0 /R 0 throughout the code, we avoid the orderings
k k /k ⊥ ¿ 1 and k ⊥ R 0 À 1 in Eqs. (2.1-2.6), and we consistently take into account toroidicity
effects in the neutral calculations. We remark that the changes implemented in GBS also
ensure that the model conserves energy apart from terms (such as collisional heat exchange
and ohmic heating) missing in the temperature equations, which should nevertheless be
negligible in the edge and SOL conditions considered in this work (collisional heat exchange
and ohmic heating), the presence of boundaries, the energy losses due to the ionization and
recombination interactions and the use of the k k /k ⊥ ¿ 1 ordering in the Poisson equation.
We highlight that a proof of the energy conservation law for the drift-reduced Braginskii model
is reported in Ref. [48].
The mass-conserving model is implemented in the GBS code and convergence tests lead to the
conclusion that mass conservation is satisfied. The comparison with the non-mass-conserving
previously implemented model highlights the significant role played by mass conservation on
the simulation results. In fact, when the mass-conserving model described in this chapter is
taken into account, plasma and neutral quantities exhibit a larger drop across the LCFS and
more pronounced differences are found between the SOL and edge regions.
The analysis of the simulation results reveal that the ion particle flux is mostly determined
by the E × B drift and balances the neutral flux, as expected. A set of simulations performed
by varying the plasma density n 0 , while keeping ν and other parameters constant, show
that higher densities lead to weaker neutral penetration across the edge, as ionization and
charge-exchange peak closer to the LCFS. As a consequence, high density simulations exhibit
a stronger neutral density drop from the SOL towards the edge-core interface. In addition, the
density scan highlights the impact of density on the radial particle fluxes, as the weaker neutral
penetration associated with higher densities leads to a stronger drop of the neutral flux across
the closed field-line region, from the LCFS to the edge-core interface. The ion flux points
opposite to the neutral flux, as imposed by mass conservation, and is mostly determined by
the E × B component over the entire domain and for all densities considered.
We highlight that the mass-conserving model presented in this chapter is implemented for
the study of limited tokamak configurations. The implementation of mass conservation in
diverted configurations [35, 60] can be performed by following the steps described in the
present chapter.
40
3 A multi-component model of plasma
turbulence and kinetic neutral dy-
namics
In this chapter, we describe a self-consistent model for the simulation of a multi-component

deuterium plasma in the tokamak boundary. The plasma dynamics are described by means
of a set of drift-reduced Braginskii equations for each species, which include electrons, deu-
terium atomic ions and deuterium molecular ions. On the other hand, deuterium atoms
and molecules are addressed by considering a kinetic advection equation for each neutral
species, which are solved by using the method of characteristics. The models describing the
plasma and neutral species are coupled through a set of collisional interactions, which include
dissociation, ionization, charge-exchange and recombination processes. In this chapter, the
derivation of the plasma and neutral models is presented in detail, including its boundary
conditions, and their numerical implementation is discussed. The results of the first GBS
simulations of a multi-component plasma are also shown and analyzed with respect to results
from the single-component plasma model discussed in Ch. 2. We highlight that the content of
the present chapter is submitted for publication to Nuclear Fusion [77].
41
Chapter 3. A multi-component model of plasma turbulence and kinetic neutral dynamics
3.1 Introduction
The plasma boundary of tokamak devices is characterized by the presence of several ion and
neutral species, interacting through a complex set of collisional reactions [78, 12]. For instance,
not only neutral atoms, but also molecules play a relevant role in the plasma dynamics at
the boundary. Molecules are generated by processes such as the plasma recycling at the
vessel walls and gas puffing. In fact, ions and electrons, which flow along the magnetic-field
lines or across them due to turbulent transport, eventually end at the vessel walls, where
they recombine and re-enter the plasma as neutral particles. These neutrals can keep the
energy of the incoming ion, which effectively undergoes a reflection process, or they can
be emitted at the wall temperature, following the absorption of the incoming ion. In the
later case, a significant fraction of the atoms associate to form molecules, which are then
reemitted back to the plasma [79]. The exact probability of a reflection process, as well as
the probability that atoms associate into molecules, depend on the physical properties of the
material constituting the limiter or divertor plates [24]. Moreover, neutral molecules can be
externally injected to fuel the plasma, reduce the heat flux to the vessel wall (by decreasing
the plasma temperature and hence enhancing volumetric recombination), or for diagnostic
purposes, providing indirect measurements of the plasma quantities.
Neutral atoms and molecules are ionized, leading to atomic and molecular ions, thus gener-
ating a multi-component plasma. Molecules and molecular ions also undergo dissociative
processes, through which they are split into mono-atomic species. Recombination, charge-
exchange, elastic and inelastic collisions also come into play. All these collisional reactions
transform neutral particles into ions and electrons and vice versa, change the temperature of
the plasma species due to the energy required to trigger ionization and dissociation processes
and also impact the velocity of the plasma species. As a consequence, due to the plasma
dynamics in the boundary being strongly influenced by the interactions with neutrals, sim-
ulations of the tokamak plasma boundary should account for its multi-component nature
and consider the multiple collisional interactions, in order to enable reliable quantitative
predictions.
Multi-component plasmas are typically described by a fluid-diffusive model, usually sim-

plifying the Braginskii equations for the plasma species by modelling cross-field transport
through empirical anomalous transport coefficients. This approach is followed by the B2
[18, 19], EDGE2D [20], EMC3 [80], SOLEDGE-2D [21] and TECXY [22] codes. Neutral particle
species can also be described by means of a diffusive fluid approach [81], e.g. in the UEDGE
code [23]. However, the validity of diffusive models is limited to small values of the neutral
mean free path, λmfp,n . Hence, these models can not be applied when λmfp,n is of the order
of the plasma gradient scale length, which is often the case in the tokamak boundary. This
leads to neutrals being more commonly modelled by means of a kinetic description valid for
all ranges of mean free path. These models, which usually rely on Monte Carlo methods, are
implemented in the DEGAS2 [24], EIRENE [25], GTNEUT [26] and NEUT2D [27] codes. In
fact, research on heat exhaust typically requires integrated neutral-plasma simulations of the
42
3.1. Introduction
tokamak boundary, which often couple above mentioned fluid-diffusive multi-component

plasma models and Monte Carlo-based models for the neutral species, e.g. B2-EIRENE [25],
EDGE2D-EIRENE [82], EMC3-EIRENE [28] and SOLPS [29]).
Aiming at shedding light on cross-field transport processes, plasma turbulence simulations in

the tokamak boundary based on fluid and gyrofluid [30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40]
and kinetic codes [41, 42, 43] have made possible significant progress in the comprehension
of the underlying mechanisms of turbulence and cross-field transport in a single-ion species
boundary plasma. On the other hand, multi-component plasma simulations including turbu-
lent transport are still in their very beginning. Progress was achieved thanks to the coupling of
the SOLEDGE2D [83] and TOKAM3X [40] codes. The resulting SOLEDGE3X code now enables
multi-component plasma simulations, addressing the neutral species by means of the EIRENE
Monte Carlo code. The investigations undertaken with SOLEDGE3X focused on the study
of carbon impurities in the tokamak boundary [66]. Progress was also achieved by coupling
the two-dimensional fluid code HESEL [39] with a one-dimensional fluid-diffusive model for
the neutral particles, including both atoms and molecules. The resulting nHESEL [45, 39, 46]
code thus enables the simulation of a single-ion plasma, featuring the interactions with three
neutral species: cold hydrogen molecules puffed into the system, warm atoms resulting from
the dissociation of the hydrogen molecules and hot hydrogen atoms generated by charge-
exchange processes. This model was used to describe fueling in the presence of gas puffs, as
well as the formation of a density shoulder in the tokamak boundary at a high gas puffing rate.
In the present chapter, we describe the development and numerical implementation in the
GBS code of a multi-component model, addressing the multi-species plasma dynamics by
means of a set of fluid drift-reduced Braginskii equations while describing the neutrals by
solving a kinetic advection equation for each species. This chapter generalizes the implemen-
tation of the interaction between the neutrals and the plasma in GBS first described in Ref. [61]
for single-ion species plasmas, which was later improved in order to verify mass-conservation,
as described in Ch. 2. While the methodology presented in this chapter can be extended to
include an arbitrary number of particle species and the corresponding more complex scenar-
ios, we consider a deuterium plasma, featuring five different species: three charged particle
species, namely electrons (e − ), monoatomic deuterium ions (D + ) and diatomic deuterium
ions (D 2+ ), and two neutral species, including deuterium atoms (D) and molecules (D 2 ). We
highlight that D− and D+ 3 ions, neglected here, may also be important in detachment condi-
tions, as they play an important role in molecular assisted recombination (MAR) processes,
such as dissociative recombination, dissociated attachment and by mutual neutralization
[81, 84, 85].
The model described in this chapter represents the first implementation of a kinetic multi-
species model that avoids the statistical noise from the Monte Carlo method. As a matter of
fact, the neutral kinetic advection equations, valid for any values of the mean free path of
neutral species, are solved by using the method of characteristics and integrating the formal
solution in the velocity space. The resulting system of coupled integral equations for the
43
density of neutral species is discretized, enabling for the density of neutral particles to be
found. The model has the potential to provide the fundamental elements required for the
description and understanding of the mechanisms taking place in the boundary, including
the fueling and gas puff imaging, where molecular dynamics plays an important role.
The results of the first simulation based on the multi-component GBS model are described in
the present chapter, shedding light on the processes related to plasma fueling. More precisely,
still considering the limited configuration and plasma parameters between the sheath-limited
and high recycling regimes, we show that molecular dissociation has an impact on the space
distribution of ionization and the plasma profiles, compared to single-component simulations.
The present chapter is organized as follows. After the Introduction, the collisional processes at
play in the multi-component deuterium plasma model implemented in GBS are presented
in Sec. 3.2. The drift-reduced Braginskii equations describing a multi-component plasma
are derived, in Sec. 3.3, extending the approach followed in the single-component version of
GBS described in Ch. 2. In Sec. 3.4, the multi-component boundary conditions applied at
the tokamak wall are introduced. The kinetic model for the neutral species is presented in
Sec. 3.5, which also describes the numerical implementation of the model, a generalization
of the approach initially developed in Ref. [61] and then improved as described in Sec. 2.2.2
for a single neutral species model. Finally, we present in Sec. 3.6 the results of the first multi-
component plasma GBS simulations, discussing the effect of the molecules on the plasma
dynamics, with respect to the results from the single-ion species plasma simulations presented
in Ch. 2. The summary follows, in Sec. 3.7.
3.2 Collisional processes in multi-component deuterium plasmas

The model presented in this chapter aims at describing a multi-component deuterium plasma.
Analogously to Refs. [81], [86] and [87], we consider a plasma featuring the e, D+ and D+ 2
species and we consider the D and D2 neutral species. D2 molecules are generated by the
association of atoms at the vessel walls and by external injection. A fraction of D2 molecules
are ionized, thus generating D+2 ions, while dissociative processes give rise to mono-atomic
ions, D , and neutrals, D, which are also ionized, leading to D+ . We highlight that the resulting
+
five-species model differs from the three-species model used in the previous GBS simulations
of a single-ion species plasma [61] considered in Ch. 2, where only mono-atomic deuterium
ions and neutrals are evolved. We remark that, by introducing the tools required to describe
the fundamental processes at play in multi-component plasmas, the model presented in this
chapter can be extended to account for more complex scenarios, featuring several plasma and
neutral species.
The plasma and neutral species are coupled by a number of collisional processes, which
include ionization, recombination, charge-exchange, dissociation, and elastic collisions be-
tween electrons and neutrals. These processes appear in the models describing the neutral
and plasma as sources and sinks of particles and heat, as well as friction terms. Table 3.1 lists
44
3.2. Collisional processes in multi-component deuterium plasmas
the collisional processes considered in our multi-component model, as well as their respective
reaction rates.
We remark that v e , v D+ and v D+2 denote the modulus of the electron, D+ and D+ 2 velocities
respectively, while their densities are represented by n e , n D+ and n D+2 . On the other hand, σiz,D
and σiz,D2 are the cross sections of the ionization of D and D2 respectively, σrec,D+ and σrec,D+2
denote the cross sections for recombination of D+ and D+ 2 with electrons, σe-D and σe-D2
refer to the cross sections of elastic collisions between electrons and D and D2 respectively,
σdiss,D2 and σdiss,D+2 represent the dissociation cross sections of D2 and D+ 2 , σdiss-iz,D2 and
σdiss-iz,D+2 are the cross sections for dissociative ionization of D2 and D+ 2 , σdiss-rec,D+
2
denotes
the cross section of dissociative recombination of D2 ions and, finally, σcx,D+ , σcx,D+2 , σcx,D-D+2
+
and σcx,D2 −D+ are the cross sections for D − D+ , D2 − D+ + +

2 , D − D2 and D2 − D charge-exchange
interactions. In the present work, the distinction between fundamental and excited states
for atoms, molecules and ions is neglected. As a matter of fact, we use the total cross section
for each process considering the sum over the accessible electronic states of the reactants
and products, following Refs. [84] and [85]. By applying momentum and energy conservation
considerations, we compute the velocity and energy of the collision products, too. Since these
also depend on the electronic states of the reactants and products, we average over the states
relevant to a given reaction, taking into account the cross section of each state. The derivation
of these values is detailed in App. B.
By considering Krook collision operators, the collision rates for ionization, recombination,
elastic collisions and dissociative processes are computed as the average over the electron
velocity distribution function, thus neglecting the velocity of the massive particle involved in
the collision (D, D2 , D+ or D+
2 ) when computing the relative velocity between the electron and
the other particle. As a matter of fact, electrons have considerably larger thermal velocity than
ions or neutrals. As for charge-exchange processes between D+ ions and the neutral species D
and D2 , given the weak dependence of the cross section on the ion-neutral relative velocity
[12], we neglect the neutral particle velocity (D or D2 ) when evaluating the relative velocity
of the colliding particles (the velocity of a neutral particle is usually smaller than the ion
velocity). Therefore, the reaction rates σcx,D+ and σcx,D2 −D+ are computed by averaging over
the distribution function of the D+ species, which we assume to be described by a Maxwellian
with temperature TD+ . We follow the same approach when computing the cross section of
charge-exchange interactions between D+ 2 ions and the D2 and D neutrals, by averaging the
cross sections σcx,D+2 and σcx,D−D+2 over the D+ 2 velocity distribution function, which we assume
to be a Maxwellian of temperature TD+2 .
The 〈vσ〉 products for most of the reactions considered in Table 3.1 are obtained from the
AMJUEL [84] and HYDEL [85] databases (precise references for each cross section listed in
Table 1 of Ref. [87]). We remark that, although these databases present the cross sections for
hydrogen plasmas, we assume in this chapter that they apply also to deuterium. We highlight
that the cross section for the e− − D elastic collisions is obtained from Ref. [88] (page 40, Table
2), while the e− − D2 elastic collision cross section is computed from Ref. [89] (page 917, Table
45
Table 3.1: Collisional processes considered and their respective reaction rates.
Collisional process Equation Reaction Frequency

− − +
νiz,D = n e v e σiz,D (v e )
®
Ionization of D e + D → 2e + D
Recombination of D+ and e− e− + D + → D νrec,D+ = n e v e σrec,D+ (v e )
®
e− − D elastic collisions e− + D → e− + D νe-D = n e 〈v e σe-D (v e )〉

e− + D2 → 2e− + D+ νiz,D2 = n e vDe σiz,D2 (v e ) E
®
Ionization of D2 2
Recombination of D+
2 and e
−
e− + D +
2 → D2 νrec,D+2 = n e v e σrec,D+2 (v e )
− − −
νe-D2 = n e v e σe-D2 (v e )
®
e − D2 elastic collisions e + D2 → e + D 2
e− + D 2 → e− + D + D νdiss,D2 = n e v e σdiss,D2 (v e )
®
Dissociation of D2
e− + D2 → 2e− + D + D+ νdiss-iz,D2 = nDe v e σdiss-iz,D2 (v
®
Dissociative ionization of D2 Ee )
Dissociation of D+
2 e− + D + −
2 → e +D+D
+
νdiss,D+2 = n e v e σdiss,D+2 (v e )
D E
Dissociative ionization of D+
2 e− + D + −
2 → 2e + 2D
+
νdiss-iz,D+2 = n e v e σdiss-iz,D+2 (v e )
D E
Dissociative recombination of D+
2 e− + D +
2 → 2D νdiss-rec,D+2 = n e v e σdiss-rec,D+2 (v e )
Charge-exchange of D+ , D D+ + D → D + D+ νcx,D = n D+ Dv D+ σcx,D+ (v D+ ) E
®
Charge-exchange of D+
2 , D2 D+ +
2 + D2 → D 2 + D 2 νcx,D2 = n D+2 v D+2 σcx,D+2 (v D+2 )
D E
Charge-exchange of D+
2 ,D D+
2 + D → D2 + D
+
νcx,D-D+2 = n D+2 v D+2 σcx,D-D+2 (v D+2 )
Charge-exchange of D2 , D+ D2 + D+ → D+ νcx,D2 −D+ = n D+ v D+ σcx,D2 −D+ (v D+ )
®
2 +D
13). The cross section for the D2 − D+ 2 charge-exchange reaction is in turn taken from the
HYDEL database (H.4, reaction 4.3.1), while the cross sections from the ALADDIN database
[90] (obtained from Refs. [91, 92]) are used for the D − D+
2 charge-exchange interaction. For all
the other reactions, the cross sections from the AMJUEL database [84] are considered. The
〈vσ〉 product for the collisional processes considered in this chapter is plotted as a function of
the temperature of the colliding particle in Fig. 3.1.
We henceforth focus on the computation of the velocity and energy of their products. For
charge-exchange interactions following the general form A + B + → A + + B , we assume that,
while A and B + exchange an electron, their velocities are not affected and energy is con-
served. As a result, the A + ion is released from the charge-exchange collision with the velocity
of A, and B is released with the velocity of B + . For the e− + D → e− + D elastic collisions,
because of the large electron to deuterium mass ratio, we consider that the D velocity is
not impacted by the collision, while the electron is emitted isotropically in the reference
frame of the massive particle following a Maxwellian distribution function, Φe [vD ,Te,e-D ] =
¤3/2
exp −m e (v − vD )2 /(2Te,e-D ) , which is centered at the velocity of the in-
£ £ ¤
m e /(2πTe,e-D )
R R
coming D particle, vD = v f D d v/ f D d v. The temperature Te,e-D is, in turn, established by
energy conservation considerations. More precisely, we observe that the average energy of
the incoming electrons consists of the sum of the kinetic energy associated with the fluid
velocity, ve , and the thermal contribution, (3/2)Te . On the other hand, the energy of the
outcoming electrons has a contribution arising from the collective velocity of the re-emitted
particles, vD , and a thermal contribution, Te,e-D . It follows that Te,e-D verifies the balance
46
3.2. Collisional processes in multi-component deuterium plasmas
Figure 3.1: 〈vσ〉 product for the collisional processes considered in this chapter. Ionization
processes, elastic collisions and charge-exchange processes are displayed on the top panel,
dissociative reactions on the bottom panel. The 〈vσ〉 product is plotted as a function of the
temperature of the colliding particle.
described by 3Te /2 + m e v e2 /2 = Te,e-D + m e v D

2
/2. The elastic collisions between electrons and
D2 are then described in a similar way. The re-emitted electrons have a distribution of veloc-
ities Φe £vD ,Te,e-D ¤ , with Te,e-D2 obtained from an analogous conservation law, expressed as
2 2
3Te /2 + m e v e2 /2 = Te,e-D2 + m e v D
2
2
/2.
Considering the electrons generated by ionization of D, we assume that they are described
by the Maxwellian distribution Φe [vD ,Te,iz(D) ] centered at the fluid velocity of the D atom vD ,
with Te,iz(D) accounting for the ionization energy loss, 〈E iz 〉, its value being presented in
Table 3.2. More precisely, Te,iz(D) satisfies the energy conservation law, 3Te /2 + m e v e2 /2 =
2
£ ¤ ®
2 Te,iz(D) + m e v D /2 + E iz,D , since the reaction gives rise to two electrons with the same prop-
erties. A similar approach is followed for the ionization of D2 , with the two emitted electrons
being described by a Maxwellian Φe £vD ,Te,iz(D) ¤ centered at the velocity of the D2 molecules,
2 2 h i
vD2 , and with temperature Te,iz(D)2 obtained from 3Te /2 + me v e2 /2 = 2 Te,iz(D2 ) + me v D2
2
/2 +
® ®
E iz,D2 , with E iz,D2 the average energy loss due to ionization of D2 (see Table 3.2). We high-
light that we neglect multi-step ionization processes when we compute the cross section for
ionization of D and D2 , adopting the same procedure for all the other reactions induced by
impacting electrons, such as the dissociative processes.
47
We also apply the same procedure used for ionization processes to describe the proper-
ties of the electrons generated by dissociative processes. The electron created by dissoci-
ation of D2 is described by the Maxwellian Φe £vD ,Te,diss(D) ¤ centered around vD2 and with
2 2
temperature Te,diss(D)2 obtained from 3Te /2 + m e v e2 /2 = Te,diss(D2 ) + m e v D
2
®
2
/2 + E diss,D 2
. Re-
garding dissociation of D+2 , the resulting electron is analogously modelled by a Maxwellian
Φh + i centered at the velocity of the D+ ion and with temperature T
2 e,diss(D)+ given
e vD ,Te,diss(D)+ 2
2 2
by the corresponding energy conservation law expressed as 3Te /2 + m e v e2 /2 = Te,diss(D+2 ) +
D E
2
me v D + /2 + E diss,D+ . On the other hand, dissociative ionization of D2 gives rise to two elec-
2
2
trons, their Maxwellian distribution function, Φe £vD ,Te,diss-iz(D) ¤ , being centered around the D2
2 2
velocity,
h vD2 , and characterized
i by a temperature Te,diss-iz(D2 ) , obtained from 3Te /2+m e v e2 /2 =
2
®
2 Te,diss-iz(D2 ) + m e v D 2
/2 + E diss-iz,D2 . Similarly, we assume that the electrons generated by
dissociative ionization of D+ 2 aredescribed by a Maxwellian Φ
h i centered at v +
e vD+ ,Te,diss-iz(D)+ D2
2 2
and with temperature Te,diss-iz(D+2 ) obtained from energy conservation, 3Te /2 + m e v e2 /2 =
h i D E
2
2 Te,diss-iz(D+2 ) + m e v D + /2 + E diss-iz,D+ .
2
2
The evaluation of the temperature of the D atoms and D+ ions released from dissociative
reactions is based on the modelling of these reactions as Franck-Condon dissociation pro-
cesses. These temperatures are summarized in Table 3.2 and rely on data from Ref. [85]. The
calculations are detailed in the App. B. We highlight that the values for the average electron
energy loss due to dissociative processes take into account the energy lost by an electron
when exciting the D2 molecule or D2 molecular ion before dissociation takes place. This
contrasts with other work (see e.g. Ref. [46]) where the energy inherent to the dissociation
process is taken into account. Our approach is justified by the fact that the effective electron
energy loss associated with a given dissociative process accounts for the energy required to
excite the molecule, which, in turn, comprises the energy cost of dissociation, the kinetic
energy of the products and the radiation emitted due to deexcitation of the products. We also
highlight that these particles are emitted isotropically in the frame of the centre of mass of
the incoming D2 or D+ 2 particle. Thus, we assume that the D atoms generated by dissociation
of D2 molecules, for example, follow a Maxwellian distribution ΦD£vD ,TD,diss(D ) ¤ . Similarly, we
2 2
describe the neutral D atoms and D+ ions produced by dissociative-ionization of D2 molecules
by the Maxwellian distributions Φ h i and Φ h
+
i respectively, with the
D vD2 ,TD,diss-iz(D2 ) D vD2 ,TD,diss-iz(D2 )
temperature TD,diss-iz(D2 ) listed in Table 3.2 and evaluated in App. B. In turn, Φ h i
D vD+ ,TD,diss
2 ( ) D+
2
and Φ +
h i are the Maxwellian distributions of D atoms and D+ ions generated by
D vD+ ,TD,diss D +
2 ( 2)
dissociation of D+ +
2 ions, where vD2 denotes the fluid velocity of the D2 ion population that
+
includes the leading order components (see Sec. 3.3). Moreover, dissociative-ionization of
D+2 gives rise to D ions that are described by a Maxwellian distribution ΦD+ v + ,T
+ h i.
D
2 D,diss-iz(D + )
2
Finally, we remark that the D atoms and D+ generated by dissociative-recombination of
2 are described by the Maxwellian distributions ΦD v + ,T
D+ h i and Φ h i
D,diss-rec(D+ )
D+ vD+ ,TD,diss-rec D+
D
2 2 2 ( 2)
respectively, with TD,diss-rec(D+2 ) denoting the average thermal energy of the reaction products.
48
3.3. The three-fluid drift-reduced Braginskii equations
Table 3.2: Average electron energy loss and average energy of reaction products for the ioniza-
tion and dissociative processes included in the model.
Collisional process e− Energy loss Temperature of products

®
Ionization of D E = 13.60eV —————————
iz,D ®
Ionization of D2 E iz,D2 = 15.43eV —————————
®
Dissociation of D2 E diss,D2 ' 14.3eV TD,diss(D2 ) ' 1.95eV
Dissociative ionization
®
of D2 (E e < 26eV) E diss-iz,D2 ' 18.25eV TD,diss-iz(D2 ) ' 0.25eV
Dissociative ionization
®
of D2 (E e > 26eV) DE diss-iz,DE2 ' 33.6eV TD,diss-iz(D2 ) ' 7.8eV
Dissociation of D+ E + ' 13.7eV TD,diss(D+2 ) ' 3.0eV
2 D diss,D2 E
Dissociative ionization of D+
2 E diss-iz,D+2 ' 15.5eV TD,diss-iz(D+2 ) ' 0.4eV
Dissociative recombination of D+
2 —————————– TD,diss-rec(D+2 ) ' 11.7eV
3.3 The three-fluid drift-reduced Braginskii equations

The plasma turbulent dynamics is described by a fluid model based on a Braginskii set of
equations for each plasma species. Their derivation starts from the kinetic equations for
e− , D+ and D+ 2 , which must include the terms arising from the neutral-plasma interactions.
These equations generalise the ones considered in the single-ion species model introduced in
Refs. [34, 61], by adding the new collisional terms listed in Table 3.1, as well as an equation
describing the D+ 2 ions. The kinetic equations yield
∂ fe ∂ fe fe fe
· ¸
+v· +a· = νiz,D n D 2Φe [vD ,Te,iz(D) ] −
∂t ∂x ∂v ne
fe nD fe
· ¸ · ¸
+
+ νe-D n D Φe [vD ,Te,en(D) ] − − νrec,D+ f e + νiz,D2 n D2 2Φe £vD ,Te,iz(D ) ¤ −
ne ne 2 2 ne
·
fe
¸ n D2 +
+ νe-D2 n D2 Φe £vD ,Te,en(D ) ¤ − − νrec,D+2 fe
2 2 ne ne
(3.1)
fe fe
· ¸ · ¸
+ νdiss,D2 n D2 Φe £vD ,Te,diss(D ) ¤ − + νdiss-iz,D2 n D2 2Φe £vD ,Te,diss-iz(D ) ¤ −
2 2 ne 2 2 ne
f
· ¸
+ νdiss-iz,D+2 n D+2 2Φ h i− e
e vD+ ,Te,diss-iz(D+ ) n
2 2 e
f fe
· ¸
e
+ νdiss,D+2 n D+2 Φ h i− − νdiss-rec,D+2 n D+2 +C ( f e ),
e vD+ ,Te,diss(D+ ) ne ne
2 2
∂ f D+ ∂ f D+ f D+
+v· +a· = νiz,D f D − νrec,D+ f D+
∂t µ ∂x ∂
¶v
nD n D2 (3.2)
− νcx,D f D+ − f D + νcx,D-D+2 f D − νcx,D2 −D+ f D+
nD + n D+
+ νdiss-iz,D2 f D2 + 2νdiss-iz,D+2 f D+2 + νdiss,D+2 f D+2 +C ( f D+ ),
49
and
∂ f D+2 ∂ f D+2 f D+
+v· + a · 2 = νiz,D2 f D2 − νrec,D+2 f D+2
∂t Ã ∂x ∂v!
n D2 nD (3.3)
− νcx,D2 f D+2 − f D2 − νcx,D2 −D+ f D2 − νcx,D-D+2 f D+
n D+2 n D+2 2
³ ´
− νdiss-iz,D+2 + νdiss,D+2 + νdiss-rec,D+2 f D+2 +C ( f D+2 ).
In Eqs. (3.1-3.3), v denotes the particle velocity, a the particle acceleration due to the Lorentz
Force, and ∂/∂x the gradient in real space and ∂/∂v in the velocity space. The C ( f e ), C ( f D+ )
and C ( f D+2 ) terms refer to the Coulomb collisions between charged particles influencing the e,
D+ and D+ 2 distribution functions, respectively.
The three-fluid Braginskii equations (for a plasma featuring e− , D+ and D+ 2 ) are obtained
by computing the first three moments of the kinetic equations for each species in the limit
ΩcD+ τD+ À 1, with ΩcD+ = eB /m D+ the cyclotron frequency (m D+ stands for the D+ ion mass
and e represents the elementary charge) and τD+ the characteristic Coulomb collision time for
D+ ions. The Braginskii equations, including the neutral-plasma interaction terms, can be
derived by following the steps presented in Ref. [48], taking the form
∂n e
+ ∇ · (n e ve ) = n D νiz,D − n D+ νrec,D+ + n D2 νiz,D2 − n D+2 νrec,D+2
∂t (3.4)
+ n D2 νdiss-iz,D2 + n D+2 νdiss-iz,D+2 − n D+2 νdiss-rec,D+2 ,
∂n D+
+ ∇ · n D+ vD+ = n D νiz,D − n D+ νrec,D+ + n D νcx,D-D+2 − n D2 νcx,D2 −D+
¡ ¢
∂t ³ ´ (3.5)
+ n D2 νdiss-iz,D2 + n D+
2
2ν diss-iz,D+
2
+ν
diss,D+
2
,
∂n D+2 ³ ´
+ ∇ · n D+2 vD+2 = n D2 νiz,D2 − n D+2 νrec,D+2 + n D2 νcx,D2 −D+ − n D νcx,D-D+2
∂t ³ ´ (3.6)
− n D+2 νdiss-iz,D+2 + νdiss,D+2 + νdiss-rec,D+2 ,
d e v eα ∂p e ∂Πeαβ £ ¤
me ne =− − − en e E α + (ve × B)α + R eα
dt ∂x α ∂x β
+ m e n D 2νiz,D + νe-D (v Dα − v eα ) + n D2 2νiz,D2 + νe-D2 v D2 α − v eα
£ ¡ ¢ ¡ ¢¡ ¢
³ ´ (3.7)
+ 2n D2 νdiss-iz,D2 v D2 α − v eα + 2νdiss-iz,D+2 n D+2 v D+2 α − v eα
¡ ¢
³ ´ ¢i
+n D+2 νdiss,D+2 v D+2 α − v eα + n D2 νdiss,D2 v D2 α − v eα ,
¡
50
d D+ v D+ α ∂p D+ ∂ΠD+ αβ £ ¡ ¢ ¤
m D n D+ =− − + en D+ E α + vD+ × B α + R D+ α
dt ∂x α ∂x β
(3.8)
h
+ m D n D (νiz,D + νcx,D + νcx,D-D+2 )(v Dα − v D+ α ) + n D2 νdiss-iz,D2 (v D2 α − v D+ α )
³ ´ i
+n D+2 2νdiss-iz,D+2 + νdiss,D+2 (v D+2 α − v D+ α ) ,
d D+2 v D+2 α ∂p D+2 ∂ΠD+ αβ h ³ ´ i

m D2 n D+2 =− − + en D+2 E α + vD+2 × B + R D+2 α
dt ∂x α ∂x β α (3.9)
+ m D2 n D2 (νiz,D2 + νcx,D2 + νcx,D2 −D+ )(v D2 α − v D+2 α ),
3 d e Te ∂v eβ
ne + p e ∇ · ve = −∇ · qe − Πeαβ +Q e
2 dt ∂x α
· µ ¶¸
3 3 4
+ n D νiz,D −E iz,D − Te + m e ve · ve − vD − n D νe-D m e ve · (vD − ve )
2 2 3
· µ ¶¸
3 3 4
+ n D2 νiz,D2 −E iz,D2 − Te + m e ve · ve − vD2 − n D2 νe-D2 m e ve · (vD2 − ve )
2 2 3
+ n D2 νdiss,D2 −E diss,D2 + m e ve · ve − vD2
£ ¡ ¢¤ (3.10)
· µ ¶¸
3 3 4
+ n D2 νdiss-iz,D2 −E diss-iz,D2 − Te + m e ve · ve − vD2
2 2 3
h ³ í
+ n D+2 νdiss,D+2 −E diss,D+2 + m e ve · ve − vD+2
· µ ¶¸
3 3 4
+ n D+2 νdiss-iz,D+2 −E diss-iz,D+2 − Te + m e ve · ve − vD+2 ,
2 2 3
3 d D+ T D+ ∂v D+ β
n D+ + p D+ ∇ · vD+ = −∇ · qD+ − ΠD+ αβ +Q D+
2 dt ∂x α
¢ m D+ ¡
· ¸
3¡ ¢2
+ n D (νiz,D + νcx,D + νcx,D-D+2 ) T D − T D+ + vD − vD+
2 2
¢ m D+ ¡
· ¸
3¡ ¢2
+ n D2 νdiss-iz,D2 TD+ ,diss-iz(D2 ) − TD+ + vD2 − vD+ (3.11)
2 2
· ³
3 ´ m +³ ´2 ¸
D
+ 2n D+2 νdiss-iz,D+2 T + + −T + + vD+2 − vD+
2 D ,diss-iz(D2 )
D
2
· ³
3 ´ m +³ ´2 ¸
D
+ n D+2 νdiss,D+2 T + + −T + + vD+2 − vD+ ,
2 D ,diss(D2 )
D
2
3 d D+2 TD+2 ∂v D+2 β

n D+2 + p D+2 ∇ · vD+2 = −∇ · qD+2 − ΠD+2 αβ +Q D+2
2 dt ∂x α
· ³ ´ m D+ ¸ (3.12)
3
+ n D2 (νcx,D2 + νiz,D2 + νcx,D2 −D+ ) TD+2 − TD+2 + 2
(vD2 − vD+2 )2 ,
2 2
where Πeαβ denotes the component of the stress tensor along the α and β directions, Re the
friction force acting on the electrons, qe the electron heat flux density, Q e the electron heat
51
generated by Coulomb collisions and d e /d t = ∂/∂t + (ve · ∇) the electron advective derivative.
The equivalent notation is used for the D+ and D+2 species.
Similarly to what was done in the derivation of the single-ion species model presented in
Ch. 2, we consider the d/dt ¿ ΩcD+ and k ⊥ ρ D+ ¿ 1 orderings, which are valid in typical
conditions of the tokamak boundary. This leads to the derivation of the drift-limit of the
Braginskii equations, keeping only leading order components in (1/ΩcD+ )d/dt in the electron
perpendicular velocity, i.e. v⊥e = v⊥e0 = vE×B + vde , with vE×B = (E × B)/B 2 the E × B drift
and vde = (B × ∇p e )/(en e B 2 ) the electron diamagnetic drift, thus neglecting electron inertia.
The D+ perpendicular velocity is analogously decomposed as v⊥D+ = v⊥D+ 0 + vpol,D+ + vfric,D+ ,
where the leading order perpendicular velocity,
v⊥D+ 0 = vE×B + vdD+ , (3.13)
is defined as the sum of the E ×B drift and the diamagnetic drift, vdD+ = (B ×∇p D+ )/(en D+ B 2 ).
The polarization drift,
d D+ n D+
µ ¶ · ¸
1 1 1 ∇G D+
vpol,D + =− ∇⊥ φ + ∇⊥ p D +
+ b × GD k −
+ , (3.14)
n D+ ΩcD+ d t B B m D+ n D+ ΩcD+ 3
is of higher order than v⊥D+ 0 in the d /d t ¿ ΩcD+ expansion, as it is shown in Ref. [48]. The
drift velocity due to friction between D+ ions and the other species,
n D νcx,D + νiz,D + νcx,D−D+2 ¡ n D νiz−diss,D2 ¡

v⊥D − v⊥D+ 0 × b + 2
¢ ¢
vfric,D+ = v⊥D2 − v⊥D+ 0 × b
n D+ ΩcD+ n D+ ΩcD+
n D+2 2νdiss-iz,D+2 + νdiss,D+2 ³ ´
+ v⊥D+2 0 − v⊥D+ 0 × b,
n D+ ΩcD+
(3.15)
is of higher order in (1/ΩcD+ )d /d t , too. This term features contributions from collisions
of D+ with D, D2 and D+ 2 particles. Assuming v D . v D , v D2 . v D and v D2 . v D , and
+ + + +
noticing that ν/ΩcD+ ¿ 1, we obtain v fric,D+ ∼ (ν/ΩcD+ )v D+ ¿ v D+ . Hence, we approximate

vD+ and vD+2 with their respective leading order components, i.e. v⊥D+ ' v⊥D+ 0 and v⊥D+2 '
v⊥D+2 0 , in Eq. (3.15). In Eqs. (3.14) and (3.15), the gyroviscous term for D+ ions, G D+ =
−η 0D+ 2∇k v kD+ +C (φ)/B +C (p D+ )/(ZD+ n D+ B ) , the D+ viscosity η 0D+ , is introduced, as well
£ ¤
as the magnetic field curvature vector k = (b · ∇) b, the gradient along the magnetic field
∇k = b · ∇, the gradient perpendicular to the magnetic field ∇⊥ = ∇ − b∇k and the magnetic
field unit vector b = B/B .
A similar approach is followed when deriving the drift-limit of the D+

2 velocity, as the d /d t ¿
52
ΩcD+2 ordering remains valid in typical tokamak boundary conditions. The perpendicular
velocity of D+
2 ions is therefore written as v⊥D2 = v⊥D2 0 + vpol,D2 + vfric,D2 , with
+ + + +
v⊥D+2 0 = vE×B + vdD+2 (3.16)
the leading order component, with vdD+2 = (B ×∇p D+2 )/(en D+2 B 2 ). We remind that vpol,D+2 is the
polarization drift velocity and vfric,D+2 denotes the drift velocity arising from friction between
D+2 ions and other species. Their respective expressions are
1 d D+2 µ n D+2 1
¶
1
· ∇G D+2 ¸
vpol,D+2 = − ∇⊥ φ + ∇⊥ p D+2 + b × G D+2 k − , (3.17)
n D+2 ΩcD+2 d t B B m D+2 n D+2 ΩcD+2 3
and
n D2 νiz,D2 + νcx,D2 + νcx,D2 −D+ ³ ´

vfric,D+2 = v⊥D2 − v⊥D+2 0 × b, (3.18)
n D+2 ΩcD+2
h i
2 gyroviscous term and η 0D2 the
with G D+2 = −η 0D+2 2∇k v kD+2 +C (φ)/B +C (p D+2 )/(n D+2 B ) the D+ +
corresponding viscosity. We note that approximation v⊥D+2 ' v⊥D+2 0 is used in Eq. (3.18).
To obtain an expression for the parallel friction forces and parallel heat fluxes and close the Bra-
ginskii equations, we apply the collisional closure developed by Zhdanov in Ref. [65], following
the formulation presented in Refs. [66, 93], which facilitates the numerical implementation.
The use of this procedure in the context of the multi-species plasma considered in the present
chapter is detailed in App. C, where we make use of the fact that the n D+2 is significantly smaller
than the n D+ , i.e. n D+2 /n D+ ¿ 1, for typical tokamak boundary conditions, which in turn allows
us to write n e ' n D+ thanks to quasi-neutrality. On the other hand, the contributions from the
perpendicular components of the heat fluxes in the terms ∇ · qe and ∇ · qD+ appearing in the
Te and TD+ equations, respectively, are evaluated by assuming the Ωce τe À 1, ωD+ τcD+ À 1
limits (typical time between collisions considerably larger than cyclotron frequency), which
significantly simplifies the expressions, following the same approach described in [17] and
in agreement with the single-ion species model [33, 34]. This approach is generalised to
evaluate the term arising from the perpendicular component of ∇ · qD+2 in the TD+2 equation,
Eq. (3.12). We also highlight that the collisional heat exchange terms, i.e. Q e , Q D+ and Q D+2 , in
Eqs. (3.10-3.12), are neglected in the derivation of the drift-limit of the Braginskii equations,
as they are assumed to be of higher order in ρ s0 /R 0 , similarly to the single-component GBS
model presented in Refs. [33, 34] and in Ch. 2 of this thesis.
Therefore, the drift-reduced Braginskii system of equations includes the continuity equation
for the electron species, the continuity equation for the D+
2 species, the vorticity equations
53
that ensures quasi-neutrality, n e = n D+ + n D+2 , and the equations for the parallel velocities and
temperature of all species. They take the form
∂n e ρ −1 2£
= − ∗ [φ, n e ] + C (p e ) − n eC (φ) − ∇ · (n e v ke b) + Dne ∇2⊥ n e + S ne
¤
∂t B B
+ n D νiz,D − n D+ νrec,D+ + n D2 νiz,D2 − n D+2 νrec,D+2 (3.19)
+ n D2 νdiss-iz,D2 + n D+2 νdiss-iz,D+2 − n D+2 νdiss-rec,D+2 ,
∂n D+2 ρ −1
∗ 2h i
=− [φ, n D+2 ] − ∇ · (n D+2 v kD+2 b) − n D+2 C (TD+2 ) + TD+2 C (n D+2 ) + n D+2 C (φ)
∂t B B
+ DnD+ ∇2⊥ n D+2 + S nD+ + n D2 νiz,D2 − n D+2 νrec,D+2 + n D2 νcx,D2 −D+ − n D νcx,D-D+2 (3.20)
2 2
³ ´
− n D+2 νdiss-iz,D+2 + νdiss,D+2 + νdiss-rec,D+2 ,
∂Ω ρ∗ £ ¢ v kD+2
· −1 ³ i´¸ v kD+ ¡
h · ³ ´¸
φ, B ΩD + 2 φ, B ωD2 ∇k B ΩD +
¤
= −∇ · + + −∇· + ∇k B ωD2 +
∂t B B B
2h i
+ n eC (Te ) + TeC (n e ) + n D+ C (TD+ ) + TD+ C (n D+ ) + n D+2 C (TD+2 ) + TD+2 C (n D+2 )
B
b×k b×k
· µ ¶ µ ¶ ¸
¡ ¢ 2 ¡ ¢
+ ∇ · j k b + ∇G D+ · +G D+ ∇ · − C G D+
B B 3B
· µ
b×k
¶ µ
b×k
¶
2 ³ ´ ¸ (3.21)
+ ∇G D+2 · +G D+2 ∇ · − C G D+2 + η 0Ω ∇2k Ω + D⊥Ω ∇2⊥ Ω
B B 3B
" #
2n D2 ¡ nD ³
· ´ ¸
νcx,D2 + νiz,D2 + νcx,D2 −D+ ωD+2 − ∇ · νcx,D + νiz,D + νcx,D-D+2 ωD+
¢
−∇·
n D+2 n D+
·n + ³
n D2
· ¸ ´¸
D2
´³
−∇· νdi-iz,D2 ωD + ∇ ·
+ 2νdi-iz,D2 + νdi,D2 ωD2 − ωD
+ + + + ,
n D+ n D+
∂v ke ρ −1 mD ∇k p e
· ¸
∗ 2
=− [φ, v ke ] − v ke ∇k v ke + ∇k φ − − ∇kG e − 0.71∇k Te
∂t B me ne 3n e
mD ¡ 1
ν v ke − v kD+ + Dv ke ∇2⊥ v ke + n D 2νiz,D + νe-D v kD − v ke
¢ £ ¡ ¢¡ ¢
−
me ne (3.22)
+ n D2 2νiz,D2 + νe-D2 v kD2 − v ke + n D2 2νdiss-iz,D2 + νdiss,D2 v kD2 − v ke
¡ ¢¡ ¢ ¡ ¢¡ ¢
³ ´³ í
+n D+2 2νdiss-iz,D+2 + νdiss,D+2 v kD+2 − v ke ,
54
+
∂v kD+ ρ −1
∗
∇k p D 2 ne
=− [φ, v kD ] − v kD ∇k v kD − ∇k φ −
+ + + − ∇kG D+ + 0.71 ∇k Te
∂t B n D+ 3n D+ n D+
ne ¡ 1 h (3.23)
−ν v kD+ − v ke + Dv kD+ ∇2⊥ v kD+ + n D (νiz,D + νcx,D + νcx,D-D+2 )(v kD − v kD+ )
¢
n D+ n D+
³ ´ i
+n D+2 νdiss-iz,D2 (v kD2 − v kD+ ) + n D+2 2νdiss-iz,D+2 + νdiss,D+2 (v kD+2 − v kD+ ) ,
∂v kD+2
" #
ρ −1 1 ∇p D +
2
= − ∗ [φ, v kD+2 ] − v kD+2 ∇k v kD+2 + −∇k φ − 2
− ∇kG D+2
∂t B 2 n D+2 3n D+2
(3.24)
n D2
+ Dv kD+ ∇2⊥ v kD+2 + (νiz,D2 + νcx,D2 + νcx,D2 −D+ )(v kD2 − v kD+2 ),
2 n D+2
∂Te ρ −1 4Te C (p e ) 5
· ¸
∗ 2Te ¡ ¢
=− [φ, Te ] − v ke ∇k Te + + C (Te ) −C (φ) − ∇ · v ke b
∂t B 3B ne 2 3
2 1.62 £ ¡ ¢ ¡ ¢¤ 2
+ n e Te ∇k Te ∇ · b + ∇k n e Te ∇k Te − 0.71Te ∇ · (v ke − v kD+ )b
3n e ν 3
Te
µ ¶
2
∇k n e + ∇k Te v ke − v kD+ + χ⊥e ∇2⊥ Te + ∇k χke ∇k Te + S Te
¡ ¢ ¡ ¢
− 0.71
3 ne
nD me nD me 2
· µ ¶¸
2 4
+ νiz,D − E iz,D − Te + v ke v ke − v kD − νe-D v ke (v kD − v ke )
ne 3 mD 3 n mD 3
¶¸ e
nD me nD me 2
· µ
2 4
+ 2 νiz,D2 − E iz,D2 − Te + v ke v ke − v kD2 − 2 νe-D2 v ke (v kD2 − v ke ) (3.25)
ne 3 mD 3 ne mD 3
nD 2 me
· ¸
2
+ 2 νdiss,D2 − E diss,D2 +
¡ ¢
v ke v ke − v kD2
ne 3 3 mD
n D2 me
· µ ¶¸
2 4
+ νdiss-iz,D2 − E diss-iz,D2 − Te + v ke v ke − v kD2
ne 3 mD 3
n D+2 ·
2 2 me ³ ´¸
+ ν + − E + + v ke v ke − v kD+2
n e diss,D2 3 diss,D2 3 m D
n D+2 ·
2 me
µ
4
¶¸
+ ν + − E + − Te + v ke v ke − v kD+2 ,
n e diss-iz,D2 3 diss-iz,D2 mD 3
55
∂TD+ ρ −1
∗ 4 T D+
· C (p e + p D+2 ) ¸
=− [φ, TD ] − v kD ∇k TD +
+ + + −C (φ) +
∂t B 3 B n D+
2TD+ h ¡ ¢ ³ ´ i
− n e ∇ · v ke b − n D+2 ∇ · v kD+2 b + v ke ∇k n e − v kD+2 ∇k n D+2 − v kD+ ∇k n D+
3n D+
10 TD+ 2 2.32 m e
r
¡ ¢
− C (TD+ ) + p ∇ · n e TD+ ∇k TD+ b
3 B 3n D+ 2ν m D
2
+ χ⊥D+ ∇⊥ TD+ + ∇k χkD+ ∇k TD+ + S TD+
¡ ¢
1
½ ³ ´· 1¡ ¢2
¸ (3.26)
+ n D νiz,D + νcx,D + νcx,D-D+2 TD − TD+ + v kD − v kD+
n D+ 3
· ¸
1¡ ¢2
+ n D2 νdiss-iz,D2 TD+ ,diss-iz(D2 ) − TD+ + v kD2 − v kD+
3
· ´2 ¸
1³
+ 2n D+2 νdiss-iz,D+2 TD+ ,diss-iz(D+2 ) − TD+ + v kD+2 − v kD+
3
· ´2 ¸¾
1³
+n D+2 νdiss,D+2 TD+ ,diss(D+2 ) − TD+ + v kD+2 − v kD+
3
and
∂TD+2
" #
ρ −1
∗ 4 TD+2 C (p D+2 ) 10 TD+2
=− [φ, TD+2 ] − v kD+2 ∇k TD+2 − C (φ) + − C (TD+2 )
∂t B 3 B n D+2 3 B
2TD+2 ³ ´ 2 0.92 m e
r
¡ ¢
− ∇ · v kD+2 b + p ∇ · n e TD+ ∇k TD+ b
3 3n D2 2ν m D
+
(3.27)
³ ´
+ χ⊥D+2 ∇2⊥ TD+2 + ∇k χkD+2 ∇k TD+2 + S TD+
2
n D2
· ¸
2
+ (νcx,D2 + νiz,D2 + νcx,D2 −D+ ) TD+2 − TD+2 + (v kD2 − v kD+2 )2 .
n D+2 3
In Eqs. (3.19-3.27) we introduce [A, B ] = b · (∇A × ∇B ), C (A) = (B /2) [∇ × (b/B )] · ∇A and the
plasma vorticity Ω = ΩD+ + 2ΩD+2 , with the D+ contribution being given by ΩD+ = ∇ · ωD+ =
∇ · n D+ /B 2 ∇⊥ φ + 1/B 2 ∇⊥ p D+ and an analogous D+ 2 contribution, ΩD2 . The system is
£¡ ¢ ¡ ¢ ¤
+
closed by the generalized Poisson equation, which is obtained by inverting the definition of
the plasma vorticity, Ω, yielding
·n + 2n D+2 ¸ · ´¸
D+ 1 ³
∇⊥ · ∇⊥ φ = Ω − ∇⊥ · ∇⊥ p D+ + 2p D+2 . (3.28)
B2 B2
We highlight that the electron gyroviscous term in Eq. (3.22) is defined by analogy with the ion
£ ¤
gyroviscous terms, G e = −η 0e 2∇k v ke +C (φ)/B −C (p e )/ (n e B ) . When writing Eq. (2.70), we
avoid the Boussinesq approximation and take into account all components of the velocity of
56
the ion species D+ and D+ 2 , including the higher order polarization and friction contributions.
On the other hand, when expressing the advective derivative for the ion species, d D+ /d t and
d D+2 /d t , we keep only the leading order components of the perpendicular velocity, v ⊥D+ 0 and
v ⊥D+2 0 , neglecting vpol and vfric . We consistently neglect the friction and polarization drifts in
the continuity equation for D+ 2 . We remark that this is still a reasonable approach within the
conditions considered in the present work, given that the density of D+ 2 ions is smaller than
the density of the main ion species by several orders of magnitude. We also note that the terms
of higher order in 1/ΩcD+2 d/dt in the perpendicular velocity of D+ 2 ions are neglected when
∇ · vD+2 is written in the temperature equations, Eqs. (3.26) and (3.27), this assumption being
required to avoid explicit time derivatives featuring in the polarization drift velocity, vpol,D+2 .
However, all terms are considered in the divergence of the perpendicular velocity of D+ ions
in Eq. (3.26), as we use ∇ · j = 0 to write ∇ · vD+ in terms of ∇ · ve and ∇ · vD+2 . Finally, when
taking the divergence of these terms, we consider ∇ · vD ¿ ∇ · vD+ to neglect the contribution
of the velocity of D atoms, which is valid since ρ s,D+ ¿ λmfp,D (with ρ s,D+ = c s,D+ /Ωc,D+ the
sound Larmor radius of D+ ions, c s,D+ = Te /m D+ the D+ ions sound speed and λmfp,D the
p
mean free path of a D atoms). This relation is also generalized to the other neutral and ion
species, namely D2 molecules and D+ 2 ions, which enables us to neglect the contribution of
the divergence of the velocities of neutral particles when compared to the divergence of ion
velocities. Similarly to Ch. 2 for the single-component plasma, S ne , S nD+ , S Te , S TD+ and S TD+ in
2
Eqs. (3.19-3.27) represent the density and temperature source terms for the different plasma
species.
We remark that dimensionless units are used in Eqs. (3.19-3.27) and in the rest of this chapter,
similarly to Ch. 2. The densities, n e , n D+ and n D+2 , are normalized to the reference value n 0 ,
while temperatures, Te , TD+ and TD+2 , are normalized to the respective reference values, Te0 ,
TD+ 0 and TD+2 0 = TD+ 0 , which are also related through the dimensionless quantity τ = TD+ 0 /Te0 .
In turn, lengths along the magnetic field are normalized to the tokamak major radius, R 0 ,
lengths in a direction perpendicular to the magnetic field are normalized to the ion sound
Larmor radius, ρ s0 = c s0 /ΩcD+ 0 , where c s0 = Te0 /m D+ is the normalized D+ ion sound speed
and ΩcD+ 0 = eB 0 /m D+ is the D+ ion cyclotron frequency at the magnetic axis, and time is
normalized to R 0 /c s0 . All the other normalizations follow, namely the parallel velocities,
v ke , v kD+ and v kD+2 , normalized to c s0 , the plasma vorticity Ω, normalized to n 0 Te0 /(ρ 2s0 B 02 ),
the perpendicular diffusion coefficients D ⊥ and conductivities χ⊥ , normalized to c s0 ρ 2s0 /R 0 ,
and finally the parallel diffusion coefficients D k and conductivities χk , normalized to c s0 R 0 .
Normalized quantities are used in the rest of the chapter, except when explicitly mentioned. We
remark that we have defined the parameter ρ ? = ρ s0 /R 0 as the ratio between the D+ ion sound
Larmor radius and the tokamak major radius R 0 . We also note that ν is the dimensionless
resistivity given by ν = (e 2 n e0 R 0 )/(m D c s0 σk ), with the parallel conductivity defined in terms of
the electron characteristic time τe as σk = e 2 n e τe /(0.51m e ).
We conclude with a couple of remarks on Eqs. (3.19-3.27). At first, we remark that the parallel
conductivity featuring in the temperature equations for electrons is expressed in the form
χk,e = χk0,e Te5/2 , where the Spitzer temperature dependence is retained, while we neglect
57
the weaker space and time dependence of the 2/(3n e ) factor, in a similarly approach to the
one followed in the single-component plasma model of GBS [33, 34] discussed in Ch. 2.
An identical procedure is followed when considering χk,D+ and χk,D+2 . This approximation
is expected not to affect significantly the results of simulations in the tokamak boundary
conditions considered in the present work, which have parameters between the sheath-limited
and conduction-limited regimes, where the contributions of conductivity are expected to
be small. We also mention that, because the D+ 2 density typically drops to very low values,
numerical issues may appear in the equations for v kD+2 and TD+2 , arising from terms featuring
a 1/n D+2 dependence. In order to develop a more robust numerical approach, we evolve the
2 ion species, ΓkD2 = n D2 v kD2 and p D2 = n D2 TD2 , instead of
parallel flux and pressure of the D+ + + + + + +
v kD+2 and TD+2 . The equations modelling the time evolution of ΓkD+2 and p D+2 are
∂ΓkD+2 ∂n D+2 ∂v kD+2

= v kD+2 + n D+2 , (3.29)
∂t ∂t ∂t
and
∂p D+2 ∂n D+2 ∂TD+2

= TD+2 + n D+2 , (3.30)
∂t ∂t ∂t
with ∂t n D+2 , ∂t v kD+2 and ∂t TD+2 being given, respectively, by Eqs. (3.20), (3.24) and (3.27). We
remark that, when presenting the results of simulations, we focus on the parallel flux, ΓkD+2 ,
and pressure, p D+2 .
3.4 Boundary conditions

The boundary conditions implemented in the single-component GBS model presented in Ch.
2 are extended in the present chapter to include the molecular ion species D+ 2 . Similarly to the
single-ion species model, the simulation domain boundary includes the limiter plates, the
outer wall and the interface with the core, which is excluded since the low plasma collisionality
does not enable a description based on a fluid model.
We start by considering the boundary conditions at the limiter plates, where the plasma
particles end as a result of the flow parallel to the magnetic field. Hence, those are the
boundary conditions that more significantly influence the simulation dynamics. Similarly to
Ch. 2, the boundary conditions are imposed at the interface between the collisional pre-sheath
(CP) and the magnetic pre-sheath (MP), which are derived from the Bohm-Chodura boundary
conditions, following the approach described in Ref. [74] in the cold ion limit and generalized
in Ref. [75] to account for finite ion temperature. In this chapter, we extend this procedure
for a multi-ion species plasma. For this purpose, we use the (y, x, z) coordinates, being z the
direction of the magnetic field, x the direction perpendicular to the magnetic field and parallel
to the limiter surface, and y the direction perpendicular to both x and z (all spatial coordinates
58
3.4. Boundary conditions
are normalized to ρ s0 while the other quantities follow the same normalizations used for Eqs.
(3.19-3.27)). We also introduce the coordinate perpendicular to the limiter plate, expressed as
s = ycosα + zsinα, where α denotes the angle between the magnetic field line and the plane
of the limiter.
We remark that, to describe the steady-state dynamics of the multi-species plasma in the
CP, we make use of the continuity equation for the D+ and D+
2 species (with quasi-neutrality
providing the electron density) and the parallel momentum equations for e− , D+ and D+ 2.
These can be written, in a steady state, as
∇ · (n D+ vD+ ) = S p,D+ , (3.31)
∇ · (n D+2 vD+2 ) = S p,D+2 , (3.32)
n e (ve · ∇)ve = −n e E − n e vD+ × B − ∇p e + Sm,e , (3.33)
¡ ¢
n D+ vD+ · ∇vD+ = n D+ E + n D+ vD+ × B − ∇p D+ + Sm,D+ (3.34)
and
³ ´
n D+2 vD+2 · ∇vD+2 = n D+2 E + n D+2 vD+2 × B − ∇p D+2 + Sm,D+2 , (3.35)
with S p,D+ and S p,D+2 the particle sources for D+ and D+

2 , and Sm,e , Sm,D and Sm,D2 the
+ +
− + +
momentum sources for e , D and D2 .
From Eqs. (3.31-3.35) and following the approach described in Ref. [75], we obtain a system
of five equations for ∂s n D+ , ∂s n D+2 , ∂s v kD+ and ∂s v kD+2 , ∂s φ for the interface between the CP
and the MP border, considering m e /m D+ ¿ 1 and assuming the isothermal approximation
for both ions and electrons. Moreover, we remark that, at the MP entrance, gradients along
the x direction are assumed weaker than gradients along s by a factor ² = ρ s0 /L n ' ρ s0 /L Te '
ρ s0 /L φ ¿ 1, with L n , L Te and L φ respectively the gradient scale lengths of n e , Te and φ along
the x direction. At the same time, we neglect finite Larmor radius (FLR) effects and, to express
the y and x velocity components of each ion species, D+ and D+ 2 , we include only the leading
order terms in (1/ΩcD )d/dt (see Eqs. (3.13) and (3.16)). This leads to
+
v y,D+ = v y,E×B + v y,dD+ , (3.36)
59
v x,D+ = v x,E×B + v x,dD+ , (3.37)
v y,D+2 = v y,E×B + v y,dD+2 (3.38)
and
v x,D+2 = v x,E×B + v x,dD+2 , (3.39)
where v y,E×B and v x,E×B are respectively the y and x components of the E × B drift velocity,
v y,dD+ and v x,dD+ are the y and x components of the D+ diamagnetic drift velocity and v y,dD+2
and v x,dD+2 are the y and x components of the D+ 2 diamagnetic velocity. The velocity of the
D+ ions along the s direction is written as v s,D+ = v kD+ sinα + v y,D+ cosα. We also define the
velocity of the D+ ions along the s direction excluding the diamagnetic contribution, that is
0 + 0
v s,D + = v s,D+ − v y,dD+ cosα, and similarly for the D2 ions, yielding v
s,D+
= v s,D+2 − v y,dD+2 cosα.
2
The system in Eqs. (3.31-3.35) is then written as
v s,D+ ∂s n D+ + n D+ sinα∂s v kD+ − ∂x n D+ cosα∂s φ = S p,D+ , (3.40)
v s,D+2 ∂s n D+2 + n D+2 sinα∂s v kD+2 − ∂x n D+2 cosα∂s φ = S p,D+2 , (3.41)
n D+ v s,D+ ∂s v kD+ + n D+ (sinα − ∂x v kD+ cosα)∂s φ + TD+ sinα∂s n D+ = S km,D+ , (3.42)
n D+2 v s,D+2 ∂s v kD+2 + n D+2 (sinα − ∂x v kD+2 cosα)∂s φ + TD+2 sinα∂s n D+2 = S km,D+2 (3.43)
and
sinαTe ∂s n e − sinαn e ∂s φ = S km,e , (3.44)
where S km,D+ = Sm,D+ · b, S km,D+2 = Sm,D+2 · b and S km,e = Sm,e · b. We make use of the quasi-
neutrality condition, n e = n D+ + n D+2 , to obtain a system of five linear equations, which is
expressed in matrix form as Mx = S, with
60
3.4. Boundary conditions
0
 
v s,D + n D+ sinα 0 0 −cosα∂x n D+
TD+ sinα n D+ v 0 + 0 0 n D+ (sinα − ∂x v kD+ cosα)
 
 s,D 
0
M= 0
 0 v s,D + n D+2 sinα −cosα∂x n D+2 ,

(3.45)
 2 

 0 0 TD+2 sinα 0
n D+2 v s,D + n D+2 (sinα − ∂x v kD+2 cosα)

2
sinαTe 0 sinαTe 0 −(n D+ + n D+2 )sinα
∂s n D+
 
 ∂s n D+2 
 
∂s v kD+ 
 
x=  (3.46)
∂s v kD+2 
 
∂s φ
and
 
S p,D+
 S p,D+2 
 
 
S=
S km,D+  .
 (3.47)
S km,D+2 
 
S km,e
Following the same approach as [74, 75], we observe that, although the source terms are
important in the CP, they become small with respect to the gradient terms at the MP entrance.
This enables the assumption that |Σj Mij Xj | À |Si |. Thus, the linear system Mx = S reduces
0
to Mx = 0 at the MP entrance. We then solve det(M) = 0 with respect to v sD + to obtain the
non-trivial solution valid at the MP entrance. For this purpose, we follow Ref. [94] to relate the
parallel velocity of the D+ +
2 ion species, v kD2 , to the parallel velocity of the D ions, v kD ,
+ +
s
m D+ v kD+
v kD+2 = v kD+ = p . (3.48)
m D+2 2
We also assume n D+2 /n e ¿ 1 (and therefore n D+ ' n e ) and keep only zero order terms in ², thus
neglecting all derivatives along the x direction. The condition det(M) = 0 then yields
0
p
v sD + =± Te F T sinα (3.49)
where the ± signs stand for the magnetic field lines entering/leaving the vessel and we have
0
defined F T = 1+τTD+ /Te . We also remark that v s,D + = v kD+ sinα, since we neglect v y,E×B cosα =
∂x φcosα, to obtain the boundary condition for v kD+ at the limiter,
61
p
v kD+ = ± Te F T . (3.50)
The expressions of the boundary conditions for the other plasma quantities then follow. As a
matter of fact, Eq. (3.42) can be inverted to express ∂s φ in terms of ∂s v kD + , which yields
p
+ ∂s v kD +
0
v sD Te
∂s φ = − = ∓ p ∂s v kD + . (3.51)
F T sinα FT
We make use of Eq. (3.44) to express ∂s n e in terms of ∂s φ, that is
ne ne
∂s n e = ∂s φ = ∓ p ∂s v kD + (3.52)
Te Te F T
and, applying n D+ = n e , we obtain
ne
∂s n D+ = n e /Te ∂s φ = ∓ p ∂s v kD + (3.53)
Te F T
Regarding the density of the D+

2 ions, we use Eq. (3.41), deriving the boundary condition given
by
p
∂s n D+2 = ∓n D+2 / Te F T ∂s v kD + . (3.54)
To derive the boundary conditions for Te , TD+ and TD+2 , we note that temperature gradients
along the direction perpendicular to the wall are small compared to the gradients of the other
physical quantities. In fact, Ref. [74, 75] shows that ∂s Te ∼ ∂s TD+ ' 0.1∂s φ. In the present
chapter, we also follow this prescription and assume ∂s Te = ∂s TD+ = ∂s TD+2 = 0.1∂s φ (we note
that our tests show that imposing ∂s Te = ∂s TD+ = ∂s TD+2 = 0 does not noticeably impact the
results of the simulation).
To obtain the boundary condition forhΩ at the MP entrance, we apply i its definition, Ω =
2 2 2 2
∇ · (n D /B )∇⊥ φ + (1/B )∇⊥ p D + ∇ · (n D2 /B )∇⊥ φ + (1/B )∇⊥ p D2 , and we write the sec-
£ ¤
+ + + +
ond order derivatives in the directions perpendicular to the magnetic field retaining only
derivatives along the y direction, making use of ∂2x ¿ ∂2y . Given that ∂y B = 0 at the limiter, the
1/B 2 factor is taken constant when computing the derivatives featuring in the definition of
Ω. We write the derivatives along the y direction in terms of derivatives along s and consider
TD+2 = TD+ (for simplicity), which finally yields
h i
Ω = −cosα ∂s (n e + n D+2 )∂s φ + TD+ ∂2s (n e + n D+2 ) + (n e + n D+2 )∂2s φ . (3.55)
62
3.5. The kinetic model for the neutral species and its formal solution
We now use Eqs. (3.52) and (3.54) to express ∂s n e and ∂s n D+ in terms of ∂s φ and start from Eq.
(3.51) to obtain the final expression of the boundary condition for Ω, that is
· p
Te
¸
1
Ω = −(n e + n D+2 )F T cos2 α ± p ∂2s v kD+ ∓ p (∂s v kD+ )2 . (3.56)
FT Te F T
Finally, we remark that the boundary condition for the electron parallel velocity is obtained
from the analysis of the electron kinetic distribution function at the MP entrance. As discussed
in Ref. [74], this yields
φ
· µ ¶¸
p
v ke = Te ±exp Λ − , (3.57)
Te
£p
where Λ = log
¤
(1/2π)(m i /m e ) ' 3.
On the other hand, at the vessel outer wall and the edge-core interface, ad hoc boundary
conditions are considered, similarly to the approach used in the single-ion species GBS
model [74, 75, 34] considered in Ch. 2. As a matter of fact, a set of first-principles boundary
conditions is yet to be derived for such boundaries. We reduce the effect of these ad hoc
boundary conditions on the results of the simulation by radially extending the simulation
domain towards the wall and the core, as in the model presented in Ch. 2. We impose
homogeneous Neumann boundary conditions to n e , n D+ , Te , TD+ , TD+2 , v ke , v kD+ and v kD+2 .
Given that the density of D+ 2 ions is expected to be very low at the edge-core interface (no D2
+
ions outflow from the core), we use Dirichlet boundary conditions at the edge-core interface
for n D+2 , setting it to a very small value (we choose n D+2 = exp(−5), and we notice that this value
has a small impact on the results), while homogenous Neumann boundary conditions are
considered at the outer wall. The boundary conditions considered for the other quantities
follow the same approach described in Ch. 2. More precisely, we also use Dirichlet boundary
conditions for the vorticity, setting Ω = 0 at both the wall and the core interface. We follow the
approach presented in Ref. [35] for the φ boundary conditions, considering φ = ΛTe at the
vessel wall and φ = φ0 at the core interface, where φ0 is a constant value chosen to prevent
large gradients of φ.
3.5 The kinetic model for the neutral species and its formal solu-
tion
We now extend the approach followed in Ref. [61] and improved in Ch. 2 for a single-neutral
species model. We consider one kinetic equation for each neutral species, namely D atoms
and D2 molecules, which compute their respective distribution functions, f D and f D2 . The
result is a set of coupled equations, yielding
63
∂ fD ∂ fD nD
µ ¶
+v· = −νiz,D f D − νcx,D f D − f D+ + νrec,D+ f D+
∂t ∂x n D+
n D2
µ ¶
+ νcx,D2 −D+ f D+ − νcx,D-D+2 f D + 2νdiss,D2 f D2 + νdiss-iz,D2 f D2 (3.58)
n D+
+ νdiss,D+2 f D+2 + 2νdiss-rec,D+2 f D+2 ,
and
Ã !
∂ f D2 ∂ f D2 n D2
+v· = −νiz,D2 f D2 − νcx,D2 f D2 − f D+2
∂t ∂x n D+2
Ã !
nD (3.59)
+ νrec,D+2 f D+2 − νcx,D2 −D+ f D2 + νcx,D-D+2 f D+
n D+2 2
− νdiss,D2 f D2 − νdiss-iz,D2 f D2 .
We then obtain the formal solution of Eqs. (3.58) and (3.59) by applying the method of
characteristics, assuming that the plasma quantities are known. This leads to
r b0 · S 0
, v, t 0 )
D (x
Z ¸
+ δ r 0 − r b0 f D (xb0 , v, t b0 )
¡ ¢
f D (x , v , t ) =
0 v
Z r0 0
(3.60)
00 J (x )
· ¸
1 00 00
× exp − νeffD (x , t )d r dr 0
v 0 J (x)
and
r b0 · S 0
, v, t 0 )
D2 (x
Z ¸
+ δ r 0 − r b0 f D2 (xb0 , v, t b0 )
¡ ¢
f D2 (x, v, t ) =
0 v
Z 0 (3.61)
1 r J (x 0 ) 0
· ¸
× exp − νeffD2 (x00 , t 00 )d r 00 dr .
v 0 J (x )
The solutions introduced in Eq. (3.60-3.61) describe the distribution functions of D and
D2 at position x, velocity v and time t as the result of the neutrals generated at a location
x0 = x −r 0 Ω, in the plasma volume or at the boundary, and at time t 0 = t −r 0 /v, where Ω = v/v
is the unit vector aligned with the neutral velocity and r 0 is the distance measured from x0
to x (the subscript "b" designates the intersection point between the domain boundary and
the characteristic starting at x with direction Ω). Since the neutrals are solved on the (R, Z )
coordinate system, with R the distance from the torus axis and Z the vertical coordinate
measured from the equatorial midplane, the integral includes the Jacobian corresponding
to this coordinate system, expressed as J (x) = R(x). We highlight that the volumetric source
64
associated with the collisional processes in Eq. (3.60) is written as
S D (x0 , v, t 0 ) = νcx,D (x0 , t 0 )n D (x0 , t 0 )Φ[vD+ ,TD+ ] (x0 , v, t 0 ) + νcx,D2 −D+ (x0 , t 0 )n D2 (x0 , t 0 )Φ[vD+ ,TD+ ] (x0 , v, t 0 )
+ νrec,D+ (x0 , t 0 )n D+ (x0 , v, t 0 )Φ[vD+ ,TD+ ] (x0 , v, t 0 ) + 2νdiss,D2 (x0 , t 0 )n D2 (x0 , t 0 )Φh i (x0 , v, t 0 )
vD2 ,TD,diss(D2 )
+ νdiss-iz,D2 (x0 , t 0 )n D2 (x0 , t 0 )Φh i (x0 , v, t 0 ) + ν

diss,D+ (x0 , t 0 )n D+2 (x0 , v, t 0 )Φh i (x0 , v, t 0 )
vD2 ,TD,diss-iz(D2 ) 2 v D+ ,T D,diss D +
2 ( )
2
+ 2νdiss-rec,D+2 (x0 , t 0 )n D+2 (x0 , v, t 0 )Φh i (x 0 , v , t 0 )

vD+ ,TD,diss-rec
2 ( )
D+
2
(3.62)
since D ions can be created in the plasma volume by D − D+ and D2 − D+ charge-exchange

interactions, recombination of D+ ions with electrons, dissociation of D2 molecules into two
D atoms, dissociative ionization of D2 into D and D+ , dissociation of D+ +
2 ions into D and D ,
+
and dissociative recombination of D2 into two D atoms.
Similarly, D2 molecules are generated in the plasma by charge-exchange collisions between D2

molecules and D+ + +
2 ions or D atoms and D2 ions, or by recombination of D2 ions with electrons.
Therefore, the volumetric source term in Eq. (3.61) is
S D2 (x0 , v, t 0 ) = νcx,D2 (x0 , t 0 )n D2 (x0 , t 0 )Φh i (x 0 , v , t 0 ) + ν

rec,D+ (x0 , t 0 )n D+2 (x0 , v, t 0 )Φh i (x 0 , v , t 0 )
vD+ ,TD+ 2 v D+ ,T D+
2 2 2 2
+ νcx,D-D+2 (x0 , t 0 )n D (x0 , t 0 )Φh i (x0 , v, t 0 ).

vD+ ,TD+
2 2
(3.63)
¤3/2
We note that Φ[vD+ ,TD+ ] (x0 , v, t 0 ) = m D+ /(2πTD+ ) exp −m D+ (v − vD+ )2 /(2TD+ ) is a Maxwellian
£ £ ¤
distribution function that describes the D+ ion population, centered at the ion velocity
vD+ (x0 , t 0 ), including only the leading order components, vD+ = v kD+ b + v⊥D+ 0 , and based on
the D+ temperature, TD+ (x0 , t 0 ). In addition, Φh i (x0 , v, t 0 ) is a Maxwellian distribution
vD+ ,TD+
2 2
describing the D+
2 ions and follows a similar definition. We remark that, when we evaluate the
average velocity of the Maxwellian distributions describing neutrals generated from D2 and
D+2 , we neglect vD2 and vD2 with respect to vD , i.e. we assume |vD2 | . |vD | and |vD2 | . |vD |.
+ + + + +
We also note that the temperature TD,diss(D2 ) is the average thermal energy of D atoms gen-
erated by dissociation of D2 molecules, presented in Table 3.2 and calculated in App. B. The
energy of the neutral D atoms generated by the other dissociative processes is computed using
a similar approach.
The effective frequencies for depletion of neutral particles are given by
νeff,D (x00 , t 00 ) = νiz,D (x00 , t 00 ) + νcx,D (x00 , t 00 ) + νcx,D-D+2 (x00 , t 00 ) (3.64)
65
and
νeff,D2 (x00 , t 00 ) = νiz,D2 (x00 , t 00 ) + νcx,D2 (x00 , t 00 ) + νcx,D2 −D+ (x00 , t 00 )

(3.65)
+ νdiss,D2 (x00 , t 00 ) + νdiss-iz,D2 (x00 , t 00 ),
since the volumetric sinks of D atoms are associated with ionization or charge-exchange with
D+ or D+ +
2 , while depletion of D2 is related to ionization, charge-exchange with D2 or D ,
+
dissociation or dissociative ionization.
There is also a contribution to the neutral distribution functions in Eqs. (3.60) and (3.61)
related to plasma recycling taking place at the boundary walls. Therefore, we now focus on the
neutral processes occurring there. Similarly to the single-ion species model discussed in Ch.
2, a fraction, αrefl (xb0 ), of the D+
2 ions that reach the boundary walls is reflected back into the
plasma, after recombination with electrons and formation of D2 neutrals. The remaining frac-
tion, 1 − αrefl (xb0 ), is absorbed and reemitted at wall temperature as D2 , also after recombining
with an electron. Similar considerations are valid when describing the D2 neutrals that reach
the boundary, i.e. the D2 molecules are assumed to be reflected or reemitted with the same
probability as D+ 2.
Turning now to the atomic species, since the wall temperature is low, a fraction, βassoc , of the
D+ and D particles absorbed at the walls associate and reenter the plasma as D2 molecules.
The remaining D+ ions and D neutrals reaching the boundaries are reflected or reemitted,
similarly to D+2 ions and D2 particles, with the same probability of reflection, αrefl (xb ). As a
0
0 0 0 0
result, the distribution functions at the vessel, f D (xb , v, t ) and f D2 (xb , v, t ), for v p = v · n̂ > 0
(with n̂ the unit vector normal to the boundary) are written as
f D (xb0 , v, t 0 ) = (1 − αrefl (xb0 ))Γreem,D (xb0

, t 0 )χin,D (xb 0
, v)
Γ 0 0
" #
out,D + (x , t )
b
(3.66)
+ αrefl (xb0 ) f out,D (xb 0
, v − 2vp , t 0 ) + Φh i (x 0 , v , t 0 )
b
vp vrefl(D+ ) ,TD+
and
f D2 (xb0 , v, t 0 ) = (1 − αrefl (xb0 ))Γreem,D2 (xb0

, t 0 )χin,D2 (xb
0
, v)
Γout,D+2 (xb , t )
0 0
" #
(3.67)
+ αrefl (xb0 ) f out,D2 (xb 0
, v − 2vp , t 0 ) + Φh i (x 0 , v , t 0 ) .
b
vp vrefl D+ ,TD+
( 2) 2
At first, we analyse the contributions of reflected particles in Eqs. (3.66-3.67). The reflected D
0
and D2 are described by the distribution functions f out,D (xb , v − 2vp , t 0 ) and f out,D2 (xb
0
,v −
0
2vp , t ), since v − 2vp is the velocity of the neutrals which are reflected when flowing towards
the wall, where vp = v p n̂ denotes the velocity along the direction normal to the wall surface.
On the other hand, we address the contributions from the D+ and D+ 2 ions reflected at the
66
walls by considering the projection of the flux of outflowing D+ and D+ 2 along the direction
normal to the boundary surface, which are given respectively by Γout,D+ (xb 0 0
) = −Γout,D+ (xb ) · n̂
and Γout,D2 (xb ) = −Γout,D2 (xb ) · n̂. These fluxes include the contributions of the plasma flow
+
0 +
0
parallel to the magnetic-field lines and the leading order perpendicular drifts, more precisely
the E × B and diamagnetic drifts, yielding
Γout,D+ (xb0 ) = n D+ v kD+ b + n D+ v⊥D+ 0 (3.68)
and
Γout,D+2 (xb0 ) = n D+2 v kD+2 b + n D+2 v⊥D+2 0 , (3.69)
We assume that the velocity distribution of the D neutrals generated by reflection of D+ ions
is described by a Maxwellian centered at the velocity, vrefl(D+ ) = vD+ − 2vp D+ , with vp D+ =
vD+ · n̂ n̂, and with temperature of the incoming D+ ions, TD+ , given by Φh i (x0 , v, t 0 ).
¡ ¢
vrefl(D+ ) ,TD+
+
Similarly, the D2 neutrals arising from reflection of D2 ions are assumed to follow ³ a Maxwellian
´
distribution, Φ h i 0 0
(x , v, t ), being vrefl(D+2 ) = vD+2 −2vp D+ , with vp D+ = vD+2 · n̂ n̂ and
v + ,T +
refl(D ) D 2 2
2
TD+2 the temperature of the incoming D+
2 ions.
We now focus on the contributions in Eqs. (3.66-3.67) that account for the reemission of
neutrals from the boundary, which are written in terms of
Γreem,D (xb0 ) = (1 − βassoc ) Γout,D (xb0 ) + Γout,D+ (xb0 )

£ ¤
(3.70)
and
βassoc £
Γreem,D2 (xb0 ) = Γout,D2 (xb0 ) + Γout,D+2 (xb0 ) + Γout,D (xb0 ) + Γout,D+ (xb0 ) .
¤
(3.71)
2
In addition to the projections of the ion fluxes to the boundary, Γout,D+2 and Γout,D+2 , Eqs. (3.70)
and (3.71) account for the projections along the direction normal to the boundary of the fluxes
of D atoms and D2 molecules outflowing to the limiter and walls, Γout,D and Γout,D2 . These are
defined based on the neutral fluxes directed towards the boundary (with v p < 0) as
Z
Γout,D (xb0 ) = − vp · n f D (xb0 , v)d v
¡ ¢
(3.72)
v p <0
and
67
Z
Γout,D2 (xb0 ) = − vp · n f D2 (xb0 , v)d v.
¡ ¢
(3.73)
v p <0
We assume that the velocity distribution of the reemitted particles follows the Knudsen cosine
law for a given wall temperature, Tw , which for the D atoms yields
2
3 mD mD v 2
µ ¶
χin,D (xb0 , v) = cos(θ) exp − , (3.74)
4π Tw2 2Tw
while the expression for D2 molecules is similarly given by
2
3 m D2 m D2 v 2
µ ¶
χin,D2 (xb0 , v) = cos(θ) exp − . (3.75)
4π Tw2 2Tw
We then follow an approach identical to the one described in Ref. [61] to obtain a set of time-
independent two-dimensional integral equations for n D and n D2 , enabling the numerical
implementation of the formal solution in Eqs. (3.60) and (3.61). First, we take advantage of
the fact that the typical neutral time of flight is shorter than the characteristic turbulence
timescales, τn ¿ τturb , a condition which is denote in Ref. [61] as the neutral adiabatic regime.
This allows for the approximation t 0 = t in Eqs. (3.60-3.67) or, equivalently, ∂t f D = 0 and
∂t f D2 = 0 in Eqs. (3.58-3.59). We also note that the neutral mean free path is typically smaller
than the characteristic elongation of turbulent structures along the magnetic field direction,
λmfp,n k k ¿ 1. Therefore, our description of neutral motion is reduced to the analysis of a set
of independent two-dimensional planes perpendicular to the magnetic field, which coincide
approximately with the poloidal planes. Finally, integrating Eqs. (3.60-3.61) over the velocity
space, we obtain a system of two coupled equations for the densities of D atoms and D2
molecules, written as
0 Z 0
1 ∞ 1 r⊥
Z ∞
S D (x⊥ , v)
Z Z ½ · ¸¾
0 00 00
n D (x⊥ ) = d A 0 d v⊥ v⊥ d vk exp − νeff,D (x⊥ )d r ⊥
D r⊥ 0 0 v⊥ v⊥ 0
0Z ∞ Z 0 (3.76)
1 r⊥
Z ∞ ½ · ¸¾
0 cosθ
Z
0 00 00
+ d ab 0 d v⊥ v⊥ d v k f D (x⊥b , v)exp − νeff,D (x⊥ )d r ⊥ ,
∂D r ⊥b 0 0 v⊥ 0
and
68
0 Z 0
1 ∞ 1 r⊥
Z ∞
S D2 (x⊥ , v)
Z Z ½ · ¸¾
0 00 00
n D2 (x⊥ ) = d A 0 d v⊥ v⊥ d vk exp − νeff,D2 (x⊥ )d r ⊥
D r⊥ 0 0 v⊥ v⊥ 0
0Z ∞ Z 0 (3.77)
1 r⊥
Z ∞ ½ · ¸¾
0 cosθ
Z
0 00 00
+ d ab 0 d v⊥ v⊥ d v k f D2 (x⊥b , v)exp − νeff,D2 (x⊥ )d r ⊥ .
∂D r ⊥b 0 0 v⊥ 0
We remark that the geometrical arguments presented in Ref. [61] are used when considering
the integral along the neutral path and the integral along the angle describing the perpendicu-
lar velocity, that is
Z r ⊥,b Z 2π Z
1
0 0
d r⊥ d ϑF (x⊥ , x⊥ )= d A0 0
0 F (x⊥ , x⊥ ), (3.78)
0 0 D r⊥
where d A 0 is the area element in the two-dimensional poloidal plane and F (x⊥ , x⊥
0
) is a generic
function. We also use
r ⊥,b 2π cosθ 0
Z Z Z
0 0 0 0
d r⊥ d ϑδ(r ⊥ − r ⊥b )F (x⊥ , x⊥ )= d a b0 0
0
F (x⊥ , x⊥b ), (3.79)
0 0 ∂D r ⊥b
with d a b0 designating a line element along the boundary of D, which we write as ∂D, and
θ 0 = arccos(Ω⊥ · n̂).
We also express the volumetric source terms featuring in Eqs. (3.62) and (3.63), S D (x0 , v) and
S D2 (x0 , v), in terms of n D and n D2 , and the distribution functions of the neutral species at the
boundary appearing in Eqs. (3.66) and (3.63), f D and f D2 , in terms of Γout,D+ , Γout,D+2 , Γout,D
and Γout,D2 . For the density of D2 molecules, n D2 , this is expressed as
Z
0 0 0 D ,D+
n D2 (x⊥ ) = n D2 (x⊥ )νcx,D2 (x⊥ 2
)K p→p 2
(x⊥ , x⊥ )d A 0
D
Z
0 0 D2 0
+ (1 − αrefl (x⊥,b ))Γout,D2 (x⊥,b )K b→p (x⊥ , x⊥,b )d a b0
∂D
βassoc
Z
+ 0
(1 − αrefl (x⊥,b )) Γout,D (x⊥,b0 D2
)K b→p (x⊥ , x⊥,b 0
)d a b0 (3.80)
∂D 2
Z
0 0 D2 ,D+ 0
+ n D (x⊥ )νcx,D-D+2 (x⊥ )K p→p 2
(x⊥ , x⊥ )d A 0 + n D2 [rec(D+2 )] (x⊥ )
D
+ n D2 [out(D+2 )] (x⊥ ) + n D2 [out(D+ )] (x⊥ ),
while for the density of D atoms, n D , this is written as
69
Z
+
0 0 D,D 0
n D (x⊥ ) = n D (x⊥ )νcx,D (x⊥ )K p→p (x⊥ , x⊥ )d A 0
D
Z
0 0 D,D+ 0
+ n D2 (x⊥ )νcx,D2 −D+ (x⊥ )K p→p (x⊥ , x⊥ )d A 0
ZD
D,diss(D
+ 2n D2 (x⊥ 0 0
)νdiss,D2 (x⊥ )K p→p 2 ) (x⊥ , x⊥ 0
)d A 0
Z D (3.81)
0 0 D,diss-iz(D2 ) 0
+ n D2 (x⊥ )νdiss-iz,D2 (x⊥ )K p→p (x⊥ , x⊥ )d A 0
D
Z
0 0 D,reem 0
+ (1 − αrefl (x⊥,b ))(1 − βassoc )Γout,D (x⊥,b )K b→p (x⊥ , x⊥,b )d a b0
∂D
+ n D[rec(D+ )] (x⊥ ) + n D[out(D+ )] (x⊥ ) + n D[diss(D+2 )] (x⊥ ).
Replacing v p in Eqs. (3.73) and (3.72), the projections of the fluxes of D2 and D along the direc-
tion normal to the boundary are written respectively as Γout,D2 (x⊥,b 0 0
R
) = − cos(θ)<0 v ⊥ cosθ f D2 (x⊥,b , v⊥ )d v⊥
and Γout,D (x⊥,b ) = − cos(θ)<0 v ⊥ cosθ f D (x⊥,b , v⊥ )d v⊥ . Then, replacing f D2 (x⊥,b , v⊥ ) and f D (x⊥,b , v⊥ )
0 0 0 0
R
by their expressions as given in Eqs. (3.66) and (3.67), these fluxes are rewritten in terms of n D ,
n D2 , Γout,D+ , Γout,D+2 , Γout,D and Γout,D2 as
Z
0 0 D ,D+ 0
Γout,D2 (x⊥,b ) = n D2 (x⊥ )νcx,D2 (x⊥ )K p→b2 2
(x⊥ , x⊥ )d A 0
D
Z
0 0 D2 0
+ (1 − αrefl (x⊥,b ))Γout,D2 (x⊥,b )K b→b (x⊥ , x⊥,b )d a b0
∂D
βassoc
Z
+ 0
(1 − αrefl (x⊥,b )) 0
Γout,D (x⊥,b
D2
)K b→b 0
(x⊥ , x⊥,b )d a b0 (3.82)
∂D 2
Z
0 0 D2 ,D+ 0
+ n D (x⊥ )νcx,D-D+2 (x⊥ )K p→b 2
(x⊥ , x⊥ )d A 0
D
+ Γout,D2 [rec(D+2 )] (x⊥ ) + Γout,D2 [out(D+2 )] (x⊥ ) + Γout,D2 [out(D+ )] (x⊥ ),
and
Z
+
0 0 D,D 0
Γout,D (x⊥,b ) = n D (x⊥ )νcx,D (x⊥ )K p→b (x⊥ , x⊥ )d A 0
D
Z
0 0 D,D+ 0
+ n D2 (x⊥ )νcx,D2 −D+ (x⊥ )K p→b (x⊥ , x⊥ )d A 0
D
Z
D,diss(D
+ 2n D2 (x⊥ 0
)νdiss,D2 (x⊥0
)K p→b 2 ) (x⊥ , x⊥ 0
)d A 0
Z D (3.83)
0 0 D,diss-iz(D2 ) 0
+ n D2 (x⊥ )νdiss-iz,D2 (x⊥ )K p→b (x⊥ , x⊥ )d A 0
D
Z
0 0 D,reem 0
+ (1 − αrefl (x⊥,b ))(1 − βassoc )Γout,D (x⊥,b )K b→b (x⊥ , x⊥,b )d a b0
∂D
+ ΓD[rec(D+ )] (x⊥ ) + ΓD[out(D+ )] (x⊥ ) + ΓD[diss(D+2 )] (x⊥ ),
We remark that that the densities and fluxes of neutral particles in Eqs. (3.80-3.83) are mul-
70
tiplied by a factor 1 − αrefl (x⊥,b

0
) in order to consider only the contribution of particles that
are reemitted at the boundary, hence excluding reflection. We note that reflection of neutral
particles is included in the definition of the kernel functions featuring in Eqs. (3.76-3.77).
We now turn to the definition of the kernel functions appearing in Eqs. (3.80-3.83). These
0
are defined as integrals in the velocity space for a given pair of source (x⊥ ) and target (x⊥ )
2 2 0D ,D+
locations. Exemplifying for K p→p (x⊥ , x⊥ ), the kernel function measures the number of D2
neutrals arriving at a location x⊥ in the plasma volume (p) as a result of collisions involving
neutralization of D+ 0
2 ions at a location x⊥ inside the plasma volume (p). Its expression is given
by
D ,D+ 0 D ,D+ 0 D ,D+ 0

2
K p→p 2
(x⊥ , x⊥ 2 2
) = K p→p,dir (x⊥ , x⊥ ) + αrefl K p→p,refl
2 2
(x⊥ , x⊥ ). (3.84)
0
which splits the contributions to n D2 arising from the direct path of length r ⊥,dir connecting x⊥
0 2 D ,D+
2 0
and x⊥ , K p→p,dir (x⊥ , x⊥ ), and the path referring to the trajectory of neutrals that are reflected
2 2 D ,D+ 0 2 2 D ,D+
2 2 D ,D+
at the boundary, K p→p,refl (x⊥ , x⊥ ). Both K p→p,dir and K p→p,refl follow the same expression,
Z 0
∞ 1 r ⊥,path
· ¸
1
Z
D2 ,D+ 0 i (x0 , v )exp 00 00
2
K p→p,path (x⊥ , x⊥ )= 0 Φ h
⊥ ⊥ − νeff,D2 (x⊥ )d r ⊥ d v ⊥ ,
0 r ⊥,path ⊥ v⊥ D+ ,TD+ v⊥ 0
2 2
(3.85)
0 0
where path = {dir, refl} and r ⊥,path denotes the distance between x⊥ and x⊥ along the path (for
0
the direct trajectory r ⊥,dir is the distance between the two points along a straight line, while
0 0
for the reflected trajectory r ⊥,refl is computed as the sum of the distance between x⊥ and the
boundary and the distance from the boundary to x⊥ ). We remark that Φ h i 0
(x⊥ , v⊥ ) is
⊥ v⊥ D+ ,TD+
2 2
2 Maxwellian distribution function, Φ

the integral along the parallel velocity of the D+ h i (x0 , v ) =
⊥ v⊥ D+ ,TD+ ⊥ ⊥
2 2
R∞ D ,D+
−∞ Φ
h i (x0 , v )d v . 2 2
We also note that K p→p,dir in Eq. (3.85) is valid if the points are
v⊥ D+ ,TD+ ⊥ k
2 2
optically connected, i.e. if the straight line connecting the two points does not cross the core
2 2 D ,D+
nor the limiter plates. Otherwise, if the points are not optically connected, K p→p,dir = 0 is
2 2 D ,D+
defined. As for K p→p,refl , we highlight that, following the same approach presented in Ref.
[61] and considered in Ch. 2, we neglect reflection at the outer walls, while ions and neutrals
can be reflected at the limiter plates. The other kernels appearing in Eqs. (3.76-3.77) follow
2 2 D ,D+
the same structure as K p→p , and they take into account possible direct and reflected paths
connecting the two points. The form of these kernels is detailed in App. D.
Turning now to the evaluation of the non-homogeneous terms appearing in Eqs. (3.80-3.83),
i.e. the terms that are not proportional to n D nor n D2 , we start by looking at the contribution
71
of the ions recycled at the wall. As a matter of fact, the contribution to the density of neutral D
atoms of reflection and reemission of D+ ions that reach the boundary and recombine with
electrons is written as
Z h
0 0
¢ D,reem 0
Γout,D+ (x⊥,b ) (1 − αrefl (x⊥,b )) 1 − βassoc K b→p
¡
n D[out,D+ ] (x⊥ ) = (x⊥ , x⊥,b )
∂D (3.86)
i
0 D,refl 0
+αrefl (x⊥,b )K b→p (x⊥ , x⊥,b ) d a b0 ,
where Γout,D+ is defined in Eq. (3.68). Similarly, the recombination of D+

2 ions with electrons at
the walls that are then either reflected or reemitted as D2 , and the recombination of D+ ions
with electrons at the walls and the following association into D2 molecules contribute to the
density of the D2 species. These contributions can be expressed as
Z h
0 0 D2 ,reem 0
n D2 [out,D+2 ] (x⊥ ) = Γout,D+2 (x⊥,b ) (1 − αrefl (x⊥,b ))K b→p (x⊥ , x⊥,b )
∂D (3.87)
i
0 D2 ,refl 0
+αrefl (x⊥,b )K b→p (x⊥ , x⊥,b ) d a b0 ,
and
βassoc D2 ,reem
Z · ¸
0 0
n D2 [out,D+ ] (x⊥ ) = Γout,D+ (x⊥,b ) (1 − αrefl (x⊥,b )) K b→p (x⊥ , x⊥,b ) d a b0 .
0
(3.88)
∂D 2
We also define the non-homogeneous contributions to the flux of neutrals at the boundary,
Γout,D and Γout,D2 , arising from the ions outflowing to the wall. Following a similar approach
to the one described for the contributions to n out,D and n out,D2 , these are expressed as
Z h
0 0 D2 ,reem 0
Γout,D2 [out,D+2 ] (x⊥,b ) = Γout,D+2 (x⊥,b ) (1 − αrefl (x⊥,b ))K b→b (x⊥,b , x⊥,b )
∂D (3.89)
i
0 D2 ,refl 0
+αrefl (x⊥,b )K b→b (x⊥,b , x⊥,b ) d a b0 ,
βassoc D2 ,reem
Z · ¸
0 0 0
Γout,D2 [out,D+ ] (x⊥,b ) = Γout,D+ (x⊥,b ) (1 − αrefl (x⊥,b )) K b→b (x⊥,b , x⊥,b ) d a b0 ,
∂D 2
(3.90)
and
72
Z h
0 0
¢ D,reem 0
Γout,D[out,D+ ] (x⊥,b ) = Γout,D+ (x⊥,b ) (1 − αrefl (x⊥,b )) 1 − βassoc K b→b
¡
(x⊥,b , x⊥,b )
∂D
i
0 D,refl 0
+αrefl (x⊥,b )K b→b (x⊥,b , x⊥,b ) d a b0 .
(3.91)
We then turn to the evaluation of the contributions to the neutral particles featuring in Eqs.
(3.80-3.83) generated by volumetric processes involving the ion species D+ and D+ 2 . The
+
contribution to the D2 density from D2 recombination processes is expressed as
Z
0 0 D ,D+ 0
n D2 [rec,D+2 ] (x⊥ ) = n D+2 (x⊥ )νrec,D+2 (x⊥ 2
)K p→p 2
(x⊥ , x⊥ )d A 0 , (3.92)
D
while the contribution to the flux of D2 to the boundary, also associated to D+

2 recombination
events, is written as
Z
0 0 D ,D+ 0
Γout,D2 [rec,D+2 ] (x⊥ ) = n D+2 (x⊥ )νrec,D+2 (x⊥ 2
)K p→b 2
(x⊥ , x⊥ )d A 0 . (3.93)
D
Similar contributions from volumetric recombination processes are considered for the D
neutral species. The contribution to the D density from D+ recombination is given by
Z
+
0 0 D,D 0
n D[rec,D+ ] (x⊥ ) = n D+ (x⊥ )νrec,D+ (x⊥ )K p→p (x⊥ , x⊥ )d A 0 , (3.94)
D
and a similar definition is considered for the flux of D,
Z
+
0 0 D,D 0
Γout,D[rec,D ] (x⊥ ) =
+ n D+ (x⊥ )νrec,D+ (x⊥ )K p→b (x⊥ , x⊥ )d A 0 . (3.95)
D
Finally, the contribution to n D from the dissociation of D+

2 ions featuring in Eq. (3.81) is
defined as
73
Z
0 0 D,diss(D+ ) 0
n D[diss(D+2 )] (x⊥ ) = n D+2 (x⊥ )νdiss,D+2 (x⊥ )K p→p 2 (x⊥ , x⊥ )d A 0
D
Z (3.96)
0 0 D,diss-rec(D+
2) 0
+ 2n D+2 (x⊥ )νdiss-rec,D+2 (x⊥ )K p→p (x⊥ , x⊥ )d A 0 .
D
2 ions gives rise to a contribution to Γout,D in Eq. (3.83)

and, similarly, the dissociation of D+
defined as
Z
0 0 D,diss(D+ ) 0
Γout,D[diss(D+2 )] (x⊥ ) = n D+2 (x⊥ )νdiss,D+2 (x⊥ )K p→b 2 (x⊥,b , x⊥ )d A 0
D
Z (3.97)
0 0 D,diss-rec(D+2) 0
+ 2n D+2 (x⊥ )νdiss-rec,D+2 (x⊥ )K p→b (x⊥,b , x⊥ )d A 0 .
D
The system of coupled kinetic equations for the neutral species is discretized on a regular
cartesian grid in the (R, Z ) plasma and written in matrix form, in order to obtain the corre-
sponding numerical solution by factorizing the matrix and solving the system. We describe
the details of the numerical implementation of the neutral model in App. E.
3.6 First simulation of a multi-component plasma with the GBS code

In this section, we present the results from the first turbulence simulations in the tokamak
boundary based on the multi-component plasma model implemented in the GBS code and
described in the present chapter, in Secs. 3.2-3.5. Analogously to the simulations reported in
Refs. [33, 61] and in Ch. 2 of the present thesis, we consider a tokamak with an infinitesimally
thin toroidal limiter at the HFS equatorial midplane, with major radius R 0 /ρ s0 = 500, and we
simulate a three-dimensional domain, considering an annular cross section that includes
both the edge and the device SOL. The radial length of the domain is S rad = 150ρ s0 and the
poloidal length at the core interface is S pol = 800ρ s0 . Given that we consider the limiter to
extend over 75ρ s0 , both the open and closed field-line regions have a radial width of 75ρ s0 ,
corresponding to half the domain size along the radial direction.
The simulation presented in this chapter considers the following parameters: q = 3.992,
n 0 = 2 × 1013 cm−3 , T0 = 20.0eV, τ = 1, Ωci = 5.0 × 107 s−1 , TW = 0.3eV, ν = 0.1, S Te = S Ti = 0.3,
η 0e = η 0D+ = 1.0, η 0Ω = 4.0, χk0,e = 0.5, χk0,D+ = 0.05, χk0,D+2 = 0.05, D kne = 0.5, D knD+ = 0.0,
2
D kv ke = 0.5, D kv kD+ = 0.0, D kv kD+ = 0.5, and D ⊥ne = 21.0, D ⊥nD+ = D ⊥Ω = D ⊥v ke = D ⊥v kD+ =
2 2
D ⊥v kD+ = D ⊥Te = D ⊥TD+ = D ⊥TD+ = 7.0. Regarding the probability of a reflection process
2 2
occurring at the limiter, we note that it has a strong dependence on the particle energy and the
properties of the wall material (see Ref. [12]). In this simulation, reflection of ions and neutrals
takes place at the limiter plates with a given probability αrefl,lim , which we take constant along
the limiter surface. We thus write the fraction of particles being reflected at the boundary as
74
3.6. First simulation of a multi-component plasma with the GBS code
(
0 αrefl,lim 6= 0 if x⊥,b
0
is located at limiter walls
αrefl (x⊥,b )= 0 (3.98)
0 if x⊥,b is located at the outer and inner boundary.
Since we choose to consider metallic boundaries, we assume a value identical to the one
adopted in Ref. [61] and in Ch. 2 of this thesis, αrefl,lim = 0.8, with the remaining fraction of the
incoming particles being absorbed and reemitted when the wall is saturated. We also assume
that most of the absorbed D atoms associate into D2 molecules at the tokamak boundary (see
Refs. [79, 95]), thus being reemitted as D2 molecules. We therefore take βassoc = 0.95.
Regarding the numerical parameters of the simulations, the plasma grid resolution is n x,p ×
n y,p × n z,p = 255 × 511 × 64, while the resolution of the neutral grid is n x,n × n y,n × n z,n = 24 ×
138 × 64. The time step is 3.75 × 10−5 R 0 /c s , with the neutral quantities being evaluated every
∆t = 0.1R 0 /c s . Although convergence studies based on the multispecies model presented in
this chapter have not been developed, convergence on plasma and neutral grid refinement
has been studied within the context of single-component simulations. The conclusions
presented in Ref. [36], which we expect to remain valid in the multi-species model presented
in this chapter, confirm that our results are converged with respect to the frequency of neutral
calculation.
Figure 3.2: Typical poloidal snapshot of the electron density (n e ) and electron temperature
(Te ) taken from the turbulent quasi-steady state of the multi-component plasma simulation
described in the present chapter.
The simulation results presented here refer to the quasi-steady state regime, reached after a
transient, when the plasma and neutral profiles oscillate around constant equilibrium values.
We highlight that, similarly to the results from single-component GBS simulations, the plasma
behaviour is turbulent in the edge and especially in the SOL, as it can be observed in the
poloidal snapshots of electron density (n e ) and temperature (Te ) presented in Fig. 3.2. We
perform toroidal and time averages of the plasma quantities evolved by Eqs. (3.19-3.27) and
(3.28) over a time interval of ∆t ' 10R 0 /c s0 . Poloidal cross section plots of these quantities
are shown in Fig. 3.3. In Fig. 3.4, we present the poloidal cross section plots of the density of
the neutral atoms and molecules, n D and n D2 respectively, and the neutral-plasma collisional
interaction terms considered in our model. The results of these multi-species simulations
75
are compared with respect to a single-component plasma simulation considering the same
parameter values. We present in Fig. 3.5 the time and toroidal averages of the plasma and
neutral quantities from the single-species simulation which we find relevant to compare with
the results from the multi-species simulation.
Figure 3.3: Cross section plots of the electron density (n e ), D+ density (n D+ ), D+ 2 density (n D2 ),
+
+ +
electron parallel velocity (v ke ), D parallel velocity (v kD+ ), D2 parallel velocity (v kD+2 ), electron
temperature (Te ), D+ temperature (TD+ ), D+ 2 temperature (TD2 ) and electrostatic potential (φ),
+
toroidal and time-averaged over an interval of ∆t = 10.1R 0 /c s0 from the quasi-steady state of
the multi-component plasma simulation.
76
Figure 3.4: Cross section plots of the neutral species densities and source terms resulting from
the neutral-plasma interaction, toroidal and time-averaged over an interval of ∆t = 10.1R 0 /c s0
from the quasi-steady state of the multi-component plasma simulation described.
We start by focusing on some general considerations on the densities of the plasma and
neutral particles. The plots in Fig. 3.3 show that the density of the molecular ion species D+
2 is
between three and four orders of magnitude smaller than the density of the main ion species
D+ , which agrees with the assumption n D+2 /n D+ ¿ 1 considered in Eqs. (3.22-3.27) for the
derivation of the parallel friction and heat flux terms and in Eqs. (3.40-3.44) to obtain the
boundary conditions at the limiter plates. We note that the density of D+ 2 peaks just inside
the LCFS close to the limiter, since most of the D2 molecules cross the open-field line region
without interacting due to the low electron densities and temperatures in the SOL and are then
dissociated and/or ionized by the denser and warmer plasma inside the LCFS. As a matter
of fact, n D+2 matches the profile of the molecular ionization source n D2 νiz,D2 shown in Fig.
3.4, which also peaks in the edge near the limiter. On the other hand, Fig. 3.4 shows that n D
77
and n D2 are similar to n D+ near the limiter plates, while being about one order of magnitude
smaller than n D+ in the rest of the SOL and up to two orders of magnitude smaller inside the
LCFS. Moreover, looking at the relative importance of D and D2 , Fig. 3.4 shows that n D2 is
larger than n D by a factor between two and three in the SOL around the limiter, while n D is
larger than n D2 inside the LCFS at the HFS equatorial midplane, as a result of the higher values
of electron density and temperature, which lead to the dissociation of D2 molecules in this
region.
We then turn our attention to the asymmetry of the plasma density and flow. As a matter of fact,
a slight up-down asymmetry in the edge region is observed in the profiles of n e and n D+ , which
are noticeably larger below the equatorial midplane than above it. This can also be observed in
the poloidal snapshot of n e in Fig. 3.2. At the same time, the profiles of v ke and v kD+ reveal that
electrons and D+ ions flow in the counterclockwise direction in the edge region (see Fig. 3.3).
Since most neutrals are ionized inside the LCFS and in the vicinity of the limiter, the plasma
particles resulting from the ionization processes flow downwards, which leads to a slightly
larger density of e− and D+ below the equatorial midplane of the device. We highlight that the
n D+ and v kD+ profiles obtained from simulations based on the single-component model of
GBS are slightly different, as illustrated in Fig. 3.5. In the single-component simulation, an
up-down asymmetry is also observed in the n D+ profile. However, this asymmetry is related to
the ionization source n D νiz being larger in the edge region below the limiter than above it, due
to larger recycling rates at the lower limiter plate. In fact, in contrast to the multispecies case,
the v kD+ is characterized by a counterclockwise parallel flow of D+ ions in the edge below the
midplane, while the parallel flow is directed clockwise above it.
Another important observation arising from the multi-component plasma simulation is the
higher recycling rates in the region above the limiter than below, which is highlighted by the
up-down asymmetry observed in the profiles of the densities of n D and n D2 in Fig. 3.4. This
suggests that the parallel flux of plasma in the SOL region towards the limiter plates is larger
above the equatorial midplane, which agrees with the up-down asymmetry shown in Fig.
3.3 for the poloidal profiles of n D+ and v kD+ . A deep investigation of the reason behind this
behaviour calls for a careful analysis of the turbulent dynamics in the SOL, taking also into
account the other components of the flux beyond the parallel flow. Nonetheless, the simulation
results suggest that this up-down asymmetry may be related to the counterclockwise flow of
plasma in the edge observed in the profile of v kD+ .
Moreover, n D+ is observed to be slightly larger in the HFS with respect to the LFS, which is
related to the existence of D+ sources in the HFS around the midplane. This result agrees
with the conclusions driven from the single-species simulation, where n D+ is also found to be
larger in the HFS as a consequence of the ionization source, n D νiz .
Focusing now on the temperature of the plasma species, we observe that the Te profile presents
a behaviour identical to the one verified in single-component plasma simulations. A clear
asymmetry between the HFS and the LFS is observed for TD+ , whose profile is also very similar
78
Figure 3.5: Cross section plots of plasma density n = n e = n D+ , ion parallel velocity v kD+ ,
ion temperature TD+ and ionization source term n D+ νiz , toroidal and time-averaged over an
interval of ∆t = 10.1R 0 /c s0 from a quasi-steady state single-component plasma simulation.
The grid sizes and simulation parameters are the same as the ones considered in the multi-
component simulations, except for the wall re-emission temperature, which is set to TW =
3.0eV, to mimick Franck-Condon dissociation processes, and D ⊥ne = 7.0.
to the results obtained from a single-component simulation presented in Fig. 3.5. As a matter
of fact, the temperature is considerably lower on the HFS compared to the LFS, which is
related to the generation of cold D+ ions inside the LCFS due to the ionization of D atoms,
dissociative processes and charge-exchange collisions. This becomes particularly important
in the region above the limiter, where larger recycling rates are observed. On the other hand,
the profile of p D+2 peaks inside the LCFS at the HFS, where the majority of the D+ 2 ions are
generated by ionization of D2 molecules generated at the limiter. The up-down asymmetry
of the p D+2 profile around the limiter plates is also related to the asymmetry of the recycling
rates. As an aside note, we highlight that, since it is strongly related to the Te profile [96], the
electrostatic potential profile obtained from the multi-component simulations is identical to
the one observed for the single-component plasma model.
Focusing on the neutral-plasma interaction terms presented in Fig. 3.4, we start by noticing
that ionization processes are in general considerably more important in the edge region at the
HFS. We also highlight that the ionization frequencies have similar profiles for both atoms and
molecules. However, n D2 νiz,D2 peaks in the vicinity of the LCFS, while n D νiz,D peaks further
79
inside the LCFS and shows a larger spread along the radial direction. In fact, D2 molecules are
generated in the SOL and are dissociated and/or ionized in the proximity of the LCFS, where
the plasma is warmer and denser. In contrast, although most D atoms are generated in the
SOL, they are also a product of dissociation of D2 molecules in the edge. This effect shifts the
maximum of n D νiz,D radially inwards and increases the radial spread of the ionization source,
when compared to single-component plasma simulations. Since n D is larger than n D2 in the
edge, the maximum of n D νiz,D is also almost two times larger than the maximum of n D2 νiz,D2 .
Regarding collisions between electrons and neutral particles, we note that the reactions
involving D2 are more important in the SOL, taking place mostly in the area surrounding the
limiter plates, where a larger number of neutral molecules are generated. Reactions with D2
are less important in the edge, as most molecules are dissociated and/or ionized due to the
higher densities and temperatures. On the other hand, elastic collisions between electrons
and D atoms peak inside the LCFS, which is because the cross sections of these reactions are
larger in the edge region due to the higher plasma density and temperature and because of the
presence of D atoms generated by molecular dissociation. We remark that elastic collisions
and charge-exchange reactions are more frequent above the limiter, which follows from the
previous discussion on the strong up-down asymmetry. As for charge-exchange reactions,
we observe that their spatial distribution is similar to the one of collision between electrons
and neutral particles. The charge-exchange reactions between the molecular species (D2 − D+ 2
collisions) are less important than the interactions between mono-atomic species (D − D+
collisions) by three to four orders of magnitude, which results from the small n D+2 to n D+ ratio.
In addition, the terms related to charge-exchange interactions between D2 molecules and
D+ ions (D2 − D+ collisions) are two orders of magnitude smaller than the ones between the
atomic species (D − D+ collisions), in the region of the domain where these interactions are
important. Finally, charge-exchange interactions between D+ 2 ions and D atoms is three orders
of magnitude less important than D − D collisions, which is a consequence of n D νcx,D−D+2
+
being proportional to n D+2 .
We finally analyse the dissociative reactions, which constitute a sink of molecular species D2
and D+ + +
2 and sources of D atoms and D ions. Dissociation of D2 and D2 , described by the
terms n D2 νdi,D2 and n D+2 νdi,D+2 respectively, which do not involve ionization nor recombination
processes, are the dominant dissociation processes, as their frequencies are similar to the
reaction rates of the ionization processes. We highlight that dissociation of D2 molecules peaks
just above the limiter plate (where most D2 molecules are generated) and in the edge region,
in the proximity of the LCFS, and then it becomes considerably smaller in the core, given
that n D2 drops rapidly across the edge. In contrast, dissociation of D+ 2 ions is very small in
the SOL, where the density of D+ 2 is negligible (at the typical electron temperature of the SOL,
ionization of D2 has a very small cross section), and is important only inside the LCFS, where
D+2 ions are generated. The n D2 νdi,D2 profile thus closely follows the n D2 profile, exhibiting
+ + +
a larger radial spread with respect to the dissociation of D2 . As for dissociative ionization of
D2 and D+ 2 , n D2 νdi−iz,D2 and n D2 νdi−iz,D2 respectively, we observe that the rates are smaller by
+ +
one to two orders of magnitude compared to the dissociation of D2 and D+ 2 and peak in the
80
edge region even further inside. This is related to the fact that the energy required to trigger
dissociative ionization processes is significantly larger than the one required to dissociate
the particles without ionizing them, as shown in Table 3.2. Therefore, these processes are
only important in the edge region, where densities and temperatures are sufficiently high
to make these cross sections significant. This is in particular the case of n D+2 νdi−iz,D+2 , since
this term is also proportional to the density of D+ 2 ions, which is relevant only inside the
LCFS. However, we remark that these reactions become considerably less important towards
the core, as very few D2 and D+ 2 cross the edge region without being dissociated. Regarding
dissociative-recombination of D+ 2 particles, n D2 νdi−rec,D2 , its amplitude is also smaller than
+ +
that of dissociation by one to two orders of magnitude and follows very closely the n D+2 profile,
as there is no energy threshold to trigger the reaction, contrarily to dissociative ionization
processes.
The results described above lead to a global picture of the main processes determining the
dynamics of D2 molecules in the tokamak boundary. Although some D2 are dissociated in the
open field-line region, most of them cross the LCFS and are dissociated into D atoms within a
short distance, as they interact with the plasma of the edge, of higher density and temperature.
The remaining D2 molecules penetrate further inwards and are ionized by the increasingly
warmer and denser plasma, giving rise to D+ 2 ions, which in turn have very short lifetimes,
+
since they are rapidly dissociated into D ions and D atoms.
We remark that, due to the low plasma density of the SOL, in the multi-component as well
as in the single-component simulations, a considerable amount of D atoms generated in the
SOL (emitted at the limiter or created by dissociation of D2 molecules) penetrate in the edge,
where ionization takes place due to the higher plasma density and temperature. However,
the presence of the D sources inside the LCFS in the multi-component simulations shifts the
spatial distribution of ionization, n D νiz , towards the core with respect to the results from single-
component simulations, as shown in Fig. 3.5. Therefore, these results highlight the importance
of considering molecular dynamics when describing the neutral-plasma interaction in the
boundary of a tokamak. A parameter scan on the electron density and temperature will allow
broader physical conclusions to be drawn from the multi-component model of the neutral-
plasma interaction presented here. In particular, increasing plasma density and temperature
is expected to reduce the penetration of both D atoms and D2 molecules, thus shifting the
dissociation and ionization sources/sinks radially outwards.
To conclude, we also present radial plots of the particle densities (Fig. 3.6) and radial fluxes
(Fig. 3.7), which are obtained by evaluating the average of these quantities in time and in
the toroidal and poloidal directions. In Fig. 3.7, we separate the contributions of the E × B,
diamagnetic and polarization drifts to the flux of the ion species, D+ and D+ 2 . On the other
hand, Fig. 3.8 presents the results from the single-component simulations. The n D+ profile in
Fig. 3.6 is similar to the one in Fig. 3.8 obtained from the single-component plasma simulation,
with a large density gradient region near the LCFS and a density shoulder forming in the far
SOL. In turn, the density of D+ 2 is small in the whole domain and peaks in the edge, across the
81
Figure 3.6: Radial profiles of the ions and neutrals species densities, averaged over the toroidal
and poloidal directions, evaluated over an interval of ∆t = 10.1R 0 /c s0 from a quasi-steady state
simulation.
LCFS, where most D2 molecules are ionized, and decreases rapidly towards the core, due to the
small penetration of D2 molecules in the warmer and denser plasma in that region. We remark
that the D+2 ions observed in the open-field line region result mostly from charge-exchange
interactions between D2 and D+ (see Fig. 3.4) and the ionization of D2 molecules reemitted
from the limiter and vessel wall.
Figure 3.7: Radial profiles of the radial flux for D+ ions (top), D+ 2 ions (middle) and neutral
species D and D2 (bottom), averaged over the toroidal and poloidal directions, evaluated
over an interval of ∆t = 10.1R 0 /c s0 from the quasi-steady state multi-component plasma
simulation. The components of the D+ and D+ 2 radial flux are discriminated.
Regarding the neutral species, we note that n D peaks in the SOL, close to the LCFS, and
82
decreases radially outwards, while in the single-component plasma simulation n D increases

monotonically radially towards the wall. This is the consequence of the dissociation of D2
molecules into D atoms in the edge and near SOL, as well as the association of D atoms into
D2 molecules at the limiter plates and the vessel wall. On the other hand, we observe that n D2
decreases monotonically from the outer wall to the interface with the core, since D2 molecules
are generated in the open-field line region as the result of recycling processes and are lost due
to dissociation and ionization processes which take place mostly in the edge and near SOL.
The dissociation of D2 molecules also affects the radial flux of D, ΓD , presented in Fig. 3.7.
In contrast with the single-component plasma simulation presented in Fig. 3.8, ΓD points
radially inwards in the edge, reversing sign in the SOL, which is a consequence of the release
of D atoms due to the dissociation of D2 molecules, of particular importance close to the LCFS.
Moreover, the D atoms reaching the outer wall associate and are reemitted as D2 molecules,
thus contributing to the outward flux of D. The multi-component plasma simulation shows
that ΓD peaks in the edge, while for a single-component model ΓD is maximum at the LCFS.
This is due to the D atoms that are generated in the edge region close to the LCFS in a multi-
component model, compensating their ionization. At the same time, we remark that the radial
flux of D2 molecules, ΓD2 , points radially inwards in the whole domain (see Fig. 3.7). As a
matter of fact, ΓD2 is roughly constant in the SOL, because the loss of D2 molecules due to
dissociation is compensated by the D2 molecules recycled at the limiter. Then, ΓD2 decreases
in the edge as a result of the molecular dissociation and ionization (due to the larger values
of plasma density and temperature in the closed field-line region), thus becoming negligible
towards the core.
Figure 3.8: Radial profiles of density (top) and radial flux (bottom) for the D+ and D species, av-
eraged over the toroidal and poloidal directions, evaluated over an interval of ∆t = 10.1R 0 /c s0
from a quasi-steady state single-component plasma situation. The components behind the
radial ion flux are discriminated. Plasma and neutral grid resolution, as well as simulation
parameters, are the same considered in Fig. 3.5.
83
Focusing now on the dynamics of the ion species, we highlight that the radial flux of D+ ions
points radially outwards across the whole domain and is mostly determined by the dominant
E × B flux except in the proximity of the core, where the diamagnetic flux dominates over the
E × B flux. The polarization drift contribution is negligible in the whole domain. We also note
that the flux increases across the edge region from the core to the LCFS, peaks in the near
SOL and gradually decreases across the SOL. This significantly differs from the profile of the
ion flux in the single-component plasma simulation (see Fig. 3.8), where the flux peaks at
the LCFS. This difference between the two models is related to the location of the ionization
source n D νiz . As a matter of fact, while the source has a smooth profile and peaks at the LCFS
in the single-component simulation, the ionization source peaks further inside the edge in
the multi-component model, leading to a sharp increase of the D+ flux in the edge, in the
proximity of the LCFS.
We remark that, as seen in Fig. 3.7, the radial flux of D+ 2 ions points radially outwards in the
SOL, but is directed radially inwards in the edge. This is a consequence of the fact that most D+
2
are generated in the vicinity of the LCFS, where the D2 molecules are ionized by the warmer
and denser plasma. The D+ 2 radial flux is determined by the balance between the inward
pointing E × B and outward pointing diamagnetic drift components in the SOL, by the E × B
flux in the edge close to the LCFS, and by the diamagnetic component towards the core.
We also note that the inward pointing D+ 2 flux, ΓD2 , is sharply peaked in the edge, in the
+
+
vicinity of the LCFS. This is because most D2 ions are generated by ionization of D2 molecules
in that region and are dissociated shortly afterwards, having travelled a negligible distance.
Indeed, the spatial location of the peak of ΓD+2 matches to the one of the n D+2 profile in Fig. 3.6.
We highlight that the contribution to the flux of D+ 2 arising from the polarization drift is not
represented in Fig. 3.7 because it is neglected in our model. Finally, ΓD+2 is three to four orders
of magnitude smaller than ΓD+ , which is a consequence of the small ratio n D+2 /n D+ . Since
the polarization drift component is expected to be small compared to the total ion flux of D+ 2
molecules, ΓD+2 , we conclude that neglecting the polarization drift terms in Eqs. (3.19-3.27)
has indeed a negligible impact on the results of the simulation.
We must highlight that the D+ 2 flux is very small compared to the flux of the main ion species,
D+ , since the density of D+2 ions is about four orders of magnitude smaller than the density
of D . This may allow to neglect the D+
+
2 ion species, by assuming that they are instantly
+
dissociated into a D atom and a D ion. In the future, we will repeat the simulation presented
here while considering instantaneous dissociation of D+ 2 to clarify the impact of taking the
finite lifetime of D+
2 ions into account.
3.7 Conclusions
In this chapter we present a multi-component plasma model for the self-consistent description
of the neutral and plasma dynamics in the tokamak boundary, which is implemented in the
GBS code. A deuterium plasma is simulated in the edge and SOL regions of a tokamak,
84
3.7. Conclusions
featuring electrons, D+ and D+ 2 ions, D atoms and D2 molecules. The models that describe
the neutrals and the plasma dynamics are coupled by means of a number of collisional
processes, which are responsible for the introduction of neutral-plasma interaction terms both
in the plasma and the neutral equations. The collisional reactions considered in this chapter
include ionization, elastic collisions between electrons and neutral particles, charge-exchange
collisions and dissociative processes. The multi-component plasma model is based upon the
Braginskii fluid equations derived in the drift-limit, which extend the single-ion species model
to account for D+ 2 ions. The closure considered in the present chapter follows the approach
developed by Zhdanov. Regarding the neutral particles, the model used in the single-neutral
species model of GBS [61] is extended to address the effect of the molecular species, D2 . As a
matter of fact, in the model presented in this chapter, the neutrals are computed by solving
two coupled kinetic equations for the D and D2 species by using the method of characteristics.
The resulting system of linear integral equations are then discretized and solved with respect
to the densities of neutral species, n D and n D2 .
The results from the first simulation based upon the multi-component model, considering a
toroidally limited plasma, are also described. These results bear some considerable differences
compared to the results from simulations carried out by using the single-ion component model
implemented in GBS. An up-down asymmetry is observed in the n e and n D+ density profiles,
with larger densities being observed below the equatorial midplane. This is related to the
counterclockwise parallel flow of the edge plasma, as reported in the profiles of v ke , v kD+ and
v kD+2 . This feature also leads to larger recycling rates and higher neutral density in the upper
side of the limiter, compared to the region below the limiter plates. In addition, the density of
the neutral atoms and molecules, n D and n D2 , is found to be about one order of magnitude
smaller than n D+ in the open-field line region and two orders of magnitude smaller in the
edge, while n D+2 is about three to four orders of magnitude smaller than n D+ , even in the edge
close to the LCFS, where n D+2 peaks.
Taking into account the molecular dynamics also allowed multi-component plasma simula-
tions to grasp the influence of D2 molecules on the plasma fuelling. In fact, most D2 particles
are generated at the limiter close to the LCFS. A large fraction of D2 molecules cross the LCFS
and reach the edge, where they are typically dissociated into atomic D as a result of the in-
teraction with the high density and temperature plasma. The D atoms thus generated and
the remaining D2 molecules are then ionized inside the edge, with the D+ 2 ions being disso-
ciated shortly afterwards as a consequence of the high electron densities and temperatures.
Therefore, the simulation results reveal an inward radial shift of the peak of the ionization of D
atoms, compared to the results from the single-ion species simulations.
The radial profiles of the densities and radial fluxes are also influenced by the inclusion of
molecules. The radial flux of D+ is observed to increase sharply in the edge, in the vicinity
of the LCFS as a result of the ionization source peaking in that region. The flux of D+ then
remains high in the proximity of the LCFS and decreases sharply in the near SOL, where the
ion sinks at the limiter are dominant over the sources of D+ . This constitutes a difference with
85
respect to the D+ flux observed in the results from a single-ion species simulation, where a
maximum is reported at the LCFS. On the other hand, the D density peaks in the SOL due to
the D2 molecules dissociated in that region. This is also the reason for the sign reversal of the
radial flux of D atoms in the far SOL, where ΓD points radially outwards. On the other hand,
the inward flux of D atoms increases radially inwards inside the LCFS, since D atoms are also
generated in the proximity of the LCFS by D2 dissociation.
To conclude, the results presented in this chapter highlight that the multi-component model
for the self-consistent description of the neutral-plasma interaction provide a description
of a deuterium plasma capturing the molecular dynamics and its overall influence on the
other plasma and neutral quantities. Therefore, the model can be used to address a multi-
component plasma and more than one neutral species at a kinetic level when carrying out
self-consistent first-principles simulations of plasma turbulence in the tokamak boundary.
The procedure described in this chapter can be extended to include other species of plasma
particles or neutrals, as well as the corresponding additional collisional processes. The model
can also be implemented in a diverted tokamak configuration (featuring one or various X-
points), where multi-component simulations are important to shed some light on the high
recycling and detachment regimes.
86
4 Numerical simulations of Gas Puff
Imaging
This chapter presents the results of gas puff imaging (GPI) studies carried out by exploiting the
multi-component model described in Ch. 3. The research reported here aims at extending the
results of previous works on the simulation of GPI to the case of a multi-component deuterium
plasma in a three-dimensional domain. The simulations are performed in a toroidally-limited
plasma with gas puff sources located at the LFS equatorial midplane. The Dα emission arising
from the excitation of D atoms and the contributions from dissociation of D2 molecules and
D+2 ions, which is observed to dominate over the other components in the proximity of the
wall, are considered. The statistical moments and the turbulence properties, computed for the
different components of the Dα emission, as well as the relevant plasma and neutral quantities,
are investigated. The correlation functions between the Dα emission rate and the plasma
and neutral quantities, namely the electron density, n e , electron temperature, Te , and density
of neutral atoms, n D , are also evaluated considering in detail each contribution to the Dα
emission and analysing the correlation functions between these quantities in the simulation
domain. The results shown in this chapter highlight the importance of considering the neutral
fluctuations when interpreting the measurements of the Dα emission rate, especially in the
edge, where the perturbations of the neutral density have a more significant impact on the Dα
emission.
87
Chapter 4. Numerical simulations of Gas Puff Imaging
4.1 Introduction
Gas puff imaging (GPI) is one of the fundamental diagnostics currently used to evaluate the
turbulent evolution of a plasma in the boundary of tokamak devices [97, 98, 99, 100]. The
GPI diagnostics is based on the injection of neutral particles at a specific toroidal location.
The neutral particles interact with the plasma via a number of collisional processes and give
rise to emission of light as a result of atomic deexcitation. Given that the plasma boundary
is a medium of low optical density, the emitted light interacts weakly with the plasma and
can be measured by one or more high temporal and spatial resolution cameras. The cameras
register the integrated light emitted along the respective lines of sight. Using an horizontal
and a vertical camera, a tomographic reconstruction of the two-dimensional light emission
profile is also made possible.
Optical filters allow for the fast cameras to select a specific spectral line of interest within
the visible range, depending on the composition of the plasma. In the case of a deuterium
plasma, measurements most often focus on the Dα line of the Balmer series [97, 99, 101, 102],
which is the result of the deexcitation of a deuterium atom from the second excited state
D∗ (n = 3) to the first excited state D∗ (n = 2). Several atomic and molecular processes account
for the presence of deuterium atoms in the D∗ (n = 3) state. These processes include impact
excitation of D atoms in the fundamental (n = 1) state and a variety of dissociative processes
of the diatomic deuterium species, D2 and D+ 2 . As a result, it is difficult in experiments to
identify the source of Dα emission, even though the majority is expected to come from impact
excitation of D atoms. Moreover, the Dα emission following each excitation process has a
complex dependence on a number of parameters, which makes it even more difficult to
interpret Dα measurements. Therefore, assumptions have to be made, based on the properties
of the diagnosed plasma, to guide the interpretation of experimental results.
The effect of molecular dynamics is often neglected when interpreting Dα emission in ex-
periments [101, 103], therefore excluding the contribution of dissociative processes in the
interpretation of the Dα emission source, which is thus ascribed to electron impact excitation.
The emission rate is thus modelled as [98, 99, 102, 64]
Dα,Exc = n e n D r α,Exc (n e , Te ), (4.1)
with n e the electron density, n D the density of D atoms and r α,Exc (n e , Te ) the emission rate
coefficient for impact excitation of D to the second excited state D∗ (n = 3). The dependence of
r α on the electron density and temperature, n e and Te respectively, is theoretically calculated,
based on a collisional-radiative model [104]. In addition, it is often assumed that Dα emission
depends weakly on the neutral density perturbations. As a consequence, n D is assumed
constant in Eq. (4.1) and n D fluctuations are not taken into account in the analysis. Finally,
electron density and temperature fluctuations are assumed strongly correlated in the SOL
region [101, 103]. With these assumptions, the electron pressure can be inferred from the Dα
88
4.1. Introduction
emission measurements, which ultimately allows for the study of boundary turbulence. A more
accurate interpretation of Dα emission is made possible by disentangling the dependence
of the Dα emission on n e , n D and Te . This can be obtained by measuring the light emission
resulting from three different spectral lines, similarly to the experiments carried out with He-
lines in the Alcator C-Mod tokamak [97] and the TJ-II stellarator [105]. However, if molecular
dissociation becomes a non-negligible source of Dα emission with respect to impact excitation,
the use of different spectral lines is not sufficient to disentangle the complexity underlying the
Dα emission.
A number of numerical simulations, based on different approaches, have been performed to

disentangle the complexity of the interpretation of GPI results over the last decades. On the
one hand, GPI simulations were carried out assuming that neutral density perturbations are
negligible, focusing therefore on the turbulent dynamics of the plasma [101, 102, 103, 106].
On the other hand, the effect of neutral fluctuations were addressed while considering time-
independent plasma profiles or using analytical models for plasma perturbations [99, 107,
108, 109]. Efforts to jointly address the plasma and neutral dynamics on GPI simulations have
been reported only in the last decade.
The first attempt of a self-consistent description of fast camera data is described in Ref. [110],
where two-dimensional simulations of plasma turbulence and a mono-energetic neutral
model are used. In Ref. [64], the GBS code is used to simulate Dα emission from GPI diag-
nostics with a model of plasma turbulence and neutral dynamics. The three-dimensional
drift-reduced Braginskii equations for a single-component deuterium plasma (D+ and e− )
are solved, coupled with a kinetic model for deuterium atoms, D, in the SOL of a limited
tokamak configuration with a toroidal limiter on the HFS equatorial plane. A gas puff source is
introduced at the LFS equatorial midplane, consisting of the injection of D atoms from the wall.
Since the model does not take into account molecular dynamics, the D atoms are injected into
the domain at a temperature of 3eV, in order to model D atoms generated by Franck-Condon
dissociation of D2 molecules. The study focuses on the Dα line of the Balmer series, reporting
on the correlation between the Dα emission, the electron density, n e , the electron tempera-
ture, Te , and the deuterium atom density, n D . Results show that n e , Te and Dα emission are
all strongly correlated, particularly in the SOL. As for n D fluctuations, different results are
reported at distances from the gas puff location smaller or larger than λmfp,D , in particular a
strong anti-correlation between n D and the plasma quantities is observed at a distance from
the gas puff location smaller than λmfp,D . In fact, positive perturbations of n e and Te lead to
higher ionization rates, thus resulting in lower n D . As a consequence, n D and Dα emission
rates are also strongly anti-correlated in that region. However, while the D atom density has
an effect on the amplitude of the Dα emission rate, according to Eq. (4.1), the impact on the
statistical moments and turbulence properties of the Dα emission remains limited. Ref. [64]
concludes that neglecting the n D fluctuations remains as a valid assumption for interpreting
GPI measurements at distances from the gas puff source smaller than λmfp,D , the main effect
of assuming constant n D being an underestimate of the plasma quantities n e , Te from the
interpretation of Dα emission. On the other hand, Ref. [64] shows that, for distances from the
89
gas puff source larger than λmfp,D , the D atom density is influenced by non-local shadowing
effects due to the interaction of the atoms with a number of structures on their way across
the plasma. As a matter of fact, when crossing positive perturbations of n e and Te , atoms are
more likely to be ionized and do not penetrate further into the plasma, thus leading to small
n D in the edge, regardless of the local values of n e and Te . This non-local shadowing effect
[64] hence weakens the anti-correlation between n D and the plasma quantities, n e and Te , as
one moves radially inwards. At the same time, the anti-correlation between n D and the Dα
emission rate decreases, becoming a positive correlation at distances from the gas puff source
larger than λmfp,D . Therefore, n D fluctuations have a more significant impact on the statistical
moments and turbulence characteristics of the Dα emission at distances larger than λmfp,D ,
where n D fluctuations should be taken into account when considering GPI measurements.
The nHESEL code [45, 39, 46] has also been used to investigate the physics behind the GPI emis-
sion. nHESEL self-consistently simulates the neutral-plasma interaction in a two-dimensional
domain, relying on a fluid model of single-component plasma turbulence coupled with a diffu-
sive model for multiple neutral species. In Ref. [46], nHESEL simulations explore the effect of
molecular dissociation on the correlation between n D and plasma quantities in the presence
of blob events. Three different neutral species are considered: thermal deuterium molecules,
D2 , injected into the system via gas puffing, thermal Helium atoms, He, and non-thermal
deuterium atoms, D, that comprise warm atoms directly generated by Franck-Condon disso-
ciation of molecular deuterium and hot atoms resulting from charge-exchange interactions
with D+ ions.
As described in Ref. [46], dissociative processes lead to volumetric sinks of D2 and sources of
D atoms, while ionization constitutes a sink of D. Since there are no volumetric sources of D2 ,
an enhancement of n e and Te due to a blob enhances dissociative processes and thus leads
to a decrease of n D2 . Therefore, n D2 is anti-correlated with n e and Te (similarly to n D in Ref.
[46]). On the other hand, the competition between dissociation and ionization events, that
constitute a source and a sink of D atoms respectively, determines the effect of blob-induced
plasma perturbations on the n D fluctuations. In Ref. [46], this balance is expressed as
S D = n D2 2νdiss(D2 ) (Te ) + νdiss-iz(D2 ) (Te ) − n D νiz (Te ) ,

£ ¤
(4.2)
where νdiss , νdiss-iz and νiz are the dissociation, dissociative-ionization and ionization rates,
all proportional to n e . If S D > 0, an increase of n e leads to an increase of S D , and therefore a
correlation between n D and n e . In contrast, if S D < 0, n D is anti-correlated with n e , similarly
to the observations in Ref. [64]. In Ref. [46] a parameter η is also introduced, defined as the
ratio between the D sources and sinks,
n D2 2νdiss(D2 ) + νdiss-iz(D2 ) n D2 2νdiss(D2 ) + νdiss-iz(D2 )

£ ¤ £ ¤
η= = , (4.3)
n D νiz nD νiz
90
4.2. The GPI diagnostics and Dα emission rate
with correlation between n e and n D found for η > 1 and anti-correlation for 0 < η < 1. The value
of η is determined partly by the n D2 /n D ratio and partly by the ratio between the reaction rates
of volumetric sources and sinks, which depend on the electron temperature. As the authors
highlight in Ref. [46], the temperature-dependent term is always larger than 1, approaching
1.7 at large electron temperature. Therefore, as long as n D2 /n D > 0.58, correlation (η > 1) is
always observed. The numerical simulation presented in Ref. [46] points out η > 1 in the
SOL, where n D2 is large, and η < 1 in the edge, because of the large temperature observed
inside the separatrix and the resulting low D2 density. The findings reported in Ref. [46] thus
suggest that n D is correlated with Dα emission in the SOL and anti-correlated in the edge.
This conclusion contrasts with the one in Ref. [64] where, by not taking into account D2
dissociation, n D perturbations are found to be anti-correlated with the plasma density and
temperature perturbations in most of the domain, particularly in the SOL.
The present chapter leverages the studies presented in Ref. [64] and extends them to the case
of a multi-component deuterium plasma by using the GBS model presented in Ch. 3. This
includes two ion species, D+ and D+ 2 , and two neutral species, D atoms and D2 molecules,
thus allowing for molecular dissociation to be taken into account. Our goal is to understand
the impact of the multi-component plasma dynamics on GPI diagnostics, extending the work
reported in Ref. [46] by using a global three-dimensional model to describe the tokamak
boundary and addressing the neutrals by discretizing a kinetic advection equation for each
species. The kinetic approach allows us to account for all values of the mean free path for the
neutral species, with no need to distinguish particles of the same species according to their
temperature and origin. In addition to Ref. [46], we also aim at evaluating the contribution to
the Dα emission from dissociation of molecular species, D2 and D+ 2 , and compare it with the
main source due to excitation of D atoms.
This chapter is structured as follows: after this Introduction, a brief description of the simula-
tion of the gas puff setup is presented in Sec. 4.2. We then present the simulation results in Sec.
4.3. These results are further discussed in Sec. 4.4, where we evaluate the correlations between
the quantities involved in the computation of the D α emission rate, and in Sec. 4.5, where we
address the impact of the fluctuations of the density of neutral species on the interpretation of
GPI results. The conclusions follow in Sec. 4.6.
4.2 The GPI diagnostics and Dα emission rate

The simulations described here leverage the self-consistent multi-component model for
plasma turbulence and kinetic neutrals in the tokamak boundary implemented in GBS and
presented in Ch. 3. We consider therefore a limited plasma, with toroidal limiter on the
equatorial HFS limiter. The GPI diagnostics is implemented by adding a low-intensity gas
puff at the equatorial midplane on the LFS. We assume that the gas puff extends over the full
toroidal direction and it is independent of the toroidal angle, to simplify our scenario and its
analysis. The gas puff injects D2 molecules at wall temperature at the outer chamber wall,
91
which does coincide with the boundary of the simulation domain. This contrasts with Ref. [64],
where the injected molecules are assumed to undergo Franck-Condon dissociation outside
the simulation domain and enter it as D atoms at a temperature of 3eV. As a matter of fact, the
results obtained from the first multi-component GBS simulation presented in Ch. 3 show that
penetration of D2 molecules across the SOL into the edge is important, at least in the plasma
conditions considered in our simulations (in the transition between the sheath-limited and
conduction-limited regimes), highlighting the importance of properly modelling the puffing
of D2 molecules from the domain boundary.
Following Ref. [64], the distribution function of the D2 molecules injected for the GPI diagnos-
tics from the domain boundary is described by
(y − y gp )2
Ã !
f D2 ,gp (y, v) = S gp exp − χin,D2 (v), (4.4)
2(∆y gp )2
where S gp is the puffing rate (equivalent to the flux of puffed molecules), y is the poloidal
coordinate along the outer wall, y gp is the poloidal coordinate of the center of the gas puff,
∆y gp is the gas puff width and χin (v) is the velocity distribution of the puffed D2 molecules,
which is assumed to follow the Knudsen cosine law,
3 m D2 m D2 v 2
µ ¶
χD2 ,in (v) = cos (θ) exp − . (4.5)
4π Tw2 2Tw
In Eq. (4.4), following the definition introduced in the previous chapters, we define the angle
θ = arccos Ω̂ · n̂ , with the unit vector Ω̂ = v/v being along the direction of the neutral velocity,
¡ ¢
and the unit vector normal to the boundary n̂ pointing towards the plasma volume. In the
present work, we consider a temperature Tw = 0.3eV, y gp = 400ρ s0 (the gas puff is located at
the equatorial outboard midplane) and ∆y gp = 20ρ s0 .
In the present thesis, similarly to Ref. [64], we focus on the local emission rate from the Balmer
Dα line. However, in addition to the Dα emission resulting from electron impact excitation of
D atoms, dissociation of molecular species D2 and D+ 2 is also taken into account as sources of
D α emission. The total Dα emission rate is hence expressed as
Dα = Dα,Exc + Dα,Diss(D2 ) + Dα,Diss(D+2 ) , (4.6)
where Dα,Exc is the contribution from the excitation of D atoms to the excited state D∗ (n = 3)
defined in Eq. (4.1), while
Dα,Diss(D2 ) = n e n D2 r α,Diss(D2 ) (n e , Te ), (4.7)
92
4.3. Simulation results
is the contribution from dissociation of D2 molecules into two D atoms, one of which in the
excited state D∗ (n = 3). Finally,
Dα,Diss(D+2 ) = n e n D+2 r α,Diss(D+2 ) (n e , Te ), (4.8)
is the contribution that joins the dissociation of D+ +

2 molecular ions into one D ion and one D
∗ +
atom in the excited state D (n = 3) and the dissociative-recombination of D2 molecular ions
into two D atoms, one of them in the excited state D∗ (n = 3). We recall that n D2 is the density of
D2 molecules and n D+2 the density of D+ 2 ions. We also remark that the emission rate coefficient
for dissociation of D2 molecules generating a D∗ (n = 3) atom, r α,Diss(D2 ) (n e , Te ), as well as
the emission rate coefficient for dissociation or dissociative recombination of D+ 2 generating
∗
a D (n = 3) atom, r α,Diss(D+2 ) (n e , Te ), are obtained from a collisional-radiative model [111].
For the emission rate coefficient r α,Diss(D2 ) (n e , Te ), we use the values made available in the
OpenADAS database, similarly to Ref. [64]. Other molecular processes accounting for Dα
emission, such as dissociative-ionization of D2 , are neglected here because their cross sections
are considerably smaller than the ones of the processes considered here.
4.3 Simulation results

The simulation of GPI presented here considers a simulation with the same parameters as the
one described in Ch. 3, where the diagnostic gas puff described in Sec. 4.2 is introduced. After
a transient phase, a quasi-steady state is reached, characterized by constant plasma profiles
resulting from the balance between sources and sinks. Our analysis focuses on a time interval
∆t = 40R 0 /c s0 during the quasi-steady state phase and targets mostly the LFS region around
the diagnostic gas puff (350ρ s0 < y < 450ρ s0 ), where Dα light emission following the injected
particles is important. Similarly to Ch. 3, we highlight that the results presented here are
expected to depend strongly on the simulation parameters considered.
We first perform time and toroidal averages of the plasma and neutral quantities relevant for
the study of Dα emission, namely the electron density, n e , the electron temperature, Te , the D
atom density, n D , the D2 molecule density, n D2 , and the D+ 2 ion density, n D2 . We also study the
+
number of particles ionized/dissociated per unit volume and unit time, i.e. n D νiz , n D2 νiz,D2 ,
n D2 νDiss(D2 ) and n D+2 νDiss(D+2 ) . The time and toroidal averages of these quantities are presented
in the poloidal cross sections shown in Fig. 4.1. We remark that the 40R 0 /c s0 time interval
over which the average is performed is considerably larger than the characteristic timescales
of turbulent events. Toroidal averaging is enabled by the axisymmetry of the considered
configuration.
We note that the cross section plots of the plasma and neutral quantities, n e , n D , D2 , n D2 ,
n D+2 and Te , are similar to the ones observed in the simulations reported in Ch. 3 for the first
multi-component plasma simulation, the difference being only in the region close to the gas
93
Figure 4.1: Poloidal cross section of time and toroidal-averaged plasma quantities (n e , n D+2 ,
and Te ), neutral particle densities (n D , n D2 ), ionization sources (n D νiz and n D2 νiz,D2 ) and
dissociation rates (n D2 νDiss(D2 ) and n D+2 νDiss(D+2 ) ).
94
Figure 4.2: Poloidal cross section of the logarithm of η and the logarithm of n D2 /n D .
puff at the LFS. As a consequence of the GPI injection, n D2 peaks in the proximity of the outer
wall at the LFS equatorial midplane, where the gas puff is located. We remark that, due to
the SOL conditions considered in this simulation, namely the low Te values observed in the
SOL at the LFS, the D2 mean free path is larger than the SOL width, λmfp,D2 ' 95ρ s0 . As a
consequence, most D2 molecules cross the SOL without interacting with the plasma and are
only dissociated/ionized inside the LCFS, which has an important impact on the Dα emission
profile. As a result of the large λmfp,D2 , the density of D atoms peaks in the edge, close to
the LCFS, and the density of D+ 2 ions also peaks further inside. In fact, as shown in Fig. 4.1,
dissociation of D2 molecules peaks inside the LCFS and ionization of D2 molecules, which
requires more energy compared to dissociation, peaks further inside, where the values of n e
and Te are higher. On the other hand, since dissociative processes have in general larger cross
sections than ionization at the same temperature (see Fig. 3.1), most D+ 2 ions are dissociated
shortly after being generated. This is confirmed by Fig. 4.1, that shows that dissociation of D+ 2
occurs where the ionization of D2 takes place. Finally, ionization of D peaks close to the core, a
consequence of the fact that most D atoms are generated in the edge region and then ionized
when they enter in contact with the warmer and denser plasma in the innermost region of the
domain.
We remark that, while the radial profiles of n e and Te are similar to the ones presented in Ref.
[64], the D atom density, n D , is different, since it peaks inside the LCFS, which is a consequence
of the dissociative processes taking place in the edge where D atoms are generated. On the
other hand, the D2 density, n D2 , peaks in the SOL close to the outer wall, while D+ 2 peaks in
the edge, where the high plasma density and temperature maximizes the ionization of D2
molecules.
We also focus on the value of the correlation parameter η, redefined from Eq. (4.2) consistently
with the multi-component plasma model under consideration, so as to take into account all
possible sources and sinks of D atoms (see the list of collisional processes described in Table
3.1), yielding
95
n D2 2νdiss(D2 ) + νdiss-iz(D2 ) + n D+2 [2νdiss-rec + νdiss ]

£ ¤
η= h i . (4.9)
n D νiz + νcx,D−D+2
Figure 4.3: The logarithm of the time and toroidal averaged D α emission and of its contribu-
tions, D α,Exc , D α,Diss(D2 ) and D α,Diss(D+2 ) , are represented in the poloidal plane.
Given the values of the reaction rates of the collisional processes, which are plotted in Fig.
3.1, for typical densities and temperatures in the plasma boundary, the term referring to the
dissociation of D2 dominates the numerator of Eq. (4.2), while ionization of D dominates over
D − D+ 2 charge-exchange in the denominator. Therefore, for typical SOL and edge parameters,
η ' (n D2 /n D )(2νDiss(D2 ) /νiz ), similarly to the expression reported in Eq. (3) from Ref. [46]. Thus,
η > 1 holds whenever n D2 /n D & 0.6 is satisfied.
The time and toroidal average of the logarithm of η and of the logarithm of n D2 /n D are plotted
in Fig. 4.2 on a poloidal cross section. Due to the considerable penetration of D2 molecules
across the LCFS, n D2 /n D > 1 is observed in the SOL and in a significant portion of the edge,
96
Figure 4.4: Five snapshots of the normalized fluctuations of each component of the Dα
emission, separated by 0.2R 0 /c s0 . The Dα,Exc , Dα,Diss(D2 ) and Dα,Diss(D+2 ) contributions and the
total Dα emission are shown in the first, second, third and fourth rows, respectively.
especially at the LFS equatorial midplane, where the diagnostic D2 gas puff is present. As a
result, η > 1 holds in most of the domain while η < 1 is verified only in a restricted portion
of the edge region closer to the core. This implies that, except for the region close to the
core, the number of D atoms generated by dissociative processes per unit time and unit
volume is larger than the number of D atoms depleted per unit time and unit volume due
to ionization events, with n D being positively correlated with n e and Te . This is in line with
the results reported in Ref. [46] and contrasts with the conclusions in Ref. [64] based on
single-component simulations carried out with GBS. Nevertheless, we highlight that results
are highly dependent on the specific SOL conditions considered, which are characterized by a
lower electron temperature than the one usually found in typical SOL plasmas. In the case of a
higher temperature SOL, we expect a shift radially outwards of the regions observed in this
simulation.
Following Ref. [64], Fig. 4.3 presents time and toroidally averaged poloidal cross sections of
the total Dα emission rate and the contributions from the three sources contributing to it, in
logarithmic scale. We highlight that the excitation of D atoms, Dα,Exc , is the main source of Dα
emission, peaking in the edge. On the other hand, the contribution from dissociation of D2 ,
Dα,Diss(D2 ) , is one to two orders of magnitude smaller than Dα,Exc . The Dα,Diss(D2 ) contribution
peaks in the edge, close to the core, because of the high plasma densities and temperatures in
this region, even though the density of D2 molecules is small. Regarding the contribution from
dissociation of D+ 2 , Dα,Diss(D2 ) , we highlight that it peaks in the edge, close to the LCFS. While
+
97
it is smaller than Dα,Exc by a factor of about five in the edge region, it becomes the dominant
source of Dα emission in the far SOL.
Typical snapshots that represent the time evolution of the Dα emission rate are presented
in Figs. 4.4, where the analysis is restricted to the GPI injection, i.e. at the LFS around
the equatorial midplane, 350ρ s0 < y < 450ρ s0 . The normalized perturbation of the total
Dα emission, computed as (Dα − 〈Dα 〉) / 〈Dα 〉, with 〈Dα 〉 the time and toroidally averaged
Dα emission rate, is considered, as well as of its three sources, computed analogously. The
snapshots are separated by time intervals of 0.2R 0 /c s0 . The Dα,Exc and Dα,Diss(D2 ) emissions
display a similar profile, while Dα,Diss(D+2 ) fluctuations, despite bearing an overall qualitative
resemblance to the other emission profiles, peak at different locations. The plots of the
fluctuations of the total Dα emission rate, follow the profiles of the snapshots of Dα,Exc in the
edge and near SOL, since this is the dominant source of Dα emission in that region, and are
similar to the plots of Dα,Diss(D+2 ) in the far SOL, where it dominates. A parameter scan on the
plasma density and temperature will extend the present study. In particular, we expect that, at
larger values of the electron density and temperature, the whole dynamics is shifted further
out towards the far SOL, as the mean free path of D2 molecules decreases.
4.4 Analysis of the correlation between the D α emission and the

plasma and neutral quantities
Since the measurements of D α emission provided by GPI diagnostics are used to infer the
properties of the plasma and neutrals, understanding the correlations between the D α emis-
sion and these properties is of crucial importance for an accurate interpretation of the GPI
measurements. We then evaluate the correlations between the D α emission rate and the
n e , n D , n D2 , n D+2 and Te (the quantities associated with the D α emission). For this purpose,
following the same procedure as in Ref. [64], we make use of the Spearman rank coefficient of
correlation [112], which varies between −1 (if two functions are totally anti-correlated) and 1
(in case they are perfectly correlated).
Fig. 4.5 shows the Spearman correlations between the quantities involved in the calculation
of the emissivity rate associated with the excitation of D atoms, D α,Exc . The electron density
and temperature are strongly correlated, especially in the edge region. On the other hand, the
correlation between the neutral density n D and the plasma quantities, n e and Te , is different
from the one observed in Ref. [64]. In fact, n D is correlated with n e in the SOL, with the
correlation increasing towards the LCFS. The correlation function peaks in the edge, close to
the LCFS, and then decreases sharply, becoming negative towards the core. A similar behavior
is found for the Spearman correlation between n D and Te , with the correlation in the SOL and
the anti-correlation in the edge being even stronger compared to the correlation between n D
and Te . This is due to the mechanisms behind the generation and depletion of D atoms in the
SOL and edge regions. In fact, in the SOL region, positive fluctuations of plasma density and
temperature enhance the dissociation of the D2 molecules injected by the gas puff from the
98
4.4. Analysis of the correlation between the D α emission and the plasma and neutral
quantities
outer wall, which results in an increase in the D atom density. Hence n e and Te are positively
correlated with n D . Around the LCFS, the density of D2 molecules decreases while the density
of D atom increases. However, the source of D atoms due to molecular dissociation is still
more important than their sink due to ionization processes, since the plasma temperature
remains lower than the ionization threshold. As a result, in agreement with Ref. [46], the
correlations between n D and n e and between n D and Te remain positive.
Figure 4.5: Radial plot of the Spearman correlation function between the quantities involved
in the Dα,Exc emission, at the LFS equatorial midplane (all quantities are averaged in time and
along the toroidal direction).
Deep in the edge, the high temperature leads to an ionization sink that dominates over
dissociation sources. Therefore, local positive fluctuations of n e and Te lead to the decrease of
D atom density, thus resulting in an anti-correlation between n D and the plasma quantities,
n e and Te . The result observed in Ref. [64] is thus recovered, but only for the warmer and
denser plasma in the edge.
Turning to the D α emission rate due to atom excitation, we observe that it is strongly correlated
with the plasma quantities n e and Te , as expected from (Eq. 4.1), in the SOL and around the
LCFS. The D α emission is also positively correlated with n D . This results from Eq. (4.1)
and from the fact that n e , Te and n D are also positively correlated with one another, which
agrees with the conclusions in Ref. [46]. On the other hand, the correlation function between
D α,Exc and the plasma quantities n e and Te falls sharply and becomes negative towards the
core. In fact, at the high densities and temperatures of this region, the density of neutral
atoms drops to very low values, reducing significantly the D α emission due to atom excitation.
As a result, the D α emission rate in Eq. (4.1) becomes more sensitive to fluctuations of
n D rather than perturbations of the plasma quantities, which explains why D α,Exc is anti-
correlated with n e and Te in this region, while strongly correlated with n D . This also highlights
the importance of taking into account the neutral density fluctuations when interpreting
99
D α emission measurements close to the core, as D α emission rates may become mostly
determined by n D fluctuations. We note that the result of the single-component simulation
in Ref. [64] is retrieved in the edge. In this region, n D is sufficiently large that it does not
constrain the D α emission. In these conditions, the n D perturbations are less important than
the fluctuations of n e and Te , leading to a positive correlation between D α,Exc and n e and Te
in most of the edge region and a negative correlation with the atom density, n D .
in the Dα,diss(D2 ) emission, at the LFS equatorial midplane (all quantities are averaged in time
and along the toroidal direction).
Fig. 4.6 shows the Spearman correlation function between the variables involved in the D α
emission rate associated with the dissociation of D2 molecules, Dα,Diss(D2 ) . It is observed that
n D2 and Te are strongly anti-correlated in the whole domain, since positive perturbations of
Te are associated with enhanced ionization of D2 molecules and hence smaller n D2 . Similarly,
n D2 is anti-correlated with n e in the edge and in the near SOL, where large plasma densities
result in a large rate of molecular dissociation and hence smaller density of D2 molecules. The
correlation becomes positive in the far SOL, closer to the gas puff source, where the density
of D2 molecules is larger. In fact, the gas puff is expected to enhance plasma fuelling, i.e.
larger densities of D2 molecules will ultimately lead to an increase of n e . These results are
similar to the ones obtained in Ref. [64] for the D α emission due to excitation of D atoms in a
single-component plasma. Regarding the D α,Diss(D2 ) , we observe that it is correlated with n e
and Te in the SOL and in most of the edge region. In particular, the D α,Diss(D2 ) − Te correlation
is strong in the SOL. Towards the core, the density of D2 molecules drops, making D α,Diss(D2 )
very small and strongly dependent on the n D2 fluctuations. Therefore, D α,Diss(D2 ) is more
sensitive to the fluctuations of n D2 than the perturbations of the plasma quantities, similarly to
the dependence of D α,Exc on n D . As a result, D α,Diss(D2 ) becomes anti-correlated with n e and
Te in the proximity of the core. In contrast, the emission rate is strongly correlated with the
density of D2 molecules near the core, but reduces sharply when one moves radially outwards,
100
quantities
as n D2 is no longer negligible and hence no longer regulates D α,Diss(D2 ) .
in the Dα,diss(D+2 ) emission, at the LFS equatorial midplane (all quantities are averaged in time
and along the toroidal direction).
Focusing now on the analysis of the D α emission due to dissociation and dissociative recom-
bination of D+ 2 ions, Dα,Diss(D2 ) (see Fig. 4.7), we start by highlighting the strong correlation
+
+
between the density of D2 ions and n e in the SOL and in most of the edge, since positive n e
perturbations enhance the ionization of D2 molecules and hence increase n D+2 in these regions.
However, since D+ 2 ions have a short lifetime, being destroyed by dissociative processes whose
cross sections are larger than the cross section of ionization of D2 molecules, the correlation
between n D+2 and n e is also strongly influenced by the density of n D2 molecules. As a matter of
fact, towards the core, n D2 drops to residual values, resulting in fewer D2 molecules that can
be ionized than D+ 2 ions that can be dissociated. As a consequence, positive fluctuations of
n e lead mostly to enhanced dissociation of D+ 2 ions and therefore negative n D2 fluctuations,
+
which turns into an anti-correlation between n D+2 and n e . The Spearman correlation function
between n D+2 and Te follows a similar behaviour, i.e. anti-correlation close to the core and cor-
relation in the edge and near SOL, but in the far SOL where anti-correlation between these two
quantities is observed. This is due to the fact that, at the typical temperatures of the far SOL,
dissociation of D+ 2 ions dominates over ionization of D2 molecules (see Fig. 3.1). Therefore,
while positive temperature fluctuations result in the generation of D+ 2 ions, they also increase
the rate of depletion of D+ 2 ions due to dissociative processes even more significantly. As a
consequence, positive Te perturbations lead to negative n D+2 fluctuations in this region.
Since the density of D+ 2 ions is very small everywhere and hence decisively constrains the
D α,Diss(D+2 ) emission rate, the D α,Diss(D+2 ) emission rate is correlated with n D+2 in the whole
domain. The correlation is particularly strong near the core, at the LCFS and in the proximity
of the outer wall. The correlation function between D α,Diss(D+2 ) and Te is positive in most of
101
the domain (higher temperatures are related to larger emissivity coefficients and increased
ionization of D2 molecules), the exception being the proximity to the core and to the outer
wall, where an anti-correlation is found between D α,Diss(D+2 ) and Te due to the anti-correlation
between n D+2 and Te . Similarly, D α,Diss(D+2 ) is correlated with n e in the SOL and in most of the
edge, where n D+2 is correlated with n e , but an anti-correlation between D α,Diss(D+2 ) and n e is
observed close to the core, where fluctuation of n D+2 and n e are also anti-correlated.
in the Dα emission, at the LFS equatorial midplane (all quantities are averaged in time and
along the toroidal direction).
Finally, Fig. 4.8 shows the correlation functions between the total D α emission rate, D α , the
electron temperature Te and the density of each of the species involved in the reactions that
emit light in the D α line. The correlation with D α can be interpreted through the analysis
of D α,Exc , since it dominates over the other sources of D α emission in the edge and near
SOL, and D α,Diss(D+2 ) , which is dominant in the far SOL. It is observed that, as for D α,Exc (see
Fig. 4.5), the D α emission is anti-correlated with n e and Te near the core, where neutrals
can hardly penetrate and a strong correlation between D α and n D is observed. For the same
reason, D α is strongly correlated with n D2 and n D+2 in this region, even though these species
contribute weakly to the overall emission via dissociative processes. In the edge, D α emission
is strongly correlated with n e and Te and the correlation between D α and n D , similarly to
D α,Exc , is negative, since ionization dominates over dissociation. The correlation between D α
and n D2 and between D α and n D+2 in the edge are explained by the fact that n D2 is strongly
anti-correlated with n e and Te , while n D+2 is strongly correlated with these plasma quantities.
In the near SOL, D α is correlated with n e , Te and n D , since dissociation dominates. On the
other hand, since the D α emission rate in the far SOL is mostly determined by D α,Diss(D+2 ) , D α
is also strongly correlated with n D+2 and n e in this region. The correlation between D α and Te
drops radially outwards, becoming anti-correlation next to the outer wall. We remark that this
102
quantities
represents a crucial difference with respect to the Spearman correlation between the D α emis-
sion rate and Te of the single-component simulation in Ref. [64] with important implications
on the interpretation of GPI measurements, suggesting a different way of interpreting GPI
measurements in the region next to the gas puff source. Moreover, the correlations of D α with
n D and n D2 near the gas puff source are determined by the correlation between the dominant
D α,Diss(D+2 ) contribution and the densities of neutral species n D and n D2 (not shown in Fig.
4.7). This is determined by the fact that D α,Diss(D+2 ) is strongly correlated with n e , thus the
correlation between D α and the densities of neutral species descends from their correlation
with n e . Fig. 4.5 shows a small correlation between n e and n D , explaining that D α and n D
are not clearly correlated. On the other hand, Fig. 4.6 shows that n e and n D2 are correlated,
thus leading to a correlation between D α and n D2 in the close proximity to the gas puff. These
conclusions agree with the results observed in Fig. 4.8.
To summarise, we can divide the domain in four different regions according to the correlation
between the D α emission rate and the plasma and neutral quantities, namely n e , Te and n D .
1) The far SOL, where D α emission is mostly determined by the contribution of dissociation of
D+2 ions and which is characterized by the D α emission being strongly correlated with n e , but
uncorrelated or even slightly anti-correlated with Te and, in general, uncorrelated with n D .
2) The near SOL and LCFS, where D α emission due to excitation of D atoms dominates over
the contributions from dissociative processes and a strong correlation between D α emission
and the plasma quantities, n e and Te , is observed. Moreover, since the sources of D due
to dissociative processes are more important than D sinks due to ionization, D α is strongly
correlated with n D (similar to the conclusions found in Ref. [46]). 3) The edge region, where
D α is also strongly correlated with n e and Te , but ionization of D dominates with respect to
dissociative processes, hence leading to an anti-correlation between D α and n D (similar to
the results reported in Ref. [64]). 4) The region of the core, where D α emission is constrained
by the residual values of the density of D atoms that can penetrate in this region, resulting in
D α being correlated with n D , but anti-correlated with n e and Te .
To illustrate the implications of these observations in Fig. 4.9, we consider snapshots of the
normalized fluctuations of the plasma and neutral quantities, namely n e , Te , n D , n D2 and n D+2 ,
and the D α emission, at a given time and toroidal location, focusing on the poloidal region
around the LFS equatorial midplane. The normalized fluctuations for the electron density,
n e , are defined as (n e − 〈n e 〉)/ 〈n e 〉, and similarly for the other quantities. The results in Fig.
4.9 confirm that n e and Te are strongly correlated in the whole domain, except in the close
proximity to the core, where the correlation is less evident. The fluctuations of the molecular
density, n D2 , are uncorrelated with n e and Te in the SOL, but clearly anti-correlated with the
perturbations of n e and Te around the LCFS and in the edge region, consistently with the
observations in Fig. 4.6. On the other hand, the fluctuations of n D+2 are strongly correlated with
the electron density and temperature fluctuations, except next to the outer wall, where n D+2 is
correlated with n e , but anti-correlated with Te (because dissociation of D+ 2 is more important
than ionization of D2 at the temperatures typically found in this region), and in the close
proximity to the core, where D2 molecules can hardly penetrate and hence ionization of D2
103
Figure 4.9: Typical poloidal snapshots of the fluctuations of n e , Te , n D , n D2 , n D+2 and the D α
emission rate at the LFS equatorial midplane, where the gas puff is located.
is small. Regarding the n D fluctuations, one can distinguish the different regions mentioned
above: the region close to the outer wall, where n D is uncorrelated with n e and Te , the edge,
104
4.5. Impact of neutral fluctuations
where the n D fluctuations are anti-correlated with the fluctuations of n e and Te , and the
LCFS with the SOL, where n D fluctuations are positively correlated with electron density and
temperature perturbations. Close to the core, the shadowing effect first reported in Ref. [64],
especially at y ' 420ρ s0 is observed. In fact, the positive perturbation of n e , Te enhances
the ionization of D atoms, thus reducing n D locally and also further inside towards the core,
since D atoms can no longer penetrate radially. As a result, n D does not increase when n e
and Te decrease. We therefore conclude that the non-local shadowing effect attenuates the
anti-correlation. Finally, we observe that the total D α emission rate, D α , is strongly correlated
with n e and Te in almost the whole domain, except for the far SOL where the link between D α
and Te is not clear. We remind that this is related to the fact that D α emission is dominated by
the contribution of dissociation of D+ 2 ions, with the total D α emission being mostly correlated
with n D+2 and n e in this region, while nothing can be said regarding the correlation between
D α and Te . We again highlight that the location and radial spread of these regions is affected
by the values of the electron density and temperature, thus leading to a shorter mean free path
of the D2 molecules puffed into the domain.
4.5 Impact of neutral fluctuations

Similarly to the procedure followed in Ref. [64], we evaluate the impact of neutral density
fluctuations on the Dα emission by using the average density of neutral particles when com-
puting the emission rates and compare them with the self-consistent Dα emission profile
that takes into account the instantaneous neutral density. This is of importance for the Dα,Exc
and Dα,Diss(D2 ) contributions, which rely on the excitation of neutral atoms and molecules,
respectively, but not for the Dα,Diss(D+2 ) contribution, which is associated with the excitation of
D+2 plasma ions. For the two excitation channels affected by the density of neutral species, the
Dα emission based on the averaged neutral density profiles can be expressed as
Dα,Exc〈nD 〉 = n e 〈n D 〉 r α,Exc (n e , Te ) (4.10)
and
®
Dα,diss,D2 〈nD 〉 = n e n D2 r α,Diss(D2 ) (n e , Te ), (4.11)
2
where we considered the time and toroidally averaged D atom density, 〈n D 〉, the average D2
®
molecule density, n D2 . The resulting total Dα emission evaluated using the average density
of neutral species is therefore given by
Dα〈nn 〉 = Dα,Exc〈nD 〉 + Dα,diss,D2 〈nD 〉 + Dα,Diss(D+2 ) . (4.12)

2
105
In Fig. 4.10, snapshots of the Dα and Dα〈nn 〉 emission is presented, as well as their relative
difference, (Dα − Dα〈nn 〉 )/Dα . The contributions of the two excitation channels involving
neutral particles are also analysed. We highlight that Dα,Exc and Dα,Exc〈nD 〉 have similar profiles,
but discrepancies of the order of 20% − 30% can be seen in the near SOL and edge regions.
The difference becomes particularly important close to the core, where it reaches values
of the order of 50%. Similar considerations hold for the comparison between Dα,Diss(D2 )
and Dα,diss,D2 〈nD 〉 . Since the Dα emission is mostly determined by the contribution from
2
excitation of D atoms in the edge and near SOL, while the dominant component in the far
SOL, Dα,Diss(D+2 ) , does not depend on the density of neutral species, the plots for the total
Dα emission are similar to those obtained for Dα,Exc . Therefore, our results support the
conclusions expressed in Ref. [64], namely that neutral fluctuations have a non-negligible
impact on the Dα emission rates.
Figure 4.10: Snapshots of the Dα,Exc contribution (top), the Dα,Diss(D2 ) component (middle) and
the total Dα emission rate (bottom). The emission rate taking into account neutral fluctuations
(left), the same quantity calculated by neglecting the atom density fluctuations (middle), and
relative difference between the two (right) are plotted.
The analysis is further extended by evaluating the effect of neutral density fluctuations on the
106
statistical moments of the Dα emission, namely the standard deviation, skewness and kurtosis,
presented in Fig. 4.11, and also the properties of turbulence, i.e. the auto-correlation time,
τauto , the radial and poloidal correlation lengths, L rad and L pol , displayed in Fig. 4.12. We also
consider the same quantities from the contributions of the Dα emission, taking into account
and excluding neutral fluctuations. These quantities play an important role on turbulence
characterization and are often used as proxy to analyse the n e and Te fluctuations.
Figure 4.11: Radial profiles of the statistical moments of each component of the Dα emis-
sion rate, representing the standard deviation (left), skewness (middle) and kurtosis (right),
computed at the LFS equatorial midplane. The first row refers to the Dα,Exc contribution, the
second row presents results for the Dα,Diss(D2 ) component, the third row displays the plots for
the Dα,Diss(D+2 ) contribution and the fourth row refers to the total Dα emission.
We therefore start our analysis by considering the statistical moments of n e and Te , as well as
of n D , n D2 and n D+2 as these affect the properties of the D α emission. The standard deviation
of n e does not change considerably throughout the domain, being of the order of 30% its
equilibrium value, as expected from previous theoretical and experimental studies. In contrast,
the standard deviation of Te in the edge is about two to three times larger than in the SOL,
having a sharp gradient in the region around the LCFS. The standard deviation of n D+2 decreases
is considerably large at the edge-core interface, decreases slightly in the proximity of the core,
107
then it increases a bit close to the LCFS and decreases significantly in the SOL, becoming
negligible close to the outer wall. Focusing on the neutral species, we note that the standard
deviation of n D is large at the edge-core interface, decreases significantly in the edge, increases
slightly around the LCFS and remains approximately constant in the SOL, being comparable to
the standard deviation of Te . Finally, the standard deviation of n D2 at the edge-core interface
is much larger than the standard deviation of the other variables and decreases sharply across
the edge, becoming negligible in most of the SOL.
Turning to the analysis of the skewness, we remark that the skewness of Te is positive and
approximately constant over the whole domain, while n e and n D have negative skewness
in the edge and positive in the SOL. This is different from the result found in Ref. [64] for
the skewness of n D , which is yet another consequence of the introduction of molecular
dynamics. In fact, in the multi-component simulation, plasma fluctuations have different
impact on n D in the edge and SOL. In the SOL, large positive perturbations of n e and Te make
dissociation of D2 molecules more frequent and hence decisively increase n D , while large
negative perturbations do not change n D significantly, which justifies the positive skewness in
the SOL. In the edge, however, D atoms have a higher density than D2 molecules. Therefore,
large positive fluctuations of n e and Te increase the level of ionization of D atoms and hence
result in a sharp decrease of n D , while large negative plasma fluctuations do not lead to
equally important positive fluctuations of n D , thus justifying the negative values of skewness
observed in the edge region. For comparison, in the single-species simulations, since there
are no volumetric sources of n D , large positive plasma perturbations are always associated
with an increase of the ionization levels and consequent sharp decrease of n D , while large
negative perturbations have a less significant impact on n D , thus suggesting negative values of
skewness in the whole domain (in reality, the plots in Ref. [64] show that skewness is negative
in the SOL, where ionization of D atoms is important, and increases radially inwards, since n D
decreases and hence the effect of the plasma perturbations on ionization of D becomes less
important). The skewness of n D2 is positive in the edge and negative in the SOL, similarly to
what was reported for the skewness of n D in Ref. [64]. In addition, the kurtosis of n e , Te and
n D have a similar behaviour, being larger in the SOL compared to the edge region, remaining
between 0 and 5.
We now turn our attention to the Dα emission. Fig. 4.11 shows that the standard deviation
of the Dα,Exc emission rate is larger than the one of n e , Te and n D in most of the domain and
has a different radial dependence. This observation is related to the complex dependency of
the emissivity coefficient r α on n e and Te , which may result in a very different behaviour of
the statistical properties of Dα,Exc compared to n e and Te . It is also found that Dα,Exc〈nD 〉 is
20% to 30% smaller than Dα,Exc in the SOL, while this relation is reversed in the edge, where
Dα,Exc〈nD 〉 is about 20% larger. The relation between the two quantities in the SOL contrasts
with the one reported in Ref. [64], which is related to the fact that molecular dynamics are
now taken into account, thus affecting the impact of neutral fluctuations on the Dα emission.
We observe that the standard deviation of Dα,Exc〈nD 〉 is larger in the SOL compared to the
edge region. The skewness of Dα,Exc and Dα,Exc〈nD 〉 is of the same order of magnitude as the
108
skewness of n e , Te and n D in the edge, but grows radially and becomes much larger than
the skewness of the other quantities in the SOL, which is also not observed in Ref. [64] (we
remark that Dα,Exc is very small in the far SOL, where the Dα emission is mostly determined
by Dα,Diss(D+2 ) ). Moreover, we notice that Dα,Exc and Dα,Exc〈nD 〉 display similar behaviour in
the SOL, but different behaviour in the edge, especially close to the core. This agrees with
the conclusion reported in Ref. [64] on the effect of neutral perturbations on Dα emission,
confirming that the impact of n D fluctuations on Dα emission increases with the distance
from the gas puff. In fact, n D fluctuations have a strong influence on the qualitative behaviour
of the Dα,Exc emission in the edge and towards the core. The interpretation of the kurtosis is
identical to that of the skewness.
Figure 4.12: Radial profiles of the turbulence properties of each component of the Dα,Exc emis-
sion rate, representing the auto-correlation time (left), poloidal correlation length (middle)
and radial correlation length (right), computed at the LFS equatorial midplane. The first row
refers to the Dα,Exc contribution, the second row presents results for the Dα,Diss(D2 ) component,
the third row displays the plots for the Dα,Diss(D+2 ) contribution and the fourth row refers to
the total Dα emission.
Providing a small contribution to the overall Dα emission, Dα,Diss(D2 ) displays very similar
standard deviation, skewness and kurtosis as Dα,diss,D2 〈nD 〉 in the SOL, with the difference
2
between these quantities being larger in the edge, especially close to the core, which resembles
109
the results obtained in for the statistical moments of the Dα emission. As a matter of fact, while
Ref. [64] considers Dα emission generated by neutral atoms that are puffed into the system
and then ionized by the plasma, here the Dα,Diss(D2 ) component of Dα emission is similarly due
to D2 molecules puffed into the system and depleted via dissociation and ionization processes.
We also highlight that the large values of the statistical moments in the SOL around r = 50ρ s0 ,
and in particular the kurtosis, are related to the fact that Dα,Diss(D2 ) is very small in this region,
and hence large values are strongly influenced by statistical fluctuations.
In contrast to Dα,Exc and Dα,Diss(D2 ) , the statistical moments of Dα,Diss(D+2 ) are of the same order
as the values of the statistical moments of n e , Te and n D+2 . We remark that, since Dα,Diss(D+2 )
dominates over the other Dα emission components in the far SOL, the small values of the
standard deviation, skewness and kurtosis of Dα,Diss(D+2 ) in the SOL ensure that the values of
these statistical moments for the total Dα emission remain small. We also highlight that the
skewness and kurtosis of Dα,Diss(D+2 ) follow those of n D+2 in the edge, while the skewness and
kurtosis decrease in the SOL and become smaller than those of the other quantities.
Finally, we analyse the statistical moments of the total Dα emission, Dα , both including and
neglecting the fluctuations of the density of neutral species, D and D2 , and the statistical
moments of all plasma and neutral quantities involved. The standard deviation of Dα is
smaller than the standard deviation of Te at the edge-core interface, increases across the edge
region, peaks at the LCFS, decreases in the SOL and remains approximately constant in the
close proximity of the edge, being similar to the standard deviation of n e in this region. We
highlight that the radial profile of the standard deviation of Dα is very different from the one
of the standard deviation of both n e and Te , but it can still be used as a proxy of the standard
deviation of n e in the far SOL. This difference is once again related to the complex dependency
of the emissivity coefficient in Dα,Exc on n e and Te , and it is also partially explained by the
non-negligible contribution of Dα,Diss(D+2 ) , which depends mostly on the profile of n D+2 . In
fact, the standard deviation of Dα follows, to some extent, the same behaviour exhibited by
the standard deviation of n D+2 . Regarding the standard deviation of Dα〈nn 〉 , although it has
a similar profile to the one of the standard deviation of Dα , we note that it is slightly larger
than the standard deviation of Dα in the proximity of the core and smaller than it in the rest
of the domain, which makes it more similar with the standard deviation of Te . Therefore,
the standard deviation of Dα〈nn 〉 could, to some extent, be used as a proxy of the standard
deviation of Te , particularly in the SOL. Regarding the skewness of Dα , we remark that it
almost verifies in the close proximity of the core, decreases slightly radially outwards and then
increases significantly in the edge and near SOL, decreasing sharply in the far SOL, where it
becomes smaller than the skewness of both n e and Te . Since it has a very different behaviour
from the skewness of n e and Te , the skewness of Dα can hardly be envisaged as a proxy for
these quantities. In fact, it has a similar profile to the skewness of n D+2 , particularly close to the
outer wall, where the contribution from Dα,Diss(D+2 ) becomes dominant. On the other hand,
the skewness of Dα〈nn 〉 has a smoother profile than the skewness of Dα , remaining closer to
the skewness of Te in most of the domain, but its behaviour is still too different from the
profiles of the skewness of n e and Te for it to be used as a proxy of these quantities. Very
110
similar observations apply to the analysis of the kurtosis of Dα and Dα〈nn 〉 , which are small in
the close proximity of the core, increase significantly when approaching the LCFS and in the
near SOL, and then decrease sharply in the far SOL, with a similar behaviour to the kurtosis of
n D+2 . Although the kurtosis of Dα and Dα〈nn 〉 has a smoother profile than the kurtosis of Dα ,
both of them have a very different behaviour compared to the kurtosis of n e and Te , which
prevent them from being used as proxies for these quantities in most of the domain.
Fig. 4.11 also shows that the neutral density fluctuations do not affect significantly the statisti-
cal moments of Dα in the SOL, but the discrepancy between the standard deviation, skewness
and kurtosis of Dα and Dα〈nn 〉 is larger in the edge and displays different behaviour close to the
core. This reinforces the conclusion that the n D fluctuations can have an important impact
on the Dα,Exc emission, particularly in the edge, and thus must be taken into account when
interpreting experimental results from GPI diagnostics, as highlighted in Ref. [64].
Focusing now on the results of the analysis of the auto-correlation time and of the poloidal
and radial correlation lengths, presented in Fig. 4.12, we note that the auto-correlation time,
τα , is the time interval such that
1
C i i (τα ) = , (4.13)
2
with C i i the auto-correlation function between two signals at the same location separated
by a τα time interval. In turn, the radial and poloidal correlation lengths, L rad and L pol , are
evaluated as
δ
L = 1.66 q , (4.14)
−l nC i j
where C i j is the cross-correlation function between the signals measured at the same time
and at the positions labelled i and j , and δ the distance between these two positions. Similarly
to Ref. [64], we observe that there is no significant dependence of the correlation lengths on
the value of δ, as long as 1.5ρ s0 . δ . 10ρ s0 , and we choose δ = 2.9ρ s0 for the evaluation of the
radial correlation length and δ = 7.8ρ s0 for the evaluation of the poloidal correlation length.
The simulation shows that the fluctuations of the neutral densities have a stronger impact on
the turbulence properties in the edge region compared to the SOL. As a matter of fact, the auto-
correlation time and the poloidal and radial correlation lengths of Dα,Exc and Dα,Exc〈nD 〉 are
approximately the same in the SOL, but considerably different in the edge. A similar conclusion
holds for the comparison of the turbulence properties of Dα,Diss(D2 ) and Dα,diss,D2 〈nD 〉 . We
2
also note that the auto-correlation time and the poloidal and radial correlation lengths for
Dα,Diss(D+2 ) , display a similar behaviour compared to the profiles of Dα,Exc and Dα,Diss(D2 ) . The
turbulence properties of the total Dα emission, are therefore marginally affected by the neutral
density fluctuations in the SOL, but strongly impacted in the edge, as shown in Fig. 4.12.
111
Since the GPI diagnostic is used to infer the turbulence properties of the plasma quantities,
namely n e and Te , it is interesting to compare the auto-correlation time and correlation lengths
of the D α emission with the ones of n e and Te , which are also presented in Fig. 4.12. The plots
of τα show that the auto-correlation time of the D α emission rate is in general longer than
the one for n e . Regarding the auto-correlation time of Te , we note that Dα can be used as its
proxy in the edge, but not in the SOL. In fact, only in the proximity of the LCFS and close to
the wall, where the D α emission is mostly determined by the contribution of dissociation of
D+ 2 , the auto-correlation time of the D α emission follows the one of Te . On the other hand,
L rad and L pol based on the D α emission provide a good estimate for the same quantities for
n e , while it underestimates the Te measurements. The same conclusions hold for D α〈nn 〉 , thus
the correlation lengths of D α,〈nn 〉 can be used to infer L rad and L pol of the electron density.
4.6 Conclusion
In this chapter we present the results of a simulation of GPI diagnostics carried out with
the multi-component GBS model introduced in Ch. 3 with the goal of supporting a proper
interpretation of the D α emission in the context of GPI diagnostics, ultimately to allow for
an accurate reconstruction of the plasma and neutral profiles and a reliable inference of the
properties of plasma turbulence. This simulation extends the study reported in Ref. [64] to
the case of a multi-component deuterium plasma, thus taking into account the dynamics
associated with molecular species, D2 and D+ 2 . It also follows up on the work presented in Ref.
[46], that we extend by considering a three-dimensional plasma and simulating the neutrals
by means of a kinetic model valid for all values of the neutral mean free path. Analogously to
Ref. [64], a toroidally symmetric source of D2 molecules at wall temperature is considered at
the LFS equatorial midplane in order to mimic a diagnostic gas puff.
With respect to the analysis in Refs. [64] and [46] the contributions from dissociation of D2 and
D+2 species to the D α emission rate is taken into account, in addition to the contribution from
the excitation of D atoms. While the D α emission is mostly determined by the contribution of
atomic excitation in most of the domain, the dissociation of D+ 2 is dominant in the far SOL,
near the gas puff source, where the contribution from atomic excitation is residual. These
results suggest, for future work, to include the contribution from recombination processes
when studying the Dα emission, especially for low SOL temperatures.
The work presented in this chapter also sheds some light on the correlations between the
fluctuations of the components of the D α emission rate, the plasma quantities and the neutral
species densities, based on the application of the Spearman correlation function. In the
configuration and parameter regime considered here, it is possible to divide the diagnosed
volume in four regions, according to the correlations between the total D α emission and
the plasma and neutral quantities determining it. In the far SOL, D α emission is strongly
correlated with n e , but no significant correlation with Te or n D is reported, which is a result
of the D α emission being mostly determined by the contribution from dissociation of D+ 2
112
4.6. Conclusion
ions. In the near SOL and around the LCFS, D α emission is strongly correlated with n e , Te and
n D , being mostly determined by excitation of D atoms. On the other hand, radially inward
across the edge region, the D α emission becomes anti-correlated with the neutral density n D ,
as a result of ionization sinks dominating over dissociation sources in that region. Finally,
the very low density of neutral atoms in the proximity of the core results in the D α emission
being correlated with n D and strongly anti-correlated with the plasma quantities n e and Te .
However, the location and radial spread of the regions found is strongly dependent on the
plasma density and temperature considered in the simulation, as the whole dynamics is
expected to shift radially outwards if higher densities and temperatures are considered.
Regarding the impact of the neutral fluctuations on the D α emission rate, the simulation results
confirm the conclusions reported in Ref. [64], that the perturbations of the neutral quantities
affect significantly the D α emission, particularly in the edge and in the proximity of the core.
This is shown for the statistical moments of the D α emission rate and the related turbulence
properties, which highlight the increasingly important role of the neutral fluctuations as one
moves radially inwards.
113
5 Summary and conclusions
The boundary region of a tokamak plays a crucial role in determining the overall performance
of the device, as it regulates the exhaust of particle and energy to the vessel walls, controls
plasma fuelling, impurity levels and Helium ashes removal [12]. In the boundary, the plasma
dynamics is significantly affected by the presence of neutral particles, which are generated by
recycling processes taking place at the surrounding walls, external injection or volumetric re-
combination processes. The GBS code enables self-consistent three-dimensional simulations
of the neutral-plasma interaction in the tokamak boundary, by using a set of drift-reduced
Braginskii equations to model plasma turbulence and a kinetic advection equation for the neu-
tral species [61]. While the plasma equations are solved by using a standard finite-difference
scheme, the numerical solution of the neutral kinetic equation relies on the application of
the method of characteristics, with the integral over the neutral path being numerically dis-
cretized, while considering the τn ¿ τturb and k k λmfp,n ¿ 1 limits. Leveraging the GBS code,
the present thesis reports on the development of a mass-conserving multi-component self-
consistent plasma and neutral model and its first applications that consider a limited plasma
configuration, with a limiter on the equatorial HFS.
Ch. 2 presents the extension of the GBS model to achieve mass conservation to leading order
in ρ s /R 0 , detailing its numerical implementation and describing the first mass-conserving
simulation results. Mass conservation is achieved by consistently considering toroidicity
effects in the neutral and plasma models and by accounting for the radial variation of the
inverse aspect ratio ² = a 0 /R 0 . Moreover, the orderings k k /k ⊥ ¿ 1 and k ⊥ R 0 À 1 are avoided
and a set of boundary conditions implemented to make the plasma recycling processes at
the walls verify mass conservation. The mass-conserving properties of the whole model are
demonstrated by performing proper convergence tests. The results from mass-conserving
simulations differ from the ones obtained by using a non-mass-conserving model, as mass-
conserving simulations are characterized by a sharper transition between the closed and
open field-line regions, noticeable in the profiles of both the plasma and neutral quantities,
thus highlighting the impact of mass conservation on the simulation results. In addition,
quantitative studies of particle fluxes performed with the mass-conserving model show that
115
Chapter 5. Summary and conclusions
the ion particle flux is mostly determined by the E ×B drift, which balances the neutral particle
flux. Finally, when performing a parameter scan on the plasma density, higher densities
result in a weaker penetration of neutral particles across the edge and consequently a more
pronounced drop of the neutral density from the SOL to the edge region. As a result, the flux
of neutrals also drops more significantly from the LCFS to the core.
In Ch. 3, the mass-conserving self-consistent model of plasma turbulence and neutral dy-
namics is extended to include the molecular dynamics in the context of a deuterium plasma,
i.e. by considering a plasma composed of electrons, D+ ions, D atoms, D2 molecules and D+ 2
ions. As a result of the introduction of the molecular dynamics, a number of new collisional
processes are taken into account, including ionization, electron-neutral elastic collisions,
charge-exchange and dissociative processes, which give rise to new collisional terms appear-
ing in the equations describing both the plasma and the neutral dynamics. The plasma model
is extended by considering a set of drift-reduced Braginskii equations for the D+ 2 ion species,
that implement the Zhdanov closure. On the other hand, the neutral model is improved to
take into account the description of D2 molecules, which is done by solving a neutral kinetic
equation for the molecular species, in addition to the atomic species, thus resulting in a system
of two coupled linear integral equations for the atomic and molecular densities.
The results from the first simulation of a multi-component plasma using GBS are presented.
The density profiles show a noticeable up-down asymmetry, which ultimately results in
larger recycling rates and higher density of neutrals above than below the limiter plates.
The simulation shows that the density of D+ 2 ions peaks inside the LCFS in the proximity of
the limiter and is between three and four orders of magnitude smaller than the density of
the main ion species, D+ . In addition, the role of D2 molecules on plasma fuelling is also
addressed. It is observed that most of the D2 particles generated in the SOL cross the LCFS
and are dissociated into D atoms in the edge. Both D atoms and D2 molecules are ionized
due to the higher plasma densities and temperatures in the edge, where D+ 2 ions have very
short lifetimes, being rapidly dissociated. An interesting result from the simulations is the shift
of the ionization of D atoms radially inward with respect to the results from single-species
simulations, as a result of the significant generation of D atoms due to the dissociation of
D2 molecules inside the LCFS. As a result of the dissociation of D2 in the region around the
LCFS and the association of D atoms at the limiter and vessel walls, the density of D atoms
peaks in the SOL region close to the LCFS, where the radial flux of D atoms reverses sign, as
opposed to the monotonic increase of the density of D atoms towards the wall observed in
single-component GBS simulations.
Ch. 4 reports on the investigation of light emission in GPI performed by exploiting the self-
consistent multi-component model developed in Ch. 3. The research presented in this
Ch. 4 constitutes a follow-up of the studies reported in Refs. [64] and [46], by featuring
three-dimensional simulations of a mass-conserving multi-component deuterium plasma
in the tokamak boundary in the presence of a diagnostic gas puff located at the equatorial
outboard midplane. The molecules puffed at wall temperature into the domain interact with
116
the boundary plasma, and this interaction results also in light emission. The studies presented
in this chapter focus on the Dα line emission and take into account the contributions from
excitation of D atoms and dissociative processes acting on D2 molecules and D+ 2 molecular
ions, with the component associated with atomic excitation being dominant in most of the
domain, except in the proximity of the gas puff source, where dissociation of D+ 2 ions is
dominant. The simulation results reveal that the fluctuations of the neutral density have a
significant impact on the D α emission rate in the edge, which is supported by the analysis of
the radial profiles of the standard deviation, the skewness, the kurtosis, the auto-correlation
time and the radial and poloidal correlation lengths.
The results presented in Ch.4 also highlight that the simulation domain can be split radially
into four different regions, according to the correlation functions between the D α emission
rate and the plasma and neutral quantities determining it, namely the electron density, n e ,
the electron temperature, Te , and the density of neutral atoms, n D . More precisely, a strong
correlation is observed between the D α emission and n e in the far SOL, but D α emission is
found not to be strongly correlated with Te or n D . In the near SOL and around the LCFS, D α
emission is strongly correlated with n e , Te and n D . In most of the edge region, D α is also
strongly correlated with both n e and Te , but anti-correlated with n D , since dissociation of
D2 molecules dominates over ionization of D atoms in the SOL, while ionization becomes
dominant in the edge. Finally, towards the core region, the very low neutral density results
in the D α emission being mostly determined by n D , which in turn translates into a strong
correlation between D α and n D , and significant anti-correlations between D α and both n e
and Te . These results provide useful insights on the interpretation of the D α emission in the
context of GPI diagnostics, as it can guide the interpretation of plasma and neutral quantities
from light emission measurements.
As a final note, we remark that the studies presented in this thesis constitute a step forward to-
wards the realistic self-consistent simulation of the complex dynamics in the boundary region
of a tokamak. While mass-conservation is an essential feature for the accurate simulations
of the neutral-plasma interaction, the ability to simulate multi-ion species and to address
the dynamics of several neutral particles is considered a key element to study relevant issues
concerning the physics of boundary plasmas, such as fuelling and GPI diagnostics.
The results published in this thesis focus on a limited plasma. The extension to diverted
configurations, more widely used in nowadays tokamaks given their advantages on reducing
the heat load on the plasma facing components, should be relatively straightforward. Since
divertor plates are placed, with respect to the limiter plates, further away from the tokamak
core, the inclusion of neutral dynamics is expected to be even more crucial in a diverted
geometry. The implementation of the multi-species model in diverted configurations will also
allow for the conduction-limited and detachment regimes to be addressed by GBS simulations.
The later will most likely require the inclusion of impurities, which in turn will call for an
extension of the model to include new charged and neutral species, by following a similar
approach to the one presented in the present thesis.
117
A Proof of mass conservation in the
formal solution of the neutral kinetic
equation
We prove that the formal solution of Eq. (2.47), presented in Eq. (2.51), satisfies the conserva-
tion relation
∇ · Γn = −n n νiz . (A1)
For the sake of simplicity, we consider an infinitely extended domain (no boundary conditions)
and plasma related quantities constant. These assumptions lead to an ion velocity distribution
constant in space, µi = µi (v), and constant collisional frequencies (νcx , νiz = const.). The
extension of the proof to the most general case is straightforward.
We start by noting that the neutral flux in Eq. (A1) is given by Γn = d 3 v f n vp in the poloidal
R
plane (the component of the neutral flux along the toroidal direction is not considered in the
λm f p,n k k ¿ 1 and ¯∇ψ × ∇ϕ¯ ¿ ¯F (ψ)∇ϕ¯ limits studied here). Making use of Eq. (2.48) to
¯ ¯ ¯ ¯
express f n , the neutral flux can be written as
∞ 2π ∞ ∞ R(xp0 ) νeff 0
Z Z Z Z µ ¶
0
Γn (xp ) = d vp vp dϑ d vt dr S(xp0 , v)exp − r (cosϑeR + sinϑeZ ),
0 0 −∞ 0 R(xp ) vp
(A2)
where S(xp0 , v) = n n (xp0 )νcx µi (v) within the hypothesis made here, and the neutral velocity vp
is decomposed in (R, Z ) coordinates as vp = v p (CosϑeR + SinϑeZ ).
We remark that the xp and xp0 locations are identified by their (R, Z ) coordinates, and they are
related by the distance r 0 and the angle ϑ as
R(xp0 ) = R(xp ) − r 0 cosϑ (A3)
and
119
Appendix A. Proof of mass conservation in the formal solution of the neutral kinetic
equation
Z (xp0 ) = Z (xp ) − r 0 sinϑ. (A4)
Eqs. (A3-A4) can be inverted to obtain the definitions of r 0 and ϑ as a function of R and Z , that
is
r 02 = (R(xp ) − R 0 (xp0 ))2 + (Z (xp ) − Z 0 (xp0 ))2 , (A5)
R(xp ) − R 0 (xp0 )
cosϑ = , (A6)
r0
and
Z (xp ) − Z 0 (xp0 )
sinϑ = . (A7)
r0
We note that, for a given location xp0 described by coordinates (R(xp0 ), Z (xp0 )), the location of
the neutral particle source xp can be described by the variables r 0 and ϑ, according to Eqs.
(A3-A4). Therefore, (R(xp0 ), Z (xp0 ), r 0 , ϑ) must be regarded as four independent variables that
fully describe each pair of source and target points.
We now apply the divergence operator to the neutral flux in Eq. (A2) at the xp location. This
can be written in (R, Z ) coordinates as
∂Γn,R ∂Γn,Z
∇ · Γn (xp ) = + . (A8)
∂R(xp ) ∂Z (xp )
Since the variables r 0 and ϑ are independent of R(xp ) and Z (xp )), the R, Z -dependence in Eq.
(A2) comes exclusively from the source S(xp0 , v), namely through the neutral density n n (xp0 ),
and from the geometric factor R(xp0 )/R(xp ). We then make use of Eqs. (A3) and (A4) and apply
the chain rule (at constant r 0 and ϑ) to write ∂/∂R(xp0 ) = ∂/∂R(xp ) and ∂/∂Z (xp0 ) = ∂/∂Z (xp ).
By taking the derivatives with respect to the R, Z coordinates of the source location xp0 , we
obtain
Z ∞ Z 2π Z ∞ Z ∞
∇ · Γn (xp ) = d vp vp dϑ d vt d r 0 νcx µi (v)
Ã 00 !0 −∞
Ã0
R(xp0 )
" ! #
∂ R(xp ) ∂ νeff 0
µ ¶
0 0
n n (xp ) cosϑ + n n (xp ) sinϑ exp − r . (A9)
∂R(xp0 ) R(xp ) ∂Z (xp0 ) R(xp ) vp
120
Since the integrals are performed with respect to r 0 and ϑ, it is convenient to express the
derivatives appearing in Eq. (A9) in terms of these variables. Again, by using the chain rule, we
obtain from Eqs. (A5-A7)
∂n n ∂r 0 ∂n n ∂ϑ ∂n n ∂n n sinϑ ∂n n
= + = −cosϑ 0 + 0 , (A10)
∂R(xp ) ∂R(xp ) ∂r
0 0 0 ∂R(xp ) ∂ϑ
0
∂r r ∂ϑ
∂n n ∂r 0 ∂n n ∂ϑ ∂n n ∂n n Cosϑ ∂n n
= + = −Sinϑ 0 − , (A11)
∂Z (xp ) ∂Z (xp ) ∂r
0 0 0 ∂Z (xp ) ∂ϑ
0
∂r r 0 ∂ϑ
Applying Eqs. (A10) and (A11) into Eq. (A9) yields
0
Ã !
∞ 2π ∞ ∞ ∂ R( x ) νeff 0
Z Z Z Z µ ¶
0 p 0
∇ · Γn (xp ) = − d vp vp dϑ d vt d r νcx µi (v) 0 n n (xp ) exp − r .
0 0 −∞ 0 ∂r R(xp ) vp
(A12)
Finally, the integral along r 0 appearing in Eq. (A12) is computed via integration by parts,
yielding
0
Ã !
∞ ∂ R(xp ) νeff 0
Z µ ¶
0 0
dr n n (xp ) exp − r
0 ∂r 0 R(xp ) vp
Z ∞ R(xp0 ) 0 νeff νeff 0
µ ¶
0
= −n n (xp ) + dr n n (xp ) exp − r (A13)
0 R(xp ) vp vp
Using Eq. (A13) to develop Eq. (A12), we can write
Z ∞ Z 2π Z ∞
∇ · Γn (xp ) = νcx n n (xp ) d vp vp dϑ d v t µi (v)
0 0 −∞
(A14)
∞ 2π ∞ Z ∞ R(xp0 ) νcx n n (x p0 )µi (v) νeff 0
Z Z Z µ ¶
0
− νeff d vp vp dϑ d vk dr exp − r .
0 0 −∞ 0 R(xp ) vp vp
R∞ R 2π R∞
Then, since 0 d v p v p 0 d ϑ −∞ d v t µi (v) = 1, the first term on the right-hand side of Eq.
(A14) yields simply νcx n n (xp ). For the second term, one recognises the neutral density at the
xp location as given by Eq. (2.51). Therefore,
∇ · Γn (xp ) = νcx n n (xp ) − νeff n n (xp ). (A15)
Since νeff = νcx + νiz , Eq. (A1) is recovered.
121
B Evaluation of the average electron
energy loss and reaction product
energies in collisional processes
The Franck-Condon principle [113, 114] states that electronic excitation occurs over a timescale
considerably shorter than the characteristic timescale associated with vibration or dissociation
of the diatomic species. In turn, the vibration or dissociation timescales are much shorter
than the electron deexcitation timescale. As a result, when an electron impacts a D2 molecule
or a D+
2 ion, an electronic excitation is observed with no significant change in the inter-atomic
distance (vertical transition). If the excited state is not stable, the molecule dissociates before
deexcitation takes place. In this case, the difference between the excitation energy and the dis-
sociation energy is converted into kinetic energy of the products (ionization and dissociative
energies are discussed in [115]). We note that the exact energies of the products of dissociation
reactions depend on the vibrational level of the D2 molecule or D+ 2 ion. Considering the
excitation of a D2 molecule in a given initial state, the set of vibrational levels accessible for the
molecule in the final state are the ones lying within the region of the potential energy surface
accessed by that particular vertical transition, known as the Franck-Condon region. The mean
energy of the reaction products is thus the average over the Franck-Condon region, taking into
account all accessible vibrational states.
In the present thesis, we model the products of dissociative reactions by considering that they
are reemitted isotropically in the reference frame of the incoming massive particle (D2 or D+ 2 ),
thus approximating their velocity distribution as a Maxwellian centered at the velocity of the
incoming D2 or D+ 2 . The temperature of the Maxwellian, together with the average electron
energy loss for each process, are obtained from the values presented in [85]. Since these
energies depend on the intermediate excited state of the D2 or D+ 2 particle, different values
are found for different channels within the same dissociative process. This requires that an
average is performed over all possible excited states, taking into account the respective cross
section of each process. We present these calculations in detail for each process, following
[85].
The energy loss and the energy of the reaction products may depend on the electronic levels (n)
and sub-levels (l ) of the reaction products, on the molecular orbital (MO) of the intermediate
state, if bonding or antibonding, and on the energy of the incident electron. The energy
123
Appendix B. Evaluation of the average electron energy loss and reaction product energies
in collisional processes
values are experimentally determined for all relevant dissociation channels. These quantities
are then averaged over all vibrational states v of the D2 molecules or D+ 2 ion and over the
Franck-Condon region, from [85].
We start by considering the dissociation of D2 molecules, i.e.
e− + D2 → e− + D + D. (B.1)
For this reaction, the values of the electron energy loss, 〈∆E e 〉, and reaction product energies,
〈E D 〉, depend significantly on the electronic state of the products. Hence, considering that
there are i = 1, ..., N electronic states of the reaction products and, associated, N different
sub-processes contributing to the dissociation of D2 , the average electron energy loss 〈∆E e 〉
is obtained by performing a weighed average of 〈∆E e 〉i , the energy loss for the sub-process i ,
based on the 〈σv〉i reaction rate, yielding
ΣiN=1 〈σv〉i 〈∆E e 〉i

£ ¤
〈∆E e 〉 = , (B.2)
ΣiN=1 [〈σv〉i ]
For simplicity, we evaluate all quantities at the reference temperature, Te = 20eV. Similarly,
the average value for the energy of the reaction products is obtained as
ΣiN=1 〈σv〉i 〈E D 〉i
£ ¤
〈E D 〉 = , (B.3)
ΣiN=1 [〈σv〉i ]
with 〈E D 〉i the average energy of the products for the sub-process i .
The values of 〈σv〉i , 〈∆E e 〉i , 〈E D 〉i are presented in Table B.1 for all sub-processes. The ad-
ditional information between brackets refers to the minimum and maximum of the range
of energies accessible to 〈∆E e 〉i and 〈E D 〉i , following the values listed in [85]. We highlight
that D(1s) denotes a D atom in the fundamental state (electron at the lowest orbital 1s), while
D∗ (2s) and D∗ (2p) denote an atom in the excited state n = 2 with the electron in an orbital of
type s or p, respectively, and D∗ (n = 3) represents an atom in the excited state n = 3. Following
[85], we assume that the energy is equally distributed over the reaction products, regardless of
the fact that their electronic states are the same. Based on the values in Table B.1, from Eqs.
(B.2) and (B.3), we obtain 〈∆E e 〉 ' 14.3eV and 〈E D 〉 ' 1.95eV, respectively, at Te = 20eV. These
are the values mentioned in Table 3.2.
Focusing now on the dissociative-ionization of D2 ,
e− + D2 → D + D+ + 2e− , (B.4)
124
Table B.1: 〈σv e 〉 product, average electron energy loss and average energy of reaction products
for each sub-process of D2 dissociation.
Reaction 〈σv e 〉i 〈∆E e 〉i 〈E D 〉i

e− + D2 → e− + D(1s) + D(1s) 3.8 × 10−9 cm3 /s 10.5eV 3eV
e− + D2 → e− + D(1s) + D∗ (2s) 5.3 × 10−9 cm3 /s 15.3eV 0.3eV
e− + D2 → e− + D∗ (2p) + D∗ (2s) 9.2 × 10−10 cm3 /s 34.6eV 4.85eV
e− + D2 → e− + D(1s) + D∗ (n = 3) 5.7 × 10−10 cm3 /s 21.5eV 5.7eV
we consider three cases. If the incoming electron has an energy E e < E th(g) , with E th(g) = 18eV,
no dissociation takes place. If E th(g) < E e < E th(u) , with E th(u) = 26eV, the electron can ionize
the molecule, resulting in an unstable D+ 2 ion, which then dissociates into a D atom and a D
+
ion. The short-lived D+2 has the electron in a bonding molecular orbital (MO) with σ-symmetry,
thus exhibiting gerade (g) symmetry (German for even) state, denoted as D+ 2 (Σg ). If E e > E th(u) ,
the intermediate D+ 2 ion has the electron in a higher-energy antibonding MO with σ-symmetry,
which exhibits ungerade (u) symmetry (German for odd), thus denoted as D+ 2 (Σu ). As a result
+
of the different energy levels of the intermediate D2 ion, the energy of the final products
will also be different, as well as the average electron energy loss. According to the results
presented in [85], these energies still depend on the energy of the incoming electron within
each sub-process. To simplify the evaluation of the 〈∆E e 〉 and the energy of the products,
we consider the energy to be evenly distributed by the reaction products (D and D+ ) and we
consider the two cases separately. For E th(g) < E e < E th(u) , all dissociative-ionization events
originate an intermediate state D+ 2 (Σg ), while for E e > E th(u) all events generate an intermediate
+
state D2 (Σu ). The values for the electron energy loss and reaction product energies being
considered for each case are evaluated for [85] and listed in Table B.2. We note that this is
just an approximation, as even with Te < E th(u) there are electrons with energies superior to
the threshold that will generate a D+ +
2 ion in a D2 (Σu ) state, and vice-versa. Nevertheless, this
approximation avoids us to evaluate 〈∆E e 〉 and 〈E D 〉 at every single value of Te .
For the dissociation of D+

2 , i.e.
e− + D + + −
2 → D +D+e , (B.5)
different sub-processes are taken into account, following an approach similar to the one
Table B.2: Average electron energy loss and average energy of reaction products for the two
cases of dissociative-ionization of D2 .
®
Reaction 〈∆E e 〉 〈E D 〉 = E D+
e− + D 2 → e− + D + (Σg ) + e− → D + D+ + 2e−
£ ¤
2 18.25eV 0.25eV
e− + D 2 → e− + D + −
→ D + D+ + 2e−
£ ¤
2 (Σu ) + e 33.6eV 7.8eV
125
Appendix B. Evaluation of the average electron energy loss and reaction product energies
in collisional processes
®
Reaction 〈σv e 〉i 〈∆E e 〉i 〈E D 〉 = E D+ i
e− + D + +
2 → D + D(1s) + e
−
1.2 × 10−7 cm3 /s 10.5eV 4.3eV
e− + D + + ∗
2 → D + D (n = 2) + e
−
1.0 × 10−7 cm3 /s 17.5eV 1.5eV
Table B.3: 〈σv e 〉 product, average electron energy loss and average energy of reaction products
for each sub-process of D+ 2 dissociation.
adopted to treat the dissociation of D2 . We perform a weighed average of the electron energy
loss and the reaction products energy by using Eqs. (B.2) and (B.3), respectively. The values of
®
〈σv e 〉i , 〈∆E e 〉i and 〈E D 〉 = E D+ i for each sub-process are presented in Table B.3. The weighed
averaged values for the electron energy loss and reaction products energy at the reference
®
temperature, Te = 20eV, yield 〈∆E e 〉 = 13.7eV and 〈E D 〉 = E D+ = 3.0eV, as listed in Table B.3.
Regarding the dissociative-ionization of D+

2 , i.e.
e− + D + + + −
2 → D + D + 2e , (B.6)
we follow [85], where the average energy of the resulting D+ ions is obtained from an average
performed over all vibrational states (v = 0 − 9) of the D+
2 ion and over the Franck-Condon
®
region. This yields E D = 0.4eV, while the average electron energy loss is 〈∆E e 〉 = 15.5eV.
+
We finally focus on the dissociative-recombination of D+

2 , which generates a D atom in the
fundamental state (electron in orbital 1s) and a D atom in an excited state (electron with
principal quantum number n ≥ 2), i.e.
e− + D + ∗
2 → D(1s) + D (n ≥ 2). (B.7)
We assume that the energy of the products is evenly distributed among the two D atoms and
is given by
µ ¶
® ® 1 Ry
E D(1s) ' E D (n≥2) '
∗ Ee + 2 , (B.8)
2 n
with Ry = 13.6eV the Rydberg unit of energy (corresponding to the electron binding energy in a
hydrogen atom in the fundamental state). Since this expression depends on the energy of the
incoming electron, E e , and the electronic level n of the excited atom, D∗ , we assume an energy
of the incident electron of E e ' 20eV, the typical value in the region around the LCFS at the
HFS, and consider that these atoms are most likely in the accessible state of lowest energy n = 2
(considering a higher excited state would not change the value of the energy of the products
® ®
by a significant amount). Under these assumptions, we get E D(1s) ' E D∗ (n≥2) ' 11.7eV.
126
C Zhdanov collisional closure
We focus on the derivation of the parallel friction forces and the parallel heat fluxes, denoted
respectively by R kα = Rα · b and q kα = qα · b for a given species α, with Rα = m α v0C α d v
R
and qα = (m α v 02 )/2v0 f α d v, where we introduce the thermal component of the velocity,

R
v0 = v − vα , with vα = v f α d v the fluid velocity of the α species, and the collision operator
R
C α = ΣβC αβ ( f α , f β ), with C αβ describing collisions of species α with species β. We consider

the collisional closure derived by Zhdanov in [65], relying on the approach proposed in [66]
and discussed in [93] for its numerical implementation.
Following [65], the parallel component of the friction forces and heat fluxes of the species α is
related to the parallel gradients of the temperature and parallel velocity of all species through
" # " #
q kα X ∇k Tβ
= Zαβ , (C.1)
R kα β w kβ
where Tβ denotes the temperature of plasma species β and w kβ is the parallel component of
the fluid velocity of species β with respect to the center of mass of the plasma, wβ = vβ − vCM ,
¡P ¢ ¡P ¢
with vCM = β n β m β vβ / β n β m β . The matrix Zαβ relates the parallel heat fluxes and
friction forces with the parallel gradients of temperature and parallel velocity. We remark that
Eq. (C.1) simplifies the general result obtained by Zhdanov [65] to the case of singly-ionized
states, neglecting possible multiplicity of charge states for the chemical species present in the
plasma.
In order to compute the matrix Zαβ , we consider the 21N -moment approximation of the
distribution function [65], thus including the moments up to the fifth order moment. We
first express Rα and qα in terms of these moments of the distribution function, namely the
first order moment, wα , the third order moment, hα = q kα , and the fifth order moment,
rα = mα /4 (c 4 − 14c 2 /γα + 35γα )c f α d c, where we introduce the velocity with respect to
R
the center of mass of the plasma, c = v − vCM , and the parameter γα = m α /(kTα ), with
Tα = (m α v 02 /2) f α d v. Since only the expressions for the parallel component of the friction
R
127
Appendix C. Zhdanov collisional closure
forces and heat fluxes are needed, we consider only the parallel component of these equations.
The heat flux, q kα , simply corresponds to the third order moment, h kα , while the friction
forces, R kα , are obtained in terms of w kα , h kα and r kα [116], yielding
q kα = h kα , (C.2)
¢ µαβ (2) h kα h kβ µαβ 2 (8) rα rβ

· µ ¶ µ ¶ µ ¶¸
(1) ¡
X
R kα = G αβ w kα − w kβ + G − + G αβ − .
β kT αβ m α n α m β n β kT mα nα mβ nβ
(C.3)
where m α and n α are respectively the mass and density of species α, µαβ = (m α m β )/(m α +m β )
(n)
is the reduced mass, and G αβ are polynomial functions of the local plasma density and
temperature, their exact expressions being presented in [65] (chapter 8.1, pp. 163-164). Eqs.
(C.2) and (C.3) can then be written in matrix form as
" # " # " #

q kα X h kβ X ∇k Tβ
= A αβ + B αβ , (C.4)
R kα β r kβ β w kβ
where the matrices A and B are defined to satisfy Eqs. (C.2) and (C.3). We now aim at
expressing the moments hα and ∇r α in terms of wα and ∇Tα . This can be achieved by solving
a system of moment equations similar to the one presented in [65] (chapter 8.1, pp. 162-163),
including the time evolution of the moments (wα , hα and ∇r α ) and the time evolution of basic
thermodynamic variables (ρ, vCM and T ). We neglect time derivatives and nonlinear terms.
For simplicity, we also assume that, for two massive particle species D+ and D+ 2 , the condition
|TD+2 − TD+ | ¿ TD+2 is fulfilled, which allows us to write TD+2 = TD+ = T . Moreover, as long as
Te /TD+ À m e /m D+ is verified, T can also be replaced by Te , following [65] (the simulation
results shown in Fig. 3.3 meet these conditions). We therefore impose TD+2 = TD+ = Te = T ,
while no assumption is made on the temperature and pressure gradients, i.e. temperature
gradients can be different from species to species [65].
The parallel projection of the system of moment equations can then be written as (see [116])
X 5 µαβ (2) ¡ (5) h kα

·
5 ¢
n kα k∇Tα = G αβ w kα − w kβ +G αβ
2 β 2 mα pα
(C.5)
(6) h kβ µαβ (9) r kα (10) r kβ
µ ¶¸
+G αβ + G αβ +G αβ ,
pβ kT pα pβ
128
X 35 µαβ 2 (8) ¡ µαβ (9) h kα (10) h β
· µ ¶ µ ¶
¢
0= G αβ w kα − w kβ + 7 G αβ +G αβ
β 2 mα mα pα pβ
(C.6)
m α (11) r kα m β (12) r kβ
¸
+ G + G ,
kT αβ p α kT αβ p β
where p α is the pressure of species α. Rewriting Eqs. (C.5-C.6) in matrix form, one obtains
" # " #
X ∇k Tγ X h kβ
P αγ = M αβ , (C.7)
γ w kγ β r kβ
which can be inverted to express the parallel third and fourth order fluid moments in terms of
the parallel gradient of temperature and relative parallel velocity as
" # " #
h kβ X X −1 ∇k Tγ
= M αβ P αγ . (C.8)
r kβ α γ w kγ
Finally, making use of Eq. (C.8) to express h kα and r kα in Eq. (C.4) in terms of the parallel
temperature gradients and relative velocities, one obtains the expressions for the parallel heat
flux and friction forces in the matrix form presented in Eq. (C.1), that is
" # " #
q kα ³
−1
´ ∇ T
k β
= A αλ M γλ P γβ + B αβ . (C.9)
R kα w kβ
Since the matrices A, B , P and M are fully determined by Eqs. (C.2), (C.3), (C.5) and (C.6),
the expressions of the parallel heat flux and friction forces can be found. Following Zhdanov
[65], these matrices can be expressed in terms of the local values of plasma quantities, namely
densities n e , n D+ and n D+2 and temperatures Te and TD+ (we again assume TD+2 = TD+ , mass
ratios and characteristic time scales τeD and τDD , with τeD defined as the inverse of the
collision frequency for momentum transfer between electrons and D+ ions, and τDD the ion
timescale defined as the inverse of the collision frequency for momentum transfer between
p
D+ ions. We retain only terms of leading order in m e /m D , while terms proportional to the
fast electron timescale τeD are neglected when compared to terms proportional to τDD , which
considerably simplifies the final expressions. We also highlight that, besides imposing the
quasi-neutrality relation n e = n D+ + n D+2 , we take into account the fact that the density of
the molecular ion species is much smaller than the density of the main ion species D+ for
typical tokamak boundary conditions, i.e. n D+2 /n D+ ¿ 1, keeping therefore only leading order
terms in n D+2 /n D+ . As a result, the friction forces between molecular ions and other species
are neglected, as well as molecular ion temperature gradient terms, while friction and thermal
force contributions involving D+ and e− species are kept in the expressions of the parallel
components of the heat fluxes and friction forces. The expressions obtained for the friction
129
Appendix C. Zhdanov collisional closure
forces and heat fluxes finally yield
3.16n e Te τeD
q ke = − ∇k Te + 0.71n e Te (v ke − v kD+ ),
me
4.52n e TD+ τDD
q kD+ = − ∇k TD+ ,
mD
1.80n e TD+ τDD
q kD+2 = − ∇k TD+ ,
mD (C.10)
0.51m e n e
R ke = −0.71n e ∇k Te − (v ke − v kD+ ),
τeD
0.51m e n e
R kD+ = 0.71n e ∇k Te − (v kD+ − v ke ),
τeD
R kD+2 = 0,
The expressions in Eqs. (C.10) can be simplified by applying the relation between the electron
and ion characteristic times,
τDD m D Te mD
µ ¶
1 1
r r
=p ∼p , (C.11)
τeD 2 m e TD + 2 me
having again assumed TD+ ∼ Te . This enables one to write τDD appearing in Eq. (C.10) in
terms of τeD . Following Braginskii’s approach [17] and considering that the the electron
characteristic time is τe = τeD , we then write Eqs. (C.10) in terms of the resistivity, defined as
[33, 34]
me R0 1
ν = 0.51 , (C.12)
m D c s0 n e τeD
The parallel friction forces and heat fluxes, as they appear in Eqs. (3.22-3.24) and Eqs. (3.25-
3.27), respectively, are therefore written in normalized units as
R ke = −0.71n e ∇k Te − νn e (v ke − v kD+ ),
R kD+ = 0.71n e ∇k Te − νn e (v kD+ − v ke ),
R kD+2 = 0,
1.62
q ke = − n e Te ∇k Te + 0.71n e Te (v ke − v kD+ ), (C.13)
ν
2.32 m e
r
q kD+ = − p n e TD+ ∇k TD+ ,
2ν m D
0.92 m e
r
q kD+2 = − p n e TD+ ∇k TD+ .
2ν m D
We note that, similarly to the single-ion species model implemented in GBS [33], the ohmic
130
heating terms are neglected.
131
D List of kernel functions
The kernels used in Eqs. (3.80-3.83) for n D2 , Γout,D2 , n D and ΓD are defined as
D ,D+ 0 D ,D+ 0 D ,D+ 0

2
K p→p 2
(x⊥ , x⊥ 2 2
2 2
(x⊥ , x⊥ ), (D.1)
D ,reem 0 D ,reem 0 D ,reem 0

2
K b→p (x⊥ , x⊥b 2
) = K b→p,dir (x⊥ , x⊥b ) + αrefl K b→p,refl
2
(x⊥ , x⊥b ), (D.2)
D ,refl 0 D ,refl 0 D ,refl 0

2
K b→p (x⊥ , x⊥b 2
) = K b→p,dir (x⊥ , x⊥b ) + αrefl K b→p,refl
2
(x⊥ , x⊥b ), (D.3)
D ,D+ 0 D ,D+ 0 D ,D+ 0

2
K p→b 2
(x⊥b , x⊥ 2 2
) = K p→b,dir (x⊥b , x⊥ ) + αrefl K p→b,refl
2 2
(x⊥b , x⊥ ), (D.4)
D ,reem 0 D ,reem 0 D ,reem 0

2
K b→b (x⊥b , x⊥b 2
) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl
2
(x⊥b , x⊥b ), (D.5)
D ,refl 0 D ,refl 0 D ,refl 0

2
K b→b (x⊥b , x⊥b 2
) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl
2
(x⊥b , x⊥b ), (D.6)
133
Appendix D. List of kernel functions
+ + +
D,D 0 D,D 0 D,D 0
K p→p (x⊥ , x⊥ ) = K p→p,dir (x⊥ , x⊥ ) + αrefl K p→p,refl (x⊥ , x⊥ ), (D.7)
D,D+ 0 D,D+ 0 D,D+ 0

K p→p2 (x⊥ , x⊥ 2
2
(x⊥ , x⊥ ), (D.8)
D,diss(D+
2) 0 D,reem 0 D,reem 0
K p→p (x⊥ , x⊥ ) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl (x⊥b , x⊥b ), (D.9)
D,diss-rec(D+
D,diss(D2 ) 0 D,reem 0 D,reem 0

D,diss-iz(D2 ) 0 D,reem 0 D,reem 0

D,reem 0 D,reem 0 D,reem 0

K b→p (x⊥ , x⊥b ) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl (x⊥b , x⊥b ), (D.13)
D,refl 0 D,reem 0 D,reem 0

K b→p (x⊥ , x⊥b ) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl (x⊥b , x⊥b ), (D.14)
+
D,D 0 D,reem 0 D,reem 0
K p→b (x⊥b , x⊥ ) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl (x⊥b , x⊥b ), (D.15)
D,D+ 0 D,reem 0 D,reem 0

K p→b2 (x⊥b , x⊥ ) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl (x⊥b , x⊥b ), (D.16)
134
D,diss(D+
D,diss-rec(D+
D,diss(D2 ) 0 D,reem 0 D,reem 0

D,diss-iz(D2 ) 0 D,reem 0 D,reem 0

D,reem 0 D,reem 0 D,reem 0

K b→b (x⊥b , x⊥b ) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl (x⊥b , x⊥b ), (D.21)
D,refl 0 D,refl 0 D,reem 0

K b→b (x⊥b , x⊥b ) = K b→b,dir (x⊥b , x⊥b ) + αrefl K b→b,refl (x⊥b , x⊥b ), (D.22)
where the kernel functions for a given path = {dir, refl} are defined as
Z 0
∞ 1 r⊥
· ¸
1
Z
D2 ,D+ 0 0 00 00
0 Φ⊥ v + ,T + (x⊥ , v⊥ )exp − νeff,D2 (x⊥ )d r ⊥ d v ⊥ ,
2
K p→p,path (x⊥ , x⊥ )= h i
0 r⊥ ⊥,D
2
D
2
v⊥ 0
(D.23)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D2 ,reem 0 0 0 00 00
K b→p,path (x⊥ , x⊥b )= 0 cosθ χ⊥,in,D2 (x⊥b , v⊥ )exp − νeff,D2 (x⊥ )d r ⊥ d v ⊥ ,
0 r⊥ v⊥ 0
(D.24)
135
Z 0
∞ 1 r⊥
· ¸
1
Z
D ,refl 0 i (x0 , v)exp − 00 00
2
K b→p,path (x⊥ , x⊥b )= 0 Φ h ν (
eff,D2 ⊥x )d r ⊥ d v⊥,
r⊥ ⊥ vrefl D+ ,TD+ v⊥ 0
0 ( 2) 2
(D.25)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D ,D+ 0 i (x0 , v )exp − 00 00
2 2
K p→b,path (x⊥b , x⊥ )= 0 cosθΦ h
⊥ ⊥ ν (
eff,D2 ⊥x )d r ⊥ d v⊥,
0 r⊥ ⊥ v⊥,D+ ,TD+ v⊥ 0
2 2
(D.26)
2 Z 0
∞ v⊥ 1 r⊥
Z · ¸
D ,reem 0 0 0 00 00
2
K b→b,path (x⊥b , x⊥b )= 0 cosθcosθ χ ( x
⊥,in,D2 ⊥b ⊥ , v )exp − ν (x
eff,D2 ⊥ )d r ⊥ d v⊥,
0 r⊥ v⊥ 0
(D.27)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D ,refl 0 i (x0 , v)exp − 00 00
2
K b→b,path (x⊥b , x⊥b )= 0 cosθΦ h ν (x
eff,D2 ⊥ )d r ⊥ d v⊥,
r⊥ ⊥ vrefl D+ ,TD+ v⊥ 0
0 ( 2) 2
(D.28)
Z 0
∞ 1 r⊥
· ¸
1
Z
+
D,D 0 0 00 00
K p→p,path (x⊥ , x⊥ )= 0 Φ (x
⊥[v⊥,D+ ,TD+ ] ⊥ ⊥ , v )exp − ν (
eff,D ⊥x )d r ⊥ d v⊥,
0 r⊥ v⊥ 0
(D.29)
Z 0
∞ 1 r⊥
· ¸
1
Z
D,D+ 0 i (x0 , v )exp − 00 00
2
K p→p,path (x⊥ , x⊥ )= 0 Φ h
⊥ ⊥ ν (
0 r⊥ ⊥ v⊥,D+ ,TD+ v⊥ 0
2 2
(D.30)
Z 0
∞ 1 r⊥
· ¸
1
Z
D,diss(D+ ) 0 i (x0 , v )exp − 00 00
K p→p,path2 (x⊥ , x⊥ )= 0 Φ h
⊥ ⊥ ν (
r ⊥, ⊥ v⊥,D+ ,TD,diss D + v⊥ 0
0 2 ( 2)
(D.31)
Z 0
∞ 1 r⊥
· ¸
1
Z
D,diss-rec(D+ ) 0 0 00 00
K p→p,path 2 (x⊥ , x⊥ )= 0 Φ⊥ v + ,T (x⊥ , v⊥ )exp − νeff,D (x⊥ )d r ⊥ d v ⊥ ,
h i
0 r⊥ ⊥,D
2 D,diss-rec(D + )
2
v⊥ 0
(D.32)
136
Z 0
∞ 1 r⊥
· ¸
1
Z
D,diss(D
2) 0 i (x0 , v )exp − 00 00
K p→p,path (x⊥ , x⊥ )= 0 Φ h
⊥ ⊥ ν (
0 r⊥ ⊥ v⊥,D2 ,TD,diss(D 2 ) v⊥ 0
(D.33)
Z 0
∞ 1 r⊥
· ¸
1
Z
D,diss-iz(D2 ) 0 i (x0 , v )exp − 00 00
K p→p,path (x⊥ , x⊥ )= 0 Φ h
⊥ ⊥ ν (
0 r⊥ ⊥ v⊥,D2 ,TD,diss-iz(D 2 ) v⊥ 0
(D.34)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,reem 0 0 0 00 00
K b→p,path (x⊥ , x⊥b )= 0 cosθ χ (x
⊥,in,D ⊥b ⊥, v )exp − ν (
0 r⊥ v⊥ 0
(D.35)
Z 0
∞ 1 r⊥
· ¸
1
Z
D,refl 0 i (x0 , v)exp − 00 00
K b→p,path (x⊥ , x⊥b )= 0 Φ h ν (
0 r⊥ ⊥ vrefl(D+ ) ,TD+ v⊥ 0
(D.36)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
+
D,D 0 0 00 00
K p→b,path (x⊥b , x⊥ )= 0 cosθΦ ( x
⊥[v⊥,D+ ,TD+ ] ⊥ ⊥ , v )exp − ν (x
eff,D ⊥ )d r ⊥ d v⊥,
0 r⊥ v⊥ 0
(D.37)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,D+ 0 i (x0 , v )exp − 00 00
2
⊥ ⊥ ν (
0 r⊥ ⊥ v⊥,D+ ,TD+ v⊥ 0
2 2
(D.38)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,diss(D+ ) 0 0 00 00
K p→b,path2 (x⊥b , x⊥ )= 0 cosθΦ⊥ v + ,T
h i (x⊥ , v⊥ )exp − νeff,D (x⊥ )d r ⊥ d v ⊥ ,
0 r⊥ ⊥,D
2 D,diss(D + )
2
v⊥ 0
(D.39)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,diss-rec(D+ ) 0 0 00 00
K p→b,path 2 (x⊥b , x⊥ )= 0 cosθΦ⊥ v + ,T
h i (x⊥ , v⊥ )exp − νeff,D (x⊥ )d r ⊥ d v ⊥ ,
0 r⊥ ⊥,D
2 D,diss-rec(D + )
2
v⊥ 0
(D.40)
137
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,diss(D
K p→b,path2 ) (x⊥b , x⊥
0
)= 0 cosθΦ h i (x0 , v )exp −
⊥ ⊥ ν (
eff,D ⊥x 00
)d r 00
⊥ d v⊥,
0 r⊥ ⊥ v⊥,D2 ,TD,diss(D 2 ) v⊥ 0
(D.41)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,diss-iz(D2 ) 0 i (x0 , v )exp − 00 00
⊥ ⊥ ν (
0 r⊥ ⊥ v⊥,D2 ,TD,diss-iz(D 2 ) v⊥ 0
(D.42)
2 Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,reem 0 0 0 00 00
K b→b,path (x⊥b , x⊥b )= 0 cosθcosθ χ ( x
⊥,in,D ⊥b ⊥ , v )exp − ν (
0 r⊥ v⊥ 0
(D.43)
Z 0
∞ v⊥ 1 r⊥
Z · ¸
D,refl 0 i (x0 , v)exp − 00 00
K b→b,path (x⊥b , x⊥b )= 0 cosθΦ h ν (
eff,D ⊥x )d r ⊥ d v⊥.
0 r⊥ ⊥ vrefl(D+ ) ,TD+ v⊥ 0
(D.44)
We remark that all velocity distributions given by a Maxwellian or a Knudsen cosine law are
integrated along the parallel velocity, that is
Z ∞
Φ i (x 0 , v ) = Φh i (x0 , v )d v ,
h
⊥ v⊥,D+ ,TD+ ⊥ ⊥ v⊥,D+ ,TD+ ⊥ ⊥ k (D.45)
2 2 0 2 2
Z ∞
0 0
Φ⊥[v⊥,D+ ,TD+ ] (x⊥ , v⊥ ) = Φ[v⊥,D+ ,TD+ ] (x⊥ , v⊥ )d v k , (D.46)
0
Z ∞
Φ i (x 0 , v ) = 0
Φv⊥,D2 ,TD,diss(D ) (x⊥ , v⊥ )d v k .,
h
⊥ v⊥,D2 ,TD,diss(D2 ) ⊥ ⊥ 2
(D.47)
0
Z ∞
Φ i (x 0 , v ) = Φh i (x0 , v )d v ,
h
⊥ v⊥,D2 ,TD,diss-iz(D2 ) ⊥ ⊥ v⊥,D2 ,TD,diss-iz(D2 ) ⊥ ⊥ k (D.48)
0
138
Z ∞
Φ i (x 0 , v ) = Φh i (x0 , v )d v ,
h
⊥ v⊥,D+ ,TD,diss ⊥ ⊥ v⊥,D+ ,TD,diss ⊥ ⊥ k (D.49)
2 (D 2+ ) 0 2 (D 2+ )
Z ∞
Φ i (x0 , v ) = Φh i (x0 , v )d v ,
h
⊥ v⊥,D+ ,TD,diss-rec ⊥ ⊥ v⊥,D+ ,TD,diss-rec ⊥ ⊥ k (D.50)
2 (D+2 ) 0 2 (D+2 )
Z ∞
Φ i (x0 , v) = Φh i (x0 , v)d v
k, (D.51)
h
⊥ vrefl(D+ ) ,TD+ 0 vrefl(D+ ) ,TD+
Z ∞
Φ i (x0 , v) = Φh i (x0 , v)d v
k, (D.52)
h
⊥ vrefl vrefl
(D+2 ) ,TD+2 0 (D+2 ) ,TD+2
Z ∞
0 0
χ⊥,in,D2 (x⊥,b , v⊥ ) = χin,D2 (x⊥,b , v⊥ )d v k , (D.53)
0
Z ∞
0 0
χ⊥,in,D (x⊥,b , v⊥ ) = χin,D (x⊥,b , v⊥ )d v k . (D.54)
0
139
E Numerical solution of the neutral
equations
The coupled neutral equations for D2 and D, Eqs. (3.80-3.83), may be discretized as a linear
matrix system, x = A x + b, with the unknown x representing the density and boundary flux of
the D2 and D species. Indicating with NP the number of points that discretize the poloidal
plane and NB the number of points discretizing the boundary, x is a vector of size 2(NP +NB ), A
is a 2(NP + NB )×2(NP + NB ) matrix and b is a 2(NP + NB ) vector that includes all contributions
not proportional to the neutral density or flux, namely the effect of recombination of D+ and
D+ +
2 with electrons, the effect of dissociative processes to which D2 ions are subject and the
contributions from the flux of D+ and D+ 2 ions to the boundary.
The matrix M , and the vectors x and b can then be written as
     
nD M 11 M 12 M 13 M 14 b1
 Γout,D 
     
 , M = M 21 M 22 M 23 M 24 
 , b = b 2  ,
  
x=
 n M (E.1)
 D2 

 31 M 32 M 33 M 34 
 b 
 3
Γout,D2 M 41 M 42 M 43 M 44 b4
where M 11 is a matrix of size NP × NP ,
+
D,D
M 11 = νcx,D K p→p , (E.2)
+
D,D
that discretizes the kernel K p→p defined in Eq. (3.84) at the spatial points where n D is evalu-
ated. The matrix
+
D,D
M 21 = νcx,D K p→b , (E.3)
141
Appendix E. Numerical solution of the neutral equations
+
D,D
has size NB × NP and discretizes the kernel K p→b defined in Eq. (D.15) at the points where ΓD
is evaluated. The other matrices appearing in the definition of M are defined similarly,
·n ¸
D+ D ,D+
M 31 = 2
νcx,D+
2 −D
2
K p→p 2
, (E.4)
nD
·n ¸
D+ D2 ,D+
M 41 = 2
νcx,D+2 −D K p→b 2
, (E.5)
nD
D,reem
M 12 = (1 − αrefl )(1 − βassoc )K b→p , (E.6)
D,reem
M 22 = (1 − αrefl )(1 − βassoc )K b→b , (E.7)
βassoc D2 ,reem
M 32 = (1 − αrefl ) K b→p , (E.8)
2
βassoc D2 ,reem
M 42 = (1 − αrefl ) K b→b , (E.9)
2
+ D,diss(D2 ) D,diss-iz(D2 )
D,D
M 13 = νcx,D2 −D+ K p→p + νdiss,D2 K p→p + νdiss-iz,D2 K p→p , (E.10)
+
D,D D,diss(D2 ) D,diss-iz(D2 )
M 23 = νcx,D2 −D+ K p→b + νdiss,D2 K p→b + νdiss-iz,D2 K p→b , (E.11)
D ,D+
M 33 = νcx,D2 K p→p
2 2
, (E.12)
142
D ,D+
M 43 = νcx,D2 K p→b
2 2
, (E.13)
M 14 = 0, (E.14)
M 24 = 0, (E.15)
D
M 34 = (1 − αrefl )K b→p
2
, (E.16)
D
M 44 = (1 − αrefl )K b→b
2
, (E.17)
The vector b is defined through the vectors b 1 and b 3 of size NP ,
b 1 = n D[rec(D+ )] (x⊥ ) + n D[diss(D+2 )] (x⊥ ) + n D[out(D+ )] (x⊥ ), (E.18)
b 3 = n D2 [rec(D+2 )] (x⊥ ) + n D2 [out(D+2 )] (x⊥ ) + n D2 [out(D+ )] (x⊥ ), (E.19)
and the vector b 2 and b 4 of size NB ,
b 2 = Γout,D[rec(D+ )] (x⊥ ) + Γout,D[diss(D+2 )] (x⊥ ) + Γout,D[out(D+ )] (x⊥ ), (E.20)
b 4 = Γout,D2 [rec(D+2 )] (x⊥ ) + Γout,D2 [out(D+2 )] (x⊥ ) + Γout,D2 [out(D+ )] (x⊥ ). (E.21)
143
It is remarked that the vector b can also be written as b = N xi , where xi refers to the densities
and boundary fluxes of the D+ and D+ 2 ion species,
 
n D+
Γout,D+ 
 
xi =  , (E.22)
 n D+2 
 
Γout,D+2
and the matrix N can be expressed as
 
N11 N12 N13 N14
 
N21 N22 N23 N24 
N =
N
, (E.23)
 31 N32 N33 N34 

N41 N42 N43 N44
with entries
+
D,D
N11 = νrec,D+ K p→p , (E.24)
+
D,D
N21 = νrec,D+ K p→b , (E.25)
D ,D+
N31 = νrec,D+2 K p→p
2 2
, (E.26)
D ,D+
N41 = νrec,D+2 K p→b
2 2
, (E.27)
D,reem D,refl
N12 = (1 − αrefl )(1 − βassoc )K b→p + αrefl K b→p , (E.28)
D,reem D,refl
N22 = (1 − αrefl )(1 − βassoc )K b→b + αrefl K b→b , (E.29)
144
βassoc D2 ,reem
N32 = (1 − αrefl ) K b→p , (E.30)
2
βassoc D2 ,reem
N42 = (1 − αrefl ) K b→b , (E.31)
2
D,diss(D+
2) D,diss-rec(D+
2)
N13 = νdiss,D+2 K p→p + 2νdiss-rec,D+2 K p→p , (E.32)
D,diss(D+
2) D,diss-rec(D+
2)
N23 = νdiss,D+2 K p→b + 2νdiss-rec,D+2 K p→b , (E.33)
D ,D+
N33 = νrec,D+2 K p→p
2 2
, (E.34)
D ,D+
N43 = νrec,D+2 K p→b
2 2
, (E.35)
N14 = 0, (E.36)
N24 = 0, (E.37)
D ,reem
N34 = (1 − αrefl )K b→p
2
, (E.38)
D ,reem D ,refl
N44 = (1 − αrefl )K b→b
2
+ αrefl K b→b
2
. (E.39)
145
We remark that a convergence study to estimate the error introduced by the discretization of
the neutral equation was carried out for a single neutral species model and it is reported in
[36].
146
Bibliography
[1] V. Smil. Energy Transitions: Global and National Perspectives, 2017.
[2] BP Statistical. BP statistical review of world energy 2020. Technical report, BP p.l.c.,
2020.
[3] J. Houghton. Global Warming: The Complete Briefing. 4th edition. Cambridge University
Press, 2009.
[4] S. Simons, J. Schmitt, B. Tom, H. Bao, B. Pettinato, and M. Pechulis. Thermal, Mechanical,
and Hybrid Chemical Energy Storage Systems. Elsevier Inc., 2021.
[5] P. A. Kharecha and J. E. Hansen. Prevented mortality and greenhouse gas emissions
from historical and projected nuclear power. Environmental Science and Technology,
47:4889, 2013.
[6] International Atomic Energy Agency and Nuclear Energy Agency. Uranium 2020 -
Resources, Production and Demand, 2020.
[7] BP Statistical. BP Statistical Review of World Energy. Technical report, BP p.l.c., 2016.
[8] European Commission. Directorate General for Research. Fusion research : an energy
option for Europe’s future. European Communities, 2005.
[9] J. Parisi and J. Ball. The Future of Fusion Energy. World Scientific Europe Publishing Ltd.,
2019.
[10] J. Reijonen, T. P. Lou, B. Tolmachoff, and K. N. Leung. Compact neutron source de-
velopment at LBNL. Charged Particle Detection, Diagnostics, and Imaging, 4510:80,
2001.
[11] J. D. Lawson. Some criteria for a power producing thermonuclear reactor. Proceedings
of the Physical Society. Section B, 70, 1957.
[12] P. C. Stangeby. The Plasma Boundary of Magnetic Fusion Devices. IOP, Bristol, 2000.
[13] Y. Yu, L. Wang, B. Cao, H. Wang, and J. Hu. Fuel retention and recycling studies by using
particle balance in EAST tokamak. In Physica Scripta, page 014070, 2017.
147
Bibliography
[14] S. M. Motevalli, S. Ostadi, and F. Fadaei. Effects of wall recycling and fuel injections rate
on particle balance behavior in the tokamak fusion. Indian Journal of Physics, 95:943,
2020.
[15] C. C. Petty and the Diii-D Team. DIII-D research towards establishing the scientific basis
for future fusion reactors. Nuclear Fusion, 59:112002, 2019.
[16] M. Z. Tokar. Modelling of detachment in a limiter tokamak as a nonlinear phenomenon

caused by impurity radiation. Plasma Physics and Controlled Fusion, 36:1819, 1994.
[17] S. I. Braginskii. Transport processes in a plasma. In Reviews of Plasma Physics, volume 1.

Consultants Bureau, New York, 1965.
[18] B. J. Braams. Radiative Divertor Modelling for ITER and TPX. Contributions to Plasma
Physics, 36:276, 1996.
[19] M. Baelmans, D. Reiter, and R. R. Weynants. New Developments in Plasma Edge Model-
ing with Particular Emphasis on Drift Flows and Electric Fields. Contributions to Plasma
Physics, 36:117, 1996.
[20] R. Simonini, G. Corrigan, G. Radford, J. Spence, and A. Taroni. Models and Numerics in
the Multi-Fluid 2-D Edge Plasma Code EDGE2D/U. Contributions to Plasma Physics,
34:368, 1994.
[21] H. Bufferand, G. Ciraolo, L. Isoardi, G. Chiavassa, F. Schwander, E. Serre, N. Fedorczak,

P. Ghendrih, and P. Tamain. Applications of SOLEDGE-2D code to complex SOL con-
figurations and analysis of Mach probe measurements. Journal of Nuclear Materials,
415:S589, 2011.
[22] R. Zagórski, H. Gerhauser, and H. A. Claaßen. Numerical simulation of the TEXTOR

edge plasma including drifts and impurities. Contributions to Plasma Physics, 38:61,
1998.
[23] A. Yu Pigarov, S. I. Krasheninnikov, W. P. West, T. D. Rognlien, J. A. Boedo, D. G. Whyte, C. J.

Lasnier, T. W. Petrie, M. J. Schaffer, and J. G. Watkins. DIII-D edge plasma simulations
with UEDGE code including non-diffusive anomalous cross-field transport. Journal of
Nuclear Materials, 313-316:1076, 2003.
[24] D. Stotler and C. Karney. Neutral Gas Transport Modeling with DEGAS 2. Contributions
to Plasma Physics, 34:392, 1994.
[25] D. Reiter, M. Baelmans, and P. Börner. The EIRENE and B2-EIRENE codes. Fusion
Science and Technology, 47:172, 2005.
[26] J. Mandrekas. GTNEUT: A code for the calculation of neutral particle transport in
plasmas based on the transmission and escape probability method. Computer Physics
Communications, 161:36, 2004.
148
Bibliography
[27] K. Shimizu, T. Takizuka, S. Sakurai, H. Tamai, H. Takenaga, H. Kubo, and Y. Miura.

Simulation of divertor detachment characteristics in JT-60 with superconducting coils.
Journal of Nuclear Materials, 313-316:1277, 2003.
[28] J. D. Lore, J. M. Canik, Y. Feng, J. W. Ahn, R. Maingi, and V. Soukhanovskii. Implementa-

tion of the 3D edge plasma code EMC3-EIRENE on NSTX. Nuclear Fusion, 52:054012,
2012.
[29] R. Schneider, X. Bonnin, K. Borrass, D. P. Coster, H. Kastelewicz, D. Reiter, V. A. Rozhansky,

and B. J. Braams. Plasma edge physics with B2-Eirene. Contributions to Plasma Physics,
46:3, 2006.
[30] B. D. Dudson, M. V. Umansky, X. Q. Xu, P. B. Snyder, and H. R. Wilson. BOUT++: A

framework for parallel plasma fluid simulations. Computer Physics Communications,
180:1467, 2009.
[31] B. D. Dudson and J. Leddy. Hermes: Global plasma edge fluid turbulence simulations.
Plasma Physics and Controlled Fusion, 59, 2017.
[32] M. Wiesenberger, L. Einkemmer, M. Held, A. Gutierrez-Milla, X. Sáez, and R. Iakymchuk.

Reproducibility, accuracy and performance of the FELTOR code and library on parallel
computer architectures. Computer Physics Communications, 238:145, 2019.
[33] P. Ricci, F. D. Halpern, S. Jolliet, J. Loizu, A. Mosetto, A. Fasoli, I. Furno, and C. Theiler.
Simulation of plasma turbulence in scrape-off layer conditions: The GBS code, simu-
lation results and code validation. Plasma Physics and Controlled Fusion, 54:124047,
2012.
[34] F. D. Halpern, P. Ricci, S. Jolliet, J. Loizu, J. Morales, A. Mosetto, F. Musil, F. Riva, T. M.

Tran, and C. Wersal. The GBS code for tokamak scrape-off layer simulations. Journal of
Computational Physics, 315:388, 2016.
[35] P. Paruta, P. Ricci, F. Riva, C. Wersal, C. Beadle, and B. Frei. Simulation of plasma
turbulence in the periphery of diverted tokamak by using the GBS code. Physics of
Plasmas, 25:112301, 2018.
[36] M. Giacomin, P. Ricci, A. Coroado, G. Fourestey, D. Galassi, E. Lanti, and D. Mancini. The
GBS code for the self-consistent simulation of plasma turbulence and kinetic neutral
dynamics in the tokamak boundary. submitted to Computer Physics Communications,
2021.
[37] M. Francisquez, B. Zhu, and B. N. Rogers. Global 3D Braginskii simulations of the

tokamak edge region of IWL discharges. Nuclear Fusion, 57, 2017.
[38] A. Stegmeir, D. Coster, A. Ross, O. Maj, K. Lackner, and E. Poli. GRILLIX: A 3D turbu-
lence code based on the flux-coordinate independent approach. Plasma Physics and
Controlled Fusion, 60, 2018.
149
Bibliography
[39] A. S. Thrysøe, M. Løiten, J. Madsen, V. Naulin, A. H. Nielsen, and J. J. Rasmussen. Plasma

particle sources due to interactions with neutrals in a turbulent scrape-off layer of a
toroidally confined plasma. Physics of Plasmas, 25, 2018.
[40] C. Baudoin, P. Tamain, H. Bufferand, G. Ciraolo, N. Fedorczak, D. Galassi, P. Ghen-

drih, and N. Nace. Turbulent heat transport in TOKAM3X edge plasma simulations.
Contributions to Plasma Physics, 58:484, 2018.
[41] E. L. Shi, G. W. Hammett, T. Stoltzfus-Dueck, and A. Hakim. Gyrokinetic continuum

simulation of turbulence in a straight open-field-line plasma. Journal of Plasma Physics,
83, 2017.
[42] C. S. Chang and S. Ku. Spontaneous rotation sources in a quiescent tokamak edge
plasma. Physics of Plasmas, 15, 2008.
[43] S. Ku, C. S. Chang, and P. H. Diamond. Full-f gyrokinetic particle simulation of centrally
heated global ITG turbulence from magnetic axis to edge pedestal top in a realistic
tokamak geometry. Nuclear Fusion, 49, 2009.
[44] J. Leddy, B. Dudson, and H. Willett. Simulation of the interaction between plasma
turbulence and neutrals in linear devices. Nuclear Materials and Energy, 12:994, 2017.
[45] A. S. Thrysøe, L. E. H. Tophøj, V. Naulin, J. J. Rasmussen, J. Madsen, and A. H. Nielsen.

The influence of blobs on neutral particles in the scrape-off layer. Plasma Physics and
Controlled Fusion, 58, 2016.
[46] A. S. Thrysøe, J. Madsen, V. Naulin, and J. J. Rasmussen. Influence of molecular dissocia-

tion on blob-induced atom density perturbations. Nuclear Fusion, 58, 2018.
[47] D. M. Fan, Y. Marandet, P. Tamain, H. Bufferand, G. Ciraolo, P. Ghendrih, and E. Serre. Ef-
fect of turbulent fluctuations on neutral particles transport with the TOKAM3X-EIRENE
turbulence code. Nuclear Materials and Energy, 18:105, 2019.
[48] A. Zeiler, J. F. Drake, and B. Rogers. Nonlinear reduced Braginskii equations with ion
thermal dynamics in toroidal plasma. Physics of Plasmas, 4:2134, 1997.
[49] P. Ricci, C. Theiler, A. Fasoli, I. Furno, K. Gustafson, D. Iraji, and J. Loizu. Methodology
for turbulence code validation: Quantification of simulation-experiment agreement
and application to the TORPEX experiment. Physics of Plasmas, 18:032109, 2011.
[50] P. Ricci, B. N. Rogers, and S. Brunner. High- and low-confinement modes in simple
magnetized toroidal plasmas. Physical Review Letters, 100:225002, 2008.
[51] P. Ricci and B. N. Rogers. Transport scaling in interchange-driven toroidal plasmas.

Physics of Plasmas, 16:062303, 2009.
[52] B. N. Rogers and P. Ricci. Low-frequency turbulence in a linear magnetized plasma.

Physical Review Letters, 104, 2010.
150
Bibliography
[53] W. Gekelman, H. Pfister, Z. Lucky, J. Bamber, D. Leneman, and J. Maggs. Design, con-
struction, and properties of the large plasma research device - The LAPD at UCLA.
Review of Scientific Instruments, 62:2875, 1991.
[54] P. Ricci and B. N. Rogers. Turbulence phase space in simple magnetized toroidal plasmas.
Physical Review Letters, 104, 2010.
[55] Federico D. Halpern, S. Jolliet, J. Loizu, A. Mosetto, and P. Ricci. Ideal ballooning modes
in the tokamak scrape-off layer. Physics of Plasmas, 20:052306, 2013.
[56] A. Mosetto, F. D. Halpern, S. Jolliet, J. Loizu, and P. Ricci. Finite ion temperature effects
on scrape-off layer turbulence. Physics of Plasmas, 22:012308, 2015.
[57] F. Riva, P. Ricci, F. D. Halpern, S. Jolliet, J. Loizu, and A. Mosetto. Verification methodology
for plasma simulations and application to a scrape-off layer turbulence code. Physics of
Plasmas, 21:062301, 2014.
[58] P. Paruta, C. Beadle, P. Ricci, and C. Theiler. Blob velocity scaling in diverted tokamaks:
A comparison between theory and simulation. Physics of Plasmas, 26:032302, 2019.
[59] C. F. Beadle and P. Ricci. Understanding the turbulent mechanisms setting the density
decay length in the tokamak scrape-off layer. Journal of Plasma Physics, 86, 2020.
[60] M. Giacomin, L. N. Stenger, and P. Ricci. Turbulence and flows in the plasma boundary
of snowflake magnetic configurations. Nuclear Fusion, 60, 2020.
[61] C. Wersal and P. Ricci. A first-principles self-consistent model of plasma turbulence and
kinetic neutral dynamics in the tokamak scrape-off layer. Nuclear Fusion, 55:123014,
2015.
[62] D. A. Russell, J. R. Myra, and D. P. Stotler. A reduced model of neutral-plasma interactions

in the edge and scrape-off-layer: Verification comparisons with kinetic Monte Carlo
simulations. Physics of Plasmas, 26, 2019.
[63] C. Wersal, P. Ricci, and J. Loizu. A comparison between a refined two-point model for
the limited tokamak SOL and self-consistent plasma turbulence simulations. Plasma
Physics and Controlled Fusion, 59:044011, 2017.
[64] C. Wersal and P. Ricci. Impact of neutral density fluctuations on gas puff imaging
diagnostics. Nuclear Fusion, 57:116018, 2017.
[65] V. M. Zhdanov. Transport Processes in Multicomponent Plasma. Plasma Physics and

Controlled Fusion, 2002.
[66] H. Bufferand, P. Tamain, S. Baschetti, J. Bucalossi, G. Ciraolo, N. Fedorczak, P. Ghendrih,

F. Nespoli, F. Schwander, E. Serre, and Y. Marandet. Three-dimensional modelling of
edge multi-component plasma taking into account realistic wall geometry. Nuclear
Materials and Energy, 18:82, 2019.
151
Bibliography
[67] A. Coroado and P. Ricci. Moving toward mass-conserving simulations of plasma tur-
bulence and kinetic neutrals in the tokamak boundary with the GBS code. Physics of
Plasmas, 28:022310, 2021.
[68] A. Mosetto, F. D. Halpern, S. Jolliet, J. Loizu, and P. Ricci. Turbulent regimes in the
tokamak scrape-off layer. Physics of Plasmas, 20, 2013.
[69] D. M. Fan, Y. Marandet, P. Tamain, H. Bufferand, G. Ciraolo, P. Ghendrih, and E. Serre.

Self-consistent coupling of the three-dimensional fluid turbulence code TOKAM3X and
the kinetic neutrals code EIRENE. Contributions to Plasma Physics, 58:490, 2018.
[70] F. D. Halpern and P. Ricci. Velocity shear, turbulent saturation, and steep plasma gradi-
ents in the scrape-off layer of inner-wall limited tokamaks. Nuclear Fusion, 57, 2017.
[71] A. Arakawa. Computational design for long-term numerical integration of the equations
of fluid motion: Two-dimensional incompressible flow. Part I. Journal of Computational
Physics, 1:119, 1966.
[72] S. Jolliet, F. D. Halpern, J. Loizu, A. Mosetto, and P. Ricci. Aspect ratio effects on limited
scrape-off layer plasma turbulence. Physics of Plasmas, 21:022303, 2014.
[73] M. Knudsen. Das Cosinusgesetz in der kinetischen Gastheorie. Annalen der Physik,
353:1113, 1916.
[74] J. Loizu, P. Ricci, F. D. Halpern, and S. Jolliet. Boundary conditions for plasma fluid
models at the magnetic presheath entrance. Physics of Plasmas, 19:122307, 2012.
[75] A. Mosetto, F. D. Halpern, S. Jolliet, J. Loizu, and P. Ricci. Finite ion temperature effects
on scrape-off layer turbulence. Physics of Plasmas, 22, 2015.
[76] F. D. Halpern, P. Ricci, S. Jolliet, J. Loizu, and A. Mosetto. Theory of the scrape-off layer
width in inner-wall limited tokamak plasmas. Nuclear Fusion, 54, 2014.
[77] A. Coroado and P. Ricci. A self-consistent multi-component model of plasma turbulence

and kinetic neutral dynamics for the simulation of the tokamak boundary. submitted to
Nuclear Fusion, 2021.
[78] C. S. Pitcher. Tokamak Plasma Interaction with Limiters. PhD thesis, University of
Toronto, 1988.
[79] C. Zhang, C. Sang, L. Wang, M. Chang, D. Liu, and D. Wang. Effect of carbon and
tungsten plasma-facing materials on the divertor and pedestal plasma in EAST. Plasma
Physics and Controlled Fusion, 61:115013, 2019.
[80] Y. Feng, F. Sardei, and J. Kisslinger. 3D fluid modelling of the edge plasma by means of a
Monte Carlo technique. Journal of Nuclear Materials, 266:812, 1999.
152
Bibliography
[81] K. Sawada and T. Fujimoto. Effective ionization and dissociation rate coefficients of
molecular hydrogen in plasma. Journal of Applied Physics, 78:2913, 1995.
[82] K. D. Lawson, M. Groth, D. Harting, S. Menmuir, D. Reiter, S. Brezinsek, G. Corrigan,

P. Drewelow, C. F. Maggi, A. G. Meigs, J. Simpson, M. F. Stamp, S. Wiesen, and contributors.
A study of the atomic and molecular power loss terms in EDGE2D-EIRENE simulations
of JET ITER-like wall L-mode discharges. Nuclear Materials and Energy, 12:924, 2017.
[83] H. Bufferand, C. Baudoin, J. Bucalossi, G. Ciraolo, J. Denis, N. Fedorczak, D. Galassi,

P. Ghendrih, R. Leybros, Y. Marandet, N. Mellet, J. Morales, N. Nace, E. Serre,
P. Tamain, and M. Valentinuzzi. Implementation of drift velocities and currents in
SOLEDGE2D–EIRENE. Nuclear Materials and Energy, 12:852, 2017.
[84] D. Reiter. The data file AMJUEL: Additional Atomic and Molecular Data for EIRENE.
Technical report, FZ, Forschungszentrum Julich GmbH FRG, 2011.
[85] R. K. Janev and Langer W. D. The Janev-Langer Hydrogen-Helium database, 1987 Con-
tents. Springer, 1987.
[86] D. P. Stotler, C. H. Skinner, R. V. Budny, A. T. Ramsey, D. N. Ruzic, and R. B. Turkot.

Modeling of neutral hydrogen velocities in the tokamak fusion test reactor. Physics of
Plasmas, 3:4084, 1996.
[87] M. Wensing, B. P. Duval, O. Février, A. Fil, D. Galassi, E. Havlickova, A. Perek,

H. Reimerdes, C. Theiler, K. Verhaegh, and M. Wischmeier. SOLPS-ITER simulations of
the TCV divertor upgrade. Plasma Physics and Controlled Fusion, 61:085029, 2019.
[88] R. K. Janev. Atomic and Molecular Processes in Fusion Edge Plasmas. Plenum Press, New
York, 1995.
[89] J. S. Yoon, M. Y. Song, J. M. Han, S. H. Hwang, W. S. Chang, B. Lee, and Y. Itikawa.

Cross sections for electron collisions with hydrogen molecules. Journal of Physical and
Chemical Reference Data, 37:913, 2008.
[90] H. K. Chung. Data for Atomic Processes of Neutral Beams in Fusion Plasma, Summary
Report of the First Research Coordination Meeting. Technical report, IAEA Nuclear Data
Section, 2017.
[91] P. S. Krstić. Inelastic processes from vibrationally excited states in slow Hˆ+ + H_2 and H+
H_2ˆ+ collisions: Excitations and charge transfer. Physical Review A - Atomic, Molecular,
and Optical Physics, 66:042717, 2002.
[92] P. S. Krstić. Vibrationally resolved collisions in cold hydrogen plasma. In Nuclear Instru-
ments and Methods in Physics Research, Section B: Beam Interactions with Materials and
Atoms, page 58, 2005.
153
Bibliography
[93] M. Raghunathan, Y. Marandet, H. Bufferand, G. Ciraolo, P. Ghendrih, P. Tamain, and

E. Serre. Generalized Collisional Fluid Theory for Multi-Component, Multi-Temperature
Plasma Using The Linearized Boltzmann Collision Operator for Scrape-Off Layer/Edge
Applications. Plasma Physics and Controlled Fusion, 29:1, 2021.
[94] D. Tskhakaya, S. Kuhn, and Y. Tomita. Formulation of boundary conditions for the
unmagnetized multi-ion-component plasma sheath. Contributions to Plasma Physics,
46:649, 2006.
[95] D. Galassi, H. Reimerdes, C. Theiler, M. Wensing, H. Bufferand, G. Ciraolo, P. Innocente,

Y. Marandet, and P. Tamain. Numerical investigation of optimal divertor gas baffle
closure on TCV. Plasma Physics and Controlled Fusion, 2020.
[96] J. Loizu, P. Ricci, F. D. Halpern, S. Jolliet, and A. Mosetto. On the electrostatic potential in
the scrape-off layer of magnetic confinement devices. Plasma Physics and Controlled
Fusion, 55:124019, 2013.
[97] J. L. Terry, R. Maqueda, C. S. Pitcher, S. J. Zweben, B. Labombard, E. S. Marmar, A. Y.

Pigarov, and G. Wurden. Visible imaging of turbulence in the SOL of the Alcator C-Mod
tokamak. Journal of Nuclear Materials, 290-293:757, 2001.
[98] R. J. Maqueda, G. A. Wurden, S. Zweben, L. Roquemore, H. Kugel, D. Johnson, S. Kaye,

S. Sabbagh, and R. Maingi. Edge turbulence measurements in NSTX by gas puff imaging.
Review of Scientific Instruments, 72:931, 2001.
[99] S. J. Zweben, D. P. Stotler, J. L. Terry, B. Labombard, M. Greenwald, M. Muterspaugh,

C. S. Pitcher, K. Hallatschek, R. J. Maqueda, B. Rogers, J. L. Lowrance, V. J. Mastrocola,
and G. F. Renda. Edge turbulence imaging in the Alcator C-Mod tokamak. Physics of
Plasmas, 9:1981, 2002.
[100] I. Shesterikov, Y. Xu, M. Berte, P. Dumortier, M. Van Schoor, M. Vergote, B. Schweer, and
G. Van Oost. Development of the gas-puff imaging diagnostic in the TEXTOR tokamak.
Review of Scientific Instruments, 84, 2013.
[101] S. J. Zweben, B. D. Scott, J. L. Terry, B. LaBombard, J. W. Hughes, and D. P. Stotler.

Comparison of scrape-off layer turbulence in Alcator C-Mod with three dimensional
gyrofluid computations. Physics of Plasmas, 16:082505, 2009.
[102] F. D. Halpern, J. L. Terry, S. J. Zweben, B. LaBombard, M. Podesta, and P. Ricci. Compar-

ison of 3D flux-driven scrape-off layer turbulence simulations with gas-puff imaging
of Alcator C-Mod inner-wall limited discharges. Plasma Physics and Controlled Fusion,
57:054005, 2015.
[103] D. Moulton, Y. Marandet, P. Tamain, P. Ghendrih, and R. Futtersack. Density and Tem-
perature Correlations in the SOL; Implications for Gas Puff Imaging of Turbulence.
Contributions to Plasma Physics, 54:4, 2014.
154
Bibliography
[104] H. P. Summers, W. J. Dickson, M. G. O’Mullane, N. R. Badnell, A. D. Whiteford, D. H.

Brooks, J. Lang, S. D. Loch, and D. C. Griffin. Ionization state, excited populations and
emission of impurities in dynamic finite density plasmas: I. The generalized collisional-
radiative model for light elements. Plasma Physics and Controlled Fusion, 2006.
[105] E. De La Cal and the Tj-Ii Team. Visualising the electron density structure of blobs and
studying its possible effect on neutral turbulence. Nuclear Fusion, 56:106031, 2016.
[106] D. A. Russell, J. R. Myra, D. A. D’Ippolito, T. L. Munsat, Y. Sechrest, R. J. Maqueda,

D. P. Stotler, and S. J. Zweben. Comparison of scrape-off layer turbulence simulations
with experiments using a synthetic gas puff imaging diagnostic. Physics of Plasmas,
18:022306, 2011.
[107] D. P. Stotler, B. LaBombard, J. L. Terry, and S. J. Zweben. Neutral transport simulations

of gas puff imaging experiments. Journal of Nuclear Materials, 313-316:1066, 2003.
[108] Y. Marandet, A. MekKoćaoui, D. Reiter, P. Börner, P. Genesio, F. Catoire, J. Rosato,

H. Capes, L. Godbert-Mouret, M. Koubiti, and R. Stamm. Transport of neutral par-
ticles in turbulent scrape-off layer plasmas. Nuclear Fusion, 51:083035, 2011.
[109] A. Mekkaoui, Y. Marandet, D. Reiter, P. Boerner, P. Genesio, J. Rosato, H. Capes, M. Koubiti,

L. Godbert-Mouret, and R. Stamm. A coarse-grained transport model for neutral parti-
cles in turbulent plasmas. Physics of Plasmas, 19:122310, 2012.
[110] Y. Marandet, P. Tamain, R. Futtersack, P. Ghendrih, H. Bufferand, P. Genesio, and

A. Mekkaoui. Influence of neutral particles on scrape-off layer turbulence with applica-
tion to the interpretation of fast camera data. Journal of Nuclear Materials, 438:S518,
2013.
[111] D. Wünderlich, M. Giacomin, R. Ritz, and U. Fantz. Yacora on the Web: Online colli-
sional radiative models for plasmas containing H, H2 or He. Journal of Quantitative
Spectroscopy and Radiative Transfer, 240:106695, 2020.
[112] C. Spearman. The Proof and Measurement of Association Between Two Things. The
American Journal of Psychology, 15:72, 1904.
[113] J. Franck. Elementary processes of photochemical reactions. Transactions of the Faraday

Society, 21:536, 1926.
[114] E. U. Condon. Nuclear motions associated with electron transitions in diatomic

molecules. Physical Review, 32:858, 1928.
[115] J. Liu, E. J. Salumbides, U. Hollenstein, Jeroen C. J. K., Kjeld S. E. Eikema, W. Ubachs,

and F. Merkt. Determination of the ionization and dissociation energies of the hydrogen
molecule. Journal of Chemical Physics, 130:174306, 2009.
[116] V. M. Zhdanov and P. N. Yushmanov. Diffusion and heat transfer in a multicomponent

completely ionized plasma. Journal of Applied Mechanics and Technical Physics, 1980.
155
André Calado Coroado
Date of birth: 09.06.1993
Email: [email protected]
Education and experience
2017-2021: PhD in Plasma Physics, Swiss Plasma Center, École Polytechnique Fédérale de
Lausanne (EPFL), Switzerland.
• Theoretical research in plasma physics and controlled fusion, based on numerical

simulations carried out on high performance computers
• Teaching assistant for numerical physics
• Supervisor of two master level student projects
2011-2016: MSc in Engineering Physics, Instituto Superior Técnico (IST), Portugal.
• Scientific initiation scholarship (2015-2016), performing research on the resonant inter-

action between energetic particles and shear Alfven waves in tokamaks.
PhD thesis
Self-consistent simulation of multi-component plasma turbulence and neutral dynamics in

the tokamak boundary, supervised by Prof. Paolo Ricci.
Master thesis
Analytical studies of energetic particles in tokamaks, supervised by Dr. Paulo Rodrigues and
Prof. Nuno Loureiro.
Languages
Portuguese (native), English (fluent), Spanish (fluent), French (intermediate), Italian (elemen-
tary)
Peer-reviewed publications
A. Coroado and P. Ricci. "Moving toward mass-conserving simulations of plasma turbulence

and kinetic neutrals in the tokamak boundary with the GBS code." Physics of Plasmas (2021)
A. Coroado and P. Ricci. "A self-consistent multi-component model of plasma turbulence and
kinetic neutral dynamics for the simulation of the tokamak boundary." Submitted to Nuclear
Fusion (2021)
157
M. Giacomin, P. Ricci, A. Coroado, G. Fourestey, D. Galassi, E. Lanti, D. Mancini. "The GBS
code for the self-consistent simulation of plasma turbulence and kinetic neutral dynamics in
the tokamak boundary." Submitted to Computer Physics Communications (2021)
P. Rodrigues and A. Coroado. "Local up–down asymmetrically shaped equilibrium model for
tokamak plasmas." Nuclear Fusion (2018)
Conference contributions (as first author)
A. Coroado, P. Ricci, C. Beadle, M. Giacomin, P. Paruta, F. Riva, C. Wersal. "First-principles

simulation of plasma fuelling in a tokamak." Poster presented at the 46th European Physical
Society Conference on Plasma Physics. Milan, Italy, 2019.
A. Coroado, P. Ricci, C. Beadle, P. Paruta, F. Riva, C. Wersal. "Numerical simulations of plasma

fuelling in tokamaks using the GBS code." Poster presented at the Joint Varenna-Lausanne
International Workshop 2018. Varenna, Italy, 2018.
A. Coroado, P. Ricci, C. Beadle, P. Paruta, F. Riva, C. Wersal. "Plasma refuelling at the SOL
simulated with the GBS code." Poster presented at the Joint Annual Meeting of SPS and ÖPG
2017. Geneva, Switzerland, 2017.
Interests
Exact sciences, environmental science, geography, writing, cycling, beach soccer.
158

PHD Thesis Andre Calado Coroado

Uploaded by

Copyright:

Available Formats

PHD Thesis Andre Calado Coroado

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PHD Thesis Andre Calado Coroado

Uploaded by

Copyright:

Available Formats

Self-consistent simulation of multi-component

plasma turbulence and neutral dynamics in the

Thèse présenté le 9 septembre 2021

acceptée sur proposition du jury:

Lausanne, EPFL, 2021

2 The mass-conserving GBS model of plasma turbulence and kinetic neutrals 13

3 A multi-component model of plasma turbulence and kinetic neutral dynamics 41

4 Numerical simulations of Gas Puff Imaging 87

5 Summary and conclusions 115

C Zhdanov collisional closure 127

D List of kernel functions 133

E Numerical solution of the neutral equations 141

Curriculum Vitae 157

1.1 Controlled nuclear fusion as an energy source

plants at the current consumption rate [6]).

1.2 The tokamak boundary

In a divertor configuration, it is possible to extend the high-recycling regime by further de-

1.3 Simulation models for tokamak boundary

1.4 Scope and outline of the present Thesis

Despite enabling self-consistent simulations of plasma turbulence and neutral dynamics in

is of particular importance, since the molecular dynamics is expected to play an important

2.2.1 A conserving model for the plasma

∇⊥ f = [∇ − b(b · ∇)] f . (2.13)

νiz = n 〈v e σiz (v e )〉 , (2.14)

νcx = n 〈v i σcx (v i )〉 , (2.15)

νrec = n 〈v e σrec (v e )〉 , (2.16)

νen = n 〈v e σen (v e )〉 , (2.17)

We perform the derivation of the geometric operators by assuming an axisymmetric magnetic

B = F (ψ)∇ϕ + ∇ψ × ∇ϕ, (2.18)

and the perpendicular derivative is defined from Eq. (2.13) as

[φ, A] = P θ∗ r [φ, A]θ∗ r + P r ϕ [φ, A]r ϕ + P ϕθ∗ [φ, A]ϕθ∗ , (2.26)

with its components and respective coefficients defined as

The curvature operator is expressed as

where the geometric coefficients are given by

one obtains, for a circular axisymmetric magnetic equilibrium,

where the geometric coefficients are given by

g ϕϕ 1 1 − ² cos θ ∗ 1 sin2 θ ∗ + (1 − ² cos θ ∗ )2 /q 2

with the geometric coefficients redefined as M r r = N r r (n/B 2 ), M r θ = N r θ (n/B 2 ) and M θ θ

2.2.2 A mass-conserving neutral model

" # " #" # " #" #

2.2.3 Mass-conserving boundary conditions

∂θ∗ Te = γe ∂θ∗ φ (2.57)

∂θ∗ Ti = γi ∂θ∗ φ (2.58)

Γi = nv ki b + Γdi + ΓE×B + Γpol,i , (2.60)

where d i /d t = ∂t + v ki b · ∇ + vE×B · ∇. In this chapter, the polarization drift contributions vpol,i

[Γi ]θlim = nv ki b θ + Γθdi + ΓθE×B ,

with the diamagnetic and E × B flux poloidal components expressed as

[Γi ]rwall = Γrdi + ΓrE×B , (2.65)

2.3 Verification of Mass Conservation

where the electron flux Γe is expressed as

Γe = nv ke b + Γde + ΓE×B , (2.69)

∇ · (Γi − Γe ) + η 0Ω ∇2k Ω + D ⊥Ω ∇2⊥ Ω = 0. (2.70)

In what follows, we use D(n, Ω) = D ⊥n ∇2⊥ n + D kn ∇2k n − η 0Ω ∇2k Ω − D ⊥Ω ∇2⊥ Ω as a simplified

mass conservation, normalized to the volume-integrated density variation, is expressed as

J d θ ∗ d φ ∆n + J d θ ∗ d φ [∇ · (Γi + Γn ) − D(n, Ω)]

with L r the radial length of the simulation domain.

Aiming at a further demonstration of convergence, we obtain a global error measurement

2.4 Simulation Results

Plasma resolution (nx p ) Neutral resolution (nx n ) l 2 (WD) l 2 (RD)