Content uploaded by Jonathan D. Toner
Author content
All content in this area was uploaded by Jonathan D. Toner on Mar 09, 2018
Content may be subject to copyright.
Soluble salts at the Phoenix Lander site, Mars: A reanalysis
of the Wet Chemistry Laboratory data
J.D. Toner
a,⇑
, D.C. Catling
a
, B. Light
b
a
University of Washington, Dept. Earth & Space Sciences/Astrobiology Program, Box 351310, Seattle, WA 98195, USA
b
Polar Science Center, Applied Physics Laboratory, University of Washington, Seattle, WA, USA
Received 9 May 2013; accepted in revised form 24 March 2014; available online 3 April 2014
Abstract
The Wet Chemistry Laboratory (WCL) on the Phoenix Mars Scout Lander analyzed soils for soluble ions and found Ca
2+
,
Mg
2+
,Na
+
,K
+
,Cl
,SO
4
2
, and ClO
4
. The salts that gave rise to these ions can be inferred using aqueous equilibrium models;
however, model predictions are sensitive to the initial solution composition. This is problematic because the WCL data is
noisy and many different ion compositions are possible within error bounds. To better characterize ion concentrations, we
reanalyzed WCL data using improvements to original analyses, including Kalman optimal smoothing and ion-pair correc-
tions. Our results for Rosy Red are generally consistent with previous analyses, except that Ca
2+
and Cl
concentrations
are lower. In contrast, ion concentrations in Sorceress 1 and Sorceress 2 are significantly different from previous analyses.
Using the more robust Rosy Red WCL analysis, we applied equilibrium models to determine salt compositions within the
error bounds of the reduced data. Modeling with FREZCHEM predicts that WCL solutions evolve Ca–Mg–ClO
4
-rich com-
positions at low temperatures. These unusual compositions are likely influenced by limitations in the experimental data used
to parameterize FREZCHEM. As an alternative method to evaluate salt assemblages, we employed a chemical divide model
based on the eutectic temperatures of salts. Our chemical divide model predicts that the most probable salts in order of mass
abundance are MgSO
4
11H
2
O (meridianiite), MgCO
3
nH
2
O, Mg(ClO
4
)
2
6H
2
O, NaClO
4
2H
2
O, KClO
4
, NaCl2H
2
O (hydro-
halite), and CaCO
3
(calcite). If ClO
3
is included in the chemical divide model, then NaClO
3
precipitates instead of NaClO
4-
2H
2
O and Mg(ClO
3
)
2
6H
2
O precipitates in addition to Mg(ClO
4
)
2
6H
2
O. These salt assemblages imply that at least 1.3 wt.%
H
2
O is bound in the soil, noting that we cannot account for water in hydrated insoluble salts or deliquescent brines. All WCL
solutions within error bounds precipitate Mg(ClO
4
)
2
6H
2
O and/or Mg(ClO
3
)
2
6H
2
O salts. These salts have low eutectic tem-
peratures and are highly hygroscopic, which suggests that brines will be stable in soils for much of the Martian summer.
Ó2014 Elsevier Ltd. All rights reserved.
1. INTRODUCTION
Soluble salts on Mars readily interact with water and
have wide-ranging implications for aqueous processes
(Squyres et al., 2004; Haskin et al., 2005). A significant
motivation for the exploration of Mars is the possibility
that Mars harbored life early in its history or may have life
in the subsurface today (Beaty et al., 2005). Because life
requires liquid water (Tosca et al., 2008; Davila et al.,
2010), the search for life has focused on aqueous alteration
minerals, such as salts, which are indirect tracers of liquid
water (Boynton et al., 2009; Niles et al., 2013). Salts also
have key properties that can stabilize liquid water in the
cold and dry conditions of Mars, such as freezing-point
depression (Brass, 1980; Kuz’min and Zabalueva, 1998;
Squyres et al., 2004; Bibring et al., 2006; Faire
´n et al.,
2009; Marion et al., 2010), slower evaporation rates due
to reduced water activity (Sears and Chittenden, 2005;
http://dx.doi.org/10.1016/j.gca.2014.03.030
0016-7037/Ó2014 Elsevier Ltd. All rights reserved.
⇑
Corresponding author. Tel.: +1 267 604 3488.
E-mail address: toner2@uw.edu (J.D. Toner).
www.elsevier.com/locate/gca
Available online at www.sciencedirect.com
ScienceDirect
Geochimica et Cosmochimica Acta 136 (2014) 142–168
Altheide et al., 2009; Chevrier et al., 2009), and hydroscop-
icity (Zorzano et al., 2009; Davila et al., 2010; Gough et al.,
2011). In addition, cryogenic salts can incorporate water
and CO
2
into their crystal structure, which will influence
water and CO
2
cycling between the Martian regolith and
atmosphere (Clark, 1978; Kahn, 1985; Niles et al., 2013).
The quantitative identification of salts on Mars is the
first step towards determining their origin and what they
mean for the evolution and habitability of Mars. Both orbi-
tal spectra and in situ measurements have identified salts.
Key salts detected from orbit include chlorides (Glotch
et al., 2008; Osterloo et al., 2008, 2010; Ruesch et al.,
2012), sulfates (Gendrin et al., 2004; Langevin et al.,
2005; Murchie et al., 2009), and carbonates (Bandfield
et al., 2003; Ehlmann et al., 2008; Niles et al., 2013). The
Mars Exploration Rovers (MERs) detected Mg, Ca, and
Fe sulfates (Squyres et al., 2006; Wang et al., 2006), indic-
ative of past acidic conditions (Hurowitz et al., 2006), and
also carbonates (Morris et al., 2010), characteristic of alka-
line environments. Comparisons of elemental abundance
measured by landers and rovers have led to the suggestion
that the soil comprises a global unit, which is a mixture of
weathered and unweathered basalt, salt, dust, and meteor-
itic material (Clark et al., 1982; McSween and Keil, 2000;
Nelson et al., 2005; Yen et al., 2005). Hence, salt composi-
tions measured in any one locality may have global implica-
tions for the evolution and habitability of Mars.
The first direct measurements of soluble salts on Mars
were made by the Wet Chemistry Laboratory (WCL) exper-
iment on the Phoenix Lander (Boynton et al., 2009; Hecht
et al., 2009; Kounaves et al., 2010a). Soluble salts were mea-
sured by adding dry soil to liquid water heated to a temper-
ature between 5 and 10 °C, and analyzing the dissolved ions
using an array of Ion Selective Electrodes (ISEs) (Kounaves
et al., 2010a). The WCL experiment identified an alkaline
soil solution consistent with previous pH inferences of
7.4–8.7 for soils in the Viking Lander biology experiments
(Quinn and Orenberg, 1993). Phoenix also found some sol-
uble sulfate (1–2 wt.%), which can be compared to 5–
9 wt.% total sulfate in typical soil inferred from previous
landers (Kounaves et al., 2010b). This suggests that most
of the soil sulfate resides in insoluble or sparingly soluble
forms. One of the most interesting findings of the WCL
experiment is that most of the soluble chloride is present
as perchlorate (ClO
4
)(Hecht et al., 2009; Kounaves
et al., 2010a). Perchlorates are among the most hygroscopic
salts (Besley and Bottomley, 1969; Gough et al., 2011) and
have eutectic temperatures as low as 75 °C(Dobrynina
et al., 1980; Pestova et al., 2005), which could stabilize
liquid water on present-day Mars (Marion et al., 2010).
To determine both the soluble ion chemistry and solid
salt precipitates on Mars, thermodynamic models have been
used, such as FREZCHEM (Marion et al., 2003, 2009, 2010,
2011). FREZCHEM suggests that a variety of parent salts
are present at the Phoenix site, including calcite (CaCO
3
),
gypsum (CaSO
4
2H
2
O), meridianiite (MgSO
4
11H
2
O),
NaClO
4
2H
2
O, KClO
4
, and Mg(ClO
4
)
2
6H
2
O, and that a
small fraction of liquid water is stable down to about
60 °C as Mg–ClO
4
-rich brine (Marion et al., 2010).
However, equilibrium model predictions are sensitive to
the initial WCL chemistry that is input into the model. This
deserves further attention because the Phoenix WCL analy-
sis is characterized by relative errors in concentration
between about 20–50% due to high levels of noise and anom-
alous signal fluctuations (Hecht et al., 2009; Kounaves et al.,
2010a,b; Quinn et al., 2011). As a result, many different solu-
tion compositions are possible within error bounds.
To better understand the soluble chemistry at the Phoe-
nix site, we reanalyzed data from the WCL experiment
using improvements to the original analyses of Hecht
et al. (2009) and Kounaves et al. (2010a,b) that include Kal-
man optimal smoothing, corrections for ion-pairs, and cor-
rections for calibrant salts. It is important that an
independent reanalysis is done because the WCL results
are the only direct measurements we have of the soluble soil
chemistry on Mars. With our revised ion concentrations
and uncertainty estimates for the WCL aqueous solution,
we apply the geochemical model FREZCHEM and a chem-
ical divide model to determine probable parent salt compo-
sitions in the Phoenix soil, i.e. the original salt precipitates
in the soil that dissolved to form the WCL solution.
2. OVERVIEW OF THE WCL EXPERIMENT
The operation and construction of the WCL experiment
has been described in detail (Kounaves et al., 2009), as well
as methods used to analyze the WCL data (Hecht et al.,
2009; Kounaves et al., 2010a,b; Quinn et al., 2011). Briefly,
the WCL instrument consists of four 40 ml beakers (labeled
cells 0 to 3), each containing an array of Ion Selective Elec-
trodes (ISEs) and other sensors for the analysis of Ca
2+
,
Mg
2+
,Ba
2+
,Na
+
,K
+
,NH
4
+
,H
+
(pH), Cl
, ClO
4
,Br
,I
,
conductivity, cyclic voltammetry, anodic stripping voltam-
metry, and chronopotentiometry. There are three ISE sen-
sors for measuring pH (pH
a
,pH
b
, and pH
irid
). Above each
WCL beaker is a 36 ml tank containing 25 ml of leaching
solution with the following dissolved ions: Ca
2+
=Mg
2+
=
Ba
2+
=Na
+
=K
+
=NH
4
+
= HCO
3
=10lM, Cl
=50
lM, Li
+
= 1 mM, and NO
3
= 1.03 mM. During operation,
the leaching solution was thawed and ejected into the WCL
beaker, which is stirred by an impeller. Following this, a cru-
cible containing salts for calibration of the ISEs was depos-
ited into the leaching solution, bringing the solution to the
composition: Ca
2+
=42lM, Mg
2+
= 34.7 lM, Ba
2+
=38
lM, Na
+
=K
+
=NH
4
+
= HCO
3
=34lM, Cl
= 190 lM,
Li
+
= 1 mM, and NO
3
= 1.1 mM. After analysis of the
calibration solution, 1cm
3
of soil sample was deposited
into the WCL beaker and the sensor array was monitored
over several hours. Once this initial analysis phase was
finished, the heater shut down and the soil-solution was
allowed to freeze in the WCL beaker. A second major phase
of the WCL analyses began by thawing the frozen soil-solu-
tion in the WCL beaker. To this thawed soil-solution was
added 4 mg of 2-nitrobenzoic acid to test for pH buffering
in the soil, followed by three crucibles containing 0.1 g of
BaCl
2
for titrimetric determination of soil SO
4
2
. For some
cells, additional thawing/freezing, soil sample addition, and
sample analysis cycles were performed.
To provide context for this paper, an overview of key
WCL events during the Phoenix mission operations is
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 143
shown in Table 1. Three soils were analyzed, named Rosy
Red, Sorceress 1, and Sorceress 2, respectively. With respect
to Sorceress 1, it is believed that acid was added on the first
sol of analysis (sol 41), just after the addition of calibration
salts and before sample addition, although the cause for
this departure from the nominal WCL experiment is
unknown (Kounaves, personal comm.). Furthermore, the
sample drawer for Sorceress 1 appeared to be only 70–
75% full, so that less soil was added to the WCL beaker
than for the other soil analyses. A fourth analysis was
attempted for soil Golden Goose, but it is believed that
no soil entered the WCL beaker. As a result, this WCL cell
was effectively run as a control blank (Kounaves et al.,
2010a). This blank analysis was fortunate, because it
revealed that BaCl
2
began leaking into the WCL beakers
following soil sample addition, causing Cl
concentrations
to increase throughout the WCL analysis. Each WCL cell
was designed as single use, but additional soil samples were
delivered to Rosy Red and Golden Goose. A total of eight
attempted sample deliveries were made, with five of them
successfully entering WCL cells.
3. METHODS
3.1. WCL analysis
We analyzed ISE potentials and temperature data from
the WCL experiment in two ways. First, we determined ion
concentrations in Rosy Red, Sorceress 1, and Sorceress 2
over the discrete calibration and analysis time windows
given in Kounaves et al. (2010a), with some modifications,
to facilitate comparison with previous estimates of ion con-
centrations. The data we used to derive ion concentrations
are tabulated in Appendix A. Second, we analyzed ISE
potentials and temperatures in Rosy Red over sols 30, 32,
34, 66, 87, and 138. We focus on Rosy Red for this more
in-depth analysis because it is the least noisy of the three
WCL soil experiments and it was analyzed over the most
sols. Modifications to the analysis of Kounaves et al.
(2010a) are described in the following subsections.
3.1.1. Kalman Smoothing
ISE potentials measured in the WCL experiment have
considerable noise, partly due to a software error that
caused the Oxidation Reduction Potential (ORP) electrode
to switch from ground to a 650 mV source whenever pres-
sure or temperature measurements were made. Previously,
noise in the WCL data had been reduced using Fourier fil-
tering to remove high frequencies (Kounaves et al., 2010a).
We use Kalman smoothing instead because (1) Kalman
smoothing does not exclude data, unlike a Fourier analysis
that removes data at an arbitrary high frequency, (2) Kal-
man smoothing outputs an uncertainty estimate of the sig-
nal, and (3) it can be shown that the Kalman filter is
mathematically the best linear filter for dealing with noisy
data where one can assume uncorrelated noise (Simon,
Table 1
Overview of WCL activities in the Phoenix mission, compiled from tables in Kounaves et al. (2010a) and Arvidson et al. (2009).
Sol Cell WCL Experiment Activity
25 0 Rosy Red (#1) – surface sample collected down to 2.5 cm depth from polygon center in the Burn Alive trench
30
a
0 Rosy Red (#1) – ISE calibration and sample delivery
32
a
0 Cell thawed
34
a
0 Acid and three BaCl
2
crucibles added
34
a
1 Sorceress 1 – sublimation lag sample collected from a soil scraped off of subsurface ice at 3 cm depth in the Snow White
trench, in a polygon center
41
a
1 Sorceress 1 – ISE calibration, acid addition, and sample delivery. Note: the sample delivery drawer appeared to be only
70–75% full
43
a
1 Three BaCl
2
crucibles added. Note: acid appears to have been added on sol 41
66
a
0 Rosy Red (#2) – surface sample collected down to 2.5 cm depth from the Rosy Red 2 trench, adjacent to the Burn
Alive trench. Sample redelivery
78
a
0 Thermal diagnostic run
87
a
0 Open loop diagnostic run
95 3 Golden Goose – subsurface sample collected from the Stone Soup trench
96
a
3 Golden Goose – attempted sample delivery, unsuccessful
101 3 Golden Goose – subsurface sample collected from same location in the Stone Soup trench
102
a
3 Golden Goose – attempted sample redelivery, unsuccessful
105 2 Sorceress 2 – sublimation lag sample collected from a soil scraped off of subsurface ice at 3 cm depth in the Snow White
trench, in a polygon center
107
a
2 Sorceress 2 – ISE calibration and sample delivery
116
a
2 Acid and 1st BaCl
2
crucibles added
127
a
2 2nd BaCl
2
crucible added
131
a
2 Cell thawed
134
a
2 3rd BaCl
2
crucible added
136 0 Rosy Red (#3) – surface sample collected down to 2.5 cm depth from the Rosy Red 3 trench, adjacent to the Burn
Alive trench
138
a
0 Rosy Red (#3) – attempted sample redelivery, unsuccessful
147
a
3 Golden Goose – attempted sample push, unsuccessful
a
Sols during which ISE potentials and beaker temperatures were monitored.
144 J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168
2006, Chapter 5). An extensive literature describes the Kal-
man filter (Simon, 2006), and its use in analytical chemistry
is reviewed in a number of papers (e.g. Brown, 1986;
Lavagnini et al., 1990; Rutan, 1991).
For a simple system that varies randomly with associated
noise, known as a local level model, the Kalman filter
describes the state of the system (x
t
) at time tas
xt¼xt1þand measurements (y
t
) made on the system as
y
t
=x
t
+g, where is the process noise and gis the measure-
ment noise. The process and measurement noise are assumed
to be independent and normally distributed with respective
covariances. Kalman smoothing recursively estimates the
true state of the system x
t
from measurements y
t
by giving
less weight to individual noisy measurements. To implement
Kalman smoothing, we use the KFAS package available in
the statistical program R (Helske, 2010; Tusell, 2011). KFAS
accounts for missing data points in a time series, which are
common in the WCL data, and estimates the variance of
the process and measurement noise using a maximum likeli-
hood method (function fitSSM in KFAS).
We applied Kalman smoothing to the raw temperature
data and minimally processed ISE potential data. Prior to
smoothing the ISE potentials, we referenced the ISE poten-
tials to one of two Li
+
ISEs and removed spurious ISE
potentials measured within one second preceding a temper-
ature/pressure measurement, as in Kounaves et al. (2010a).
We then removed obvious outliers. For the analysis of the
WCL data over discrete time windows, we estimated the
noise characteristics of the temperature and ISE potentials
over the duration of each discrete time window (exact time
windows are specified in Appendix A). We then applied
Kalman smoothing, which computes both the smoothed
signal and error estimates, the latter based on the noise level
in the data. Because smoothed temperatures and ISE poten-
tials vary over these time intervals, the values we use for cal-
culating ion concentrations are the average values over the
time intervals. Uncertainties in these average values are
derived by adding in quadrature the uncertainty in the
smoothed values over the time interval and the average esti-
mated uncertainty in the smoothed values output by
KFAS. For the time-varying analysis of Rosy Red, we esti-
mated the time-varying process and measurement noise in
the temperature and ISE potential data over five minute
time windows. We then applied Kalman smoothing using
the time-varying noise characteristics.
3.1.2. Reanalysis of ISE calibration slopes
Initially, it was thought that the ISE sensors could be
calibrated with a two-point in situ calibration using ISE
potentials measured in the leaching solution and the cali-
bration solution (Kounaves et al., 2009); however, later sen-
sitivity analyses indicated that this method of calibration
would introduce large errors into the analysis because the
concentrations in the two dilute solutions occupy only a
small part of the dynamic range (Grunthaner et al., 2009).
It was determined that the best calibration method is a
one-point calibration using ISE potentials measured in
the calibration solution and ISE slopes previously deter-
mined during five Earth-based calibrations. In our data
reduction, we find that the errors associated with the
Earth-based calibration slopes have the largest influence
on final concentration errors. Although five Earth-based
calibrations were performed, we exclude the JPL functional
test calibration from our calculations, as did Kounaves
et al. (2010a), because the JPL functional test slopes are
consistently lower than in the other four calibrations.
To better characterize uncertainties in the Earth-based
calibration slopes, we applied least squares linear fits to
ISE potential vs. log activity data measured in the four
Earth-based calibrations (Grunthaner et al., 2009). We then
calculated final ISE calibration slopes as the weighted mean
of the Earth-based calibration slopes, using the error in the
linear fits as the weights (Table 2). Errors in the final cali-
bration slopes are calculated as the unbiased standard devi-
ation of the weighted mean. The pH sensor slopes are
derived from pre-flight calibrations that used two test solu-
tions saturated with 5% CO
2
(Grunthaner et al., 2009).
Errors in the pH sensor slopes are determined by propagat-
ing errors in the two individual pH ISE measurements to
the slope. The calibration slope for the ClO
4
ISE sensor
was measured on a spare flight beaker at a temperature
of 7 °C(Kounaves et al., 2010a). We assume that the error
in the ClO
4
ISE sensor slope is 0.7 mV dec
1
, near the aver-
age error for the other ISE sensors.
3.1.3. Ionic strength
The conductivity sensor, used for measuring ionic
strength, failed during the Rosy Red analysis. We calculate
the ionic strength of Rosy Red by calculating an initial
ionic strength, accounting for ions already present in the
calibration solution, and iteratively refine this value by
using the resulting concentrations to calculate a new ionic
strength until convergence. The ionic strength we use is
the effective ionic strength, which accounts for ion pairs
in solution. For Sorceress 1 and Sorceress 2, we use ionic
strength values determined from conductivity measure-
ments in Kounaves et al. (2010a).
3.1.4. Debye–Hu
¨ckel model
To determine activity coefficients from ionic strength, we
used the temperature dependent Debye–Hu
¨ckel ion-associ-
ation model. For a given ion i with hydrated radius a
i
and
charge z
i
, the Debye–Hu
¨ckel activity coefficient c
i
is given
by:
log ci¼Az2
iffiffiI
p
1þBaiffiffiI
pð1Þ
with temperature dependent parameters A and B. We
incorporated this temperature dependence into our WCL
data reduction using the empirical equations:
A¼1:824928 106q1=2
0ðTÞ3=2ð2Þ
B¼50:3ðTÞ1=2ð3Þ
where is the dielectric constant of water, T is temperature
in Kelvin, and q
0
is the density of water (Langmuir, 1997).
The temperature dependence of is given by:
¼2727:586 þ0:6224107T 466:9151 ln T
52000:87=Tð4Þ
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 145
3.1.5. Ion-pair corrections
The WCL ISE sensors measure the activity of unpaired
ions, but MgSO
4
0
and CaSO
4
0
ion-pairs can be significant in
solution (Kounaves et al., 2010b), resulting in an underesti-
mation of Mg
2+
and Ca
2+
concentrations and an overesti-
mation of unpaired SO
4
2
concentrations. We corrected for
ion-pairs using the phreeqc.dat database in the aqueous
geochemical program PHREEQC (Parkhurst and Appelo,
1999) by adding ions until the unpaired species activity
equaled that determined in the WCL analysis. The values
of the ion pair concentrations are given later in Section 4.
3.1.6. Calibration/leaching solution ions
Prior to the addition of soil samples, the WCL solution
contained ions from salts dissolved in a calibration solu-
tion. To correct for ion concentrations already present from
the WCL calibration solution, we subtracted these ions
from the final results. This simple correction, which was
previously neglected, has a significant effect on low concen-
tration ions such as Cl
and Ca
2+
, which is described in
Section 4.
3.1.7. Calibration of the pH ISE Sensors
Kounaves et al. (2010a) calibrated pH ISE sensors using
ISE potentials measured after the addition of leaching solu-
tion, but before the sample drawer had opened, at which
point the pH was assumed to equal 5.1. This method could
be used only for a few sensors because most of the pH ISEs
had unstable signals prior to the first drawer opening. As a
result, pH could not be determined for the pH
irid
sensor in
Rosy Red and Sorceress 1, and none of the Sorceress 2 pH
ISE sensors could be analyzed. To obtain values from the
Rosy Red and Sorceress 1 pH
irid
sensors, we calibrated
these sensors to the pH
a
sensors, which can be calibrated
as in Kounaves et al. (2010a). We did this by referencing
the pH
irid
sensors to the pH measured after the addition
of calibration salts (specific time intervals are given in
Appendix A); the Rosy Red and Sorceress 1 pH
a
sensors
indicate that the pH after the addition of calibration salts
was 5.3 and 3.8 respectively. For the Rosy Red pH
irid
sen-
sor, we used a later analysis interval than for all the other
analytes. This is because the pH
irid
sensor appears to
respond more sluggishly than the pH
a
sensor and is still
decreasing during the analysis interval used for other ions
(see graphs in Appendix B). A similar slow response does
not seem to affect either the Sorceress 1 or 2 pH
irid
sensors.
Because none of the Sorceress 2 pH ISE sensors could be
calibrated as in Kounaves et al. (2010a), we calibrated these
sensors by assuming that the pH after the addition of cali-
bration salts is the same as for Rosy Red after the addition
of calibration salts (pH = 5.3). This assumption is reason-
able because the composition of the calibration salts was
the same for each WCL cell and the same number of drawer
opening events occurred prior to the addition of calibration
salts. However, the first drawer opening for Sorceress 2
occurred an hour after the solution thawed, which may
have allowed much more CO
2
to be expelled during the first
drawer opening. Such a CO
2
release would have raised the
pH of the calibration solution. As a result, the pH for Sor-
ceress 2, calibrated as described above, may be a minimum
value (i.e. the actual value may be somewhat higher). We
note that the pH of Sorceress 1 is also a minimum pH
because acid had been added to Sorceress 1 before the soil
sample addition (Kounaves, personal comm.).
3.2. Calculating ion concentrations
Procedures for calculating ion concentrations from the
WCL data are discussed in Kounaves et al. (2010a). Briefly,
ion concentrations are calculated using the equation:
CS¼1
c
10
ES
SMEC
SMþlogðaCÞ
hi ð5Þ
where C
s
(molal) is the sample ion concentration, cis the
ion activity coefficient, E
s
(mV) is the potential measured
after soil sample addition, E
C
(mV) is the potential mea-
sured during the calibration interval, a
C
is the ion activity
in the calibration solutions, and S
M
(mV dec
1
) is the tem-
perature corrected ISE slope. ISE slopes measured at tem-
perature T
E
on Earth (S
E
) are corrected for the
temperature of the WCL solution on Mars (T
M
) using the
equation:
SM¼SE
TM
TEð6Þ
where temperatures are in Kelvin. The Earth-based calibra-
tions were done at ‘room temperature’, but the precise tem-
perature was not recorded. We assume here that T
E
is
22 °C.
Several additional corrections are necessary to calculate
NH
4
+
, ClO
4
, and Ca
2+
concentrations. The NH
4
+
ISE sen-
sor is also sensitive to K
+
. This is corrected for by assuming
that K
+
contributes an additional activity of 0:15aKþto the
measured NH
4
+
activity. Because perchlorate was not
expected on Mars, ClO
4
was not included in the calibration
solution. As a result, the ClO
4
ISE sensor was calibrated to
its lower activity detection limit of 10
6
(i.e. a
C
=10
6
in
Eq. (6). Finally, the Ca
2+
ISE sensor is also sensitive to
Table 2
ISE slopes (mV dec
1
) calculated from four Earth-based calibra-
tions with ±1rerror. The JPL functional test calibration was
excluded because it was consistently different from the four other
calibrations performed. All calibrations were performed at ‘room
temperature’ (the precise temperature was not recorded, but we
assume a value of 22 °C), except for the ClO
4
sensor, which was
calibrated at 7 °C.
Rosy Red – flight
unit 20
Sorceress 1 –
flight unit 18
Sorceress 2 –
flight unit 22
pH
a
55.97 ± 0.46 59.35 ± 0.40 58.11 ± 0.34
pH
b
57.23 ± 0.44 59.77 ± 0.43 55.79 ± 0.30
pH
irid
52.40 ± 0.39 56.43 ± 0.33 52.40 ± 0.42
Mg 28.68 ± 0.65 28.12 ± 0.75 29.09 ± 0.47
Ca 29.13 ± 0.84 28.86 ± 0.58 29.06 ± 0.28
K 59.01 ± 0.60 59.07 ± 1.50 58.96 ± 0.38
Na 53.54 ± 0.97 52.89 ± 0.45 53.70 ± 1.12
NH
4
59.59 ± 0.74 59.23 ± 1.05 59.77 ± 0.80
Cl 54.91 ± 0.29 54.34 ± 0.23 53.98 ± 0.31
ClO
4a
62.00 ± 0.70 62.00 ± 0.70 62.00 ± 0.70
a
The ClO
4
sensor was calibrated only once, so the error is
assumed.
146 J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168
ClO
4
ions and has a Nernstian response to ClO
4
with a
slope of 29 mV dec
1
. To correct the measured sample
ISE potential for the bias introduced by ClO
4
, we use the
equation (Kounaves et al., 2010a):
ECa2þ
corr:¼ECa2þS
2½logðaClO
4Þlogð9:5x106Þ ð7Þ
where ECa2þ
corr:is the corrected Ca
2+
ISE potential, S is
29 mV dec
1
, and 9.5 10
6
is the lower activity limit
for the ClO
4
bias on the Ca
2+
ISE sensor. We assume an
error of 0.7 mV dec
1
for S.
Error propagation for any given function q (with vari-
ables x,y, etc.) is determined in quadrature using the stan-
dard error propagation equation:
r2
q¼@q
@xrx
2
þ@q
@yry
2
þ ð8Þ
We use symmetric errors, as opposed to the asymmetric
errors used in Kounaves et al. (2010a), because the
observed asymmetry in the ISE potential and temperature
data is small (see Appendix B). Eq. (9) assumes that all var-
iable are uncorrelated; however, temperatures are corre-
lated with ISE potentials due the temperature dependence
of ISE slopes. We do not include this correlation in our
error analysis because the additional error is negligible.
For example, the correlation coefficient between tempera-
ture and K
+
ISE potential in Rosy Red is low, 0.05 and
0.03 for the calibration and analysis intervals respectively.
Accounting for this correlation changes the final concentra-
tion errors by <0.1%. Partly, the poor correlation is due to
noise in the temperature and ISE potential data. Addition-
ally, a non-Nernstian temperature response in ISE poten-
tials may have arisen if temperature variations in the soil
solution were damped relative to those in the beaker wall,
within which the temperature sensor was embedded. This
inference is supported by measurements as WCL beakers
warm above 0 °C where a plateau in temperature might
be expected as ice melts endothermically, but none is seen
(e.g., cell 3, sol 107, 10:30 LMST).
4. WCL ANALYSIS RESULTS
Final concentration values resulting from our reanalysis
of the Rosy Red, Sorceress 1 and 2 WCL experiments using
the methods described above indicate that Phoenix soils are
alkaline, rich in Mg
2+
,Na
+
, ClO
4
, and SO
4
2
, low in Ca
2+
,
Cl
, and K
+
, and deficient in NH
4
+
(Table 3). Our analysis
of Rosy Red is generally consistent with Kounaves et al.
(2010a,b); however, several ions in Rosy Red are higher
in concentration than in Kounaves et al. (2010a) due to
the higher ionic strength we calculate (0.017 m vs.
0.008 m), and our Ca
2+
and Cl
concentrations are lower
due to the correction for ions from the calibration salts.
Ions in the Rosy Red calibration solution account for
21% of the measured Ca
2+
and 32% of the Cl
. By account-
ing for MgSO
4
0
ion-pairs in Rosy Red, the total Mg concen-
tration increases by 20%. In contrast to Rosy Red, ion
concentrations for Sorceress 1 and 2 are significantly differ-
ent from values in Kounaves et al. (2010a,b), often outside
of ±1r. In general, ion concentrations between Rosy Red,
Sorceress 1 and 2 vary in our analysis, which suggests that
soluble salt compositions in Phoenix soils are more hetero-
geneous than previously thought.
Soil pH values determined in our analysis are generally
consistent with previous inferences of an alkaline soil
(Table 3). In Rosy Red and Sorceress 1, pH values deter-
mined for previously unanalyzed pH
irid
sensors (the method
we use is described in Section 3.1.7) are somewhat different
from pH values determined using the calibration method in
Kounaves et al. (2010a). This may be due to problems with
the pH
irid
sensor (the pH
irid
sensor is different from the
polymer-based pH
a
and pH
b
sensors, and we note that
the pH
irid
sensor appears to respond sluggishly in Rosy
Red); however, consistent values between all three pH sen-
sors are obtained for Sorceress 2. The measured pH of Sor-
ceress 2 is near 6.8, but this value may be a minimum due to
the possible increase in the calibration pH from CO
2
out-
gassing during drawer openings. The pH of Sorceress 1
may also be a minimum because acid was added to the cell
before the soil sample addition, which may have neutralized
some of the sample alkalinity. Hence, the only accurate
WCL pH measurement with its own internal calibration
was measured by the Rosy Red pH
a
ISE sensor, giving a
pH of 7.7.
Differences in ion concentrations and errors between this
study and Kounaves et al. (2010a,b) for Sorceress 1 and 2
are likely due to differences in the treatment of tempera-
tures, ISE potentials, Earth-based calibration slopes, and
errors. A sensitivity analysis on Rosy Red, in which Mg
2+
concentrations are calculated by varying each measured
parameter by ±2r, reveals that changes in the Earth-based
calibration slope have the greatest effect on calculated ion
concentrations (Table 4). However, our calibration slopes
are similar to Kounaves et al. (2010a) and changing them
to the precise values used in Kounaves et al. (2010a) does
not resolve differences in concentration values, which indi-
cates that variations in other measured parameters are
affecting the final concentrations. In Rosy Red, the noise
level was relatively low and ISE potentials were stable, so
that we obtained similar results regardless of our different
smoothing method. In contrast, Sorceress 1 and 2 are char-
acterized by much greater noise. As a result, the Kalman
smoothing we use may have caused our analysis to differ
significantly from Kounaves et al. (2010a), which used Fou-
rier filtering to remove noise at an arbitrary, unspecified
cut-off frequency.
Our analysis of errors accounts for random, uncorre-
lated errors, such as electronic signal noise, but we cannot
account for systematic errors due to shifts in calibration
or fluctuations in the ORP. Other possible sources of sys-
tematic error include the possibility that calibration salts
only partially dissolved in Sorceress 1, leakage of BaCl
2
reagent into the WCL cells, ionic interferences in the ISE
potentials, and uncertainties in the conductivity analysis
(Kounaves et al., 2010a). Systematic errors due to BaCl
2
leakage can be evaluated by examining the change in the
Cl
ISE potential over time. Furthermore, ionic interfer-
ences have been estimated and can be accounted for, such
as the influence of K
+
on the NH
4
+
ISE sensor (Lukow
and Kounaves, 2005) and the bias introduced in the Ca
2+
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 147
ISE potential by ClO
4
(Kounaves et al., 2010a). Systematic
errors due to fluctuations in the ORP appear to be perva-
sive throughout the WCL analyses. With respect to Rosy
Red, the signal on sol 30 behaves largely as expected for
a nominal WCL experiment and appears to be minimally
affected by the ORP (Fig. 1A). In contrast, Sorceress 1
and 2 on sols 41 and 107 respectively are characterized by
large and anomalous signal fluctuations before the addition
of calibration salts and after the soil analysis (Kounaves
et al., 2010a). Quantifying systematic errors in these analy-
ses would be difficult, but would increase the errors for ion
concentrations in Sorceress 1 and 2.
Unfortunately, systematic errors due to shifts in the cal-
ibration are difficult to detect because a two-point calibra-
tion of the right dynamic range could not be performed
repeatedly on Mars, but the method of calibration using
the Earth-based calibration slope is expected to cause an
additional relative error of only 2–4% in ion activity for
concentrations similar to the WCL solutions (Grunthaner
et al., 2009). This is small compared to the errors in Table 3.
However, it is likely that errors in the pH ISE slopes are
underestimated. This is because the pH ISE slopes were
determined using only a single two-point calibration and
the measured WCL pH values are beyond the calibration
range (the two calibration pHs were near 5.0 and 6.7). If
repeated, more complete calibrations had been done, we
could be more confident of the pH ISE slopes and the esti-
mated error in the pH ISE calibration slopes would proba-
bly be larger; however, even if the calibration slope of the
Rosy Red Ph
a
sensor is varied by ±5 mV dec
1
, the calcu-
lated pH varies only by approximately ±0.25. The pH val-
ues are not used in any of the modeling later in this paper,
so they do not affect our conclusions.
We extended our analysis of Rosy Red by analyzing ISE
potentials and temperatures over sols 30, 32, 34, 66, 87, and
138 (Fig. 2). During sol 30, ISE potentials change sharply
upon addition of calibration salts and sample, and attain
relatively constant values at other times. Ion concentrations
calculated from potentials after the addition of sample
remain within error bounds of the values in Table 3, with
the exception of Cl
, due to leakage of BaCl
2
reagent,
and the pH. The Cl
ISE appears to be affected by the
BaCl
2
leakage only after the addition of soil sample, which
suggests that the soil analysis was minimally affected by
BaCl
2
leakage. The concentration of Cl
at the end of sol
30 is 3.9 mM, indicating that 3.5 mM of Cl
had leaked
into the WCL cell by the end of sol 30. Assuming that this
Cl
was from the BaCl
2
leak, this corresponds to 9.1 mg of
BaCl
2
or 3% of the total BaCl
2
in all three crucibles (0.3 g
BaCl
2
total). The pH
a
ISE increases slowly after the addi-
tion of soil, from a value of 16.6 mV during the analysis
window to a value to 18 mV at the end of sol 30. For a tem-
perature corrected ISE slope of 53.4 mV dec
1
, this indi-
cates a decrease in the pH by 0.65 over a period of five
hours. Such slow drifts in pH are common in measurements
of soil pH and downward drifts in pH have been noted to
occur in oxide-rich soils after addition of base due to the
buffering capacity of OH groups on oxide surfaces, which
gradually desorb protons (Onoda and De Bruyn, 1966).
Alternatively, Quinn et al. (2011) suggested that the BaCl
2
leak caused the slow decrease in pH. The addition of BaCl
2
to the WCL solution should precipitate insoluble BaCO
3
,
which would decrease the solution alkalinity and lower
the pH. However, using PHREEQC, we model a pH
decrease of only 0.01 in Rosy Red due to BaCl
2
leakage,
Table 3
Total ion concentrations (mM) and pH in Rosy Red, Sorceress 1, and Sorceress 2 from this study and from previous analyses by Kounaves
et al. (2010a,b).
Rosy Red Sorceress 1 Sorceress 2
This study Previous This study Previous This study Previous
pH
a
7.67 ± 0.08 7:74þ0:11
0:11 >7.32 ± 0.08 7:62þ0:18
0:12 >6.52 ± 0.06
a
–
pH
b
–
b
–
b
>7.40 ± 0.08 7:61þ0:16
0:12 >6.86 ± 0.09
a
–
pH
irid
8.30 ± 0.08
a
– >7.00 ± 0.12
a
– >6.80 ± 0.13
a
–
Ca
2+
0.16 ± 0.07 0:55þ0:75
0:34 0.45 ± 0.18 0:42þ0:76
0:31 0.09 ± 0.04 0:60þ0:79
0:34
Mg
2+
2.91 ± 0.85 2:90þ1:90
1:20 6.22 ± 2.23 2:20þ2:00
1:10 1.31 ± 0.42 3:70þ3:00
1:70
Na
+
1.46 ± 0.33 1:40þ0:65
0:48 3.52 ± 0.45 1:10þ0:60
0:38 0.99 ± 0.28 1:40þ1:00
0:61
K
+
0.33 ± 0.05 0:36þ0:29
0:17 0.50 ± 0.17 0:17þ0:20
0:10 0.17 ± 0.03 0:39þ0:32
0:17
NH
4
+
0.02 ± 0.01 0:04þ0:04
0:01 -0.03 ± 0.01 – 0.00 ± 0.01 0:03þ0:06
0:03
Cl
0.39 ± 0.04 0:60þ0:14
0:12 0.79 ± 0.14 0:24þ0:11
0:13 0.24 ± 0.03 0:47þ0:21
0:11
ClO
4
2.89 ± 0.54 2:70þ1:40
0:95 2.11 ± 0.50 2:10þ0:86
1:20 2.72 ± 0.57 2:20þ2:20
0:81
SO
4
(total) 4.17 ± 3.47 4:80þ1:50
1:50 –
b
–
b
4.97 ± 1.44 5:90þ1:50
1:50
CaSO
4
0
0.05 – –
b
–
b
0.02 –
MgSO
4
0
0.56 – –
b
–
b
0.28 –
a
pH values calculated by calibrating to the pH
a
ISE sensors as described in Section 3.1.7.
b
No value is given due to an ISE sensor failure.
Table 4
A sensitivity analysis of Mg
2+
concentrations (mM) in Rosy Red.
Values for temperature, ISE potential, ionic strength, and slope are
varied from the original value by ± 2r. The difference between the
2rand +2rvalues, D2r, is shown in the far right column.
Measurement 2rOriginal +2rD2r
Calibration temperature 2.92 2.91 2.81 0.11
Calibration ISE potential 3.11 2.91 2.64 0.47
Calibration slope 3.87 2.91 2.21 1.66
Analysis temperature 2.83 2.91 2.90 0.07
Analysis ISE potential 2.54 2.91 3.23 0.69
Ionic strength 2.50 2.91 3.11 0.61
148 J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168
far less than the observed 0.65 pH decrease. In our model-
ing, we assume that 1.75 mM BaCl
2
had leaked into the
WCL cell (based on the observed Cl
increase over sol
30), a pCO
2
of 4 mbar, and an alkalinity of 2 mM, adjust-
ing SO
4
2
for charge balance. The observed pH decrease of
0.65 can be replicated in PHREEQC only if 50 times more
BaCl
2
had leaked into the cell, which suggests that the pH
decrease was not caused by the BaCl
2
leak.
After sol 30, ISE potentials have large and anomalous
variations from sol to sol (Fig. 2). We speculate that the
large offsets in ISE potentials over successive sols are due
to changes in the ISE calibration induced by repeated
freeze–thaw cycles in the WCL beaker (see Fig. 2, lower
panel). Changes in ISE calibration upon freeze/thaw cycles
imply that absolute ion concentrations cannot be accurately
determined after the initial sol of analysis. For most ions,
ISE potentials increase following the previous sol. In partic-
ular, the Cl
ISE potential is higher during sol 32 and the
beginning of sol 34 than at the end of sol 30, perhaps sug-
gesting that Cl
had decreased in concentration (note: there
is an inverse relationship between potential and concentra-
tion for anions). Near the rapid increase in Ba
2+
upon addi-
tion of the second BaCl
2
crucible, indicated by the spike in
Ba
2+
ISE potential, the Cl
ISE potential is used to infer
the SO
4
2
content of Rosy Red (Kounaves et al., 2010b);
however, the Cl
ISE potential at this point is similar to
its value at the end of sol 30. Furthermore, ISE potentials
on sol 34 appear to be adversely affected by fluctuations
in the ORP, more so than on other sols (Fig. 2 and
Fig. 1B). These observations indicate that Cl
ISE poten-
tials used to calculate SO
4
2
concentrations are inaccurate.
To account for additional errors in the SO
4
2
measurements
in Rosy Red due to systematic errors in the Cl
ISE
potentials on sol 34, we add an additional error of
±15 mV to the Cl
ISE potential on sol 34. We base this
additional systematic error on the uncertainty in the Cl
ISE potential near the end of sol 34, keeping in mind that
there may be additional error because of shifts in the
calibration. No additional error is added to the SO
4
2
analysis in Sorceress 2.
The accuracy of aqueous chemical analyses is commonly
checked by calculating the charge balance; a charge balance
near zero indicates that all major species have been accu-
rately accounted for. The charge balances of Rosy Red,
Sorceress 1 and 2 are 12.7%, 71.8%, and 53.1% respec-
tively. For reference, charge balances in Earth-based aque-
ous analyses are typically less than ±10%, and a charge
balance less than ±5% is considered a good aqueous anal-
ysis (Fritz, 1994). Sorceress 1 has a positive charge balance,
indicating that there is a large anion deficiency in the anal-
ysis. This makes sense because SO
4
2
, a significant compo-
nent in Rosy Red and Sorceress 2, could not be analyzed
in Sorceress 1 due to a failure of the Ba
2+
ISE (Kounaves
et al., 2010b). Charge balances for both Rosy Red and Sor-
ceress 2 are negative, indicating either excess anions in the
analysis or that cation concentrations are deficient. The
presence of another major cation besides Ca
2+
,Mg
2+
,
Na
+
,orK
+
is unlikely because these ions account for
nearly all of the positive charge in most natural waters. Fur-
thermore, Quinn et al. (2011) concluded that Fe
2+
and
other redox active species are not present in significant con-
centrations based on oxidation–reduction potential mea-
surements. Ferric sulfates (Fe
2
(SO
4
)
3
) have been found on
Mars (Johnson et al., 2007; Lane et al., 2008) and may con-
tribute Fe
3+
to solution; however, substantial Fe
3+
is only
soluble in acidic waters, which is inconsistent with the alka-
line pH of Phoenix soils. Given that all major cations in the
WCL solutions were probably measured, the negative
charge balance in Rosy Red and Sorceress 1 suggests that
either Cl
, ClO
4
,orSO
4
2
concentrations were overesti-
mated in the WCL analyses, especially considering that
unanalyzed anions were probably present in the WCL solu-
tions, as discussed in the following paragraph. We suggest
that SO
4
2
concentrations are overestimated because the
SO
4
2
concentrations are either close to or exceed the total
cation charge and the SO
4
2
analysis is subject to the most
error.
Although it is unlikely that another major cation is pres-
ent in the WCL solutions, it is likely that some anions were
-20
-10
0
10
20
30
40
50
60
70
10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00
Mg2+ ISE Potential (mV)
LMST
Sample Addition
Calibration Crucible
Rosy Red
sol 30
A
180
200
220
240
260
280
300
320
340
11:00 12:00 13:00 14:00 15:00 16:00 17:00
Cl ISE Potential (mV)
LMST
BaCl2Additions
Acid Addition
Rosy Red
sol 34
B
Fig. 1. Examples of relatively clean and noisy ISE data from the
Rosy Red analysis. Gray data points are ISE potentials referenced
to a Li
+
ISE, and red lines are Kalman smoothed values. Drawer
opening events are indicated by gray dashed vertical lines, and
other major events are indicated by black dashed vertical lines.
Additional graphs juxtaposing raw and Kalman smoothed data on
sol 30 are presented in Appendix B. (A) Mg
2+
ISE potentials on sol
30. The calibration and sample analysis intervals are indicated by
black dots. (B) Cl
ISE potentials on sol 34 showing data affected
by ORP fluctuations. The increase in the Ba
2+
ISE potential used
to flag the endpoint of the SO
4
2
titration occurred upon addition of
the second BaCl
2
crucible. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version
of this article.)
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 149
not analyzed. The presence of HCO
3
is suggested by the
rapid increase in pH upon sample addition (Hecht et al.,
2009; Kounaves et al., 2010a) and the existence of carbon-
ate minerals in the Phoenix soil (Boynton et al., 2009; Sutter
et al., 2012). Estimates of alkalinity concentrations in Phoe-
nix soils are on the order of 1–2 meq L
1
(Marion et al.,
2010; Kounaves et al., 2010a,b). Other anions that may
be present in the WCL solution are chlorate (ClO
3
)
(Hanley et al., 2012) and nitrate (NO
3
), which are com-
monly associated with each other and perchlorate on Earth.
The perchlorate:chlorate mass ratio varies widely in terres-
trial deserts, but is approximately 1:1 in caliche from the
Atacama where perchlorate is closest to the high abundance
on Mars (Rao et al., 2010). Furthermore, both chlorate and
nitrate have recently been found in high abundance relative
to perchlorate in Mars meteorite EETA79001 (Kounaves
et al., 2014). We consider the possible presence of ClO
3
,
and briefly consider NO
3
, later in our modeling results.
5. MODELING WCL SOLUTION COMPOSITIONS
AND PARENT SALT ASSEMBLAGES
5.1. Possible WCL solution compositions
The relative proportions of ions measured in the WCL
experiment strongly influence what salts precipitate from
solution during freezing or evaporation (Marion et al.,
2010; Hanley et al., 2012) and have implications for soil
minerals in equilibrium with the solution (Boynton et al.,
2009; Hecht et al., 2009; Kounaves et al., 2010a,b). Due
to uncertainties in the ion concentrations inferred from
WCL (Table 3), we examine the spread of possible solution
compositions in Rosy Red within error bounds to deter-
mine variations in the predicted parent salts. We choose
Rosy Red for this analysis because it is clearly more robust
than the Sorceress 1 or 2 analyses. Seven ions were mea-
sured during the Rosy Red analysis (Ca
2+
,Mg
2+
,Na
+
,
K
+
,Cl
,SO
4
2
, and ClO
4
), excluding other ions present
in the calibrant solution. If we assume that there are five
possible values for each ion within error bounds (the mea-
sured value, ±r/2, and ±r), then there are 5
7
or 78,125
unique ionic combinations possible.
For each of the 78,125 possible solutions, we calculate
alkalinity from charge balance and implement ion-pair cor-
rections in PHREEQC. The ion-pair corrections were facil-
itated by using the RATES and KINETICS functions to
adjust unpaired ion concentrations, and organizing the
PHREEQC input code in Excel
Ò
. From these possible solu-
tions, we exclude solutions if the alkalinity is less than zero,
implying a strongly acidic soil solution. An acidic soil solu-
tion conflicts with the measured pH around 7.7. 27,512 out
of the 78,125 possible solutions have an alkalinity greater
Fig. 2. Graphs of ISE potentials (mV) referenced to a Li
+
ISE and temperature (°C) measured over sols 30, 32, 34, 66, 86, and 138. Sols are
separated by vertical black lines, drawer open/close events are indicated by gray dashed lines, and events such as calibrant, sample, acid, and
BaCl
2
addition are indicated by black dashed lines. Note: ISE potentials increase with concentration for cations, and decrease with
concentration for anions.
150 J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168
than zero. Within these 27,512 solutions, Ca
2+
,Na
+
,K
+
,
Cl
, and ClO
4
concentrations are uniformly distributed
between the +r,+r/2, WCL, r/2, and rvalues
(Table 5). In contrast, Mg
2+
and SO
4
2
concentrations have
asymmetrical distributions; higher Mg
2+
and lower SO
4
2
concentrations are more frequent in the possible solutions.
No possible solutions have SO
4
2
concentrations greater
than the WCL analysis (4.24 mM), which places an upper
bound on SO
4
2
concentrations in Rosy Red. Average sol-
ute concentrations in the possible WCL solutions (including
ion-pair contributions) are: Ca
2+
= 0.186, Mg
2+
= 3.459,
Na
+
= 1.472, K
+
= 0.329, Cl
= 0.407, SO
4
2
= 1.511,
ClO
4
= 2.743 mM, and Alk. = 2.919 meq.
5.2. Modeling with FREZCHEM
Parent salt compositions at the Phoenix site have been
inferred from numerical equilibrium models such as FREZ-
CHEM (Marion et al., 2010) and Geochemist’s Work-
bench
Ò
(Hanley et al., 2012). These equilibrium models
assume that salts are initially dissolved in solution and
are then precipitated as the solution is concentrated by
either freezing or evaporation. The presence of liquid water
at the Phoenix site, both past and present, is suggested by
the detection of 3–5 wt.% carbonates by the Thermal and
Evolved Gas Analyzer (TEGA) (Boynton et al., 2009),
the distribution of perchlorate salts throughout the soil col-
umn (Cull et al., 2010), and observations (albeit controver-
sial) of deliquescence on the Phoenix Lander struts (Renno
´
et al., 2009). After the formation of a liquid solution, salts
may precipitate from solution as the temperature decreases
and, once temperatures drop below the eutectic, the solu-
tion will freeze. Evaporation is an alternative pathway for
salt precipitation and produces different salt assemblages
than freezing (Marion et al., 2010). However, there are sev-
eral observations which suggest that salts at the Phoenix
site precipitate during freezing: (1) concentrated solutions
of ClO
4
have low water activities and will resist evapora-
tion (Chevrier et al., 2009), (2) soil solutions will be buffered
against evaporation by overlying soil layers and ground ice,
and (3) the presence of ice near the surface indicates that
vapor transport in the soil is either extremely slow or is in
steady state (Mellon et al., 2008, 2009; Smith et al., 2009).
Equilibrium models also assume that salts in the soil are
in equilibrium with each other and with brines. Potentially,
salts in Martian soils could be in disequilibrium due to the
limited mobility of brines in cold-dry soils. However, the
distribution of perchlorate salts in Phoenix soils suggests
that salts have been redistributed throughout the soil by
downward percolating brines (Cull et al., 2010). Aqueous
transport in the soil would promote equilibration by allow-
ing salts to chemically interact with each other. Although
brine transport in cold-dry soils is slow (Ugolini and
Anderson, 1973; Hagedorn et al., 2010; Toner and
Sletten, 2013), the assumption of equilibrium in the Phoenix
soil is not unreasonable given an estimated soil age at the
Phoenix site of 600 Ma (Heet et al., 2009) and inferred con-
ditions of periodic moisture (Boynton et al., 2009).
To determine how salt phases and brine compositions
will change during freezing, we use FREZCHEM to model
equilibrium freezing of the average composition of possible
WCL solutions (Figs. 3 and 4). This equilibrium modeling
is different from fractional crystallization modeling done
by Marion et al. (2010) because our model allows salts to
redissolve into solution at a lower temperature after they
precipitate. Between 0 and 35 °C, our modeling results
are similar to modeling done by Marion et al. (2010); the
solution evolves towards a primarily Na-ClO
4
-rich compo-
sition at low temperatures, with lesser components of Mg
2+
and Cl
.K
+
concentrations quickly decrease and stay low
due to precipitation of insoluble KClO
4
.Ca
2+
concentra-
tions are also low due to the precipitation of CaSO
4
2H
2
O
and CaCO
3
, but steadily increase with decreasing tempera-
ture. Hydromagnesite is the dominate sink for alkalinity.
SO
4
2
concentrations rapidly decrease on freezing when
MgSO
4
11H
2
O precipitates, which is the most prevalent
soluble salt phase by weight in the Phoenix soil according
to our model.
Following Marion et al. (2010), alkalinity and carbonate
phases were removed below 20 °C because these phases
cause convergence failures and, as noted by Marion et al.
(2010), carbonate chemistries are only valid in FREZ-
CHEM down to 22 °C. Marion et al. (2010) also removed
minor Ca
2+
,K
+
, and SO
4
2
from solution; however, we
retain these ions. To conserve charge balance, alkalinity is
assumed to precipitate as CaCO
3
, removing an equivalent
amount of Ca
2+
from solution. Below 35 °C, the solution
evolves towards a Mg–Ca–ClO
4
-rich composition and is
distinct from the solution compositions modeled in
Marion et al. (2010).At36.2 °C and 43.7 °C, NaClO
4-
2H
2
O and MgCl
2
12H
2
O begin precipitating respectively,
with NaClO
4
2H
2
O later transitioning to NaClO
4
H
2
Oat
45.2 °C. These salts remove Cl
and Na
+
from solution,
causing the resulting solution to become enriched in
Mg
2+
,Ca
2+
, and ClO
4
. Surprisingly, Ca
2+
does not com-
bine with SO
4
2
at low temperatures to precipitate CaSO
4-
2H
2
O; instead, SO
4
2
is precipitated with Mg
2+
as
MgSO
4
11H
2
O. The formation of Ca
2+
-rich compositions
did not occur in modeling done by Marion et al. (2010)
because Ca
2+
was removed from solution at 20 °C.
Table 5
The percentage of Rosy Red solutions with a given ion concentration within ±1rof the WCL analysis in which alkalinity > 0.
Ca
2+
Mg
2+
Na
+
K
+
Cl
ClO
4
SO
4
2
+r20.7 26.7 21.3 20.2 19.8 18.0 0.0
+r/2 20.3 22.7 20.6 20.1 19.9 18.6 0.0
WCL 20.0 22.5 20.0 20.0 20.0 19.8 4.0
r/2 19.6 16.6 19.4 19.9 20.1 21.2 39.3
r19.3 11.4 18.7 19.8 20.2 22.4 56.8
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 151
Solution compositions and salt phases are only shown to
70 °CinFig. 3, but modeling at lower temperatures indi-
cates that the solution continues evolving towards a more
Ca-ClO
4
-rich composition and does not freeze until below
100 °C. This is much lower than the individual eutectics
of either Mg(ClO
4
)
2
(68.2 °C) or Ca(ClO
4
)
2
(74.4 °C).
A
B
Fig. 3. (A) Ion concentrations modeled during freezing of the average composition of possible WCL solutions. (B) Salt precipitates modeled
during freezing of the average composition of possible WCL solutions. Precipitates in B occur at the same temperature indicated by the arrows
in A.
KClO4
4.2%
CaCO3
0.8% hydromagnesite
9.3%
MgSO4·11H2O
52.6%
MgCl ·12H O
5.8%
Mg(ClO ) ·6H O
5.0%
Ca(ClO ) ·6H O
3.1%
NaClO42
422
422
22
·H O
19.2%
Fig. 4. The average proportion of salt phases by weight in Rosy Red inferred from FREZCHEM.
152 J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168
Although we expect that some eutectic depression will
occur in mixed salt systems, a eutectic depression below
100 °C is probably an artifact of the model. Similarly
low eutectic temperatures occur when other mixtures of
aqueous Ca(ClO
4
)
2
–Mg(ClO
4
)
2
are modeled in FREZ-
CHEM. These anomalously low eutectic depressions may
be caused by the parameterization of the Mg(ClO
4
)
2
6H
2
O
solubility product in FREZCHEM to the 68 °C eutectic
determined by Dobrynina et al. (1980) and Pestova et al.
(2005).Stillman and Grimm (2011) have suggested that
the 68 °C Mg(ClO
4
)
2
eutectic is incorrect, likely due to
supercooling, and should be about 10 °C higher at
57 °C. Our own lab experiments confirm that solutions
of Mg(ClO
4
)
2
readily supercool and that 57 °C is the true
eutectic (Toner et al., 2014).
There are several other reactions in Fig. 3 that run coun-
ter to experimental and modeling predictions for freezing
seawater (Marion et al., 1999), surface waters (Toner and
Sletten, 2013), and FREZCHEM modeling at other sites
on Mars (Marion et al., 2009). Specifically, modeling done
here predicts that MgCl
2
12H
2
O precipitates instead of
NaCl2H
2
O, despite the lower eutectic temperature of
NaCl2H
2
O(21.3 °C) compared to MgCl
2
12H
2
O
(33 °C), and the high concentrations of both Na
+
and
Cl
above 35 °C. Similar reversals occur with respect to
hydromagnesite vs. CaCO
3
, MgSO
4
11H
2
O vs. CaSO
4
2H
2-
O, and MgSO
4
11H
2
O vs. Na
2
SO
4
10H
2
O precipitation; the
more soluble minerals hydromagnesite and MgSO
4
11H
2
O
preferentially precipitate over less soluble CaCO
3
, CaSO
4-
2H
2
O, and Na
2
SO
4
10H
2
O. This effect is most apparent
below 50 °C, where the solution is rich in Ca
2+
. Typically,
freezing of surface waters results in Ca
2+
-depleted solutions
because HCO
3
and SO
4
2
combine with Ca
2+
to precipitate
CaCO
3
and CaSO
4
2H
2
O from solution (Marion et al.,
2009; Toner and Sletten, 2013). However, the WCL solu-
tion evolves to a Ca
2+
-rich composition in spite of the ini-
tially high proportions of HCO
3
and SO
4
2
relative to Ca
2+
.
The peculiar low temperature evolution of Ca–ClO
4
-rich
brine modeled in FREZCHEM is caused by high Mg
2+
activities relative to Ca
2+
activities, which results in simul-
taneous ‘salting out’ and ‘salting in’ effects for salts of these
ions. ‘Salting out’ occurs when the activity coefficient of an
ion increases, which can cause that ion to precipitate out of
solution as a salt; similarly, ‘salting in’ occurs when the
activity coefficient for an ion decreases, causing that ion
to either stay in solution or to dissolve into solution from
a precipitated phase. For a mixed Ca(ClO
4
)
2
–Mg(ClO
4
)
2
brine, FREZCHEM calculates activity coefficients (c) for
Ca
2+
and Mg
2+
using the Pitzer equation:
ln cM¼z2
MFþmað2BMa þZCMaÞ
þX
c½mcð2UMc þmaWMcaÞþjzMjmcmaCca ð9Þ
where, m is a molal concentration, subscript M stands for
either Ca
2+
or Mg
2+
, subscript c stands for a cation differ-
ent from M, a stands for ClO
4
,z
M
is the cation charge, B
Ma
and C
Ma
are cation-ClO
4
interaction parameters, U
Mc
is
the Ca
2+
–Mg
2+
interaction parameter, and W
Mca
is the
Ca
2+
–Mg
2+
–ClO
4
interaction parameter. F is a modified
Debye–Hu
¨ckel term and Z is an equation parameter. By
separating this equation into Debye–Hu
¨ckel, cation-ClO
4
,
Ca
2+
–Mg
2+
, and Ca
2+
–Mg
2+
–ClO
4
interaction compo-
nents, we can visualize the various contributions to ln c
M
in Ca(ClO
4
)
2
–Mg(ClO
4
)
2
solutions at varying temperatures
and compositions (Fig. 5).
Fig. 5 shows that cation-ClO
4
and Ca
2+
–Mg
2+
interac-
tions greatly increase at lower temperatures, causing Ca
2+
activity coefficients to become small relative to Mg
2+
.In
a 3 molal solution dominated by Mg
2+
at 60 °C,
cMg2þ=cCa2þis 5000, whereas at 25 °C, cMg2þ=cCa2þis 2.
Such large relative differences in modeled Mg
2+
activities
relative to Ca
2+
at low temperatures causes salts of Mg
2+
to strongly precipitate from solution (the ‘salting out’ effect)
and salts of Ca
2+
to dissolve into solution (the ‘salting in’
effect). We note that the strong influence of Ca
2+
–Mg
2+
interactions on activity coefficients in FREZCHEM is sur-
prising because Pitzer (1991) indicates that these interac-
tions should have only a slight effect on activity
coefficients. Furthermore, Silvester and Pitzer (1978) found
that the temperature dependence of interaction parameters
is small, but in FREZCHEM temperature dependencies can
be large. For example, the change in the B1
Mg;ClO4Pitzer
parameter with temperature is +0.0045 °C
1
at 25 °C based
on experimental heat of dilution data (Silvester and Pitzer,
1978), which is similar to other salts. Assuming that this
change can be extrapolated over a reasonable temperature
range, the value of B1
Mg;ClO4should decrease by 0.1 from
25 °Cto0°C; however, in FREZCHEM, B1
Mg;ClO4increases
by 2.2 over this temperature range.
Very high cMg2þ=cCa2þratios in the presence of ClO
4
suggest the possibility that at low temperatures Ca–Mg–
ClO
4
-rich solutions could form because Mg
2+
will act as
a stronger sink for SO
4
2
and HCO
3
than Ca
2+
. This pre-
diction is consistent with a recent, more in-depth analysis
of the WCL Ca
2+
ISE signal response and experimental
results of ISE signals in mixed Ca(ClO
4
)
2
–Mg(ClO
4
)
2
solu-
tions (Kounaves et al., 2012). Kounaves et al. (2012) found
that a transient signal in the Ca
2+
ISE suggests that about
60% of the ClO
4
in the soil was present as Ca(ClO
4
)
2
. The
salting in and salting out effects found here provide a
possible theoretical framework for how a Ca–ClO
4
-rich
solution might have evolved from a solution with initially
high concentrations of SO
4
2
and HCO
3
.
However, activity coefficients calculated by FREZ-
CHEM are only as accurate as the experimental data used
to parameterize the model. Binary Pitzer parameters for
Mg(ClO
4
)
2
and Ca(ClO
4
)
2
used in FREZCHEM are calcu-
lated from the freezing point depression of ice in varying con-
centration salt solutions (Marion et al., 2010). The probable
underestimation of the Mg(ClO
4
)
2
eutectic in Dobrynina
et al. (1980) and Pestova et al. (2005) suggests that the ice-
solution datasets used to parameterize FREZCHEM may
be inaccurate. To test how sensitive the Pitzer model is to
experimental errors, we calculate Pitzer parameters and
Mg
2+
activity coefficients by assuming that the concentration
of salt in equilibrium with ice varies from FREZCHEM sys-
tematically up to ± 10% (Fig. 6). Following Marion et al.
(2010), this is done by fitting the Pitzer parameters B
Ma
and C
Ma
to the equation P(T) = P
298.15
+ A(298.15 T)
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 153
using a sum-least-squares fitting method, where P(T) is the
temperature-dependent Pitzer parameter, P
298.15
is the value
at 298.15 K, T is the temperature in Kelvin, and A is the con-
stant to be fitted. Mg
2+
activity coefficients are then calcu-
lated using the new Pitzer parameters. The results of this
sensitivity analysis indicate that activity coefficients are
highly sensitive to errors in the experimental data, particu-
larly at low temperatures (Fig. 6). A decrease in the experi-
mental salt concentration values in equilibrium with ice
produces an exponential increase in modeled activity coeffi-
cients. Given that freezing point depressions in Dobrynina
et al. (1980) and Pestova et al. (2005) may be too low due
to metastability in Mg(ClO
4
)
2
solutions (Stillman and
Grimm, 2011), the concentration of salt in equilibrium with
ice at a given temperature would be higher (the blue lines in
Fig. 6B, which correspond to activity coefficients in
Fig. 6A). The possible experimental errors and inconsisten-
cies mentioned above point to a need for new, highly accurate
experimental work on freezing-point depressions in pure
Mg(ClO
4
)
2
and Ca(ClO
4
)
2
solutions, and tests of the FREZ-
CHEM model in mixed salt systems containing ClO
4
.
5.3. Chemical divide modeling
Ideally, all of the 27,512 possible WCL solutions could
be evaluated for salt precipitates and brine compositions
using FREZCHEM. However, for reasons given above,
the use of perchlorate parameters in mixed salt systems
needs further experimental evaluation before we can be
confident of model predictions. In addition, FREZCHEM
sometimes fails to converge below about 20 °C, which
has been attributed to the presence of minor species at
low temperatures (Marion et al., 2010). We find that con-
vergence failures in FREZCHEM below 20 °C are caused
when ClO
4
concentrations become high in the presence of
Mg
2+
. In some cases, the activity coefficient of Mg
2+
-15
-10
-5
0
5
10
15
20
020406080100
Contribution to ln(γMg)
% Mg
Debye-Hückel
Cation-Anion
Cation-Cation
Ternary
ln(γ)
25°C
-15
-10
-5
0
5
10
15
20
0 20 40 60 80 100
Contribution to ln(γCa)
% Mg
25°C
-15
-10
-5
0
5
10
15
20
020406080100
Contribution to ln(γMg)
% Mg
–60°C
-15
-10
-5
0
5
10
15
20
0 20 40 60 80 100
Contribution to ln(γCa)
% Mg
–60°C
Fig. 5. Contributions to ln c
Ca
and ln c
Mg
in 3 m perchlorate solutions containing varying proportions of Ca
2+
and Mg
2+
at 25 °C and
60 °C, determined using Pitzer parameters from FREZCHEM. The different components of ln c
M
are: (1) the Debye–Hu
¨ckel component,
Z2
MF, (2) the cation-ClO
4
interaction component, mað2BMa þZCMaÞþPcjzMjmcmaCca , (3) the Ca
2+
–Mg
2+
interaction component,
Pc2UMcmc, and (4) the Ca
2+
-Mg
2+
-ClO
4
interaction component, PcmcmaWMca.
1
10
100
1000
-70 -50 -30 -10 10
Mg2+ Activity Coeff.
Temperture (°C)
-10%
-5%
-2%
+2%
+5%
+10%
FREZCHEM
value
A
-70
-60
-50
-40
-30
-20
-10
0
012345
Temperature (°C)
Mg(ClO4)2(molal)
-10%
-5%
-2%
+2%
+5%
+10%
FREZCHEM
B
Fig. 6. (A) The variation in Mg
2+
activity coefficients with temperature for 3 m solutions of Mg(ClO
4
)
2
that arise if salt concentrations from
experimental ice–solution data used to parameterize the Pitzer model are varied by up to ±10%. (B) Deviations from the FREZCHEM
Mg(ClO
4
)
2
phase diagram assuming that the concentration of salt in equilibrium with ice varied up to ±10%.
154 J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168
increases by a factor of around 10
5
. These convergence
problems could potentially be resolved by revised
Mg(ClO
4
)
2
Pitzer parameters.
Because of the problems in numerical models, we use a
chemical divide model (Hardie and Eugster, 1970; Eugster
and Jones, 1979; Drever, 1982) to evaluate the parent salts
of possible Rosy Red solutions. A chemical divide model
has previously been applied to fluids derived from basalt
weathering on Mars (Tosca and McLennan, 2006) and pri-
mordial Martian solutions (King et al., 2004). In a chemical
divide model, the evolution of brines as they are concen-
trated is determined using a decision tree that considers
the relative solubility of salts and ratios of cations to anions
as the solution evolves. For example, if a solution is charac-
terized by K
+
< ClO
4
, then precipitation of insoluble
KClO
4
will consume K
+
early during freezing and the solu-
tion will evolve towards a K
+
-depleted composition. The
chemical divide model assumes that K
+
completely precip-
itates with an equivalent concentration of ClO
4
, and the
resulting solution is further evaluated for other chemical
divides. If, on the other hand, K
+
> ClO
4
, then the solution
will become depleted in ClO
4
after the precipitation of
KClO
4
. Because the chemical divide model removes chem-
ical species by precipitation reactions, chemical divides
between species are evaluated stoichiometrically in equiva-
lent concentrations (e.g. 2[Mg
2+
] is compared to [Cl
]).
5.3.1. Choosing the chemical divides
In a chemical divide model, the relative solubility of salts
defines the order in which salts are removed upon concen-
tration. Studies of evaporating closed basin lake waters
on Earth have demonstrated the validity of this technique
(Hardie and Eugster, 1970; Eugster and Jones, 1979). Fewer
studies have been done on freezing waters, but in freezing
seawater, salts with higher eutectic temperatures precipitate
before salts with lower eutectic temperatures (Gitterman,
1937; Nelson and Thompson, 1954; Herut et al., 1990). This
suggests that a chemical divide model for freezing waters
can be based on relative eutectic temperatures; however, a
potential complication to this is that salts may redissolve
into solution at temperatures lower than their eutectics
due to changes in ion activities or solubility products with
temperature. FREZCHEM suggests that Na
2
SO
4
10H
2
O,
CaCO
3
, and CaSO
4
2H
2
O will redissolve into solution
when WCL solutions are frozen to low temperatures, but
these effects are likely influenced by an overestimation of
Mg
2+
activities, as discussed previously. Furthermore, with
the exception of Na
2
SO
4
10H
2
O, experimental freezing of
seawater indicates that salts do not redissolve into solution
at lower temperatures once they have precipitated (Marion
et al., 1999).
In a system containing the ions Cl
,SO
4
2
, HCO
3
, and
ClO
4
, the eutectic temperature of salts in order of decreas-
ing temperature is generally given by: carbonates > sul-
fates > chlorides > perchlorates (Table 6). KClO
4
(0.18 °C), NaHCO
3
,(2.76 °C), and KHCO
3
(6.38 °C)
are notable exceptions to this rule, although highly soluble
Na–K-carbonates are unlikely to form because HCO
3
pref-
erentially precipitates with Ca
2+
or Mg
2+
first. K
+
will be
consumed early during freezing because it is a minor soil
constituent and is strongly precipitated as insoluble KClO
4
from a Phoenix WCL solution (Marion et al., 2010). Solu-
ble Ca
2+
is also minor in the Phoenix WCL solution and
will be consumed by precipitation in CaCO
3
first, followed
by precipitation of any remaining Ca
2+
in CaSO
4
2H
2
O.
Although CaCO
3
and CaSO
4
can precipitate in various
hydrated states, such as ikaite (CaCO
3
6H
2
O), anhydrite
(CaSO
4
), or basanite (CaSO
4
1.5H
2
O), this does not affect
the continuing evolution of brines in a chemical divide
model because the ratio of cations to anions in different
hydrated salts is the same. Mg
2+
will then combine with
any remaining alkalinity after CaCO
3
precipitation, which
quantitatively removes alkalinity from solution. Assuming
that anhydrous magnesite (MgCO
3
) does not form on Mars
during freezing due to kinetic inhibition at low tempera-
tures (Langmuir, 1965; Marion et al., 2010), FREZCHEM
predicts that Mg
2+
will precipitate with HCO
3
as hydro-
magnesite. However, because hydrous MgCO
3
minerals
are poorly characterized at low temperatures, and the
hydrated state depends on the pCO
2
and pH, we denote
the hydrous MgCO
3
mineral in our chemical divide model
as MgCO
3
nH
2
O.
After the precipitation of minor Ca
2+
,K
+
, and alkalin-
ity in KClO
4
, CaCO
3
, CaSO
4
2H
2
O, and MgCO
3
nH
2
O,
the remaining solution will have a Mg–Na–Cl–ClO
4
–SO
4
composition. This simplifies the chemical divide model con-
siderably. Based on the relative eutectic temperatures of
salts, the order of salt precipitation from a Mg–Na–Cl–
Table 6
Eutectic temperatures (°C) and solid phase eutectic salts for binary salt systems modeled in FREZCHEM, except where otherwise noted.
Ca
2+
Mg
2+
Na
+
K
+
CO
3
2
CaCO
3
0.006 Hydromagnesite 0.2 NaHCO
3
2.8 KHCO
3
6.4
SO
4
2
CaSO
4
2H
2
O0.03 MgSO
4
11H
2
O3.6 Na
2
SO
4
10H
2
O1.2 K
2
SO
4
1.5
Cl
CaCl
2
6H
2
O49.4 MgCl
2
12H
2
O33.0 NaCl2H
2
O21.3 KCl 10.8
ClO
3
Ca(ClO
3
)
2
6H
2
O41
b
Mg(ClO
3
)
2
6H
2
O69
b
NaClO
3
23
b
KClO
3
3
b
ClO
4
Ca(ClO
4
)
2
6H
2
O74.4 Mg(ClO
4
)
2
6H
2
O
e
57
a
NaClO
4
2H
2
O34.3 KClO
4
0.18
NO
3
Ca(NO
3
)
2
4H
2
O28.7
c
Mg(NO
3
)
2
9H
2
O31.9
d
NaNO
3
17.6 KNO
3
2.9
a
Stillman and Grimm (2011).
b
Hanley et al. (2012).
c
Bassett and Taylor (1912).
d
Ewjing et al. (1933).
e
Marion et al. (2010) is uncertain whether this salt has 6 or 8 waters of hydration. We assume a hydrated state of 6H
2
O because this is
consistent with the hydrated state at 25 °C.
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 155
ClO
4
–SO
4
solution will be given by Na
2
SO
4
10H
2
O,
MgSO
4
11H
2
O, NaCl2H
2
O, MgCl
2
12H
2
O, NaClO
4
2H
2-
O, and Mg(ClO
4
)
2
6H
2
O, in order from first to last.
Although this precipitation sequence indicates that Na
2-
SO
4
10H
2
O precipitates before MgSO
4
11H
2
O, once tem-
peratures decrease to the point where NaCl2H
2
Oor
NaClO
4
2H
2
O begin precipitating, Na
+
concentrations will
decrease. This will cause any Na
2
SO
4
10H
2
O that had pre-
cipitated to redissolve into solution. The SO
4
2
released by
Na
2
SO
4
10H
2
O will then combine with Mg
2+
to precipitate
MgSO
4
11H
2
O. Because Mg
2+
concentrations are high in
the WCL solutions, this reaction will continue until either
all of the Mg
2+
has been precipitated as MgSO
4
11H
2
O
or all of the Na
2
SO
4
10H
2
O has dissolved. A similar effect,
in which Na
2
SO
4
10H
2
O redisolves into solution after
NaCl2H
2
O begins precipitating, occurs in seawater
(Marion et al., 1999). As a result, Na
2
SO410H
2
O and
MgSO
4
11H
2
O can be thought of as effectively precipitating
after NaCl2H
2
O and NaClO
4
2H
2
O in a chemical divide
model. Possibly, mixed Na
2
SO
4
MgSO
4
salts could precip-
itate from solution, such as bloedite (Na
2
SO
4
MgSO
4
4H
2-
O); however, we do not consider such salts in our low
temperature modeling because bloedite is unstable at low
temperatures (Marion and Farren, 1999).
5.3.2. Chemical divide model results
When the chemical divide model outlined above is
applied to all of the 27,512 possible Rosy Red WCL solu-
tions, only two, similar evolution pathways are followed
(Fig. 7). Ca
2+
is removed early as both CaCO
3
and some-
times CaSO
4
2H
2
O, followed by K
+
as KClO
4
.
Here, the
chemical divide model indicates that the possible solutions
diverge slightly. If 2Ca
2+
> Alk., then alkalinity is com-
pletely consumed by CaCO
3
precipitation and excess
Ca
2+
precipitates as CaSO
4
2H
2
O; on the other hand, if
2Ca
2+
< Alk., then Ca
2+
is completely consumed by CaCO
3
precipitation and excess alkalinity precipitates as MgCO
3-
nH
2
O. Following the precipitation of Ca
2+
and alkalinity,
all of the possible WCL solutions in the chemical divide
model have a similar Mg–Na–Cl–ClO
4
composition and
follow the same evolution pathway.
The most common WCL evolution pathway (92.3% of
the possible solutions) results in a final parent salt compo-
sition of MgSO
4
11H
2
O, MgCO
3
nH
2
O, Mg(ClO
4
)
2
6H
2
O,
NaClO
4
2H
2
O, KClO
4
, NaCl2H
2
O, and CaCO
3
, in order
of mass abundance (Table 7 and Fig. 8A). All of the possi-
ble WCL solutions precipitate KClO
4
, CaCO
3
, MgSO
4
11H
2
O, NaCl2H
2
O, NaClO42H
2
O, and Mg(ClO
4
)
2
6H
2
O. MgSO
4
11H
2
O is the most prevalent salt phase pres-
ent in the Phoenix soil and comprises 1.2 wt.% of the soil on
average. The weight fraction for CaCO
3
is much lower than
estimates of 3–5 wt.% CaCO
3
based on a high temperature
CO
2
release measured in TEGA (Boynton et al., 2009). This
is not surprising because CaCO
3
is sparingly soluble and
would have only partially dissolved into solution. In con-
trast, highly soluble salts such as Mg(ClO
4
)
2
6H
2
O and
NaClO
4
2H
2
O would have completely dissolved and repre-
sent the actual concentration of these salts in the soil.
On a mole fraction basis, MgSO
4
11H
2
O is the
dominant sink for SO
4
2
in the possible WCL solutions,
averaging 99.4 mol% of the SO
4
2
. MgCO
3
nH
2
O is the
dominant sink for alkalinity, averaging 87.8 mol% of the
alkalinity, followed by CaCO
3
(averaging 12.2 mol%).
The inferred dominance of the MgCO
3
nH
2
O form of car-
bonate is consistent with orbital IR observations of Mar-
tian dust from the Mars Global Surveyor Thermal
Emission Spectrometer (Bandfield et al., 2003) as well as
outcrops of magnesium-rich carbonate observed with the
Compact Reconnaissance Imaging Spectrometer for Mars
(CRISM) (Ehlmann et al., 2008) and instruments on the
Spirit Rover (Morris et al., 2010). Of the perchlorate salts,
KClO
4
averages 12.0 mol%, NaClO
4
2H
2
O averages
38.8 mol%, and Mg(ClO
4
)
2
6H
2
O averages 49.2 mol% of
the ClO
4
.Cl
is completely precipitated as NaCl2H
2
O, a
mineral that is present in all of the possible WCL solutions
and has not been previously considered at the Phoenix site.
How might the existence of the chlorate (ClO
3
) ion in
the Phoenix soil affect the parent salt chemistry? Based on
analyses of ClO
3
/ClO
4
ratios on Earth by Rao et al.
(2010), we consider molar equivalent amounts of ClO
3
and ClO
4
on Mars. We incorporate this possibility into
our Rosy Red chemical divide model by adding in ClO
3
to the possible solutions so that ClO
3
= ClO
4
, and adjust
the solution for charge balance by removing alkalinity.
WCL Solution
K < ClO4
Chemical Divide Model
Ca-Mg-Na-Cl-SO4-ClO4-HCO3
2Ca < Alk. 2Ca > Alk.
2Ca < SO4
2Mg > Alk.
Mg-Na-Cl-SO4-ClO4-HCO3Ca-Mg-Na-Cl-SO4-ClO4
Mg-Na-Cl-ClO4
Mg-Na-ClO4
KClO4
CaCO3
MgCO3·nH2OCaSO4·2H2O
NaCl·2H2O
Na > Cl
NaClO4·2H2O
Mg(ClO4)2·6H2O
Mg-Na-Cl-SO4-ClO4
MgSO4·11H2O
Mg > SO4
Fig. 7. The evolution pathways for possible Rosy Red WCL
solutions in a chemical divide model based largely on the relative
eutectic temperature of component salts. Only pathways taken by
the possible WCL solutions are shown.
156 J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168
With the addition of ClO
3
ion, a total of 13,901 possible
WCL solutions have an alkalinity greater than zero. Apply-
ing the chemical divide model based on relative eutectic
temperatures, the modeled parent salt assemblages are sim-
ilar to assemblages modeled in the absence of ClO
3
(Table 7
and Fig. 8B), with several exceptions. All of the Na
+
precip-
itates as NaClO
3
instead of NaClO
4
2H
2
O due to the higher
eutectic temperature of NaClO
3
(23 °C) compared to
NaClO
4
2H
2
O(34.3 °C). This frees all of the ClO
4
to pre-
cipitate with Mg
2+
as Mg(ClO
4
)
2
8H
2
O. Furthermore,
some of the residual ClO
3
after the precipitation of NaClO
3
precipitates as Mg(ClO
3
)
2
8H
2
O. The lower alkalinity
caused by the addition of ClO
3
in the possible WCL
solutions also lowers the amount of Mg
2+
that precipitates
as MgCO
3
nH
2
O, which frees more Mg
2+
to precipitate as
Mg(ClO
4
)
2
6H
2
O and Mg(ClO
3
)
2
6H
2
O.
We do not explicitly model the possible presence of NO
3
here because its concentration at the Phoenix site is not
known. Qualitatively, the inclusion of nitrates in a chemical
divide model would have the greatest impact on chloride
salts because the eutectic temperatures of Mg
2+
and Na
+
chlorides are similar, but slightly lower than, Mg
2+
and
Na
+
nitrates. As a result, Na
+
in the WCL solutions would
precipitate as NaNO
3
instead of NaCl2H
2
O, and any
remaining Cl
after the precipitation of NaCl2H
2
O would
precipitate as MgCl
2
12H
2
O.
The results of our chemical divide modeling indicate the
equilibrium salt composition if the soil solution is frozen
below its eutectic, i.e. there is no brine component in Table 7.
Above the eutectic, a eutectic solid will melt to form a
mixture of salt, ice, and brine. Although a chemical divide
model cannot predict precise brine compositions above the
eutectic, the general composition of brines can be inferred
by considering the solubility and eutectic temperatures of
the various salt components in the eutectic mixture. The
salts having the lowest eutectic temperatures in our chemical
divide modeling are Mg(ClO
4
)
2
6H
2
O(57 °C) and
Mg(ClO
3
)
2
6H
2
O(69 °C), indicating that liquid brine
Table 7
Results from the chemical divide modeling both without ClO
3
in the WCL solutions and assuming equal concentrations of ClO
3
and ClO
4
.
Values given are the % occurrence of salt phases in possible Rosy Red solutions and the average weight% ð100 gsalt g1
soilÞassuming a soil
mass of 1 g and a soil density of 1 g cm
3
. The standard deviation of the weight% in the possible solutions is also given. To calculate the
weight fraction of MgCO
3
nH
2
O, the molecular weight of hydromagnesite is assumed (365.3 g).
Salt phase Without ClO
3
ClO
3
= ClO
4
% occurrence Weight % % occurrence Weight %
KClO
4
100 0.114 ± 0.011 100 0.114 ± 0.011
CaCO
3
100 0.044 ± 0.015 100 0.046 ± 0.015
CaSO
4
2H
2
O 7.7 0.004 ± 0.015 15.6 0.007 ± 0.020
MgCO
3
nH
2
O 92.3 1.171 ± 0.788 84.4 0.602 ± 0.522
NaCl2H
2
O 100 0.096 ± 0.007 100 0.096 ± 0.007
NaClO
3
– – 100 0.288 ± 0.063
NaClO
4
2H
2
O 100 0.422 ± 0.094 – –
MgSO
4
11H
2
O 100 1.197 ± 0.785 100 0.752 ± 0.495
Mg(ClO
3
)
2
6H
2
O – – 100 0.591 ± 0.153
Mg(ClO
4
)
2
6H
2
O 100 0.559 ± 0.177 100 0.966 ± 0.141
MgSO4·11H2O
33.2%
KClO4
3.2%
Mg(ClO4)2·6H2O
15.5%
MgCO3·nH2O
32.5%
CaCO3
1.2%
CaSO4·2H2O
0.1%
A
NaClO4·2H2O
11.7% NaCl·2H2O
2.7%
MgSO4·11H2O
21.5%
KClO4
3.3%
Mg(ClO4)2·6H2O
28.0%
NaCl2·2H2O
2.8%
MgCO3·nH2O
17.5%
CaCO3
1.2% CaSO4·2H2O
0.2%
Mg(ClO3)2·6H2O
17.2%
NaClO3
8.3%
B
Fig. 8. (A) The average proportion of salt phases by weight in Rosy Red inferred from the chemical divide model without ClO
3
, and (B)
assuming that the ClO
3
concentration is the same as the ClO
4
concentration in the initial solution.
J.D. Toner et al. / Geochimica et Cosmochimica Acta 136 (2014) 142–168 157