Available online at www.sciencedirect.

com

ScienceDirect
Advances in Space Research 57 (2016) 1585–1603
www.elsevier.com/locate/asr

A parametric sizing model for Molten Regolith Electrolysis reactors

to produce oxygen on the Moon
Samuel S. Schreiner a,⇑,1, Laurent Sibille b,2, Jesus A. Dominguez b,3, Jeffrey A. Hoffman a,4
a
Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, United States
b
NASA Kennedy Space Center, FL 32899, United States

Received 9 August 2015; received in revised form 4 January 2016; accepted 7 January 2016
Available online 14 January 2016

Abstract

We present a parametric sizing model for a Molten Regolith Electrolysis (MRE) reactor that produces oxygen and molten metals
from lunar regolith. The model has a foundation of regolith material property models validated using data from Apollo samples and
simulants. A multiphysics simulation of an MRE reactor is developed and leveraged to generate a database linking reactor design
and performance trends. A novel design methodology is created which utilizes this database to parametrically design an MRE reactor
that can (1) sustain the required current, operating temperature, and mass of molten regolith to meet a desired oxygen production level,
(2) operate for long periods of time by protecting the reactor walls from the corrosive molten regolith with a layer of solid ‘‘frozen”
regolith, and (3) support a range of electrode separations to enable operational flexibility. Mass, power, and performance estimates
for an MRE reactor are presented for a range of oxygen production levels. Sensitivity analyses are presented for several design variables,
including operating temperature, regolith feedstock composition, and the degree of operational flexibility.
Ó 2016 COSPAR. Published by Elsevier Ltd. All rights reserved.

Keywords: Molten Regolith Electrolysis; In situ resource utilization; Lunar oxygen production; Moon; Molten oxide electrolysis; Parametric modeling

1. Introduction This paradigm has resulted in prohibitive launch costs on

1.1. In-situ resource utilization (ISRU) overview
One of the most significant barriers to space exploration
is the fact that of the material resources required for a mis-
sion must be transported from Earth. The rocket equation
(Turner, 2008) describes how a small increase in payload
mass results in a dramatic increase launch system mass.
⇑ Corresponding author.
E-mail address: schr0910@umn.edu (S.S. Schreiner).
1
NASA Space Technology Research Fellow, MIT Department of
Aeronautics and Astronautics.
2
Surface Systems Group, ESC-5, NASA Kennedy Space Center.
3
VENCORE-ESC, NASA Kennedy Space Center.
4
Professor of the Practice, MIT Department of Aeronautics and
Astronautics.
the order of $110,000/kg to the lunar surface (Diaz et al.,
2005), in 2015 dollars. To enable sustainable, affordable
exploration of the solar system, the reliance on Earth’s
resources must be reduced.
In-situ resource utilization (ISRU) is ‘‘the collection,
processing, storing and use of materials encountered in the
course of human or robotic space exploration that replace
materials that would otherwise be brought from Earth”
(Sacksteder and Sanders, 2007). One form of ISRU is pro-
ducing oxygen from lunar regolith, the top layer of loose
material covering the lunar surface. Oxygen is a major
component of launch vehicle, spacecraft, and lander masses
– 80% of launch vehicle mass is fuel and oxygen, which
itself is 7/8 oxygen by weight (Altenberg, 1990). At the
same time, oxygen is one of the most abundant lunar
resources – lunar regolith is more than 40% oxygen by

http://dx.doi.org/10.1016/j.asr.2016.01.006
0273-1177/Ó 2016 COSPAR. Published by Elsevier Ltd. All rights reserved.
1586 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
weight (Schreiner et al., 2016). The production of this valu-
able resource outside of Earth’s gravity well can support
lunar surface activities and enable orbital refueling to
reduce the amount of resources that must be launched from
Earth.
1.2. Molten Regolith Electrolysis (MRE) overview
Over twenty different techniques to produce oxygen
from lunar regolith have been proposed in the literature
(Taylor and Carrier, 1993). These techniques range from
solid–gas reactions, such as Hydrogen Reduction of Ilme-
nite (Gibson and Knudsen, 1985), to molten-gas reactions,
such as Carbothermal Reduction (Balasubramaniam et al.,
2010), and even direct dissociation via pyrolysis (Senior,
1993). A range of electrochemical techniques have also
been proposed. Some, such Fluxed Magma Electrolysis
(Keller and Taberaux, 1991), require fluxing reagents to
be mixed with molten regolith to support the electrochem-
ical process. The FFC–Cambridge process (Chen et al.,
2000; Schwandt et al., 2012) directly reduces the solid oxi-
des in lunar regolith using a supporting molten salt.
A promising electrochemical technique, called Molten
Regolith Electrolysis (MRE) requires no additives, but
rather performs direct electrolysis on raw molten lunar
regolith (Kesterke, 1971). In the MRE process (also known
as Molten Oxide Electrolysis), granular lunar regolith is fed
into a reactor and heated to a molten state. Molten lunar
regolith is conductive enough to sustain electrolysis, in
which a voltage is applied across two electrodes immersed
in the molten regolith to drive a current. Oxygen gas is pro-
duced at the anode and molten iron, silicon, aluminum,
and titanium, are produced at the cathode.
There is a strong impetus to explore the feasibility of an
ISRU system based on the MRE process. A study compar-
ing oxygen production methods, including water extraction
from lunar polar craters, identified MRE as having the
most favorable power consumption and the second most
favorable mass throughput (Mason, 1992). MRE does
not require additional process materials (such as hydrogen,
methane, or fluxing reagents) or the systems required to
recycle them, leading to a decrease in system mass and
complexity. MRE can extract up to 95% of the oxygen
from lunar regolith, which decreases regolith throughput
requirements and reactor mass and power. With post-
reactor processing, the byproducts of MRE (molten iron,
silicon, aluminum, titanium, and glassy slag) can be used
to produce infrastructure, spare parts and even solar arrays
on the lunar surface (Landis, 2005; Curreri et al., 2006). In
fact, MRE is also being developed for environmentally-
friendly metal production on Earth with a zero-carbon
footprint (Gmitter, 2008; Vai et al., 2010).
1.3. Historical development of MRE
The process of electrolytically extracting metals from
molten ores was first patented by Aiken (1906) and Carr
(1963) first suggested using such a process to produce
oxygen from lunar soil. Kesterke (1971) conducted the
earliest experimental work, using an iridium anode and
silica-carbide cathode to electrolytically produce 170 g of
oxygen and significant amounts of iron, titanium, alu-
minum, and magnesium from a lunar simulant. In the
ensuing decades, MRE underwent considerable laboratory
development (Lindstrom and Haskin, 1979; Jarrett et al.,
1980; Carroll, 1983; Haskin et al., 1992).
Under the Vision for Space Exploration, further MRE
development continued at MIT (Kim et al., 2011; Wang
et al., 2011), Kennedy Space Center (Sibille et al., 2009),
Ohio State University (Standish, 2010), and Marshall
Space Flight Center (Curreri et al., 2006). This work
demonstrated many suitable anode materials, including
iridium, platinum, rhodium (Gmitter, 2008), iridium-
plated graphite (Paramore, 2010), 50–50 iridium/tungsten
alloys (Vai et al., 2010), and iron–chromium alloys
(Allanore et al., 2013). Reactor designs were scaled up from
0.1 A with 0.3 cm2 electrodes to 10 A with 10 cm2 elec-
trodes (Vai et al., 2010; Sirk et al., 2010), making progress
from laboratory-scale testing towards technology demon-
stration levels.
1.4. Molten regolith containment: joule heated cold wall
operation
Perhaps the most challenging design consideration in an
MRE reactor is the containment of corrosive molten rego-
lith. The longest laboratory experiments lasted on the order
of a few hours (Paramore, 2010) before molten regolith
eroded through the inner crucible. Paramore (2010) noted
that ‘‘fortunately, this frustration [crucible failure] is purely
an artifact of laboratory-scale experimentation”.
To solve the challenge of molten regolith containment,
an MRE reactor can be designed similarly to industrial
Hall–Heroult reactors to support joule heated cold wall
operation (Carroll, 1983; Sibille and Dominguez, 2012).
In joule heated cold wall operation, the current flowing
through the molten regolith generates joule heating (I 2 R)
to maintain a central molten core, and the thermal gradi-
ents are designed such that the reactor walls are protected
from the molten core by a layer of solid ‘‘frozen” regolith
to enable long duration operation as shown in Fig. 1.
The current streamlines are depicted in red and simplified
anodic and cathodic reactions are indicated. Oxygen is pro-
duced at the anode and exits the reactor through a port
(not shown).
1.5. Motivation for MRE reactor modeling
Although a number of analyses have examined the util-
ity of lunar oxygen production (Simon, 1985; Joosten and
Guerra, 1993; Duke et al., 2003), the accuracy of their find-
ings depends heavily on the mass and performance of the
ISRU systems involved. Sherwood and Woodcock (1993)
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
Fig. 1. Schematic of a Molten Regolith Electrolysis reactor that produces
oxygen gas at the anode and molten metals at the cathode. The central
molten core is insulated by solid ‘‘frozen” regolith around the reactor
perimeter.
conducted an extensive economic analysis and calculated a
cost of $18,370/kg (2015 dollars) to produce oxygen on the
lunar surface using the Hydrogen Reduction of Ilmenite
process. However, they noted that, ‘‘the sensitivities [of
their economic model] are modest, except for the mass of
production hardware”. When the ISRU system mass was
varied by a factor of two, the oxygen production cost
varied from $12,570/kg to $29,857/kg (2015 dollars).
Simon (1985) showed that, after the Earth-to-Moon trans-
portation cost, the power required for lunar ISRU had the
biggest impact on its economic feasibility. To understand
the appropriate applications of lunar ISRU, it is imperative
to accurately model the mass, power and performance of
such systems.
Furthermore, there is a large impetus to model MRE
reactors. Altenberg (1990) noted that two of the primary
issues associated with MRE were (1) the lack of specified
optimal process conditions, feed rate, and feedstock
requirements and (2) a poor understanding of the meaning-
ful design parameters and oxygen extraction efficiency.
Teeple (1994) surmised that ‘‘the electrolysis techniques [in-
cluding MRE], involve high temperatures, so one would
expect high plant masses”. Uncertainty on the operating
conditions and mass and power of an MRE reactor can
be reduced through modeling.
Although parametric models to predict the mass
and power of reactors utilizing other lunar ISRU processes
have been developed (Hegde et al., 2009; Balasubramaniam
et al., 2010), a similar model of suitable fidelity for a lunar
MRE reactor does not yet exist in the literature. This deficit
has prevented quantitative comparisons between MRE and
other processing techniques (Chepko, 2009). A substantial
amount of work modeling electrochemical reactors exists in
the literature (Inman and White, 1978; Gerogiorgis and
Ydstie, 2005; Kar and Evans, 2008; Kennedy, 2012), but
this work has primarily focused on terrestrial reactors
which differ from lunar reactors in a number of ways
(Carroll, 1983). MRE reactors operate on raw molten
regolith without the fluxing reagents commonly used in
1587
terrestrial reactors, which results in significantly different
reactor behavior. The differences in process chemistry
between lunar and terrestrial techniques significantly affects
reactor design (Inman and White, 1978), inhibiting the
ability to translate models for terrestrial reactors to lunar
applications (Poizeau and Sadoway, 2011). Furthermore,
in lunar applications the thermal behavior of the external
reactor surface will consist only of radiative heat transfer
and may include multi-layer insulation, which will decrease
surface emissivity as compared to terrestrial reactors.
There has been some preliminary simulation of lunar
MRE reactors in the literature. Sibille and Dominguez
(2012) demonstrated that a joule heated cold wall MRE
reactor appears feasible with lunar regolith feedstock, but
they studied a point design rather than developing a para-
metric model capable of studying reactor performance
across a range of production levels. Previous work (both
simulation and experimental) has studied reactors that
use tens of amperes of current, while production-level reac-
tors will need to use on the order of kiloamperes. The
proper method for scaling up MRE reactor designs to these
higher production levels remains to be explored.
1.6. MRE reactor model overview
To this end, we develop a sizing model for an MRE
reactor that parametrically generates reactor design and
performance estimates for a set of inputs, including oxygen
production level, operating temperature, and regolith feed-
stock type. This model can be used to (1) quantitatively
compare MRE to other lunar ISRU techniques, (2) study
how MRE reactors scale to higher production levels, and
(3) improve our understanding of optimal process condi-
tions and feedstock requirements.
The MRE reactor model has a foundation of lunar rego-
lith material property models which were validated using
data from Apollo samples and regolith simulants
(Schreiner et al., 2016). The reactor model calculates the
amount of regolith (Section 2) and current (Section 3)
required to meet the desired oxygen production level. A
multiphysics simulation, which models the electrochemical
and thermodynamic behavior of an MRE reactor, is devel-
oped and leveraged to create an extensive tradespace of
reactor designs (Section 4).
As described in Section 5, the reactor model parametri-
cally generates a reactor design that (1) sustains the
required current, operating temperature, and mass of mol-
ten regolith to meet a desired oxygen production level, (2)
ensures that the reactor walls are insulated from the molten
core by a layer of solid lunar regolith during joule heated
cold wall operation (see Section 1.4), and (3) enables a
range of viable electrode separations to support flexible
reactor operation. The methodology for estimating reactor
power and mass is presented in Section 6. The mass, power,
and performance estimates are given for a range of oxygen
production levels in Section 7, including sensitivity analyses
for several design variables.
1588 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603

2. Calculating regolith processing requirements
Lunar regolith is primarily comprised of plagioclase,
pyroxene, olivine, and ilmenite, and each of these minerals
are composed of oxides, including iron(II) oxide (FeO), sil-
ica (SiO2), alumina (Al2O3), titania (TiO2), magnesia
(MgO), and calcium oxide (CaO). Data from Schreiner
et al. (2016) for the oxide compositions of three different
types of lunar regolith were used to characterize the rego-
lith feedstock in the MRE reactor model. Fig. 2 shows
the composition data from Schreiner et al. (2016) with a
map of the lunar surface with the lunar High-Ti Mare (yel-
low), Low-Ti Mare (cyan) and Highlands (older forma-
tions in red, younger formations in blue), from the
Clementine UVVIS instrument (Lucey et al., 2000).
The MRE process utilizes a current flowing through
molten regolith to electrochemically produce oxygen. The
reactions at the interface between the electrodes and the
raw molten regolith are presented below in simplified form
using the reduction of iron oxide (FeO) as an example
(Kesterke, 1971; Colson and Haskin, 1990):
½Net Cathode Reaction Fe2þ þ 2e ! Fe0
ð1Þ
½Net Anode Reaction 2O2 ! O2 þ 4e
Colson and Haskin (1990) noted that, due to the slow
kinetics of O2 formation from silicate polymer chains,
the actual anodic reaction will primarily involve the oxida-
tion of polymer chains to produce oxygen in the neutral
state, but Eq. (1) still represents the correct stoichiometry.
The MRE model calculates the regolith processing
requirement according to the reactions shown in Eq. (1)
adapted for each oxide specie in lunar regolith. The
amount of oxygen extracted per kilogram regolith, termed
the ‘‘oxygen extraction yield” (kO2 ), is calculated as:
X  
mO2 MW O2
k O2 , ¼ ðwi Þ ðrmol;i Þðefrac;i Þ; ð2Þ
mregolith i
MW oxide;i
where mO2 is the mass of oxygen produced, mregolith is the
mass of regolith processed, wi is the weight percent of oxide
i in lunar regolith from Schreiner et al. (2016), MW oxide;i is
the molecular weight of oxide i; rmol;i is the number of moles
of oxygen in one mole of oxide i, and efrac;i is the fraction of
oxide i that is reduced in each batch (between 0 and 1, to
account for the fact that certain oxides may be partially
reduced or not reduced at all). In the MRE reactor model,
Eq. (2) is used to calculate the amount of regolith that must
be processed to meet a given oxygen demand.
The fraction of each specie that can be electrolyzed
(efrac;i ) is dependent upon the operating temperature,
because the liquidus temperature of molten regolith gener-
ally increases as the composition changes during electroly-
sis. Fig. 3 shows the liquidus temperature throughout the
batch electrolysis calculated using phase diagrams from
Slag Atlas (Allibert et al., 1995). To capture the time-
varying behavior throughout a batch, the model assumes

Fig. 2. The composition of three different types of lunar regolith: High-Titanium Mare (yellow), Low-Titanium Mare (cyan), and Highlands (older rock in
red, younger rock in blue), from Schreiner et al. (2016). Composition data from Apollo and Luna missions (Stoeser et al., 2010) and imagery data from
Clementine UVVIS instrument (Lucey et al., 2000). (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603 1589

Fig. 3. As electrolysis progresses, the composition and properties of the molten regolith vary. (top) The liquidus temperature, calculated from Slag Atlas
(Allibert et al., 1995), for Low-Ti Mare lunar regolith initially increases as electrolysis progresses. As the operating temperature increases, the reactor can
extract more oxygen per unit regolith. (bottom) Highlands regolith initially has a higher liquidus temperature than Mare, but as electrolysis progresses the
liquidus temperature for Mare regolith rises dramatically higher.

that oxide species are discretely reduced in the order of MgO and TiO2 leads to a general decrease in liquidus tem-
increasingly negative Gibbs Free Energy. As depicted in perature. Finally, as Al2O3 is reduced the liquidus temper-
the top plot, after FeO is reduced, the liquidus temperature ature first decreases and then increases to a plateau for a
increases as SiO2 is reduced. After SiO2, the reduction of mixture dominated by the remaining CaO.
1590 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
The bottom plot in Fig. 3 compares the liquidus temper-
ature curves for Mare and Highlands lunar regolith, which
vary due to differences in composition. Although the High-
lands regolith has a higher liquidus temperature early in the
batch, the liquidus temperature for Mare regolith quickly
rises above that of Highlands. This is due to the fact that
Mare regolith has more MgO while Highlands regolith
has more Al2O3 (see Fig. 2).
Electrolysis is allowed to progress until the liquidus
temperature comes within a 50 K safety margin of the
operating temperature. At this point fresh regolith is added
while the leftover slag and molten metals are removed, and
the batch process begins again. As shown in the top plot of
Fig. 3, with an operating temperature of 1900 K, the reac-
tor can reduce approximately 1/2 of the SiO2 for High-Ti
Mare regolith (kO2 ¼ 0:19 kg O2/kg regolith) and High-
lands regolith (kO2 ¼ 0:13) before needing to start a new
batch with fresh regolith. If the operating temperature is
raised to 2100 K, the reactor can now reduce 5/6 of the
SiO2 for High-Ti Mare (kO2 ¼ 0:24), and all of the SiO2,
MgO, and TiO2 for Highlands (kO2 ¼ 0:35) Thus, higher
operating temperatures increase the oxygen extraction yield
and enable a more diverse set of oxides to be reduced. For
all species, efrac;i had a maximum of 0.95 to account for the
fact that when the concentration of a specie becomes too
low, the specie will no longer be preferentially reduced in
favor of more abundant species.
In addition to avoiding solidification of the molten core,
there is another important aspect in the batch operation of
an MRE reactor. As FeO and SiO2 are reduced, the residual
melt moves towards the spinel stability field and a MgAl2O4
spinel begins to precipitate (Colson and Haskin, 1990). This
spinel must be periodically removed from the reactor to
enable further regolith processing. Future work can investi-
gate methods for managing spinel production, either
through mechanical or electrochemical means.
Using the fraction electrolyzed (efrac;i ) of each oxide in
lunar regolith, the amount of molten metal produced can
also be calculated:
 
MW metal;i
mmetal;i ¼ ðmregolith Þðwi Þ ðefrac;i Þ; ð3Þ
MW oxide;i
where mmetal;i is the mass of metal produced from oxide i
and MW metal;i is the molecular weight of the product metal
i.
3. Calculating reactor current
The MRE reactor model also calculates the current
required to meet the desired oxygen production level. Fara-
day’s law (Faraday, 1834), with an adjustment for the
expected current efficiency, is used to calculate the number
of coulombs (C) required to produce a given mass of
oxygen:
    
nF mO2 nF
C ¼ ðnO2 Þ ¼ ; ð4Þ
g
I MW O2 g
I
where nO2 is the desired moles of diatomic oxygen, n is the
number of electrons required to reduce diatomic oxygen
(=4), F is Faraday’s constant, and g
I is the average current
efficiency defined as the ratio of the oxygen mass produced
to the theoretical prediction of Faraday’s law.
Using Eq. (2) to substitute for mO2 and including the
oxide current efficiencies in the summation yields:
X ðwi Þðrmol;i Þðefrac;i Þ
C ¼ ðmregolith ÞðnF Þ ; ð5Þ
i
ðMW oxide;i ÞðgI;i Þ
where gI;i is the expected current efficiency while oxide i is
the primary specie being reduced.
From experimental work conducted at MIT, the
expected current efficiency while electrolyzing iron-
bearing molten mixtures is 30–60% (Sirk et al., 2010) due
to the redox cycling of multivalent ions (Yen, 1977; Chen
et al., 2000). As a conservative estimate, the lower bound
of 30% is used for FeO. A current efficiency of 50% is used
for TiO2 due to a similar redox cycling phenomena (Chen
et al., 2000). Na2O, P2O5, K2O, and MgO may form gas-
eous products when reduced (Lindstrom and Haskin,
1979; Sirk et al., 2010), which will likely recombine with
some of the product O2 gas above the melt. This cyclic
behavior is captured with an estimated current efficiency
of 50%, though this effect can be mitigated with an oxygen
collection tube near the anode (Sirk et al., 2010). For all
other species in lunar regolith, a current efficiency of close
to 100% can be expected (Sibille et al., 2010) but a value of
95% is used as a conservative estimate.
Using the number of coulombs predicted by Eq. (5), the
average current (I avg ) over an entire batch can be
calculated:
C
I avg ¼ ; ð6Þ
ðtbatch  tfeed Þ
where tbatch is the total batch time and tfeed is an assumed
5 min downtime between each batch during which electrol-
ysis is not performed, to be conservative.
The model assumes that the reactor will be operated in a
constant power mode to minimize peak power. Constant
power is achieved by varying the current (about an average
value of I avg ) as the voltage changes due to different species
being electrolyzed and the conductivity of the melt chang-
ing throughout the batch. For instance, for an oxygen pro-
duction level of 500 kg/year, the model predicts a required
operating voltage of around 0.96 V while FeO is the pri-
mary object of electrolysis and 5.61 V while SiO2 is the pri-
mary object of electrolysis. The respective operating
current for these two species is 1470 A and 250 A. Both
of these operating conditions result in a power consump-
tion of 1.4 kW.
4. Multiphysics reactor simulation
A multiphysics simulation, presented in Schreiner et al.
(2015), was developed to predict the coupled electrochemical
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603 1591

and thermodynamic behavior of an MRE reactor. Fig. 4
shows a cross-sectional side view of a simulated reactor,
which is revolved around the dashed axis on the left to gen-
erate a 3D reactor design (shown in Fig. 1). The rainbow line
illustrates the phase transition between the inner core of mol-
ten regolith and the outer solid ‘‘frozen” regolith layer that
insulates the reactor walls. Current streamlines in the molten
regolith and electrodes are depicted in black.
Oxygen is produced at the anode, travels through the
overhead cavity, and leaves the reactor through an exit
port. Due to the fact that this study was limited to an
axis-symmetric simulation, this exit port was not included.
Furthermore, modeling oxygen gas flow in the simulation
dramatically impeded convergence and was thus omitted
to enable the large parametric sweeps described below.
There are four primary modes of heat transfer in the
simulation: conduction (the blue arrows in Fig. 4),
surface-to-surface radiation in the overhead cavity
(magenta arrows), radiation to ambient on the exterior sur-
face (black arrows), and radiation in participating media
(red arrows). Radiation in participating media models the
emission, absorption, and scattering of radiation within
the high temperature regolith (Arndt et al., 1979; Sibille
and Dominguez, 2012).
As presented in Schreiner et al. (2015), data on the
Enthalpy of Formation, Gibbs Free Energy, specific heat
and latent heat of melting for lunar regolith summarized
in Schreiner et al. (2016) were leveraged to add heatsinks
for the endothermic chemical electrolysis and the heating
of newly added regolith in molten region.
The multiphysics simulation is a steady-state approxima-
tion of the time-varying behavior of the reactor. Although
the thermal performance of the reactor will vary over time
as different oxide species are reduced, this variance is some-
what mitigated by the constant power operational concept.
As described in Section 3, to maintain constant power the
current can be varied inversely with voltage (Ptotal = I  V
is constant). Heat is generated in the molten region by resis-
tive heating from the over-voltage above the decomposition
potential (Vover) multiplied by the current (Pheat = -
I  Vover). Constant power operation does not ensure that
the heating power in the reactor is constant over time, but
it does reduce the variation in this value. Even as heating
power is held relatively constant, the endothermic heatsink
will vary over time as different oxide species are reduced.
Thus, the simulation presented in this section represents a
steady-state approximation of the time-varying behavior
of the reactor. Future work will need to study the complex,
time-varying behavior of an MRE reactor.
5. Using the simulation to guide reactor design
There are four primary criteria that must be satisfied by
an MRE reactor design. The first two, the required molten
mass and current in the reactor, are derived from the oxy-
gen production level using Eqs. (2) and (6), respectively.
The third, operating temperature, is a user input. The
fourth criteria is that the reactor design can sustain the
joule heated cold wall operation described in Section 1.4.
To generate a reactor design that meets these performance
criteria, the reactor diameter, wall thermal conductivity,
and electrode separation must be carefully designed. This
section gives an overview of a novel design methodology
developed to address this challenge.
5.1. Effects of molten mass and operating temperature: the
cutoff line
The multiphysics simulation described in Section 4 was
leveraged to generate a database of over 100,000 reactor

Fig. 4. A cross-section view of the cylindrical MRE reactor in the multiphysics simulation. The heat fluxes modeled are shown with different colored
arrows, scaled logarithmically by the same factor. The 3D reactor geometry is generated by revolving this cross section around the axis of symmetry on the
left. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
1592 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
designs (consisting of unique combinations of reactor
diameter, electrode separation, wall thermal conductivity,
current, and regolith feedstock type) with associated per-
formance characteristics (molten mass within the reactor,
operating temperature, and whether or not the joule heated
cold wall condition was satisfied). A small sample of the
data for a representative reactor design, with 2 kA of cur-
rent and a wall thermal conductivity of 5.0 W/m-K, is
shown in Fig. 5. The mass of molten regolith in the reactor
(left) and operating temperature (right) are plotted over a
range of reactor diameters, where each line represents a dif-
ferent electrode separation. The molten mass on the left of
Fig. 5 first rises with the diameter squared because all of the
regolith within the reactor is molten. Beyond a certain
inflection point, as reactor diameter increases solid regolith
begins forming at the outer wall of the reactor, decreasing
the mass of molten regolith in the reactor as discussed by
Schreiner (2015).
Data points in Fig. 5 with a red ‘‘X” overlaid represent
infeasible designs, in which the reactor wall temperature
rises above 1400 K and becomes too close to the melting
temperature of lunar regolith circa 1500 K. Lower cutoff
temperatures were evaluated by Schreiner (2015) and found
to be less optimal in terms of both reactor mass and power.
The reactor design is chosen using the line that separates
infeasible and feasible design regions, termed the ‘‘cutoff
line” as shown in Fig. 5. The choice to design reactors on
the cutoff lines was made after an initial analysis revealed
that those designs had the lowest heat loss per kilogram
of molten regolith within the reactor (Schreiner, 2015).
Fig. 5. The molten mass in an MRE reactor (left) and the operating temperature (right) depend on reactor diameter and electrode separation. Red X’s
indicate infeasible designs for which the side wall temperature comes within 100 K of the melting temperature of regolith (1500 K). The ‘‘cutoff line”
separates the infeasible and feasible design regions. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)
It is critical to note that the cutoff lines illustrated in
Fig. 5 are for a fixed current and wall thermal conductivity.
The range of designs simulated in the parametric sweep
described in Section 4 included reactor currents from
250 A to 3 kA and wall conductivities from 0.05 W/m-K
to 5 W/m-K. It was observed that current and wall conduc-
tivity also affect reactor performance and the cutoff lines,
but for the sake of conciseness, these plots are not included
here. The reader is directed to Schreiner (2015) for a more
detailed discussion of the effects of current and wall ther-
mal conductivity on reactor performance.
5.2. Designing the electrode separation
As is evident in the left-hand plot in Fig. 5, as the reac-
tor diameter increases the electrode separation must also
increase (moving from the green line towards the black
one) in order to stay on the cutoff line and maintain joule
heated cold wall operation. The database generated using
the reactor simulation was fit with an empirical equation
for the electrode separation that enables joule heated cold
wall operation for a given reactor design of diameter (D),
current (I) and wall thermal conductivity (k wall ):
0 1
B e3 ðD  e2 Þ C
De ¼ e1 I e8 ðk wall Þ 7 exp @ 
e
e 6 A; ð7Þ
e4
I e5 kwall þ1
where ei are regression coefficients (shown in Table 1)
which depend upon regolith type.
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603 1593

Table 1
The regression coefficients for Eq. (7), which predicts the appropriate
electrode separation needed to maintain thermal equilibrium in an MRE
reactor.
Regolith e1 e2 e3 e4
Highlands 0.0011 0.1945 2.5738 0.6487
Mare 0.0008 0.0458 4.6638 1.3147
Regolith e5 e6 e7 e8
Highlands 0.4656 0.3522 0.0570 0.4348
Mare 0.5827 0.3243 0.0064 0.5857

The electrode separation scaling equation (Eq. (7)) is

plotted in Fig. 6 against a small sample of the data from
the multiphysics simulation. The required electrode separa-
tion increases with reactor diameter, decreases with current
(left plot), and increases with wall conductivity (right plot)
to ensure that the reactor does not generate too much heat
and violate the joule heated cold wall constraint described
in Section 1.4.

5.3. Bounding reactor diameter and electrode separation

The cutoff lines in Fig. 5 prove useful for defining bounds

on the reactor design, as shown in Fig. 7. Succinctly, (1) the
molten mass requirement sets a minimum feasible reactor
diameter, Dmin , (2) the operating temperature sets a maxi-
mum feasible reactor diameter, Dmax , and (3) Dmax and
Dmin depend on the reactor current and the wall thermal
conductivity. The cutoff line represents the minimal reactor
diameter that can meet the required amount of molten rego-
lith in the reactor while ensuring that the molten regolith
core is insulated from the reactor walls by a layer of
‘‘frozen” solid regolith (left plot in Fig. 7). As shown in
Fig. 5, decreasing the reactor diameter (to the left of the cut-
off line) crosses into the infeasible region.
As can be seen in the right plot Fig. 7, the cutoff line rep- Fig. 6. The required electrode separation for a given reactor design. The
resents the maximum reactor diameter that can meet the electrode separation is affected by current (top) and wall thermal
required operating temperature in the reactor while ensuring conductivity (bottom). A regression model, illustrated by the solid line,
was fit to the multiphysics data, depicted by the data points connected by
that no molten regolith touches the reactor wall. As shown in
dashed lines (R2=0.975, RMSE = 0.0032 m).
the right plot of Fig. 5, increasing the diameter to the right of
the cutoff line crosses into the infeasible region, meaning that
molten regolith will touch the reactor wall.
Fig. 7 depicts a scenario in which 40 kg of molten mass in The molten mass and operating temperature data from
the reactor and a current of 2.5 kA are required to meet the the multiphysics simulation (of which only a small portion
desired oxygen production level and the user sets an operat- is shown in Figs. 5–7) were fit with empirical regression
ing temperature of 2250 K. The 40 kg of molten mass equations to predict the molten mass (M m ) and operating
imposes a minimum reactor diameter of 1.27 m for a current temperature (T op ) for a given reactor design:
of 2.5 kA (left plot in Fig. 7). The 2250 K operating temper- 0 1
m ð D  m Þ
ature imposes a maximum feasible diameter of 1.4 m for a M m ¼ I m6 m4 exp @  A;
1 5
ð8Þ
current of 2.5 kA (right plot in Fig. 7). As described in Sec- I m3
m2 þ kwall þ0:15
1
tion 5.1, the cutoff lines are a function of wall thermal con- 0 1
ductivity, so the bounds on reactor diameter can be t ð D  t Þ
T op ¼ t1 þ I t5 t2 exp @  A;
3 7
manipulated by varying that design parameter. Due to the ð9Þ
fact that there is a one-to-one, monotonic mapping between I t6 1 þ kwalltþ0:154

diameter and electrode separation, the bounds on reactor

where mi and ti are regression coefficients which depend
diameter also translate into bounds on the electrode separa-
upon the type of regolith as shown in Table 2.
tion, calculated by substituting Dmin and Dmax into Eq. (7).
1594 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603

Fig. 7. An example of how the MRE model calculates the bounds on reactor diameter. In this example (1) for a given reactor current of 2.5 kA, (2) the
required molten mass in the reactor (40 kg) sets a lower bound on the reactor diameter of 1.27 m (left plot) and (3) the desired operating temperature
(2250 K) imposes an upper bound on reactor diameter of 1.4 m (right plot). This example is for a reactor operating on High-Ti Mare regolith with a wall
thermal conductivity of 5.0 W/m-K.

Eqs. (8) and (9) can be algebraically manipulated to Due to the fact that the operating temperature (T op ) is
derive expressions for the minimum and maximum diame- set by the user, and the molten mass (M m ) and current (I)
ter bounds illustrated in Fig. 7: are fixed for a given oxygen production level and batch
   time (via Eqs. (2) and (6), respectively), the thermal con-
I m3 Mm 1
Dmin ¼ m5 þ log m2 þ ð10Þ ductivity of the wall is the only free parameter in Eq.
m1 m4 I m6 k wall þ 0:15
   (12). By substituting Eqs. (10) and (11) into Eq. (12), a
I t6 T op  t1 t4 closed-form expression for the required wall thermal con-
Dmax ¼ t7  log 1 þ ð11Þ
t3 I t5 t 2 k wall þ 0:15 ductivity can be derived:
 m  
T t
I m3 Ut3 log mM4 Im  I t6 m1 t4 log Iopt5 t2 1
6

5.4. Enabling operational flexibility: the design margin k wall ¼ ;

b
 
During reactor operation, the electrode separation must T op  t1
where b  I t6 m1 log  m1 t3 t7 . . . þ Um1 m5 t3
be varied to control the thermal behavior of the reactor. To I t5 t 2
 
quantify operational flexibility, a new design variable was Mm
introduced, termed the ‘‘design margin” (U): þ I m3 Um2 t3 log  0:15 ð13Þ
m4 I m6
Dmax ðT op ; I; k wall Þ The thermal conductivity prescribed by Eq. (13), with the
U, ; ð12Þ
Dmin ðMM; I; k wall Þ maximum diameter from Eq. (11), (1) satisfies the molten
which represents the ratio of the maximum acceptable mass, current and operating temperature constraints, (2)
diameter from Eq. (11) to the minimum acceptable sets the maximum and minimum diameter bounds within
diameter from Eq. (10). Imposing the constraint that a certain range of one another as specified by the design
U P 1 ensures that the diameter bounds are properly margin, and (3) ensures that the reactor can support joule
ordered and U ¼ 1 results in only a single acceptable value heated cold wall operation (no molten regolith touches the
for reactor diameter. reactor wall).

Table 2
The regression coefficients for Eqs. (8)–(11), which utilize data from the multiphysics simulation to predict the molten mass and operating temperature for
an MRE reactor.
Regolith m1 m2 m3 m4 m5 m6 t1 t2 t3 t4 t5 t6 t7
Highlands 11.0 5.31 0.160 4.99 0.343 0.435 1525 5387 0.351 0.0289 0.0105 0.215 5.09
Mare 9.73 3.98 0.226 5.60 0.500 0.485 1598 2200 0.593 0.0434 0.0390 0.419 1.54
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
5.5. Determining wall thickness and MLI design
Although the thermal conductivity of the reactor wall is
treated as a design variable in the multiphysics simulation
and design methodology described above (doing so greatly
decreased the time required to simulate a large design
space), the true thermal design variables are the wall mate-
rial, wall thickness and number of layers of multilayer insu-
lation (MLI). The MRE reactor model designs a reactor
with real refractory and insulation materials such that its
thermal resistance is equivalent to the thermal resistance
of the reactor from the multiphysics simulation, which has
a 3 cm thick wall, an outer wall emissivity of 0.05, the ther-
mal conductivity prescribed in Eq. (13), and the maximum
diameter prescribed from Eq. (11). The thermal resistance
of the reactor includes that of the reactor wall (Rcylinder )
and that of the radiation from the reactor wall through lay-
ers of MLI to a distant ambient temperature (Rradiation )
(Kreith et al., 2011):
Rtotal ¼ Rcylinder þ Rradiation ;
where:
1 of two to be conservative. The refractory material was cho-
Rcylinder ¼  2
 ;
ð Din
þDxÞ ðHin þ2DxÞ
 density of 2.8 g/cm3 and a thermal conductivity of 1.4 W/
2pk eff 2
Dx
þ
ln 1þ2Dx
D in
   1
Rradiation ¼  r T 2out þ T 21 ðT out þ T 1 Þ A ;
where k eff and Dx are the effective thermal conductivity and
thickness, respectively, of the combined refractory and
insulation layers, Din is the reactor inner diameter, H in is
the reactor inner height (set to 1.5 times the electrode sep-
aration), T out is the temperature between the reactor surface
and MLI, T 1 is the ambient temperature, A is the reactor
outer surface area, r is the Stefan–Boltzmann constant,
and  is the effective emissivity of the layers of MLI
(Donabedian et al., 2002):
 1
 2N 1 1
 ¼ N 1þ þ ; ð17Þ
MLI surf 1
where N is the number of MLI layers, MLI is the emissivity of
each MLI layer, taken to be 0.03 for Mylar (Donabedian
et al., 2002), and surf and 1 are the emissivities of the
reactor outer surface and environment, respectively (taken
to be 1.0). Although Donabedian et al. (2002) assert that
the emissivity in Eq. (17) is a dramatic underestimate, recent
work by Hatakenaka et al. (2013) demonstrated a new pin
attachment method for MLI that allows for hardware
performance to approach the theoretical predictions of
Eq. (17).
Although the ambient temperature will dramatically
vary through the lunar cycle, it was taken to be 298 K after
an initial analysis revealed low sensitivity to this parameter.
Varying the ambient temperature between 3 K and 400 K
resulted in less than a 2% change in the reactor thermal
performance compared to the 298 K case, to first order.
T out in Eqs. (15) and (16) was calculated using the radiative
1595
heat transfer equation for the predicted heat loss from the
reactor (Q_ heat-loss ), discussed Section 6:
!1=4
Q_ heat-loss þ Ar1 ðT 1 Þ4
T out ¼ ð18Þ
Ar
The MRE model leverages Eqs. (13)–(18) to calculate
the required thermal resistance as a design target. This
design target is then translated into an equivalent reactor
with MLI and real insulation and refractory materials
according to the following methodology. First a mini-
mum wall thickness is set as a user input. Gmitter
(2008) utilized two 1 cm thick crucible walls, so a mini-
mum wall thickness of 2 cm was used for this study.
Next, layers of MLI are added in an attempt to meet
the required thermal resistance design target. If this fails
to achieve the required thermal resistance (the effect of
adding layers of MLI has diminishing returns), the wall
thickness is also increased to achieve the thermal resis-
tance design target.
ð14Þ
The mass of the MLI was taken to be 0.02 kg/m2 per
layer (Plachta and Kittel, 2002), which includes a factor
ð15Þ sen to be CW Hi Al Kastite C (high alumina)TM,5 with a
m-K. Aeroguard HDTM,6 was selected as the insulation
ð16Þ material, for its excellent thermal properties at high tem-
peratures (a wall thermal conductivity of 0.02 to 0.06 W/
m-K depending upon temperature) and low density of
0.2 g/cm3. Titanium was selected as the outer structural
material for its tolerance of high temperatures. Future
work can examine additional materials to further optimize
MRE reactor design.
A thickness ratio of 3:1 (insulation:refractory thick-
ness) was used on the reactor bottom and side walls
and 6:1 was used on the reactor top, as the contact
between molten regolith and the roof of the reactor will
be minimal. The structural thickness is set such that the
hoop stress resulting from the operating pressure of
101.3 kPa is less than the yield stress of the material with
a safety factor of 3.5.
6. Estimating reactor mass and power
The MRE reactor model estimates reactor mass using
the geometry of the reactor described above. With the
diameter, height, and thickness of the structural, insula-
tion, and refractory layers determined, the mass of the
three cylindrical reactor shells can be calculated.
The mass of the anode and cathode are also estimated
using the simple shaft and plate geometry shown in
Fig. 1. Iridium and molybdenum were chosen for the anode
and cathode materials, respectively, as they have significant
5
http://www.alliedmineral.com/.
6
http://www.microthermgroup.com/landingpage/assets/TDS_AERO-
GUARD_V1-EN.pdf.
1596 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603

experimental heritage operating with lunar simulant feed- Table 3

stock (Paramore, 2010). The diameter of the cathode plate The regression coefficients for Eq. (20), which predicts the expected heat
loss from an MRE reactor based off of data from the multiphysics
is set to 15% of the reactor diameter, after an initial anal- simulation.
ysis revealed that this ratio provided a good balance
Regolith h1 h2 h3 h4 h5 h6
between larger molten masses and higher operating tem-
peratures. The anode diameter is designed to be 19% less Highlands 1.59  106
8.29 1.08 1.15 222 0.0357
Mare 1.76  106 8.84 1.17 1.14 189 0.0345
than the cathode diameter, using the same ratio as Sibille
and Dominguez (2012). The cathode and anode shafts
are designed to be long enough to pass through the reactor
wall. The diameters of the anode and cathode shafts are set
to 2 cm plus 251 th of the plate diameter.
The total reactor power is estimated using the sum of
four different terms:
P total ¼ Q_ regheatup þ P ðDGÞ þ Q_ endo þ Q_ heat-loss ð19Þ
The power required to heat the regolith from ambient tem-
perature (298 K) to the operating temperature (Q_ regheatup ),
is calculated using the specific heat and latent heat of melt-
ing for lunar regolith from Schreiner et al. (2016). The heat
required to offset the endothermic electrolysis reaction
(Q_ endo ) is calculated from the difference between the
Enthalpy of Formation and Gibbs Free Energy provided
in Schreiner et al. (2016). The power required to perform
the electrolysis chemical reaction is calculated from the
Gibbs Free Energy data from Schreiner et al. (2016).
The radiative heat loss to the environment (Q_ heat-loss ) is
predicted using regression equations that were fit to the
data from the multiphysics simulation:
0  1
h2
exp Dþh
Q_ heat-loss ¼ h1 @ A þ h5 k wall ð20Þ
3

h4 þ kwall1h6

where k wall is given in Eq. (13), D is the diameter of the

reactor given by Eq. (11) and hi are regression coefficients
given in Table 3, which depend upon the regolith type.
Fig. 8 shows data generated by the multiphysics simula-
tion (dashed lines with data points) and the predictions of
Eq. (20) (solid lines). Although it may seem counterintu-
itive that reactor current does not significantly affect the
heat loss, this is because the electrode separation decreases
as current increases to maintain the joule heated cold wall
constraint as shown in Fig. 6. The MRE model shows that,
for a fixed diameter and wall conductivity, the thermal
topology required by the joule heated cold wall condition
imposes a certain heat loss on the reactor which does not
Fig. 8. The reactor heat loss does not significantly depend on current
significantly depend on the current.
(top), but is affected by wall thermal conductivity (bottom) when designing
Conversely, reactor heat loss does depend significantly on the cutoff lines. A nonlinear regression model (solid lines) was fit to the
on the wall thermal conductivity. Lower wall thermal data from a multiphysics simulation (dots with dashed lines) with
conductivities intuitively result in less heat loss, as a R2 = 0.997, RMSE = 0.51 kW.
direct result of the associated decrease in the required
electrode separation observed in Fig. 6. Fig. 8 demon- 7. Results: MRE performance and design trends
strates that by increasing the insulation of the reactor wall,
the electrode separation required to maintain joule heated 7.1. Oxygen extraction yield and current efficiency
cold wall operation decreases, resulting in a decrease in
resistive heating inside the reactor and therefore radiative The oxygen extraction yield (kO2 ) is the kilograms of
heat loss. oxygen that can be extracted per kilogram regolith, as
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
defined in Eq. (2). Fig. 9 shows how kO2 increases with
operating temperature for various types of lunar regolith.
As discussed in Section 2, higher operating temperatures
enable an MRE reactor to electrolyze more of the oxides
in lunar regolith before the melting temperature of the mol-
ten regolith core approaches the operating temperature,
risking core solidification and reactor shutdown. Thus by
raising the operating temperature, one can extract more
oxygen per kilogram regolith and increase kO2 . Using com-
position data from Schreiner et al. (2016), the maximum
value for kO2 is 0.454 for Highlands, 0.417 for High-Ti
Mare, and 0.423 for Low-Ti Mare regolith.
For Highlands regolith, kO2 exhibits a sharp rise around
2000 K, at which point the reactor can reduce all of the
SiO2 (see Fig. 3: the SiO2 peak occurs around 1950 K, add-
ing a 50 K safety margin yields 2000 K). For Mare regolith,
a jump in kO2 occurs around 2200 K. This jump is not due
to electrolyzing all of the SiO2, but is rather due to a rever-
sal in the electrolysis order of SiO2 and MgO. At tempera-
tures above 2200 K, the Gibbs Free Energy for MgO rises
above that of SiO2 (Schreiner et al., 2016), meaning that
MgO will be preferentially reduced over SiO2. In reality,
it is likely that both MgO and SiO2 will be concurrently
reduced at temperatures circa 2200 K, but for simplicity,
the MRE reactor model assumes that oxide species are dis-
cretely reduced in the order of increasingly negative Gibbs
Free Energy.
The right plot in Fig. 9 shows how the current efficiency
(gI ) rises with temperature as more SiO2 (and other species)
can be electrolyzed with a higher current efficiency than
FeO, increasing the average current efficiency. For High-
lands regolith, gI increases from 74% at an operating tem-
perature of 1850 K to 82% above 2000 K. For Mare
regolith, gI increases from 57% at 1850 K to 72% at
2200 K. The drop in current efficiency for Mare regolith
Fig. 9. The oxygen extraction yield (left) and current efficiency (right) of an MRE reactor generally rise with operating temperature, due to the fact that
more oxygen can be electrolytically extracted from lunar regolith at higher temperatures.
1597
near 2200 K is due to MgO also being electrolyzed, with
a low current efficiency of 50% (see Section 3).
The current efficiency estimates presented here are lower
than many numbers in the literature, which are often
around 90–100% (Sibille et al., 2010; Vai et al., 2010;
Sirk et al., 2010). Those estimates are for melts without
FeO, TiO2, K2O, Na2O, which is why dramatically higher
current efficiencies are measured. Realistic current efficien-
cies over an entire batch of raw regolith, in which FeO,
TiO2, and other oxides may be reduced, will be lower.
7.2. The effect of operating temperature
Fig. 10 shows the how operating temperature affects
reactor mass (left) and power (right) over a range of oxygen
production levels for High-Ti Mare regolith. Higher oper-
ating temperatures increase the oxygen extraction yield (see
Fig. 9), which reduces the regolith throughput requirement
and reactor size. Higher operating temperatures also result
in a larger temperature drop that must be accommodated
before reaching the reactor wall and a lower density for
molten regolith (Schreiner et al., 2016). Although the latter
two effects drive larger reactor designs, they are outweighed
by the increase in oxygen extraction yield and subsequent
decrease in reactor size shown in the left plot in Fig. 10.
At lower temperatures (1850–2150 K), reactor power
increases with operating temperature due to the larger tem-
perature drop, and therefore heat loss, between the reactor
core and walls. At higher temperatures, this effect is out-
weighed by the steep increase in oxygen extraction yield
near 2200 K (see Fig. 9 for High-Ti Mare) and the increase
in electrical conductivity which decreases the resistive heat-
ing that must exit the reactor as heat loss.
Fig. 10 also illustrates that, for a given batch time and
design margin, a minimum achievable oxygen production
1598 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
Fig. 10. Operating temperature affects both the mass (left) and power (right) for an MRE reactor. Each line represents a different operating temperature
and all designs have a design margin of 1.5 and a batch time of 8 h with High-Ti Mare feedstock.
level exists. Below a certain oxygen production threshold,
the reactor size becomes too small to maintain joule heated
cold wall operation. As operating temperature increases, it
becomes more difficult to design a small reactor and the
minimum oxygen production threshold increases. For
High-Ti Mare with an operating temperature above
2200 K, this effect is outweighed by the dramatic increase
in oxygen extraction yield seen Fig. 9, and the minimum
oxygen production threshold decreases with increasing
operating temperature.
7.3. The effect of design margin
The trends in operating temperature presented above
are for reactor designs with a fixed design margin of 1.5.
This means that the wall thermal conductivity is designed
such that the maximum diameter set by the operating tem-
perature is 1.5 times the minimum diameter set by the mol-
ten mass requirement (see Section 5.4). Fig. 11
demonstrates that increasing the design margin results in
higher reactor mass and power, and increases the maxi-
mum production rate for a single reactor. The maximum
production levels for design margins of 1.0 and 1.1 are
5000 kg O2/year and 9500 kg O2/year, respectively. Above
these production rates, the MRE reactor model is unable to
produce a feasible reactor design which could sustain joule
heated cold wall operation.
Perhaps the most important effect of design margin is on
the range of acceptable electrode separation values. The
design margin controls the acceptable bounds on reactor
diameter and because there is a one-to-one mapping
between reactor diameter and electrode separation, this
also translates into bounds on the electrode separation
(see Section 5.3). As described in Section 5.4, during reac-
tor operation a control system will need to regulate the
electrode separation in order to maintain thermal equilib-
rium in the reactor. The left-hand plot in Fig. 12 demon-
strates how increasing the design margin enables a range
of electrode separations that (1) generate enough heat to
maintain the central molten core and (2) maintain the joule
heated cold wall operation (no molten regolith touches the
reactor wall) by not generating excessive heat.
The multiphysics simulation described in Section 4 was
leveraged to validate the range of electrode separations
enabled by a design margin greater than 1. The right side
of Fig. 12 shows reactor designs with the maximum (top)
and minimum (bottom) electrode separations for an oxy-
gen production of 4000 kg O2/year with a design margin
of 1.2. The multiphysics simulation confirmed that both
reactor designs can support joule heated cold wall opera-
tion by insulating the central molten core from the reactor
walls with a layer of solid regolith.
7.4. Sensitivity to regolith type/composition
The MRE reactor model has a strong dependence on
regolith type. Of the regolith material properties presented
in Schreiner et al. (2016), composition, density, specific
heat, electrical conductivity, current efficiency, latent heat
of melting, endothermic power, and electrolysis power all
depend upon regolith composition. Furthermore, different
sets of multiphysics data were generated for Highlands
and Mare regolith (the differences between High-Ti and
Low-Ti Mare were not significant enough to warrant sepa-
rate runs with the multiphysics simulation). This resulted in
separate sets of regression coefficients for the appropriate
electrode separation (Table 1), molten mass and operating
temperature (Table 2), and heat loss (Table 3).
Fig. 13 shows how the regolith type affects the mass and
power of an MRE reactor for operating temperatures of
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603 1599

Fig. 11. Larger design margins enable a range of acceptable electrode separations, but this comes at the cost of increased reactor mass (left) and power
(right). Larger design margins also increase the maximum production level for a reactor. All data is for an operating temperature of 1850 K and a batch
time of 18 h with High-Ti Mare feedstock.

Fig. 12. (left) Increasing the design margin opens up a larger range of feasible electrode separation values to enable operational flexibility. (right) At a
production level of 4000 kg O2/year, a design margin of 1.2 enables an acceptable electrode separation range of 1.9 cm to 3.8 cm, validated using the
multiphysics simulation.

1850 K (top row), 2000 K (middle row), and 2300 K (bot-
tom row). From these results, it appears that the general
increase in regolith liquidus temperature during electrolysis
(see Fig. 3) plays a large role in the regolith type dependence.
That is, the type of regolith strongly affects how much oxy-
gen can be extracted from lunar regolith before having to
stop to avoid solidification of the central molten core.
At 1850 K (top row of Fig. 13), one can extract less oxy-
gen per kilogram Highlands regolith compared to Mare
regolith so Highlands reactors must process more regolith,
which increases reactor power. Furthermore, Highlands
regolith has more SiO2, which must be electrolyzed at a
higher voltage than FeO and subsequently drives the power
requirement higher. Reactor mass remains relatively simi-
lar between Highlands and Mare because Highlands rego-
lith has a lower electrical conductivity (Schreiner et al.,
2016), which increases resistive heating in the central core
and requires less insulation around the reactor wall.
7.5. Molten metal production
The trend in liquidus temperature reverses around
2000 K (middle row of Fig. 13), at which point an MRE
reactor can now extract more oxygen from Highlands rego-
lith compared to Mare. This results in a dramatic decrease
in reactor mass (33%) and power (60%) for Highlands reac-
1600 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603

Fig. 13. The type of regolith feedstock strongly affects reactor mass and power at midrange operating temperatures circa 2000 K (middle row), where
Highlands reactors are significantly less massive and require less power than Mare reactors. At the low operating temperatures circa 1850 K (top row), and
high operating temperatures circa 2300 K (bottom row), the affect of regolith type is reduced, yet Highlands reactors are slightly less massive at high
production levels and require slightly more power across all production levels.
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
Fig. 14. The amount of metal produced by an MRE reactor operating on
Mare (top) and Highlands (bottom) regolith producing 10 mT O2/yr. As
operating temperature increases, more oxides can be reduced to produce
more molten metal and leave less leftover slag.
tors compared to Mare for operating temperatures circa
2000 K. At operating temperatures above 2300 K, reactor
mass and power become similar for both Highlands and
Mare reactors because the oxygen extraction yield is simi-
lar in that regime.
Fig. 14 illustrates the estimated metal produced from an
MRE reactor, as discussed in Section 2, as a function of
operating temperature for Mare (left) and Highlands
(right) regolith. As temperature increases, the metal pro-
duction increases while the mass of leftover slag decreases
by approximately 35% from 1850 K to 2300 K (for Mare)
or 2000 K (for Highlands). Reactors processing Highlands
regolith can produce aluminum at lower operating
temperatures because the liquidus temperature peak occurs
around 2000 K for Highlands regolith compared to 2250 K
for Mare (see Fig. 3). Due to differences in regolith compo-
sition (Schreiner et al., 2016), reactors processing Mare
regolith can produce more iron and titanium per kilogram
1601
regolith, while reactors processing Highlands regolith can
produce more silicon and aluminum per kilogram regolith.
The data in Fig. 14 is for an oxygen production level of
10,000 kg O2/year. At higher operating temperatures
(>2000 K for Highlands and >2300 K for Mare), the
cumulative amount of metal produced is almost equal to
the oxygen production level. It is important to note that
the metals produced via MRE will alloy at the bottom of
the reactor. Post-reactor processing will be required if the
metals are to be separated, though this processing should
be feasible without significant additional energy expendi-
ture, as the metal alloy will already be molten.
8. Conclusions
We have developed a parametric sizing model for a Mol-
ten Regolith Electrolysis reactor to produce oxygen and
molten metals from lunar regolith. The reactor model is
grounded in a foundation of regolith material property
models presented in Schreiner et al. (2016), which are val-
idated using data from Apollo samples and lunar simu-
lants. The oxygen production level drives the regolith
processing requirement and reactor current. A multiphysics
simulation of an MRE reactor is developed and utilized to
generate a database of reactor designs and performance
characteristics. A novel design methodology is developed
that leverages this database to parametrically design
MRE reactors that can meet the required regolith process-
ing rate and reactor current. The predicted mass and power
of an MRE reactor are presented over a range of oxygen
production levels and the effects of operating temperature,
regolith type, and design margin (degree of operational
flexibility) are explored.
Our model predicts that an MRE reactor is able to
achieve oxygen extraction yields (kilograms of oxygen per
kilogram regolith) on the order of 0.15 at low operating
temperatures and 0.375 at high operating temperatures.
We also observe that increasing the operating temperature
increases the predicted current efficiency, from 72.5% at
low temperatures to 82% at higher temperatures for High-
lands regolith. For Mare regolith, predicted current effi-
ciencies start around 57.5% at low temperatures and rise
to 70% at higher temperatures.
Higher operating temperatures also appear to always
reduce reactor mass and sometimes reduce power
(Fig. 10), but this effect has a strong dependence on rego-
lith type (Fig. 13). Prior to this work it was unclear whether
higher operating temperatures would indeed result in smal-
ler reactor sizes due to the mixed effect of operating tem-
perature on regolith material properties and reactor
thermal topology. Higher operating temperatures also
enable MRE reactors to produce more molten metal per
kilogram regolith and open up the possibility of producing
aluminum and titanium.
Design margins greater than 1.0 enable a range of feasi-
ble electrode separations (Fig. 12), which allows the reactor
thermal behavior to be controlled by varying electrode sep-
1602 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
aration and resistive heating during operation. This opera-
tional flexibility comes at the cost of increased reactor mass
and power. Increasing the design margin also expands the
maximum production level of a single reactor (Fig. 11).
Regolith type (and therefore composition) also has an
impact on MRE reactor design and performance
(Fig. 13). For midrange operating temperatures circa
2000 K, Highlands reactors are 33% less massive and
require 60% less power than Mare reactors. At the low
and high ends of the operating temperature design space,
the effect of regolith type becomes less significant. Never-
theless, in these regimes Highlands reactors are marginally
less massive at higher production levels and require slightly
more power across all production levels. Although MRE
reactors appear to generally perform slightly better with
Highlands regolith, which covers 84% of the lunar nearside
and 99% of the lunar farside (Heiken et al., 1991), their per-
formance does not degrade significantly with Mare regolith.
The MRE model provides initial evidence that MRE is rea-
sonably feedstock insensitive, allowing it to support surface
exploration across a large variety of lunar locations.
Our results indicate that MRE reactors scale reasonably
with production level and that operational flexibility can
indeed be parametrically designed into a reactor. We
emphasize that the mass and power results presented here
are for an unoptimized MRE reactor. Future work can
evaluate the optimization of MRE reactor design variables
in light of a more complete ISRU system, including a
power system and the oxygen liquefaction and storage sys-
tem. This MRE reactor model will enable the quantitative
comparison of MRE to other lunar oxygen production
methods, as well as the quantitative analysis of the feasibil-
ity of MRE in the broader context of space exploration
missions utilizing lunar ISRU.
Acknowledgment
This work was supported by a NASA Space Technology
Research Fellowship (Grant #NNX13AL76H). Any opin-
ions, findings, and conclusions or recommendations
expressed in this material are those of the author and do
not necessarily reflect the views of NASA.
References
Aiken, R.H., 1906. Process of Making Iron from the Ore. US Patent
816,142.
Allanore, A., Yin, L., Sadoway, D.R., 2013. A new anode material for oxygen
evolution in molten oxide electrolysis. Nature 497 (7449), 353–356.
Allibert, M., Gaye, H., Geiseler, J., Janke, D., Keene, B.J., Kirner, D.,
Kowalski, M., Lehmann, J., Mills, K.C., Neuschutz, D., Parra, R.,
Saint-Jours, C., Spencer, P.J., Susa, M., Tmar, M., Woermann, E.,
1995. Slag Atlas. Verlag Stahleisen, GmbH.
Altenberg, B., 1990. Processing lunar in-situ resources. Technical report,
Bechtel Group, Inc., Technical Research and Development Project Job
90634-002.
Arndt, J., Flad, K., Feth, M., 1979. Radiative cooling experiments on
lunar glass analogues. In: Lunar and Planetary Science Conference
Proceedings, vol. 10, pp. 355–373.
Balasubramaniam, R., Gokoglu, S., Hegde, U., 2010. The reduction of
lunar regolith by carbothermal processing using methane. Int. J.
Miner. Process. 96 (1), 54–61.
Carr, B., 1963. Recovery of water or oxygen by reduction of lunar rock.
AIAA J. 1 (4), 921–924.
Carroll, W.F., 1983. Research on the Use of Space Resources. JPL
Publication, 83-36.
Chen, G.Z., Fray, D.J., Farthing, T.W., 2000. Direct electrochemical
reduction of titanium dioxide to titanium in molten calcium chloride.
Nature 407 (6802), 361–364.
Chepko, A., 2009. Technology selection and architecture optimization of
in-situ resource utilization systems (PhD thesis). Massachusetts Insti-
tute of Technology.
Colson, R.O., Haskin, L.A., 1990. Lunar oxygen and metal for use in
near-earth space: magma electrolysis. In: NASA Space Engineering
Research Center for Utilization of Local Planetary Resources.
Curreri, P., Ethridge, E., Hudson, S., Miller, T., Grugel, R., Sen, S.,
Sadoway, D., 2006. Process demonstration for lunar in situ resource
utilizationmolten oxide electrolysis. NASA Marshall Space Flight
Center. MSFC Independent Research and Development Project.
Project No. 5-81.
Diaz, J., Ruiz, B., Blair, B., Harsch, M., Duke, M., Parrish, C., Lueck, D.,
Mueller, R., Whitlow, J., 2005. Space transportation architectures and
refueling for lunar and interplanetary travel and exploration (starlite).
Technical Report NASA Grant NAG9-1535, Center for the Commer-
cial Applications of Combustion in Space, Colorado School of Mines.
Donabedian, M., Gilmore, D.G., Stultz, J., Tsuyuki, G.T., Lin, E.I., 2002.
Satellite Thermal Control Handbook: Fundamental Technologies.
Duke, M.B., Diaz, J., Blair, B.R., Oderman, M., Vaucher, M., 2003.
Architecture studies for commercial production of propellants from
the lunar poles. In: AIP Conference Proceedings. AIP Publishing, pp.
1219–1226.
Faraday, M., 1834. Xxv. Experimental researches in electricity – seventh
series. The London and Edinburgh Philosophical Magazine and
Journal of Science, 5(27):161–181.
Gerogiorgis, D., Ydstie, B., 2005. Multiphysics CFD modelling for design
and simulation of a multiphase chemical reactor. Chem. Eng. Res.
Des. 83 (6), 603–610.
Gibson, M.A., Knudsen, C.W., 1985. Lunar oxygen production from
ilmenite. In: Lunar bases and space activities of the 21st century, pp.
543–550.
Gmitter, A.J., 2008. The influence of inert anode material and electrolyte
composition on the electrochemical production of oxygen from molten
oxides (PhD thesis). Massachusetts Institute of Technology.
Haskin, L.A., Colson, R.O., Lindstrom, D.J., Lewis, R.H., Semkow, K.
W., 1992. Electrolytic smelting of lunar rock for oxygen, iron, and
silicon. In: Lunar Bases and Space Activities of the 21st Century, vol.
1, pp. 411–422.
Hatakenaka, R., Miyakita, T., Sugita, H., Saitoh, M., Hirai, T., 2013.
Thermal performance and practical utility of a mli blanket using
plastic pins for space use. In: 43rd International Conference on
Environmental Systems.
Hegde, U., Balasubramaniam, R., Gokoglu, S., 2009. Development and
validation of a model for hydrogen reduction of jsc-1a. In: 47th
Aerospace Sciences Meeting. AIAA, NASA/TM2009-215618.
Heiken, G.H., Vaniman, D.T., French, B.M., 1991. The Lunar source-
book: A user’s guide to the Moon. CUP Archive.
Inman, D., White, S., 1978. The production of refractory metals by the
electrolysis of molten salts; design factors and limitations. J. Appl.
Electrochem. 8 (5), 375–390.
Jarrett, N., Das, S.K., Haupin, W.E., 1980. Extraction of oxygen and
metals from lunar ores. Space Sol. Power Rev. U.S. 1 (4).
Joosten, B.K., Guerra, L.A., 1993. Early lunar resource utilization: a key
to human exploration. In: AIAA Space Programs and Technologies
Conference. AIAA, 93-4784.
Kar, P., Evans, J.W., 2008. A model for the electrochemical reduction of
metal oxides in molten salt electrolytes. Electrochim. Acta 54 (2), 835–
843.
 S.S. Schreiner et al. / Advances in Space Research 57 (2016) 1585–1603
Keller, R., Taberaux, A.T., 1991. Electrolysis of lunar resources in molten
salt. Resources of Near-Earth Space: Proc. Second Annual Symp. UA/
NASA SERC.
Kennedy, M.W., 2012. Electric slag furnace dimensioning. In: Interna-
tional Smelting Technology Symposium: Incorporating the 6th
Advances in Sulfide Smelting Symposium. John Wiley & Sons Inc,
pp. 279–290.
Kesterke, D.G., 1971. Electrowinning oxygen from silicate rocks. Tech-
nical report, U.S. Dept. of the Interior, Bureau of Mines. Report of
Investigations 7587.
Kim, H., Paramore, J., Allanore, A., Sadoway, D.R., 2011. Electrolysis of
molten iron oxide with an iridium anode: the role of electrolyte
basicity. J. Electrochem. Soc. 158 (10), E101–E105.
Kreith, F., Manglik, R., Bohn, M., 2011. Principles of Heat Transfer,,
Cengage Learning.
Landis, G.A., 2005. Materials refining for solar array production on the
moon. NASA Technical Memorandum. NASA/TM2005-214014.
Lindstrom, D.J., Haskin, L.A., 1979. Electrochemistry of lunar rocks. In:
Fourth Princeton/AIAA Conference on Space Manufacturing Facil-
ities, pp. 79–1380.
Lucey, P.G., Blewett, D.T., Jolliff, B.L., 2000. Lunar iron and titanium
abundance algorithms based on final processing of clementine ultra-
violet–visible images. J. Geophys. Res. Planet. (1991–2012) 105 (E8),
20297–20305.
Mason, L., 1992. Beneficiation and comminution circuit for the produc-
tion of lunar liquid oxygen. In: Engineering, Construction, and
Operations in Space III, pp. 1139–1149.
Paramore, J.D., 2010. Candidate anode materials for iron production by
molten oxide electrolysis (PhD thesis). Massachusetts Institute of
Technology.
Plachta, D., Kittel, P., 2002. An updated zero boil-off cryogenic propellant
storage analysis applied to upper stages or depots in an leo environ-
ment. In: 38th AIAA/ASME/SAE/ASEE Joint Propulsion Conference
and Exhibit. AIAA, 2002-3589.
Poizeau, S., Sadoway, D.R., 2011. Towards a design tool for self-heated
cells producing liquid metal by electrolysis. Light Metals 2011, 387–392.
Sacksteder, K.R., Sanders, G.B., 2007. In-situ resource utilization for
lunar and mars exploration. In: The 5th AIAA Aerospace Sciences
Meeting and Exhibit. AIAA, 2007-345.
Schreiner, S.S., 2015. Molten regolith electrolysis reactor modeling and
optimization of in-situ resource utilization systems (Master’s thesis).
Massachusetts Institute of Technology.
Schreiner, S.S., Sibille, L., Dominguez, J.A., Hoffman, J.A., Sanders, G.
B., Sirk, A.H., 2015. Development of a molten regolith electrolysis
reactor model for lunar in-situ resource utilization. In: AIAA SciTech
Conference – 8th Symposium on Space Resource Utilization.
Schreiner, S.S., Dominguez, J.A., Sibille, L., Hoffman, J.A., 2016.
Thermophysical property models for lunar regolith. Adv. Space Res.
Schwandt, C., Hamilton, J.A., Fray, D.J., Crawford, I.A., 2012. The
production of oxygen and metal from lunar regolith. Planet. Space Sci.
74 (1), 49–56.
1603
Senior, C.L., 1993. Lunar oxygen production by pyrolysis. In: Lewis, J.S.,
Matthews, M.S., Guerrieri, M.L. (Eds.), Resources of Near-Earth
Space. University of Arizona Press, Tucson, AZ, pp. 179–197.
Sherwood, B., Woodcock, G.R., 1993. Cost and benefits of lunar oxygen:
economics, engineering, and operations. In: Lewis, J.S., Matthews, M.
S., Guerrieri, M.L. (Eds.), Resources of Near-Earth Space. University
of Arizona Press, Tucson, AZ, pp. 199–227.
Sibille, L., Dominguez, J.A., 2012. Joule-heated molten regolith electrol-
ysis reactor concepts for oxygen and metals production on the moon
and mars. In: 50th AIAA Aerospace Sciences Meeting including the
New Horizons Forum and Aerospace Exposition. AIAA, 2012-0639.
Sibille, L., Sadoway, D.R., Sirk, A., Tripathy, P., Melendez, O., Standish,
E., Dominguez, J.A., Stefanescu, D.M., Curreri, P.A., Poizeau, S.,
2009. Recent advances in scale-up development of molten regolith
electrolysis for oxygen production in support of a lunar base. In: 47th
AIAA Aerospace Sciences Meeting. American Institute of Aeronautics
and Astronautics (AIAA).
Sibille, L., Sadoway, D., Tripathy, P., Standish, E., Sirk, A., Melendez,
O., Stefanescu, D., 2010. Performance testing of molten regolith
electrolysis with transfer of molten material for the production of
oxygen and metals on the moon. In: AIAA: 3rd Symposium on Space
Resource Utilization.
Simon, M.C., 1985. A parametric analysis of lunar oxygen production. In:
Lunar Bases and Space Activities of the 21st Century.
Sirk, A.H., Sadoway, D.R., Sibille, L., 2010. Direct electrolysis of molten
lunar regolith for the production of oxygen and metals on the moon.
ECS Trans. 28 (6), 367–373.
Standish, E., 2010. Design of a molten materials handling device for
support of molten regolith electrolysis (PhD thesis). The Ohio State
University.
Stoeser, D., Rickman, D., Wilson, S., 2010. Design and Specifications for
the Highland Regolith Prototype Simulants NU-LHT-1M and-2M.
Citeseer.
Taylor, L.A., Carrier III, W.D., 1993. Oxygen production on the moon:
an overview and evaluation. In: Lewis, J.S., Matthews, M.S., Guer-
rieri, M.L. (Eds.), Resources of Near-Earth Space. University of
Arizona Press, Tucson, AZ, pp. 69–108.
Teeple, B.S., 1994. Feasibility of producing lunar liquid oxygen (PhD
thesis). Massachusetts Institute of Technology.
Turner, M.J., 2008. Rocket and Spacecraft Propulsion: Principles, Practice
and New Developments. Springer Science & Business Media.
Vai, A., Yurko, J., Wang, D., Sadoway, D., 2010. Molten oxide
electrolysis for lunar oxygen generation using in situ resources. In:
The Minerals, Metals, and Materials Society.
Wang, D., Gmitter, A.J., Sadoway, D.R., 2011. Production of oxygen gas
and liquid metal by electrochemical decomposition of molten iron
oxide. J. Electrochem. Soc. 158 (6), 51–54.
Yen, C.-A.F., 1977. Electrical conductivity in the FeO Fe2O3–Al2O3–
SiO2 system (PhD thesis). Massachusetts Institute of Technology.