Communication

pubs.acs.org/JACS

Physical Organic Approach to Persistent, Cyclable, Low-Potential

Electrolytes for Flow Battery Applications
Christo S. Sevov,†,‡ David P. Hickey,†,§ Monique E. Cook,†,‡ Sophia G. Robinson,†,§ Shoshanna Barnett,†,‡
Shelley D. Minteer,*,†,§ Matthew S. Sigman,*,†,§ and Melanie S. Sanford*,†,‡
†
Joint Center for Energy Storage Research, 9700 S. Cass Avenue, Argonne, Illinois 60439, United States
‡
Department of Chemistry, University of Michigan, 930 N. University Avenue, Ann Arbor, Michigan 48109, United States
§
Department of Chemistry, University of Utah, 315 South 1400 East, Salt Lake City, Utah 84112, United States
*
S Supporting Information

ABSTRACT: The deployment of nonaqueous redox flow

batteries for grid-scale energy storage has been impeded by
a lack of electrolytes that undergo redox events at as low
(anolyte) or high (catholyte) potentials as possible while
exhibiting the stability and cycling lifetimes necessary for a
battery device. Herein, we report a new approach to
electrolyte design that uses physical organic tools for the
predictive targeting of electrolytes that possess this
combination of properties. We apply this approach to
the identification of a new pyridinium-based anolyte that
undergoes 1e− electrochemical charge−discharge cycling at
low potential (−1.21 V vs Fc/Fc+) to a 95% state-of-
charge without detectable capacity loss after 200 cycles. Figure 1. CV of 1+ (0.1 M TBAPF6 in CH3CN at 100 mV/s).

persistent organic electrolytes that undergo stable cycling in

R edox-flow batteries (RFBs) offer a potential solution to
the unmet challenge of the large-scale integration of
intermittent renewable energy sources into the electrical grid.1
acetonitrile at low potentials.
We selected anolyte candidate 1+ as a test case for our design
strategy. A recent report demonstrated that this acylpyridinium
In an RFB, solvated anolyte and catholyte molecules are derivative meets a number of techno-economic metrics for
pumped over inert electrodes to charge/discharge the battery nonaqueous RFB applications, including high solubility (1.5 M
and are then stored in separate reservoirs. This design principle in acetonitrile), low molecular weight per electron transferred
means that RFBs can be inexpensively scaled by simply adding (142 g/mol of e−), and multiple redox events at relatively low
more of the electroactive compounds to the external potentials [−1.08 and −1.78 V vs Fc/Fc+ (Fc = ferrocene),
reservoirs.2 Techno-economic modeling has implicated redox- Figure 1].6 However, the application of this molecule in
active organic molecules (ROMs) as particularly attractive nonaqueous RFBs remains fundamentally limited by the short
electrolytes for RFBs due to their relatively low cost and lifetime of the radical 1,7 which is formed upon the 1e−
molecular weights, as well as the feasibility of achieving high reduction of 1+. For example, 0.29 M solutions prepared
solubility and significant electrochemical potential differences from isolated samples of 1 decay by ∼7% over 24 h at room
between the anolyte and catholyte.3 However, even if all of temperature; furthermore, ∼20% capacity fade is observed after
just 75 charge−discharge cycles between 1+ and 1. We
these techno-economic metrics are met, ROM-based electro-
hypothesized that modeling tools based on physical organic
lytes will not be viable unless they are long-lived in both the parameters could be used to identify second generation
charged and discharged states and undergo stable electro- anolytes with enhanced persistence while maintaining the low
chemical cycling. redox potential associated with the parent molecule. Notably,
Many ROMs, particularly those that store considerable although such tools have been successfully applied in the fields
amounts of energy by undergoing redox reactions at low of drug development,8 catalyst design,9 and mechanistic
(anolytes) or high (catholytes) potentials, have poor shelf-lives study,10 they have not yet been utilized for the development
in the charged state and undergo fast decomposition during of energy storage materials.
charge−discharge cycling.4 To date, no general strategies have We implemented a five-step workflow to develop a
been developed for improving/optimizing the lifetime (herein quantitative correlation between the structures of 4-acylpyr-
referred to as persistence) and cyclability of ROMs without
concomitantly decreasing their redox potential window.4f,5 Received: January 5, 2017
Herein, we introduce a physical organic approach to designing

© XXXX American Chemical Society A DOI: 10.1021/jacs.7b00147

J. Am. Chem. Soc. XXXX, XXX, XXX−XXX
Journal of the American Chemical Society Communication

idinium derivatives and their redox potentials and persisten- Conversely, the less electron-rich N-aryl derivatives 6 and 7
ce.9a,11 First, a training set of 12 acylpyridine radicals was exhibit enhanced lifetimes but more positive reduction
synthesized, and the rate of decomposition was measured for potentials. These data are all consistent with a ground state
each. Second, a set of physical-organic parameters was electronic stabilization of the radical (which is reported to be
identified to relate the experimental decomposition rates to localized at the 4-position of the pyridine)13 by electron
the electronic and steric properties of each molecule in the withdrawing substituents. Furthermore, they reinforce the
training set. Third, a mathematical relationship was established general paradigm that persistence and potential are inversely
to identify the key structural and electronic features responsible correlated.14 Although distal to the primary site of radical
for persistence. Fourth, virtual extrapolation of the mathemat- density, steric hindrance at nitrogen itself has been previously
ical model was used to predict new anolytes with improved demonstrated to improve the persistence of related pyridine
properties.10a Finally, the predicted molecules were synthesized radicals.15 Indeed, upon systematically increasing the size of the
and validated, ultimately leading to new anolytes with nitrogen substituent (1, 8, 9, 10), we observe significant
dramatically enhanced persistence as well as lower redox enhancements in radical persistence. For example, the tert-
potentials. butyl-substituted benzoylpyridine 10 exhibits a 100-fold slower
Radicals 1−12 were selected for the initial training set, rate of decomposition relative to the N-methyl analogue 1. In
because they represent a range of steric and electronic variation this case, the dramatic improvement in radical lifetime does not
at different sites on the acylpyridinium core (Figure 2). The occur at the expense of potential, providing promising evidence
decomposition of each was monitored at 70 °C in CD3CN via that persistence and redox potential can be decoupled through
NMR spectroscopy. As an example, a 0.30 M solution of 1 the judicious modification of appropriate substituents.
degraded through three half-lives in a 12 h period with a second We next focused on generating a predictive model for the
order rate constant of 6.6 × 10−4 M−1 s−1 (Figure 2, top).12 A virtual evaluation of new pyridinium anolytes. Using the data
number of trends are apparent from the training set. For above, we calculated thermodynamic properties and identified
example, radicals bearing electron-donating substituents (e.g., computationally derived structural descriptors for the training
2) generally have more negative potentials but also faster rates set anolytes (Figure 3a). As substituent size appeared to be
of decomposition than electron deficient analogues (e.g., 4). influential, initial correlations were attempted using a variety of
traditional steric parameters (i.e., Charton and Sterimol).9b
Poor trends were observed in all cases (R2 < 0.60), prompting
the examination of alternative descriptors. This ultimately led
to the identification of a parameter describing the substituent
height out of the pyridine ring plane (Hst). This effectively
accounts for the combined steric effects of the substituents at
nitrogen, at C2, and at C6. Using a linear regression algorithm

Figure 2. Decomposition studies of 1−16 in CD3CN at 70 °C. Plot of Figure 3. (a) Illustration of the electronic and steric descriptors. (b)
[1] (blue, left axis) and 1/[1] (red, right axis) vs time (top). Redox Plot of predicted vs experimentally determined free energies for
potentials and relative rates of decomposition of 1−16 (bottom). decomposition of training (●) set and validation (○) set.

B DOI: 10.1021/jacs.7b00147
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX
Journal of the American Chemical Society Communication

to explore a range of parameters, a statistical correlation (R2 = are predicted to have enhanced redox potentials and
0.92) was found when Hst was used in combination with persistence, (ii) have substituents that add minimal molecular
computed electrochemical reduction potentials (Figure 3b). weight, and (iii) are synthetically accessible.
This simple relationship illustrates that reduction potential and Among the modeled anolyte candidates, 17 was selected for
radical persistence can indeed be decoupled via the experimental testing, as it was predicted to exhibit high
incorporation of appropriate substituents at nitrogen, at C2, persistence as well as a potential lower than that of any
and/or at C6. derivative examined to date (predicted to be −1.23 V versus
With a tool for quantitative prediction in hand, we next Fc/Fc+, Figure 3a).16 Compound 17 also incorporates the first
sought to test the extrapolative limits of the model by examples of substituents at the C2 and C6 positions of the
identifying a radical with exceptional persistence (defined for pyridine ring, thereby providing a key test of the extrapolative
our purposes as t1/2 ≥ 1 month at 70 °C). We conducted a capabilities of the model. As predicted, 17 exhibits the lowest
directed virtual screen of 18 compounds, focusing on experimental potential (−1.21 V vs Fc/Fc+) and one of the
maximizing ΔG‡dec according to the physical descriptors in longest radical lifetimes (krel = 0.56). The model’s ability to
the equation. This screen suggested that N-xylyl-substituted 16 predict accurately the decomposition barrier of 17 (ΔG‡pred =
should exhibit 3 orders of magnitude greater persistence than 28.4 kcal/mol, ΔG‡exp = 28.2 kcal/mol) represents an
the parent radical 1 and should thus persist for months at 70 electronic extrapolation of nearly 100 mV beyond the training
°C. Experimentally, we found that 16 exhibits such high set, while simultaneously demonstrating its ability to account
persistence that it exceeds the capabilities of our standard for a new substituent pattern.
protocol to experimentally measure a rate of decomposition, Most important for the identification of desirable RFB
necessitating that measurements be performed over 1 week at electrolytes is that the model accounts for nonintuitive effects
70 °C. This example highlights an extrapolation in rate constant of chemical structure on persistence. For example, the N-
of over an order of magnitude from the limit of the original methyl analog 18 was predicted and experimentally validated to
training set, corresponding to 2.3 kcal/mol greater ΔG‡dec. To exhibit an approximately 20-fold lower persistence than the N-
understand better the origin of the observed persistence, we ethyl anolyte 17. This is in contrast to the N-methyl and N-
obtained a single crystal X-ray structure of 16. As illustrated in ethyl derivatives 1 and 8, respectively, which display nearly
Figure 4a, the solid state structure of 16 shows that the identical lifetimes. Because the reduction potentials for 17 and
perpendicular xylyl substituent protects the C2/C6 positions of 18 are essentially the same, the ∼2 kcal/mol enhanced stability
the pyridine ring, which presumably slows the homocoupling of of 17 arises from a difference in kinetic stability, reflected by the
two pyridine radicals.15a parameter Hst (Et = 9.06 Å vs Me = 5.16 Å). This appears to
Consistent with the existing paradigm, the high persistence of result from gearing between the Et group and the 2,6
anolyte 16 mainly results from a ground-state stabilization, substituents, which forces this Et group above the plane of
which leads to an undesirably high reduction potential of −0.94
the pyridine ring.
V vs Fc/Fc+. However, the utility of our parameter-based
Finally, we sought to establish whether the high persistence
modeling tool is that we can now rapidly identify compounds
of 17 translates to stable electrochemical cycling, a necessity for
with high charged-state persistence without sacrificing low
applications in RFBs. Thus, we subjected 17+ to 1e− charge−
potential. As depicted in Figure 4, plotting the lifetimes of
discharge cycling in a symmetric electrochemical cell and
charged anolytes 1−16 as a function of their potential reveals a
compared the results to those of 1+, the progenitor of this
quadrant diagram with corners that represent combinations of
high/low potential and high/low persistence. The most study. Importantly, this experimental design was selected in
desirable anolytes reside in the upper left quadrant. Targeting order to isolate the charge−discharge reactions to those of the
this region, we virtually evaluated a variety of molecules that (i) anolyte. This allows an evaluation of the anolyte’s electro-
chemical cycling stability without complications that can arise
in a complete redox flow battery (e.g., catholyte and membrane
compatibility as well as crossover).17 Galvanostatic cycling was
performed on a 10 mM solution of 17+ with reticulated vitreous
carbon electrodes at 2.5C (Figure 5a). Upon the 1e− reduction
of 17+ to 17, an average state-of-charge of 95% was reached,
and the anolyte discharged with 98% Coulombic efficiency
(Figure 5b). No loss in storage capacity was observed for 17+
after 200 cycles, and the cycling potentials are consistent with
the redox potential measured by CV. These results are in
marked contrast to cycling data for 1+, which shows 35%
capacity loss over the same time. The excellent cyclability of
17+ at even lower potentials than the parent molecule 1+
highlights the utility of this approach for ROM development/
optimization.
In summary, we demonstrate that physical organic modeling
of an organic electrolyte can deliver new ROMs that possess a
previously elusive combination of properties for nonaqueous
RFB applications. Importantly, although this approach was
successfully demonstrated for the development of 17+, it is not
Figure 4. (a) X-ray structures of 16+ and 16. (b) Quadrant diagram of specific to this class of electrolyte. As such, we anticipate that it
radical half-life as a function of redox potential. should prove more broadly applicable to the design and
C DOI: 10.1021/jacs.7b00147
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX
Journal of the American Chemical Society Communication

Electrochim. Acta 2009, 54, 5605. (d) Bockman, T. M.; Kochi, J. K. J.

Org. Chem. 1990, 55, 4127. (e) Wei, X.; Cosimbescu, L.; Xu, W.; Hu, J.
Z.; Vijayakumar, M.; Feng, J.; Hu, M. Y.; Deng, X.; Xiao, J.; Liu, J.;
Sprenkle, V.; Wang, W. Adv. Energy Mater. 2015, 5, 1400678.
(f) Assary, R. S.; Zhang, L.; Huang, J.; Curtiss, L. A. J. Phys. Chem. C
2016, 120, 14531. (g) Wei, X.; Xu, W.; Huang, J.; Zhang, L.; Walter,
E.; Lawrence, C.; Vijayakumar, M.; Henderson, W. A.; Liu, T.;
Cosimbescu, L.; Li, B.; Sprenkle, V.; Wang, W. Angew. Chem., Int. Ed.
2015, 54, 8684. (h) Brushett, F. R.; Vaughey, J. T.; Jansen, A. N. Adv.
Energy Mater. 2012, 2, 1390. (i) Duan, W.; Vemuri, R. S.; Milshtein, J.
D.; Laramie, S.; Dmello, R. D.; Huang, J.; Zhang, L.; Hu, D.;
Vijayakumar, M.; Wang, W.; Liu, J.; Darling, R. M.; Thompson, L.;
Smith, K.; Moore, J. S.; Brushett, F. R.; Wei, X. J. Mater. Chem. A 2016,
4, 5448. (j) Lin, K.; Chen, Q.; Gerhardt, M. R.; Tong, L.; Kim, S. B.;
Eisenach, L.; Valle, A. W.; Hardee, D.; Gordon, R. G.; Aziz, M. J.;
Marshak, M. P. Science 2015, 349, 1529.
(5) (a) Gerhardt, M. R.; Tong, L.; Gómez-Bombarelli, R.; Chen, Q.;
Marshak, M. P.; Galvin, C. J.; Aspuru-Guzik, A.; Gordon, R. G.; Aziz,
M. J. Adv. Energy Mater. 2016, 1601488. (b) Weng, W.; Huang, J.;
Shkrob, I. A.; Zhang, L.; Zhang, Z. Adv. Energy Mater. 2016, 6,
1600795. (c) Narayana, K. A.; Casselman, M. D.; Elliott, C. F.; Ergun,
Figure 5. (a) Selected charge−discharge curves for the cycling of 17+. S.; Parkin, S. R.; Risko, C.; Odom, S. A. ChemPhysChem 2015, 16,
(b) Comparison of cycling data for anolytes 1+ and 17+. 1179. (d) Chénard, E.; Sutrisno, A.; Zhu, L.; Assary, R. S.; Kowalski, J.
A.; Barton, J. L.; Bertke, J. A.; Gray, D. L.; Brushett, F. R.; Curtiss, L.
A.; Moore, J. S. J. Phys. Chem. C 2016, 120, 8461. (e) Carino, E. V.;
optimization of new ROM derivatives for applications in RFBs Diesendruck, C. E.; Moore, J. S.; Curtiss, L. A.; Assary, R. S.; Brushett,
as well as in other electrochemical devices.

■
F. R. RSC Adv. 2015, 5, 18822.
(6) Sevov, C. S.; Brooner, R. E. M.; Chénard, E.; Assary, R. S.;
ASSOCIATED CONTENT Moore, J. S.; Rodríguez-López, J.; Sanford, M. S. J. Am. Chem. Soc.
*
S Supporting Information 2015, 137, 14465.
The Supporting Information is available free of charge on the (7) 1− and 1+ persist as 0.70 M solutions for over 4 days.
ACS Publications website at DOI: 10.1021/jacs.7b00147. (8) Verloop, A.; Tipker, J. Pharmacochem. Libr. 1977, 2, 63.
(9) (a) Sigman, M. S.; Harper, K. C.; Bess, E. N.; Milo, A. Acc. Chem.
Data for C20H18NO, BF4 (CIF) Res. 2016, 49, 1292. (b) Harper, K. C.; Bess, E. N.; Sigman, M. S. Nat.
Data for C20H18NO (CIF) Chem. 2012, 4, 366.
Experimental procedures, characterization, and spectral (10) (a) Milo, A.; Neel, A. J.; Toste, F. D.; Sigman, M. S. Science
and electrochemical data (PDF) 2015, 347, 737. (b) Niemeyer, Z. L.; Milo, A.; Hickey, D. P.; Sigman,

■
M. S. Nat. Chem. 2016, 8, 610.
AUTHOR INFORMATION (11) Harper, K. C.; Vilardi, S. C.; Sigman, M. S. J. Am. Chem. Soc.
2013, 135, 2482.
Corresponding Authors (12) Consistent with decomposition by radical homocoupling, the
*mssanfor@umich.edu immediate products could not be characterized as they react further to
*minteer@chem.utah.edu form intractable product mixtures.
*sigman@chem.utah.edu (13) (a) Kosower, E. M.; Poziomek, E. J. J. Am. Chem. Soc. 1964, 86,
ORCID 5515. (b) Itoh, M.; Nagakura, S. Bull. Chem. Soc. Jpn. 1966, 39, 369.
(14) (a) Huang, J.; Pan, B.; Duan, W.; Wei, X.; Assary, R. S.; Su, L.;
Shelley D. Minteer: 0000-0002-5788-2249 Brushett, F. R.; Cheng, L.; Liao, C.; Ferrandon, M. S.; Wang, W.;
Matthew S. Sigman: 0000-0002-5746-8830 Zhang, Z.; Burrell, A. K.; Curtiss, L. A.; Shkrob, I. A.; Moore, J. S.;
Melanie S. Sanford: 0000-0001-9342-9436 Zhang, L. Sci. Rep. 2016, 6, 32102. (b) Schon, T. B.; McAllister, B. T.;
Notes Li, P.-F.; Seferos, D. S. Chem. Soc. Rev. 2016, 45, 6345.
The authors declare no competing financial interest. (15) (a) Hermolin, J.; Levin, M.; Kosower, E. M. J. Am. Chem. Soc.

■
1981, 103, 4808. (b) Kosower, E. M. In Preparative Organic Chemistry;
ACKNOWLEDGMENTS Springer: Berlin, Heidelberg, 1983; p 117.
(16) 17+ has a solubilty of greater than 0.70 M in acetonitrile.
This work was supported by the Joint Center for Energy (17) (a) Montoto, E. C.; Nagarjuna, G.; Hui, J.; Burgess, M.; Sekerak,
Storage Research (JCESR) a Department of Energy, Energy N. M.; Hernández-Burgos, K.; Wei, T.-S.; Kneer, M.; Grolman, J.;
Innovation Hub. We thank Dr. J. M. Kampf for X-ray Cheng, K. J.; Lewis, J. A.; Moore, J. S.; Rodríguez-López, J. J. Am.
crystallography, and the NSF (CHE-0840456) for X-ray Chem. Soc. 2016, 138, 13230. (b) Burgess, M.; Moore, J. S.; Rodriguez-
instrumentation. Lopez, J. Acc. Chem. Res. 2016, 49, 2649. (c) Nagarjuna, G.; Hui, J.;

■
Cheng, K. J.; Lichtenstein, T.; Shen, M.; Moore, J. S.; Rodríguez-
REFERENCES López, J. J. Am. Chem. Soc. 2014, 136, 16309. (d) Liu, Y.; Lu, S.; Wang,
H.; Yang, C.; Su, X.; Xiang, Y. Adv. Energy Mater. 2016, 1601224.
(1) Gong, K.; Fang, Q.; Gu, S.; Li, S. F. Y.; Yan, Y. Energy Environ. Sci. (e) Wild, A.; Strumpf, M.; Häupler, B.; Hager, M. D.; Schubert, U. S.
2015, 8, 3515. Adv. Energy Mater. 2016, 1601415. (f) Winsberg, J.; Hagemann, T.;
(2) (a) Dunn, B.; Kamath, H.; Tarascon, J.-M. Science 2011, 334, 928. Janoschka, T.; Hager, M. D.; Schubert, U. S. Angew. Chem., Int. Ed.
(b) Perry, M. L.; Weber, A. Z. J. Electrochem. Soc. 2016, 163, A5064. 2017, 56, 686. (g) Janoschka, T.; Martin, N.; Hager, M. D.; Schubert,
(3) Darling, R. M.; Gallagher, K. G.; Kowalski, J. A.; Ha, S.; Brushett, U. S. Angew. Chem., Int. Ed. 2016, 55, 14427. (h) Winsberg, J.; Stolze,
F. R. Energy Environ. Sci. 2014, 7, 3459. C.; Muench, S.; Liedl, F.; Hager, M. D.; Schubert, U. S. ACS Energy
(4) (a) Soloveichik, G. L. Chem. Rev. 2015, 115, 11533. (b) Hicks, R. Lett. 2016, 1, 976.
G. Org. Biomol. Chem. 2006, 5, 1321. (c) Chen, Z.; Qin, Y.; Amine, K.

D DOI: 10.1021/jacs.7b00147
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX