J. Electrochem. Soc.-2017-Monroe-E3547-51
J. Electrochem. Soc.-2017-Monroe-E3547-51
J. Electrochem. Soc.-2017-Monroe-E3547-51
JES FOCUS ISSUE ON MATHEMATICAL MODELING OF ELECTROCHEMICAL SYSTEMS AT MULTIPLE SCALES IN HONOR OF JOHN NEWMAN
Does Oxygen Transport Affect the Cell Voltages of Metal/Air
Batteries?
Charles W. Monroez
Department of Engineering Science, University of Oxford, Oxford OX1 3PJ, United Kingdom
Simultaneous transport of an electrolyte and dissolved oxygen is analyzed with Newman’s concentrated-solution theory to assess
how nonuniform oxygen distributions might impact the voltages of metal/air batteries. For a solution comprising a neutral solvent,
a simple salt, and oxygen, the Onsager–Stefan–Maxwell transport equations are inverted, yielding flux-explicit laws for oxygen,
anion, and cation transport that distinguish the effects of individual diffusion and migration driving forces. Along with the ionic
conductivity, electrolyte diffusivity, oxygen diffusivity, and cation transference number, a migration coefficient and a cross-diffusion
coefficient are identified, which respectively account for the effects of electro-osmotic drag on oxygen and diffusional drag between
salt and oxygen. A derived current/voltage relation reveals how oxygen gradients can in principle affect the cell potential; significant
diffusion potential can arise from oxygen if it experiences electro-osmotic drag. Prior models are proven to follow from an assumption
that cross-diffusion and electro-osmosis are both negligible, or, equivalently, that oxygen/ion interactions are weak. Experiments
to quantify the novel transport properties are discussed, along with quantitative estimates of the cross diffusivity and migration
coefficient.
© The Author(s) 2017. Published by ECS. This is an open access article distributed under the terms of the Creative Commons
Attribution 4.0 License (CC BY, http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse of the work in any
medium, provided the original work is properly cited. [DOI: 10.1149/2.0521711jes] All rights reserved.
Manuscript submitted April 3, 2017; revised manuscript received June 14, 2017. Published July 5, 2017. This paper is part of the
JES Focus Issue on Mathematical Modeling of Electrochemical Systems at Multiple Scales in Honor of John Newman.
A standard liquid electrolyte for a metal/oxygen battery is made aprotic Li/O2 cells typically require a preconditioning interval, in
up of a neutral solvent, dissolved oxygen, and a salt with a single which the system is held at open circuit while being exposed to oxy-
type of anion and a single type of cation. Given that four electro- gen at fixed pressure until the cell voltage equilibrates.11–14 During
chemical species comprise these electrolytes, one expects on the basis preconditioning, which can last for hours,12–14 the open-circuit poten-
of the Onsager mass-transport theory1 that six independent transport tial can vary across tens or hundreds of millivolts. It is likely that some
coefficients are needed to describe all the possible pairwise interac- part of this variation arises from a diffusion potential: since salt flux
tions associated with diffusion and migration. A recent paper by the is negligible and the interfacial kinetics of oxygen absorption into the
author’s group2 applies the Onsager–Stefan–Maxwell formalism to liquid electrolyte is probably fast, relaxation of the oxygen distribu-
lithium/oxygen batteries, but assumes two of the six transport prop- tion within the cell could be the source of the slow voltage change.
erties are negligible. Most other state-of-the-art continuum models This hypothesis is supported by the observation that far shorter pre-
of metal/oxygen-battery electrolytes incorporate only four transport conditioning is needed when Li/O2 cells are thinner.11
coefficients.3–7 We set out here to address this issue by proposing By inverting the Onsager–Stefan–Maxwell laws, both a MacInnes
convenient flux laws that account for dissipative interactions between equation and flux laws in Nernst–Planck form (describing the excess
oxygen flow and ion flow. fluxes of oxygen, cations, and anions relative to solvent explicitly,
Macroscopic transport properties can yield information about mi- using terms that distinguish the impacts of individual diffusion and
croscopic interactions, and vice versa. By establishing relationships migration driving forces) can be derived. The flux-explicit laws nat-
between the set of macroscopic transport coefficients and the Stefan– urally include six macroscopic transport properties bijectively related
Maxwell diffusivities, one can determine what conditions must be met to the six Stefan–Maxwell coefficients.
to reduce a complete transport model to the four-coefficient form used Most existing models include four of the six properties that ther-
in most simulations. modynamic completeness requires: oxygen diffusivity, electrolyte dif-
Following Newman,8 flux-explicit transport equations can be de- fusivity, ionic conductivity, and cation transference number. But the
rived from Onsager–Stefan–Maxwell constitutive laws by a sequence flux-law inversion process also suggests natural definitions for a ‘mi-
of linear transformations. A general implementation of Newman’s gration coefficient’, which determines how the flow of ionic current
approach, which also shows how to identify and count independent can drive oxygen flux, and for a ‘cross-diffusion coefficient’, which
macroscopic transport properties for multicomponent electrochemi- quantifies how salt gradients can drive oxygen flux (and how oxygen
cal systems, was recently presented by Monroe and Delacourt.9 That gradients can drive salt flux). The MacInnes equation for the four-
inversion procedure yields a ‘MacInnes equation’ (extended form of species electrolyte further shows how oxygen migration may manifest
Ohm’s law8,10 ) expressing the current density i in terms of neutral- a diffusion potential associated with oxygen gradients.
species activity gradients, as well as the gradient of a single ionic After the theory is developed, it is applied to analyze the precon-
electrochemical potential μi that can be related to the potential of a ditioning of planar lithium/oxygen cells. An estimate of the migration
reference electrode. The portion of the electrode-potential drop aris- coefficient is provided, along with lower and upper bounds on the
ing from spatial salt-activity differences is commonly called ‘dif- cross diffusivity. The analysis suggests that significant overpotential
fusion potential’ (or ‘concentration overpotential’). Considering the can arise from non-uniform oxygen distributions in Li/O2 cells, and
complete set of proper mass-transport driving forces from irreversible that oxygen flux due to migration could be comparable to the migra-
thermodynamics suggests that diffusion potential might arise from tory fluxes of cations and anions.
differences in oxygen concentration, as well as from differences in
salt concentration.
Macroscopic Transport Properties
There is reason to believe that diffusion potentials from oxygen
exist in some metal/oxygen batteries. For instance, experiments with Monroe and Delacourt’s procedure to develop thermodynamically
complete transport laws in Nernst–Planck form9 begins with the ar-
rangement of species indexes in an ordered list. (In this context, ‘ther-
z
E-mail: [email protected] modynamically complete’ means that the flux laws incorporate all
Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
E3548 Journal of The Electrochemical Society, 164 (11) E3547-E3551 (2017)
Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
Journal of The Electrochemical Society, 164 (11) E3547-E3551 (2017) E3549
equation involving oxygen gradients, 5, 10, and 11 that → 0 and Xeo → 0. Thus the migration coeffi-
cient and cross-diffusivity vanish if and only if the Stefan–Maxwell
1 − t+0 κ
i = − κ ∇μ ++ e + (z + − z − ) κ ∇μ
∇μ o. [12] diffusional resistances associated with microscopic ion/oxygen inter-
F z+ F z + ν+ F z+ z− actions vanish. Prior models therefore result from neglecting drag
To arrive at this result the isothermal, isobaric, electroneutral Gibbs- forces that ions could exert on oxygen.
Duhem equation In all known air-battery electrolytes the maximum oxygen concen-
tration is extremely small in comparison to the concentrations of both
0 + co ∇μ
c0 ∇μ o + ce ∇μ
e=0 [13] salt and solvent,18 which is a limit worth exploring. In this case one
has been used, in which μe = ν+ μ+ + ν− μ− . The inversion process is not justified in neglecting the cross-diffusion terms in Equations 15
further yields equations in Nernst–Planck form that describe the excess and 16; although do ∼ ∇c o , the fact that co ce does not require
fluxes of oxygen, anions, and cations:
that ∇co ∇ce . (Indeed, at the beginning of oxygen exposure
under open-circuit conditions, these two gradients are of the same
o − co v0 = Xeo co de + Do do − (z + − z − ) i,
N [14] order of magnitude.) Since ∝ co , the molar migration coefficient
cT F z+ z− ,
˜ defined as
− − c− v0 = De ν− de + Xeo c− do +
N
1 − t+0
i, [15] ˜ = cT ,
[21]
cT F z− co
remains finite as co → 0 if the Stefan–Maxwell diffusivities are all
+ − c+ v0 = De ν+ de + Xeo c+ do + + i.
t0
N [16] finite. Incorporation of ˜ into the oxygen flux law (Equation 14)
cT F z+
shows that the de , do , and i terms are all of O (co ); no term can be
Here the current density i, related to the fluxes through Faraday’s neglected on the grounds that co is small.
law i = F i z i N i , is taken to drive migration; the solvent velocity It would be thermodynamically inconsistent to incorporate a con-
v0 = N 0 /c0 drives convection. Two thermodynamic forces driving centration overpotential associated with oxygen gradients (such as the
o term in Equation 12) without placing a corresponding migration
∇μ
diffusion of the neutral-solute components, de and do , appear:
term in the flux law for oxygen (such as the i term in Equation 14).
e
cT ce ∇μ o
cT co ∇μ Significant modification of all the transport laws may be needed to
de = − and do = − . [17]
c0 νe RT c0 RT account for effects of the oxygen distribution on cell potential.
Assuming a reference electrode reversible only to cations is used
Note that di has been expressed with units of molar-concentration to measure the voltage drop across the cell interior (so that ∇μ + =
gradient here for convenience.
F z + ∇), equation 12 yields a law that governs the change in electrode
Under isothermal, isobaric conditions, the Gibbs phase rule re-
potential ,
quires that the component chemical potentials depend at most on ce
and co . Thus the thermodynamic driving forces can be expressed as
linear combinations of ∇c o , through thermodynamic factors
e and ∇c = i + νe RT c0
− ∇ ˜ ce do .
1 − t+0 de − [22]
κ F z + ν+ cT ce cT
χi j (ce , co ),
The term with do scales as −∇c
o . Thus large oxygen gradients can
de = −χee ∇c
e − χeo ∇c
o,
induce an electric field—even if co is very low.
do = −χoe ∇c
e − χoo ∇c
o. [18]
16,17 Oxygen Diffusion Potential During Lithium/Oxygen
Assuming standard constitutive laws
Preconditioning
μe (T, p, ce , co ) = μθe (T, p) + νe RT ln ( f e ce ) ,
Equation 22 rationalizes the voltage response observed during a
μo (T, p, ce , co ) = μθo (T, p) + RT ln ( f o co ) , [19] standard preconditioning experiment — formally, an open-circuit hold
where μiθ is the chemical potential in a primary reference state, f e of a planar air-battery cell initially equilibrated under argon, whose
is the mean molar activity coefficient of the electrolyte, and f o is the ambient gas undergoes a sudden change from argon to fixed-pressure
activity coefficient of oxygen on a molar basis, one can write generally oxygen at time t = 0. Immediately after the cell is exposed to oxygen
that (time t = 0+ ), the oxygen concentration in the liquid at the gas
pos
interface, co , will rapidly reach the saturated liquid-phase oxygen
cT ∂ ln f i concentration cosat if the kinetics of oxygen absorption into the liquid
χi j = δi j + ci , [20]
c0 ∂c j T, p,ck= j is fast in comparison to oxygen diffusion rates. During this very early
time, the cell potential will rise quite rapidly from the equilibrium
in which δi j is the Kronecker delta. For an ideal neutral component i, f i potential achieved at open circuit under argon to a value somewhat
is independent of composition; in the limit of infinite dilution, cT ≈ c0 . above or below the equilibrium potential under oxygen.
Thus the force driving mass transport of neutral-solute component i At sufficiently short times after the kinetic relaxation associated
in an ideal, dilute air-battery electrolyte is di ≈ −∇c
i. with oxygen sorption, diffusing oxygen will have reached its equilib-
rium concentration within the liquid at the gas interface, but will not
have penetrated through the separator, so the oxygen concentration at
Consequences of Ion/Oxygen Interactions neg
the anode/separator interface, co , will not equal cosat . Whenever such a
Existing models of air-battery electrolytes tend to use the equa- difference exists within the cell, a concentration overpotential arising
tions from Nernst–Planck theory or the theory of concentrated binary from oxygen may affect the cell voltage. If present, this overpotential
electrolytes to describe the ion fluxes, and additionally adopt Fick’s would be expected to relax on a diffusional timescale, which — un-
law to describe the oxygen flux. This theoretical structure is (uniquely) like the kinetic relaxation — would depend on the cell thickness. The
achieved by taking and Xeo to vanish in Equations 12 and 14–16, discussion that follows analyzes the diffusional relaxations of oxygen
leaving a theory with four transport coefficients. and salt during preconditioning, with the aim of determining whether
Given a system in which the species concentrations are all finite, oxygen migration is significant in lithium/oxygen cells.
the conditions Xeo → 0 and → 0 imply through Equations A3 and Liu et al.2 listed a set of thermodynamic parameters that de-
A4 (in the Appendix) that 1/Do+ → 0 and 1/Do− → 0; conversely, scribe 1 M LiTFSI in dimethoxyethane — a standard electrolyte
the conditions 1/Do+ → 0 and 1/Do− → 0 imply through equations for lithium/oxygen batteries. At room temperature the saturated
Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
E3550 Journal of The Electrochemical Society, 164 (11) E3547-E3551 (2017)
oxygen concentration in this electrolyte is cosat = 0.0021 M. The 1 M LiTFSI in dimethoxyethane therefore simplify to
partial molar volumes of solvent and salt are V 0 = 0.1043 M−1 and c
V e = 0.0210 M−1 , respectively. With these parameters one finds in o = − T Do ∇c
N o [27]
the absence of oxygen that c0 = 9.4 M and cT = 11.4 M, and that the c0
volume fraction occupied by solvent is c0 V 0 = 0.98. An assumption ce Xeo
N+ = −ν+ cT V 0 De χee ∇ce +
∇co . [28]
that saturation with oxygen negligibly affects liquid volume implies cT
that V o → 0. These values confirm that c0 and ce are both of the order (Angled brackets around a quantity indicate the value in an oxygen-
of cT , whereas co cT . free equilibrated state.) Oxygen therefore does obey a form of Fick’s
The assumption that co cT simplifies the transport properties law in the special case at hand, but only because the open-circuit
substantially. As one might expect, Equations 6, 8, and 9 show in the condition removes any contribution that migration might make to
limit that the electrolyte diffusivity, cation transference number, and the oxygen flux. Despite the null applied current, oxygen can still
ionic conductivity match their values in the absence of oxygen, contribute to the diffusion potential; the MacInnes equation during
preconditioning becomes
lim De = De0 , lim t+0 = t+ø
0
, and lim κ = κø . [23]
co cT co cT co cT
F z + ν+ cT χee 1 − t+0 c0
∇ = e−
∇c o,
˜ ∇c [29]
Interestingly, the oxygen diffusivity in this limit differs substantially νe RT ce
from Do0 :
in which the second term on the right may be substantially larger in
1 1 νe ce magnitude than the first.
lim = + . [24] Insertion of Equation 27 into Equation 25 produces the familiar
co cT Do Do0 c0 Deo
diffusion equation
Thus one straightforward way to evaluate the strength of the dif- ∂co cT Do ∂ 2 co
fusional resistance that electrolyte and oxygen exert on each other, = , [30]
1/Deo , would be to measure how much the apparent oxygen diffusiv- ∂t c0 ∂ x 2
ity reduces upon the addition of salt to the pure solvent. whereas inserting Equation 28 into Equation 25 yields a more unusual
Liu et al. also presented Stefan–Maxwell diffusivities for 1 M balance for the dissolved electrolyte,
LiTFSI in dimethoxyethane that yield De = 5.65 × 10−10 m2 s−1 ,
t+0 = 0.43, and κ = 36.9 mScm−1 ;2 Lu et al. report for the same 1 ∂ce ∂ 2 ce ce Xeo ∂ 2 co
= De χee 2 + . [31]
electrolyte that Do = 7.3 × 10−10 m2 s−1 .19 The values of ˜ and Xeo cT V 0 ∂t ∂x cT ∂ x 2
depend primarily on oxygen/salt interactions and remain unknown.
Given the low saturated concentration of oxygen, it is reasonable Here x denotes the position normal to the negative electrode (anode),
to treat it as a thermodynamically ideal solute, so that f o = 1, imply- assumed to be a slab of lithium metal.
ing that χoe = 0 and χoo = cT /c0 . For simplicity we assume χeo = 0 A standard preconditioning experiment begins with initial
as well (i.e., salt activity is unaffected by oxygen content), which al- conditions
lows χee to be estimated from properties of oxygen-free electrolytes. co (0, x) = 0 and ce (0, x) = ce . [32]
Generally χee is of order unity; typical values are c0 V 0 χee = 0.68
for 2.24 M LiPF6 in a mixed-carbonate solvent and c0 V 0 χee = 3.1 The boundary conditions during the open-circuit hold after oxygen
for 0.85 M LiPF6 in propylene carbonate.20 For the purposes of the exposure are
present analysis we take χee = 5 (independent of composition), an co (t, L) = cosat , No (t, 0) = 0, and N+ (t, 0) = 0, [33]
atypically large value that should amplify the concentration overpo-
tential contributed by salt gradients. where x = L corresponds to the position of the liquid/gas interface.
Finally, it is reasonable to assume that the variations in c0 and ce Since lithium ions are neither produced nor consumed within the
are of similar order as the variations of co during a preconditioning resting cell, one can also write an auxiliary condition
experiment. Thus oxygen transport can be thought of as inducing a
first-order regular perturbation to the system that governs transport 1 L
ce (t, x) d x = ce , [34]
in oxygen’s absence, while the transport properties, thermodynamic L 0
factors, and partial molar volumes can be assumed constant and equal which holds throughout the preconditioning period.
to their values in the limit co → 0. Equations 30 through 34 comprise a well-posed description of
The transient cell response is governed by four species material mass transport during preconditioning. The transient oxygen and elec-
balances of the form trolyte distributions yielded by solving this governing system can then
∂ci be inserted into Equation 29 to evaluate the voltage relaxation that ac-
= −∇ i.
·N [25] companies oxygen absorption into the liquid electrolyte.
∂t
Solving the problem is more straightforward than it appears be-
Under the justifiable assumption that the electrolyte maintains local cause the transient oxygen balance is decoupled from the salt balance.
electroneutrality, two of these balances can be replaced with equations The difference in oxygen concentration across the cell is found to vary
that respectively express the local continuity of charge and volume,21 with time after exposure to pure oxygen gas as
Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
Journal of The Electrochemical Society, 164 (11) E3547-E3551 (2017) E3551
× −
0
ν−
π2 De Do − νe c2e co Xeo2
Do (2n − 1)2 − 4V 0 c0 De χee (2m − 1)2 νe c
m=1 n=1 D0− = T , [A2]
2 Do − cco Xeo t+0 + De − νce ce Xeo
⎫ T T
exp − π (2m−1)4Lc2T De χee t
2
∞
∞ ⎬
ν+ co
× [36] De Do − νe c2e co Xeo2
νe c0
2⎭
Do (2n − 1) − 4V 0 c0 De χee (2m − 1)
2 cT
m=1 n=1 Do+ = , [A3]
cT Xeo 1 − t+ + De
co 0
for the time evolution of the difference in salt concentration across the
ν− co
separator. νe c0 De Do − νe c2e co Xeo2
cT
The prefactor on the right of Equation 36 involves the single un- Do− = , [A4]
c Xeo t+ − De
co 0
and 24, this inequality places the upper bound on Xeo . The lower = − +
D+− RT κ ν+ ν− ce Do0
bound arises from inspection of Equation 24, which shows that Deo
1 c0 co
leads to increased diffusional drag on oxygen by the electrolyte. Salt − + . [A6]
ce ν− D0+ + ν+ D0− ν− Do+ + ν+ Do−
exerts the least additional drag in the limit Deo → ∞, a condition that
equation 11 shows to correspond with Xeo → 0.
Integration of Equation 29 demonstrates that the potential drop References
across the separator during preconditioning follows
1. L. Onsager, Annals of the New York Academy of Sciences, 46, 241 (1945).
F z + ν+ c0 χee 1 − t+0 ˜ 2. J. Liu, S. Khaleghi Rahimian, and C. W. Monroe, Physical Chemistry Chemical
= ce − co . [38] Physics, 18, 22840 (2016).
νe RT cT ce cT 3. P. Albertus, G. Girishkumar, B. McCloskey, R. Sánchez-Carrera, B. Kozinsky,
J. Christensen, and A. Luntz, Journal of the Electrochemical Society, 158, A343
The greatest potential drop occurs at t = 0. Experiments using lithium (2011).
perchlorate in diglyme — an ethereal electrolyte similar to LiTFSI in 4. P. Andrei, J. Zheng, M. Hendrickson, and E. Plichta, Journal of the Electrochemical
dimethoxyethane — suggest that (0) is of the order of −100 mV.22 Society, 159, A770 (2012).
Inserting all of the property values listed in this section, and assuming 5. B. Horstmann, T. Danner, and W. Bessler, Energy & Environmental Science, 6, 1299
(2013).
the maximum possible value for Xeo , one finds that the ce term in 6. U. Sahapatsombut, H. Cheng, and K. Scott, Journal of Power Sources, 227, 243
Equation 38 is more than three orders of magnitude smaller than the (2013).
term. The potential drop can be explained by the co term if 7. K. Yoo, S. Banerjee, and P. Dutta, Journal of Power Sources, 258, 340 (2014).
˜ ≈ 1.1 × 104 ; the corresponding migration coefficient is = 1.9. 8. J. Newman and K. E. Thomas-Alyea, Electrochemical Systems, 3rd Edition John
Wiley & Sons, Inc.: Hoboken, New Jersey, 2004.
On the basis of this estimate, we hypothesize that the electro-osmotic 9. C. W. Monroe and C. Delacourt, Electrochimica Acta, 114, 649 (2013).
effect could be significant in lithium/oxygen cells, and encourage its 10. D. A. MacInnes, The Principles of Electrochemistry, 2nd Ed. Dover Publications,
consideration in air-battery modeling and characterization. It is worth Inc.: New York, New York, 1961.
noting that the prefactor of the current density in Equation 14 is of 11. Y.-C. Lu, H. Gasteiger, M. Parent, V. Chiloyan, and Y. Shao-Horn, Electrochemical
and Solid-State Letters, 13, A69 (2010).
the same order as those in Equations 15 and 16. Thus the fluxes of 12. B. McCloskey, D. Bethune, R. Shelby, G. Girishkumar, and A. Luntz, The Journal
oxygen, cations, and anions due to migration all appear to have similar of Physical Chemistry Letters, 2, 1161 (2011).
magnitudes. 13. B. Adams, C. Radtke, R. Black, M. Trudeau, K. Zaghib, and L. Nazar, Energy &
Liquid electrolytes with minimal ion/oxygen drag interactions will Environmental Science, 6, 1772 (2013).
14. H. Geaney, J. O’Connell, J. Holmes, and C. O’Dwyer, Journal of the Electrochemical
exhibit apparently lower overpotentials when operated at higher rates, Society, 161, A1964 (2014).
a range where oxygen concentration polarization across the cell is 15. T. F. Fuller and J. Newman, Journal of the Electrochemical Society, 139, 1332 (1992).
expected to be significant. Targeting electrolytes with low migration 16. E. A. Guggenheim, Thermodynamics: an Advanced Treatment for Chemists and
coefficients could be a route by which to improve the energy efficien- Physicists, 5th Ed. North-Holland Publishing Company: Amsterdam, 1967.
17. R. A. Robinson and R. H. Stokes, Electrolyte Solutions, 2nd Ed. (Revised) Butter-
cies of air-battery systems. worths: London, U.K., 1968.
18. J. Read, Journal of the Electrochemical Society, 153, A96 (2006).
19. Y.-C. Lu, D. G. Kwabi, K. P. C. Yao, J. R. Harding, J. Zhou, L. Zuin, and
Y. Shao-Horn, Energy and Environmental Science, 4, 2999 (2011).
Acknowledgment 20. J. Liu and C. W. Monroe, Electrochimica Acta, 167, 357 (2015).
21. J. Liu and C. W. Monroe, Electrochimica Acta, 135, 447 (2014).
This work was supported by the US National Science Foundation, 22. Personal communication of data gathered by Xiangwen Gao, 22 January 2016. From
grant no. CBET 1336387. Lee Johnson, Peter Bruce Group, Department of Materials, University of Oxford.
Downloaded on 2017-11-09 to IP 190.111.30.254 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).