On the thermal stability of LiPF6

Article abstract of DOI:10.1016/j.tca.2005.09.006

The results of a comprehensive study of the thermal stability of the salt LiPF₆, using both accelerating rate (ARC) and differential scanning (DSC) calorimetry, are presented. Pressure monitoring during ARC experiments permits also the study of endothermic processes. The origins of apparently inconsistent results and conflicting interpretations in previous reports in the literature are explicated. In a confined volume, LiPF₆(s) melts reversibly at 467 K with a heat of melting of 2.0 ± 0.2 kJ mol ^-1. Reversible decomposition to PF₅(g) and LiF(s) commences with melting, but the autogenic development of PF₅(g) pressure makes the temperature profile of decomposition a function of volume and sample size. The heat of this reaction at constant volume, ΔU _r, as determined by a variety of methods is in the range 60 ± 5 kJ mol^-1, and is approximately temperature independent in range 490-580 K.

Full text of DOI:10.1016/j.tca.2005.09.006

Thermochimica Acta 438 (2005) 184–191
On the thermal stability of LiPF6
Ella Zinigrada, Liraz Larush-Asrafa, Josef S. Gnanarajb,
Milon Sprechera, Doron Aurbacha,∗
a Department of Chemistry, Bar-Ilan University, Ramat-Gan 52900, Israel
b Chemical Engineering, Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609, USA
Received 24 July 2005; accepted 3 September 2005
Available online 5 October 2005
Abstract
The results of a comprehensive study of the thermal stability of the salt LiPF6, using both accelerating rate (ARC) and differential scanning
(DSC) calorimetry, are presented. Pressure monitoring during ARC experiments permits also the study of endothermic processes. The origins of
apparently inconsistent results and conflicting interpretations in previous reports in the literature are explicated. In a confined volume, LiPF6(s)
melts reversibly at 467 K with a heat of melting of 2.0 0.2 kJ mol−1. Reversible decomposition to PF5(g) and LiF(s) commences with melting,
but the autogenic development of PF5(g) pressure makes the temperature profile of decomposition a function of volume and sample size. The heat
of this reaction at constant volume, ꢀUr, as determined by a variety of methods is in the range 60 5 kJ mol−1, and is approximately temperature
independent in range 490–580 K.
© 2005 Elsevier B.V. All rights reserved.
Keywords: ARC; DSC; Thermal decomposition; Hexafluorophosphate
1. Introduction
years, calorimetric tools such as differential scanning calorime-
try (DSC) and accelerating rate calorimetry (ARC) were widely
used in this field in order to study the thermal stability of Li bat-
teries and their components [3,4]. The crucial role of LiPF6 in
the determination of the thermal behavior of these systems has
been clearly demonstrated. Consequently, there is no doubt that
the thermal behavior of LiPF6 itself is interesting and important,
and thus deserves special attention.
Indeed, the thermal stability of LiPF6 has been studied in
recent years using various methods and approaches [5–9]. These
included calorimetric measurements at constant volume [6,8,9]
constant pressure [5,6], dynamic or isothermal conditions, dif-
ferential scanning calorimetry with and without gas removal, in
closed or open crucibles, and thermogravimetric analysis (TGA)
under nitrogen flow [5]. It was found that LiPF6 is not sta-
ble at elevated temperatures and decomposes to LiF and PF5,
demonstrating one [6] or two [8] endothermic peaks in DSC at
constant volume, and more that two peaks [6] at normal pres-
sure. Since the thermal decomposition of LiPF6 is accompanied
by the development of PF5 gas, the onset of these endothermic
processes depends on the test conditions. In an open crucible
undernitrogenflowatatmosphericpressure, theisothermalTGA
measurements show a weight loss related to only one process
at temperatures as low as 343 K [5]. In closed crucibles during
There is no doubt that the recent development and commer-
cialization of high energy density, rechargeable Li-ion batteries
is one of the most important successes in modern electrochem-
istry. These battery systems are conquering the market, and
their range of application continues to expand. In the mean-
time, a huge amount of scientific work is being devoted to this
field by thousands of research groups throughout the world,
because of the complexity of these systems, whose operation
involves highly complicated Li intercalation reactions, interfa-
cial charge transfer reactions, passivation and corrosion phe-
nomena, and simultaneous surface and bulk side reactions. At
present, all the commercial Li-ion batteries comprise electrolyte
solutions based on a LiPF6 salt and alkyl carbonate solvents.
Despite its reactivity with Li-C anodes and LixMOy cathodes
(M = transition metals such as Mn, Ni, Co, etc.), LiPF6 was cho-
sen as a compromise because its nonaqueous solutions are highly
conductive, and it is less dangerous and poisonous than other
possible suitable salts such as LiClO4 or LiAsF6 [1,2]. In recent
∗
Corresponding author. Fax: +972 3 5351250.
E-mail address: aurbach@mail.biu.ac.il (D. Aurbach).
0040-6031/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.tca.2005.09.006

E. Zinigrad et al. / Thermochimica Acta 438 (2005) 184–191
185
DSC measurements, the onset of the first endothermic process
is close to 468 K [6,8,9]. It appears as a peak and is reversible
[6,8].
The XRD patterns of the solid product were measured using
the Advanced D8 diffractometer from Bruker Inc. The patterns
measured at RT showed typical peaks of LiPF6 and LiF (well-
known XRD) patterns.
Though the endothermic decomposition of LiPF6 at elevated
temperatures has been intensively explored, the observations
reported in the literature regarding the onset of its decompo-
sition and the dependence of heat flow on T in DSC experiments
are not consistent. There is furthermore no one clear and unified
explanation for the endothermic behavior observed. Thus, some
[8] have observed a first endothermic onset at 453 K in sealed
crucibles, and have suggested that the peak is connected to the
melting point of the salt. Du Pasquir et al. [9] obtained the first
endothermic peak near 463 K but did not discuss it. The sec-
ond endothermic peak in their measurements was attributed to
the melting point of LiPF6, but the onset was observed at about
513 K. Gavritchev et al. [6] attributed the first endothermic peak
at about ∼468 K in LiPF6/DSC measurements to the decompo-
sition of LiPF6, and not to any phase transition. In experiments
with sealed crucibles, they did not find an additional endotherm
below 513 K. However, it was shown [6] that the decomposi-
tion of LiPF6, when it occurs, is suppressed by the autogenic
development of PF5 pressure.
3. Results and discussion
3.1. ARC measurements
Fig. 1 shows temperature and pressure changes (in time) mea-
sured during ARC experiments with 1.9 g of LiPF6 heated up to
548 K. The ARC construction and software is designed to study
exothermic processes by temperature changes, but not endother-
mic processes. Nevertheless, by a thorough examination of the
pressure changes during ARC experiments, and by careful anal-
ysis of the temperature behavior, it is possible to use ARC for
the study of endothermic processes as well.
Typical results from these ARC measurements are presented
in Fig. 1a and b. Fig. 1b enlarges the results related to the
450–490 K temperature range. The arrow in Fig. 1b shows
an unusual temperature drop of 2.09 ◦C during a waiting and
searching stage, between 460 and 465 K. There were no signif-
icant pressure changes in this range. The normal temperature
decrease in all the other steps was only by 0.65–1.25 ◦C. This
abnormal phenomenon can be explained only by an endothermic
process (melting), which was also observed by DSC measure-
ments, as discussed below. In the following steps, the decompo-
The present work aims at providing a better and more com-
plete picture of the thermal behavior of LiPF6, and to explain
relevant calorimetric responses. To this purpose, we used two
separate tools:
1. DSC measurements of hermetically sealed crucibles loaded
with different amounts of LiPF6.
2. ARC measurements of LiPF6 with a special arrangement in
whichthepressuredevelopedisrigorouslymeasured, inaddi-
tion to the temperature changes.
2. Experimental
An accelerating rate calorimeter from Arthur D Little Inc.,
Model 2000, and a differential scanning calorimeter from Met-
tler Toledo Inc., Model DSC 822, were used.
IntheARCtestsabout2 gofsaltwereloadedintothetitanium
flask (8.2 ml volume) in an argon filled glove box (VAC Inc.) and
were transferred to the ARC under a purified Ar atmosphere. The
salt was heated from 313 to 548 and 623 K, respectively, in 5 ◦C
increments at the rate of 2 ◦C min−1 in the search for self-heating
at a sensitivity threshold of 0.02 ◦C min−1. The controller was
programmed to wait 15 min for the sample and the calorimeter
temperatures to equilibrate, and then to search during 20 min for
a temperature increase of 0.02 ◦C min−1. After the ARC exper-
iments, the reaction vessel (a titanium flask) was cooled with
liquid nitrogen until the pressure was slightly above the atmo-
spheric pressure. The gas was then released through a specially
designed high-pressure valve.
DSC tests were conducted in high pressure, gold-plated stain-
less steel crucibles 30 ␮l in volume, over a temperature range
of from 303 to 623 and 673 K, respectively. The crucibles were
filled with 1–7 mg of the salt, and then sealed in a glove box
under an argon atmosphere. The heating/cooling rates were 0.5,
Fig. 1. (a) Typical ARC response during heating process of LiPF6. (b) The chart
emphasizes the behavior in the temperature range 450–490 K.
1, 2, 5, 8, 10, 20 and 30 K min−1
.

186
E. Zinigrad et al. / Thermochimica Acta 438 (2005) 184–191
This means that for the ARC vessels’ volume, which is
8.2 cm3, the limit of the molar volume of the PF5 gas that
can be formed (i.e., due to the full decomposition of the salt)
is equal to 8.2/0.0125 = 656 cm3 mol−1. The highest value of
the second virial coefficient B at 600 K, based on the Hand-
book of American Institute of Physics [10] for a heavy gas such
as Xe, is 19.6 cm3 mol−1. If we presume that the coefficient B
at 513 K taken for heavy gases is, for example, 40 cm3 mol−1
,
which should be considered as an exaggeration for our case, we
obtain:
B
40
=
= 0.061.
(4)
Vm
656
In our case, the molar volumes of the PF5 gas produced are
relatively high, and thereby, the second term in Eq. (2) is rela-
tively small. Hence, the real gas equation of state converges to
that of an ideal gas.
Fig. 2. In each step of the ARC measurements, an equilibrium pressure is mea-
sured, T, P vs. t.
Thenumberofmolesofthegas(includingbothArandPF5)in
the vessel’s volume of 8.2 cm3 at 548 K and at 137 psi (9.32 atm)
(Fig. 2) is equal to:
sition of LiPF6 is accompanied by gas evolution:
LiPF6 (liquid) ↔ LiF (solid) + PF5 (gas)
(1)
PV
RT
9.32 × 0.0082
0.082 × 548
n =
=
= 1.7 mmol,
(5)
The heat of the endothermic processes cannot be calculated
by ARC measurements.
The pressure changes in the cooling stage is presented in
Fig. 2.
The significant pressure growth that starts above 473 K is, of
course, related to the salt decomposition and PF5 formation (Eq.
(1)). The pressure measured at 548 K was 137 psi. The amount
of salt decomposed could be calculated by the measurements
of pressure, provided that the equation of the state of the gas is
known (e.g., for a perfect gas, PV = nRT, where n is the number
of gas moles). Real gases of course deviate from the perfect gas
law. The compression factor Z (Z = PVm/RT) can be used in the
framework of a virial equation of state:
where R = 0.082 L atm mol−1 K−1
.
TheinitialpressureofArinthebombis1 atm, andthesaltvol-
m
ρ
1.9
ume is:V =
=
= 0.67cm3,where ρ = 2.838 cm3 is the
2.838
density of LiPF6 [11].
Therefore the amount of Ar (moles) in the bomb is:
PV
RT
1 × (0.0082 − 0.00067)
0.082 × 298
nAr
=
=
= 0.31 mmol.
(6)
The pressure reduction due the cooling of the vessel from 548
to 483 K should have been less than what was actually observed.
1.7 mmol of gas occupying a volume of 8.2 cm3 results at 483 K:
nRT
V
0.0017 × 0.082 × 483
0.0082
ꢀ
ꢁ
P =
=
= 8.21 atm = 120.7 psi.
(7)
B
C
Vm2
PVm = RT 1 +
+
+ · · ·
,
(2)
Vm
where B and C are the second and the third virial coefficients,
and Vm is the molar volume. The compression factor Z can be
defined as the ratio:
However, the measured pressure at 483 K was only 41 psi.
This difference is due to the reverse reaction 1. On the other
hand, the pressure actually measured at 483 K is too high for a
case where all the PF5 reacts with LiF to form LiPF6. At 483 K,
the partial pressure expected for the argon in the reaction vessel
should not be higher than that for 0.31 mmol of gas:
Vm
Vm0
Z =
,
(2a)
where Vm0 is the molar volume of a perfect gas.
nRT
V
0.00031 × 0.082 × 483
0.0082
Based on this equation, the measured pressure values can be
used for calculating Vm, and hence, n, if V is known. In general,
C/Vm2 ꢀ B/Vm, and hence the third term in the equation above,
can be neglected. The virial coefficients depend of course on the
temperature.
P =
=
= 1.49 atm = 22 psi.
(8)
We should note that the above evaluations can suffer the fol-
lowing errors:
In our case, the number of LiPF6 moles in the sample, and
hence the maximal moles of PF5 gas formed, are:
1. The titanium bomb expands upon heating and hence its vol-
ume is also temperature dependent.
2. The masses of the solid LiF and liquid LiPF6 also occupy
some volume. We do not have information on the specific
density of liquid LiPF6.
m
1.9
nLiPF
=
=
= 12.5 mmol,
(3)
6
M
152
where m = 1.9 g is the weight of the sample, and M = 152 the
molecular mass of LiPF6.

E. Zinigrad et al. / Thermochimica Acta 438 (2005) 184–191
187
In any event, taking into account the specific density of solid
LiPF6 and LiF (2.838 and 2.635 g cm−3, respectively) and the
thermal expansion of Ti, αl = 8.6 × 10−6 K−1, the errors in the
above calculations are estimated to be less then 1%.
Hence, the difference of 19 psi between what was measured
and the partial pressure of argon is due to the remaining PF5
that did not react with LiF to re-form LiPF6. This residual gas
corresponds in amount to the LiF that we found by XRD anal-
ysis showed the presence of the LiF [11] in the solid mixture
remaining after the heating and cooling cycle of LiPF6. This
finding correlates with the conclusions from the DSC measure-
ments (see next section) that show that the PF5 and LiF formed
by thermal decomposition of LiPF6 at elevated temperatures
in closed vessels (as is the case for the ARC and DSC mea-
surements reported herein), did not, when cooled, recombine in
entirety to form the original amount of LiPF6. This is presum-
ably because upon cooling, the first-formed LiPF6 is a liquid
that covers the remaining LiF crystallites. For these to react,
the PF5 must diffuse through the layer of LiPF6 whose thick-
ness increases as the reverse reaction proceeds. A time much
longer than that available during cooling in these measurements
is needed for all the PF5 to recombine with the coated LiF.
During heating in the ARC experiments, we measured the
equilibrium pressure at the end of the searching time, and at
each temperature range, we waited for a time sufficient to allow
the compounds in the reaction vessel to equilibrate (Fig. 2). We
also showed that in these experiments we could use the approxi-
mations of a perfect gas for the pressure values that we measure.
If LiPF6 and LiF are assumed to be pure phases in the tem-
perature ranges measured (i.e., there is no unreacted PF5 or
LiF in the molten LiPF6), then the pressure of the PF5 mea-
sured at different temperatures (total pressure measured minus
the partial pressure of argon) would reflect the equilibrium con-
stant of the decomposition reaction at each temperature (Kp).
PPF5 versus T is plotted in Fig. 3a and ln P(PF5) versus 1/T is
plotted in Fig. 3b (for ARC experiments with 1.9 g of LiPF6
heated up to 623 K). The relatively good linear correlation thus
obtained, allows the calculation of an approximate heat of reac-
tion at constant volume (ꢀUr) using a non-calorimetric method
by using the integral form of the van’t Hoff isochoric equa-
tion, ln K = −ꢀUr/RT − const. ꢀUr – nearly invariant in the
temperature range measured (503–578 K) – was found to be
64.9 0.3 kJ mol−1 e. This value is close to that reported by
Gavritchev et al. [6] as ꢀH (69 kJ mol−1) for the same reaction
(Eq. (1)) in the same temperature range, carried out under con-
stant pressure conditions. The difference, ꢀH − ꢀUr, is very
close to RT (n = 1), where T is the average temperature in the
relevant temperature range.
Fig. 3. The equilibrium pressure of PF5 that is formed by thermal decomposition
of LiPF6, [LiPF6 (l) ↔ LiF (s) + PF5 (g)], as a function of temperature related
to the ARC measurements in Fig. 2: (a) Peq. vs. T; (b) ln (PPF5) vs. 1/T.
sample, the LiPF6 melting temperature in the second cycle is
slightly below that in the first cycle.
The melting heat of LiPF6 measured in the temperature inter-
val 443–473 K is 2.02 0.18 kJ mol−1
.
In order to understand the nature of the endothermic process
measured by DSC at 468 K, we conducted experiments in which
LiPF6 was heated in closed quartz ampoules, whose temperature
was monitors. We observed visual changes in the morphology of
the heated samples during the heating process. At temperature
close to 467 K, we could clearly observe morphological changes
that looked like liquid layers that were formed and covered the
solid crystals of the salt. We could also observe gas formation
and some bubbling. Thereby, we concluded that the endother-
mic process at 467 K is melting of LiPF6, which is followed by
its decomposition from the liquid state, thus forming solid LiF
crystals (clearly visible in these experiments) and PF5, as heating
proceeds. In any complete heating-cooling cycle, the total heat
of crystallization was always less than the heat of melting. How-
ever, the ratio of the two is a function of the mass of the sample
(see Fig. 6a). Thus, as noted in the discussion of the ARC results,
the reformation of LiPF6, which decomposed to PF5 + LiF upon
heating, is never fully completed upon cooling. From the ratio
of the heat of crystallization to that of melting in the same cycle
one can calculate the fraction of the original LiPF6 which has
been reformed as well as the amount of residual LiF in the that
experiment. From the data in Fig. 6a, it may be seen that in these
3.2. DSC measurements
Figs. 4 and 5 present typical DSC heating/cooling curves for
several samples of LiPF6 of different weight, including curves
for second and third consecutive heating/cooling cycles (Fig. 5).
These curves show well-defined melting/crystallization peaks
at 467.4 0.4 K (endothermic) and 460.2 0.4 K (exothermic),
respectively (Tables 1 and 2). In multiple cycles on the same

188
E. Zinigrad et al. / Thermochimica Acta 438 (2005) 184–191
Fig. 4. DSC curves of LiPF6 samples at heating rate/cooling rate of 1 K min−1, weighing 1–1.074 mg, 2–2.676 mg, 3–3.645 mg, 4–5.862 mg, 5–13.875 mg,
6–26.700 mg: (a) heating; (b) cooling; (c) lower curve–11.577 mg, upper curve–11.078 mg at heating rate of 1 and 20, respectively.
experiments only 13–50% of the original LiPF6 was reformed,
depending on the size of the sample. Fig. 6b presents the data
in terms of the mass of residual LiF as a function of mass of the
original LiPF6 sample.
Upon the first heating at the lower rate, the endothermic
decomposition process is not separated from melting and the
shape of the curves and peak depends on the mass of the sample
(Figs. 4a and 5). Fig. 4c shows typical DSC curves at higher
heating rates.
Fig. 7 demonstrates that the greater the sample mass, the
higher the peak temperature of the endothermic decomposi-
tion. A rationalization of this finding will be presented below.
It should however be noted at this point that the sample mass
that Gavritchev et al. [6] used, corresponds to ∼20 mg in our
tests. It is clear from Fig. 6 that a decomposition peak of 20 mg
(132 ␮mol) LiPF6 in a sealed crucible, such as we used, is at
temperature higher than 513 K. Gavritchev et al. [6] worked with
sealed crucibles in the range 298–513 K. It is clear therefore why
they did not observe a decomposition peak, and detected only the
reversible melting/crystallization peak near 463 K, which they
misidentified.
Upon repeated cycling of the same sample at the heating
rate 1 K min−1 the melting and the decomposition peaks are
separatedandthe decompositionpeaksin the subsequent heating
cycles become progressively narrower and more symmetrical
(Fig. 5).
Distinguished separation between the melting and the decom-
position peaks upon the first heating was observed at high heat-
ing rates. Fig. 4c shows DSC curves of LiPF6 samples at heating
rates of 1 and 20 K min−1. The masses of the samples are very
close as indicated in Fig. 4c. As the rate increases from 1 up to
20 K min−1, the melting peak shifts to higher temperatures by
Fig. 5. The first, second and third DSC consecutive heating and cooling curves
of a LiPF6 sample, weighing 3.645 mg.

E. Zinigrad et al. / Thermochimica Acta 438 (2005) 184–191
189
Table 1
Parameters related to the LiPF6 melting and decomposition, appearing in the DSC curves at heating rate of 1 K min−1
No.
Weight
(mg)
Weight
(␮ mol)
Melting peak
onset (K)
Melting peak
temperature (K)
ꢀUr of melting
Decomposition peak
temperature (K)
ꢀUr of decomposition
(kJ mol−1
)
(kJ mol−1
)
1
2
3
4
5
6
7
1.074
1.453
2.676
2.386
3.645
3.696
5.862
7.07
9.56
467.46
467.23
467.01
466.21
466.84
466.51
467.06
467.92
468.07
467.36
466.57
467.48
466.90
467.78
2.26
2.15
2.32
2.12
2.35
1.97
1.62
535.70
520.00
562.50
564.30
570.40
574.30
557.42
50.54
54.88
61.75
46.05
65.07
57.39
54.91
17.61
15.70
23.98
24.32
38.57
Average
466.90 0.33
467.44 0.42
2.02 0.18
55.79 4.80
ca. 7 K, while the shift of decomposition peak is ca. 24 K. Com-
paring DSC curves obtained at different heating rates (Fig. 4a
and c) shows that the first endothermic peak does not relate to
the consequent endothermic decomposition. The melting heat of
LiPF6 measured in the temperature range 445–503 K at heating
rate 20 K min−1 is 1.8 kJ mol−1. This result correlates with the
average value measured at a heating rate of 1 K min−1 that was
estimated without deconvolution between the melting and the
decomposition peaks (Table 1).
The cooling curves show the exotherm peaks resulting from
thereformationofLiPF6 fromPF5 andLiF. Thesearemuchmore
symmetric in shape than the respective endotherm peaks upon
heating, both in the case of sample cycled once (Fig. 4b) and in
the case of repeated cycling (Fig. 5). As for the endotherms, so
alsofortheexotherms, theironsetandpeaktemperaturearefunc-
tion of the mass of the original LiPF6 samples (Figs. 4b and 8);
the larger the sample, the higher the temperatures. For each sam-
ple the quantity of heat absorbed upon decomposition of LiPF6
was significantly larger than that released during LiPF6 refor-
mation upon cooling, and the ratio Qreformation/Qdecomposition is
dependent on the mass of the sample. This is in keeping with the
conclusion presented above regards incomplete reformation of
LiPF6 upon cooling. In fact the extent of reformation calculated
for any sample from Qreformation/Qdecomposition corresponds to the
extent calculated from Qcrystallization/Qmelting. This is shown in
Fig. 6a.
unknown, as are the solubility of PF5 and LiF in liquid LiPF6.
The latter solubility is probably very small, in keeping with its
high melting point (1143 K). The endothermicity measured upon
heating above 473 K reflects not only the net heat of decompo-
sition of LiPF6 in the liquid phase (and the specific heats of the
components). It also includes the outcome of secondary thermal
processes such as the exothermic formation of a crystalline LiF
solidphase(possiblyviaanamorphousstage), andtheseparation
of PF5 bubbles into the gas phase. These may be accompanied
by supersaturation phenomena. It is therefore not surprising that
the first cycle DSC heating curves frequently show poor repro-
ducibility. A specific example is curve 1 in Fig. 4a, exemplifying
the finding for a very small sample (1.074 mg). Two poorly
resolved peaks appear. The existence of a second peak cannot
be attributed to the further decomposition of PF5 to PF3 + F2,
since the equilibrium constant for this reaction at this tempera-
ture is negligible (∼10−55) [10]. Rather it is suggested that in
the case of such a small sample, disconnected micro droplets are
formed upon LiPF6 melting, and these suffer different secondary
behavior.
In a sealed DSC crucible of given volume the larger the LiPF6
sample, the smaller the fraction which must decompose to reach
equilibrium pressure of PF5 and repress further net decompo-
sition at that temperature. To continue, and eventually attain
substantially complete decomposition as per Eq. (1), the tem-
perature must be progressively raised. This explains the shift of
the endotherm peak covering complete decomposition to higher
temperatures as the mass of the sample is increased (Fig. 4a).
Similarly, upon cooling the greater PF5 pressure leads to the
The rationalization of the above findings is of necessity qual-
itative since quantitative data for key phenomena, such as the
equilibrium constant for Eq. (1) as a function of temperature, are
Table 2
Parameters related to the LiPF6 reformation and crystallization, appearing in the DSC curves at heating/cooling rate of 1 K min−1
No.
Weight
(mg)
Weight
(␮ mol)
Reformation
peak onset (K)
Reformation peak
temperature (K)
ꢀUr of reformation
(kJ mol−1)
Crystalization
peak onset (K)
Crystalization peak
temperature (K)
ꢀUr of
crystallization (mJ)
1
2
3
4
5
6
7
1.074
1.453
2.676
2.386
3.645
3.696
5.862
7.07
9.56
496.00
503.26
525.96
522.47
542.79
541.95
573.59
479.98
487.00
513.89
510.52
532.81
532.90
557.42
68.55
49.62
69.41
63.91
66.91
59.03
45.18
460.73
461.06
461.49
460.40
461.45
460.45
460.76
459.70
460.14
460.89
459.61
461.04
460.09
460.41
2.14
6.36
17.61
15.70
23.98
24.32
38.57
14.14
16.00
25.04
22.46
30.00
Average
60.37 7.80
460.90 0.36
460.27 0.43

190
E. Zinigrad et al. / Thermochimica Acta 438 (2005) 184–191
Fig. 8. The dependence of the onset of the reverse reaction LiF + PF5 → LiPF6
on the samples’ mass in DSC measurements.
The secondary thermal phenomena are thus smoother, and the
observed exotherm peaks in later cycles are more symmetric
and reproducible (Fig. 5). Somewhat similar effects lead to the
relative symmetric shapes of the exothermic peaks upon cooling.
The incomplete reformation of LiPF6 upon cooling in the
DSC experiments is rationalized on the basis of the necessity
of PF5 to diffuse to the surface of the LiF crystallites through
a growing layer of LiPF6 liquid, as discussed in detail for the
ARC results. Support for this postulate was adduced by a DSC
experiment on a 3.90 mg (25.66 ␮mol) sample of LiPF6 in which
the cooling after heating to 623 K, was extended over 48 h at
473 K, permittingampletimeforthereversereaction. Inthiscase
it was found that LiPF6 was reformed to the extent of 95%, rather
than the ca. 47% found for the usual experimental conditions.
Only in the cases of samples 1–4 (Fig. 4a) do the DSC (heat-
ing) curves clearly show that complete decomposition of the
LiPF6 has taken place by 623 K. From the size of the sam-
ple in these cases the number of moles of PF5 produced is
available. Assuming the ideal gas law as on approximation,
the pressure of PF5 in the sealed crucible at the temperature of
onset of the reverse reaction (reformation of LiPF6) upon cool-
ing (Figs. 4b and 8) was calculated for each of the five samples.
The data are plotted in Fig. 9. Extrapolation of the curve to lower
pressure indicates that at a PF5 pressure of 0–1 atm the refor-
mation of LiPF6 should set in below 423 K, below the melting
point of LiPF6. Consequently it is now clear why Gavritchev
et al. [6] in their experiments involving removal of (most of)
gaseous products of decomposition of LiPF6, found upon cool-
ing only one exotherm of LiPF6 reformation (from residual PF5
in the crucible) at or below ∼421 K, and did not see a separate
exotherm for LiPF6 crystallization [6].
Fig. 6. (a) The extent of the reverse LiPF6 reformation as a function of the
samples’ mass in DSC experiments: (ꢀ) crystallization heat/melting heat; (ꢀ)
crystallization heat/heat of reformation reaction; (ꢁ) melting heat/heat of
decomposition reaction. (b) The LiF mass that remained after cooling as a func-
tion of the samples mass (calculated from the DSC measurements).
onset of the exotherm (LiPF6 reformation) at a higher tempera-
ture, the larger the sample of LiPF6, which has been decomposed
(Figs. 4b and 8).
Since the reformation of LiPF6 upon cooling is incomplete,
in repeated cycling of the same sample the LiF crystallites
remaining from the previous cycle act as seed for the crys-
tallization of LiF (and possibly as catalyst for release of PF5
gas) formed by LiPF6 decomposition in the subsequent cycle.
From the DSC data, one arrives at an absolute value of ca.
55.8 4.8 kJ mol−1 for the internal energy change (ꢀUr) for the
decomposition of LiPF6, and of ca. 60.4 7.8 kJ mol−1 for its
reformation reaction, in the liquid phase (Table 2). The mass
of LiPF6, which was formed by the reformation reaction upon
cooling (mref), is equal to: mref = Qcr/qm, where Qcr is the crystal-
lizationheatandqm thespecificmeltingheat. Thespecificheatof
the reformation reaction (qref) was estimated as: qref = Qref/mref
where Qref is the heat of the reformation reaction. ꢀUr of LiPF6
,
Fig. 7. The dependence of the endothermic decomposition peak’s position on
the sample mass in DSC experiments.

E. Zinigrad et al. / Thermochimica Acta 438 (2005) 184–191
191
LiF(s) and PF5 (g), under conditions of constant volume. The
onset of melting is fixed at 466.9 0.3 K, heat of melting is
2.0 0.2 kJ mol−1. Upon cooling crystallization occurs and on
exotherm peak onset is observed at 460.9 0.4 K.
Decomposition commences with melting, but in DSC mea-
surements, because of the limiting crucible volume and the
pressure of PF5 produced during decomposition, an increase
in the mass of the LiPF6 sample leads to a shift to higher tem-
perature of the peak of the endotherm of decomposition, and of
its completion. In the reverse direction, upon cooling, the larger
the sample of LiPF6 decomposed, the higher the temperature of
onset of reformation of LiPF6 (l) from LiF(s) and PF5(g). At
rates of cooling even as low as 0.5 ◦C min−1, this reformation
reaction is not completed, and the larger the original sample, the
more LiF remains after cooling to ambient temperature.
The heat of the endothermic LiPF6 (l) decomposition at
constant volume, calculated from the ARC pressure mea-
surements (by a non-calorimetric method using integral form
Fig. 9. The calculated PF5 pressure vs. the onset temperature of the reverse
reaction.
of the van’t Hoff isochoric equation), is 64.9 0.3 kJ mol−1
.
The change in the internal energy for the decomposition and
the reformation reactions measured in the DSC experiments
is 55.8 4.8 kJ mol−1/60.4 7.8 kJ mol−1, respectively (which
allowing for the inherent differences in the techniques and for
the approximations in the calculation, is in reasonable agreement
with the value derived from ARC). Both ARC and DSC exper-
iments indicate that ꢀUr for LiPF6 decomposition/formation is
approximately temperature invariant in the temperature range
measured (490–580 K).
References
[1] J. Barthel, J.J. Gores, Handbook of Battery Materials, Wiley-VCH,
Weinheim, NY, 1999, p. 457 (Part III, Chapter 7).
[2] D. Aurbach, in: W.A. Van Schalkwijk, B. Scrosati (Eds.), Advances in
Li-ion Batteries, Kluwer Academic, Plenum Publishers, NY, London,
Moscow, 2002, p. 7 (Chapter 1).
[3] J. Jian, J.R. Dahn, Electrochem. Solid State Lett. 6 (2004) A180.
[4] E.P. Poth, D.H. Goughty, J. Power Sources 128 (2004) 308.
[5] X. Zhang, P.N. Ross Jr., R. Kostecki, F. Kong, S. Sloop, J.B. Kerr, S.
Striebel, E.J. Cairns, F. McLarnon, J. Electrochem. Soc. 148 (5) (2001)
A463–A470.
[6] K.S. Gavritchev, G.A. Sharpataya, A.A. Smagin, E.N. Malyi, V.A.
Matyukha, J. Therm. Anal. Calorim. 73 (2003) 71–83.
[7] K. Tasaki, K. Kanda, S. Nakamura, M. Ue, J. Electrochem. Soc. 150
(12) (2003) A1628–A1636.
Fig. 10. The heat of LiPF6 decomposition/reformation vs. peak temperature for
7 samples: (ꢀ) heat of decomposition reaction; (ꢁ) heat of reformation reaction.
decomposition was calculated approximately as the difference
between the total heat related to the endothermic process and
the heat of melting.
Fig. 10 presents the values of ꢀUr of decomposition and of
reformation of various samples of LiPF6. It may be seen that
the value is essentially temperature independent over the range
480–575 K. A similar conclusion was reached above on the basis
of ARC measurements over the range (503–578 K).
[8] G. Gerardine, R. Botte, E. White, Z. Zhang, J. Power Sources 97–98
(2001) 570–575.
4. Conclusions
[9] A. Du Pasquier, F. Disma, T. Bowmer, A.S. Gozdz, G. Amatucci, J.-M.
Tarascon, J. Electrochem. Soc. 145 (2) (1998) 472–477.
[10] D.E. Grey (Ed.), Handbook of the American Institute of Physics,
McGraw-Hill, New York, 1972.
The results of ARC and DSC experiments and their com-
prehensive analysis demonstrate the reversibility of both the
melting of LiPF6 and of its decomposition from liquid state to
[11] jCPDS-ICDD, 82-0784.