Computational mathematics and physics of
fusion reactors
Paul R. Garabedian*
Courant Institute of Mathematical Sciences, New York University, 251 Mercer Street, New York, NY 10012
Contributed by Paul R. Garabedian, October 3, 2003
Theory has contributed significantly to recent advances in mag-
netic fusion research. New configurations have been found for a
stellarator experiment by computational methods. Solutions of a
free-boundary problem are applied to study the performance of
the plasma and look for islands in the magnetic surfaces. Mathe-
matical analysis and numerical calculations have been used to
study equilibrium, stability, and transport of optimized fusion
reactors.
M agnetic fusion is a widely accepted but not completely
developed response to the energy crisis, and it is funded
internationally at more than $1 billion per annum. This requires
confining ionized hydrogen at high temperatures by means of a
strong magnetic field. As an alternate to the axially symmetric
tokamak, there is now increasing interest in the stellarator
concept for confinement of the fusion plasma in toroidal geom-
etry that is nontrivially three-dimensional. Mathematical anal-
ysis has produced a theory of these configurations that can be
implemented in a practical way through high-performance com-
puting (1).
The large helical device (LHD) is a major stellarator exper-
iment in Toki, Japan, that has achieved a peak ion temperature
near 3 keV (1 eV ⫽ 1.602 ⫻ 10⫺19 J) at low density and has
achieved an average ␤ limit of 3.2% at high density, where ␤ is
the average fluid pressure p divided by the magnetic pressure
B2兾2. The Wendelstein 7 (W7)-AS is a smaller experiment in
Garching, Germany, that has modular coils and has reached a ␤
of 3.3%. Neither of these stellarators seems to have significant
disruptions (2, 3).
The European fusion community has commenced construc-
tion of the W7-X experiment, which will have superconducting
modular coils and a major radius exceeding 5 m. The US
Department of Energy is making plans for the National Compact
Stellarator Experiment at the Princeton Plasma Physics Labo-
ratory in Princeton. These new devices were designed by running
equilibrium, stability, and transport codes with effectiveness that
is made possible by the latest computer technology and improved
mathematical methods. We shall describe the codes and apply
them to reactor studies.
Equilibrium
Most configurations for magnetic fusion use toroidal geometry
to confine the orbits of ions and electrons on closed flux surfaces
inside the plasma. The Kolmogorov–Arnold–Moser theory gives
conditions for the existence of nested families of these magnetic
surfaces (4). It has been shown by similar methods that contin-
uously differentiable solutions of the equilibrium problem can-
not be found when the geometry of the magnetic field is
genuinely three-dimensional (5). Nevertheless, stellarator exper-
iments have become increasingly successful in confining fusion
plasmas. Mathematical analysis shows that an appropriate re-
sponse to this anomaly is to look for weak solutions of the
equilibrium equations in a conservation form that captures
current sheets and islands but does not require the magnetic field
to be either differentiable or continuous (6).
The variational principle of magnetohydrodynamics (MHD)
leads to a method of accelerated steepest descent to compute
www.pnas.org兾cgi兾doi兾10.1073兾pnas.2436390100
weak solutions of the equilibrium problem for both stellarators
and tokamaks (7, 8). This has been implemented in a family of
high-performance codes including BETA, VMEC, and NSTAB.
The VMEC code is widely accepted by the fusion community
and has become an integral part of the diagnostics in major
stellarator experiments (9). Work remains to be done on free-
boundary versions to reconstruct the plasma from observations
and calculate solutions accurately enough to detect the presence
of significant islands among the flux surfaces.
The codes have been run to study the LHD experiment, which
has performed surprisingly well because of high shear near the
edge of the plasma. They have played an important role in the
design of the W7-X experiment at Greifswald in Europe and
APPLIED PHYSICAL
the National Compact Stellarator Experiment at the Princeton
SCIENCES
Plasma Physics Laboratory (10, 11). The codes are mathematical
tools that have enabled physicists to exploit a concept of
two-dimensional quasisymmetry of the magnetic field providing
good drift surfaces of trapped particles. Moreover, bifurcated
solutions with three-dimensional asymmetries have been calcu-
lated for tokamaks that simulate observed physical phenomena
(12). In many cases, three-dimensional computations seem to
model tokamak physics better than results from two-dimensional
theory (12, 13).
Stability
The failure of differentiability in solutions of the three-
dimensional equilibrium equations leads to difficulties in apply-
ing linear stability theory to stellarators because calculation of
coefficients that are singular is required. The mathematics are
harder than the two-dimensional analysis of tokamaks, and it is
not always easy to distinguish between equilibrium and stability
limits on ␤. Local estimates of stability such as the Mercier
criterion or the ballooning theory turn out to be unrealistic for
stellarators, and experiments have been found to exceed pre-
dictions by those methods (1–3).
Because it is based on the MHD variational principle, the
NSTAB code furnishes tests for linear and nonlinear ideal MHD
stability that do not depend on unjustified integration by parts
and differentiation (8). An intuitively plausible mountain-pass
theorem can be applied that asserts that, between any two
minima of the energy landscape, a saddle point can be found
representing an unstable equilibrium. Runs of the NSTAB-
iteration scheme can be used to search for bifurcated solutions
that indicate whether a configuration is in linearly or nonlinearly
unstable equilibrium, because the unexpected appearance of
several solutions of the problem establishes instability.
This method has been applied to the LHD experiment to
explain why surprisingly high values of ␤ were observed after the
plasma passed through states that were linearly unstable but
nonlinearly stable. Examples of such calculations over 5 or 10
field periods, with the pressure p given as a function of the
Abbreviations: LHD, large helical device; W7, Wendelstein 7; MHD, magnetohydrodynam-
ics; MHH2, modular Helias-like Heliac 2.
*E-mail: garabedian@cims.nyu.edu.
© 2003 by The National Academy of Sciences of the USA
PNAS 兩 November 25, 2003 兩 vol. 100 兩 no. 24 兩 13741–13745
Fig. 1. Flux surfaces at four cross sections over the full torus of a bifurcated
LHD equilibrium at ␤ ⫽ 0.025 with the magnetic axis located at major radius
R ⫽ 3.6 m. For a pressure profile p ⫽ p0(1 ⫺ s), and with bootstrap current, the
global m ⫽ 1, n ⫽ 1 mode of this solution is linearly unstable but nonlinearly
stable. Harmonics up to degree 40 in the toroidal angle were used in the
computation.
toroidal flux s, are displayed in Figs. 1 and 2. The theory is also
consistent with recent measurements (3) of ␤ in the W7-AS
experiment (compare with Fig. 3). It seems to be more realistic
Fig. 2. Poincaré map of the flux surfaces at four cross sections over five field
periods of a bifurcated LHD equilibrium at ␤ ⫽ 0.02 with the magnetic axis
shifted inward to a position with major radius R ⫽ 3.6 m. For a triangular
pressure profile p ⫽ p0(1 ⫺ s0.7) similar to observed values of the electron
temperature, the solution is linearly unstable but nonlinearly stable.
13742 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.2436390100
Fig. 3. Cross sections of the flux surfaces over one field period of a bifurcated
W7-AS equilibrium at ␤ ⫽ 0.03 with net current bringing the rotational
transform into the interval 0.6 ⬎ ␫ ⬎ 0.5. An MHD mode with ballooning
structure appears in the solution, which became stable after some of the net
current was removed. These predictions were confirmed by observations of
␤ ⫽ 0.033 in the experiment.
than computations of local criteria for the LHD displaying the
dependence of a ballooning eigenvalue on the poloidal transit of
a magnetic line that was examined (compare with Fig. 4).
Accuracy of the NSTAB code adequate to address finer issues
of stability has been achieved by coupling a delicately arranged
difference scheme in the radial flux coordinate s to a spectral
Fig. 4. Values of ␤ observed in the LHD experiment exceed a pessimistic limit
found by integrating the ordinary differential equation of local ballooning
theory.
Garabedian
treatment of dependence on poloidal and toroidal angles within
the flux surfaces (7, 8). Convergence studies and comparisons
with physical observations have been performed to validate the
theory, and the method has been applied to design configura-
tions for a new stellarator experiment (14).
Transport
The three-dimensional problem of transport in stellarators is not
easy to handle by asymptotic expansions. However, it can be
formulated quite effectively as a linearized Fokker–Planck equa-
tion in many independent variables. One can apply the Monte
Carlo method in that analysis, which solves a drift kinetic
equation by alternating between a random walk to model the
collision operator and an ordinary differential equation solver to
track the orbits of charged particles (15). Separation of variables
shows that there are solutions that decay exponentially in time,
and they lead to a fast algorithm determining the confinement
times of the ions and electrons independently (16).
Quasineutrality requires that there be no difference between
the confinement times and distribution functions of the two
species. That can possibly be achieved at low collision frequen-
cies by iterating in a systematic fashion on the choice of the
electrostatic potential, which may be allowed to vary within the
magnetic surfaces (13). A flow in the background could be
adjusted to make the collision operator conserve momentum,
but numerical experiments show that, if the flow is not thermal,
it does not have enough effect on the collision operator to
account for quasineutrality (1). Small oscillations of the elec-
trostatic potential along the magnetic lines serve that purpose
and also model the ambipolar electric field and turbulence
tolerably well. A parallel version of the TRAN computer code
implementing this concept has been programmed for the IBM
Special Purpose computer at the National Energy Research
Computing Center by using OPENMP software.
Recent high-temperature shots in the LHD experiment have
produced observations of the energy confinement time at low
collision frequencies that can be simulated by the Monte Carlo
calculations we have described (2). Careful comparisons estab-
lish that the numerical method gives realistic results. Oscillations
of the electric potential along the magnetic lines model anom-
alous transport, so there is good agreement with measured values
(1). Similar agreement with experimental data for tokamaks is
obtained when bifurcated equilibria without perfect two-
dimensional symmetry are used in the computations (13).
For axially symmetric tokamaks, steady solutions obtained
from asymptotic expansions are used in the two-dimensional
theory of transport to predict intrinsic ambipolarity. This is not
easy to reconcile with steady solutions we have calculated by the
Monte Carlo method from exponential decay of the distribution
function, which have unequal sources of ions and electrons (1).
The TRAN code estimates how long an ion or an electron stays
in the plasma. Tracking orbits of the electrons is what reveals the
importance of quasineutrality in these calculations of transport.
Because the mean free path of the particles is orders of mag-
nitude longer than the circumference of the torus, it is hard to
see how conservation of momentum could make the numbers of
ions and electrons agree to several significant digits in a real
volume of plasma.
Coils
When running equilibrium, stability, and transport codes to
design stellarator experiments, one usually starts by looking for
Fourier coefficients specifying the boundary of a plasma with
good physical properties (7). After that it becomes necessary to
use a line-tracing code such as NESCOIL to find a set of coils
that generate a corresponding external magnetic field according
to the Biot–Savart law (17, 18). This is not a well posed
mathematical problem, because the coils may not satisfy phys-
Garabedian
ically reasonable constraints unless care is exercised in choosing
the coefficients that define the original equilibrium. It is advis-
able to limit the degree of the harmonics that are used and to
avoid shapes for the separatrix that are unnecessarily compli-
cated so that the final coils will furnish a robust equilibrium that
continues to have good Kolmogorov–Arnold–Moser surfaces
after realistic changes are made in the geometrical parameters of
the configuration. The coils become excessively twisted within
the control surface where they are wound if that is located too
far from the plasma. This problem is as important as physics
issues that sometimes receive more attention.
Unwanted islands are suppressed by manipulating resonant
harmonics in the Biot–Savart law. This seems to succeed if the
term in the magnetic spectrum with the same resonance as an
undesirable island turns out to be small in the original equilib-
rium. Such islands can be controlled by auxiliary shaping coils or
by modification of the larger modular coils that generate the
external magnetic field, which should be only moderately
twisted. Numerical calculations show that the quality of the
Kolmogorov–Arnold–Moser surfaces may improve if the coor-
dinates defining the coils are smoothed. One can also address
these issues for finite ␤ by running free-boundary codes (19).
APPLIED PHYSICAL
Design
The modular Helias-like Heliac 2 (MHH2) is a compact stel-
SCIENCES
larator designed by computational methods to perform well at
reactor conditions (1, 14). Sixteen moderately twisted modular
coils define an external magnetic field with rotational transform
near 2兾5 that has robust magnetic surfaces. There are two field
periods, and the aspect ratio of the plasma is 3.5. The nominal
width of a square cross section of the coil structure is a quarter
of the plasma radius. Quasiaxial symmetry of the magnetic
spectrum provides excellent confinement at low collision fre-
quencies. The shape of the separatrix produces a good magnetic
well and assures nonlinear stability up to an average ␤ of 4% for
ideal MHD modes (14). According to our transport theory, an
MHH2 experiment could be constructed with a major radius of
2 m that might achieve a peak ion temperature of 10 keV in a
magnetic field of 2 T at an electron density of 1.5 ⫻ 1013 cm⫺3.
The configuration can be operated effectively as a hybrid and can
be conceived of as an advanced tokamak.
It is a challenge to predict the performance of future exper-
iments by making numerical calculations. With that in mind, we
examined the National Compact Stellarator Experiment pro-
posal for a proof of principle stellarator experiment at the
Princeton Plasma Physics Laboratory, which will have a plasma
of aspect ratio 4.5 with three field periods. To get adequate space
between the separatrix and the first wall, the filaments defining
the modular coils become kinky, which may lead to deterioration
of the magnetic surfaces in the edge region. Consequently, we
ran the NSTAB and TRAN computer codes for an example with
less complicated geometry. Nonlinear stability of the alternate
configuration is favorable, and the ␤ limit exceeds 4%, but
islands may appear when the rotational transform ␫ crosses 3兾7,
3兾6, and 3兾5 near the separatrix.
Free-Boundary Code
We describe next a free-boundary algorithm for the design and
analysis of stellarator and tokamak equilibria. An important
application is the reconstruction of plasma geometry from
available diagnostics and observations. Correct mathematical
formulations of the problem are required not just to meet
acceptable standards of self-consistent physics but also to arrive
at computer simulations of nonlinear phenomena that can be
correlated with experimental observations to make predictions
going beyond those provided by standard scaling laws (20).
Some of the mathematical difficulties in understanding plasma
confinement in three dimensions have been explained already. A
PNAS 兩 November 25, 2003 兩 vol. 100 兩 no. 24 兩 13743
reliable free-boundary code is needed for the engineering
analysis of coil sets, and a satisfactory treatment of islands and
current sheets at resonant surfaces in discontinuous solutions of
the equilibrium problem is required. We present a mathematical
method to attack these questions by solving the free-boundary
value problem numerically.
The NSTAB code applies the MHD variational principle

冕冕冕 冉 1 2
2
冊
B ⫺ p dV ⫽ minimum [1]

to calculate weak solutions of the equilibrium equations that

satisfy the relations

B ⫽ ⵜs ⫻ ⵜ␪ ⫽ ⵜ␾ ⫺ ␨ⵜs [2]

for the magnetic field B in a plasma with fixed boundary defined

by the formula

r ⫹ iz ⫽ e iu 冘 ⌬ mne ⫺imu⫹inv, [3]

where r, v, and z are cylindrical coordinates and u is a poloidal

angle (8). In this theory the functions s, ␪, and ␾ comprise an
invariant system of flux coordinates and ␨ is an additional
Clebsch potential that specifies the distribution of current on the
family of nested toroidal surfaces s ⫽ constant. The strength of
the field in the plasma has a Fourier series representation
Fig. 5. Poincaré section of a free-boundary calculation for the MHH2
1
B2
⫽ 冘 B mncos共m␪ ⫺ 关n ⫺ ␫ m兴␾兲, [4]
stellarator displaying the control surface for the coils, the known shape of
the plasma at ␤ ⫽ 0, filaments that simulate currents in the plasma driven
by the pressure, and magnetic lines computed at finite ␤ in the scrape-off
layer of the solution, where the rotational transform is crossing the reso-
the coefficients Bmn ⫽ Bmn(s) of which comprise the magnetic nant value ␫ ⫽ 2兾5.
spectrum. Elementary transformation of the equilibrium equa-
tions shows that ␨ has a related representation
filaments modeling the Pfirsch–Schlueter current and the boot-
␨ ⫽ p⬘共s兲 冘 B mn
n ⫺ ␫m
sin共m␪ ⫺ 关n ⫺ ␫m兴␾兲 [5] strap current (cf. ref. 18).
In practice the method exploits line tracing to recalculate the
shape of the separatrix and eliminate islands from the solution.
with small denominators that vanish at resonant surfaces cor- This problem is not well posed because of the Kolmogorov–
responding to rational values of the rotational transform ␫. In this Arnold–Moser theory, which means that only an approximate
formulation, ⵜs ⫻ ⵜ␨ is the current density (16). The singularities answer should be sought (5). However, the approach is effective
occurring at rational surfaces explain why it is important to work
for the design and analysis of quasisymmetric stellarators if just
with weak solutions of the problem.
a few terms in the Fourier series for the Clebsch potential ␨ and
On an outer control surface, let ␨0 stand for a distribution of
a sparse set of flux surfaces s ⫽ sj suffice to give reliable results.
current with level curves that are filaments for a system of
Fig. 5 displays a Poincaré section of the magnetic lines in a
modular coils that generate the external magnetic field control-
computation for the MHH2 stellarator by a version of the
ling a stellarator equilibrium, and let ␨1, . . . , ␨k be corresponding
method that uses just one internal flux surface s ⫽ s1 in the
values of the Clebsch potential ␨ that describe distributions of
surface current on a discrete set of magnetic surfaces s ⫽ sj inside Biot–Savart law and one coefficient B10 in the spectral repre-
the plasma. The NSTAB code provides an accurate calculation sentation of ␨. In the reconstruction of the free boundary, the
of the magnetic spectrum Bmn when the shape factors ⌬mn location of the magnetic axis determined by the NSTAB code
defining the separatrix are known, and therefore it also provides plays a significant role. Finally, let us observe that the form of
a way to compute ␨j. Moreover, the Biot–Savart law the Biot–Savart law we have presented can also be applied more
directly to find modular coils for a prescribed equilibrium

冕冕
冘
k
ⵜ␨j ⫻ N
B⫽ⵜ⫻ ␦sj d␴
4␲r
j⫽0
gives an approximation of the magnetic field that is valid both
inside and outside the plasma, where the double integrals are
evaluated over the control surface and the discrete flux surfaces
s ⫽ sj, and ␦sj is an increment in sj.
To solve the free-boundary problem, what is required is to
adjust the Fourier coefficients ⌬mn such that the shape of the
plasma becomes compatible with magnetic lines in the solution.
In a preparatory code implementing this concept, which has
features in common with NESCOIL, the surface integrals from
the Biot–Savart law are approximated by line integrals over
configuration when ␤ is positive.
[6]
Reactor Studies
A reactor study of compact stellarators is currently being un-
dertaken by the US Department of Energy. Principal configu-
rations under consideration as candidates for the study are the
National Compact Stellarator Experiment with three field pe-
riods and the MHH2 with two field periods. Here we review
results from runs of our computer codes that were performed in
connection with this project.
As we have seen, NSTAB calculations show that the LHD
stellarator is linearly unstable but remains nonlinearly stable at
the ␤ of 3.2% that has been achieved experimentally. At a peak
ion temperature of 3 keV observed in the LHD, the TRAN code

13744 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.2436390100 Garabedian

predicts an energy confinement time of 160 ms that agrees with
the measured value (1). This validates our model of anomalous
transport based on considerations of quasineutrality. On the
other hand, computations of the energy confinement time ␶E in
an LHD reactor study call for major radius 25 m and plasma
radius 4.5 m at average T ⫽ 12 keV, n ⫽ 1.2 ⫻ 1014 cm⫺3, and
B ⫽ 5 T. That is consistent with the low ion temperatures
achieved thus far for a hydrogen plasma in the experiment (2).
Good correlation of our computations with observations in
the LHD has been applied to assess equilibrium, stability, and
transport for a compact MHH2 stellarator. According to our
transport theory, an MHH2 reactor could be constructed with a
major radius of 7.5 m and a plasma radius of 2.2 m that would
achieve average temperature T ⫽ 16 keV in a magnetic field of
5 T at average electron density n ⫽ 1014 cm⫺3. The energy
confinement time is ⬎2 s if the drop in the radial electric
potential is twice the temperature. Dependence of ion, electron,
and ␣-particle transport on the radial electric field and the mirror
coefficients ⌬01 and B01 has been examined, and transport can be
improved by using the mirror term to raise the radial electric
field.
The ␤ limit of the MHH2 power plant is 4%, and 16 only
moderately twisted coils provide robust magnetic surfaces.

APPLIED PHYSICAL
Above the ␤ limit a ballooning mode appears in the equilibrium

that is wall-stabilized and can be detected by long runs of the
NSTAB code (compare with Fig. 6). The gap between the
separatrix and central filaments defining the coils exceeds 140
cm everywhere, and there is room for a maintenance port
between any pair of adjacent coils. This fusion reactor is
designed to deliver 1 GW of electric power.
A principal issue for stellarators is the quality of the magnetic
surfaces, especially near the edge of the plasma. For compact
stellarators the rotational transform per field period is usually
large, so low-order resonances may become dangerous. In the
MHH2 reactor the rational surface ␫ ⫽ 2兾5 appears in the
scrape-off layer at low values of ␤, and the rational surface ␫ ⫽
2兾4 may emerge near the magnetic axis when ␤ becomes larger
so that there is significant bootstrap current. The resonant terms
B42 and B52 in the magnetic spectrum are both relatively small,
which means that the widths of the corresponding islands are not
expected to be large (1).
SCIENCES
Fig. 6. Four cross sections of the flux surfaces over one field period of a
wall-stabilized MHH2 equilibrium at ␤ ⫽ 0.05 with pressure p ⫽ p0(1 ⫺ s1.5)1.5
and with hybrid net current bringing the rotational transform into the interval
0.58 ⬎ ␫ ⬎ 0.52. The ballooning mode that has become visible in the solution
after 200,000 cycles of an accelerated iteration scheme shows that a ␤ limit has
been reached.
We have described results that illustrate the role of mathe-
matical and computational methods in magnetic fusion. To
complete the research it would be desirable to construct an
experiment for a quasiaxially symmetric stellarator such as the
MHH2 with two field periods and only moderately twisted coils.
I thank Frances Bauer, Long-Poe Ku, Farrokh Najmabadi, Shoichi
Okamura, and Harold Weitzner for helpful advice. This work was
supported by U.S. Department of Energy Grant DE-FG02-86ER53223.

1. Garabedian, P. (2002) Phys. Plasmas 9, 137–149.
2. Komori, A. & Yamada, H. (2000) Plasma Phys. Control. Fusion 42, 1165–1177.
3. Weller, A. & Geiger, J. (2001) Phys. Plasmas 8, 931–956.
4. Kolmogorov, A. (1954) Dokl. Akad. Nauk SSSR 98, 527–530.
5. Grad, H. (1967) Phys. Fluids 10, 137–154.
6. Garabedian, P. (1998) Commun. Pure Appl. Math. 51, 1019–1033.
7. Bauer, F., Betancourt, O. & Garabedian, P. (1984) Magnetohydrodynamic
Equilibrium and Stability of Stellarators (Springer, New York).
8. Taylor, M. (1994) J. Comput. Phys. 110, 407–418.
9. Hirshman, S. & Betancourt, O. (1991) J. Comput. Phys. 96, 99–109.
10. Hirshman, S. & Ku, L. (1999) Phys. Plasmas 6, 1858–1864.
11. Nuehrenberg, J. & Zille, R. (1988) Phys. Lett. 129A, 113–116.
12. Garabedian, P. (1997) Plasma Phys. Control. Fusion 39B, 129–134.
13. Garabedian, P. & Taylor, M. (1992) Nucl. Fusion 32, 265–270.
14. Garabedian, P. (2002) Proc. Natl. Acad. Sci. USA 99, 10257–10259.
15. Boozer, A. & Kuo-Petravic, G. (1981) Phys. Fluids 24, 851–859.
16. Bauer, F., Betancourt, O., Garabedian, P. & Wakatani, M. (1987) The Beta
Equilibrium, Stability and Transport Codes (Academic, Boston).
17. Anderson, D. & Garabedian, P. (1994) Nucl. Fusion 34, 881–885.
18. Merkel, P. (1987) Nucl. Fusion 27, 867–871.
19. Cooper, W. (2000) Phys. Plasmas 7, 2546–2553.
20. Weitzner, H. (1971) Phys. Fluids 14, 658–670.

Garabedian PNAS 兩 November 25, 2003 兩 vol. 100 兩 no. 24 兩 13745