See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/358516569

Continuous electrochemical refrigeration based on the Brayton cycle

Article  in  Nature Energy · April 2022

DOI: 10.1038/s41560-021-00975-7

CITATIONS READS

3 213

3 authors, including:

Aravindh Rajan
Palo Alto Research Center
10 PUBLICATIONS   37 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Aravindh Rajan on 17 February 2022.

The user has requested enhancement of the downloaded file.

 Articles
https://doi.org/10.1038/s41560-021-00975-7

Continuous electrochemical refrigeration based

on the Brayton cycle
Aravindh Rajan1, Ian S. McKay2 and Shannon K. Yee   1 ✉

Zero-global-warming-potential cooling technologies can mitigate the climate change effects attributed to the use of conven-
tional vapour compression refrigeration. In this work, we conceptualize an electrochemical refrigeration cycle and demonstrate
a proof-of-concept prototype in continuous operation. The refrigerator is based on the Brayton cycle and draws inspiration
from redox flow battery technologies. A peak coefficient of performance of 8.09 was measured with a small temperature drop
of 0.07 K. A peak cooling load of 0.934 W with a coefficient of performance of 0.93 was measured, however, with only a mod-
est measured temperature drop of 0.15 K. This is still much lower than the theoretical maximum temperature drop of 2–7 K for
these electrolytes. This work could inspire research into high-cooling-capacity redox-active species, multi-fluid heat exchanger
design and high-efficiency electrochemical refrigeration cell architectures.

D
riven by rising regional temperatures1, growing regional
populations2 and accelerating urbanization3, the global
cooling demand is expected to triple by 2050 (refs. 4,5).
In a business-as-usual scenario, nearly all of this demand will be
met using hydrofluorocarbons employed in vapour compres-
sion units6. Hydrofluorocarbons have global warming poten-
tials several times higher than that of carbon dioxide, and
vapour-compression-associated emissions are projected to be as
high as 9 GtCO2 equivalents per year (that is, 15% of CO2 equivalent
emissions) by 2050 (refs. 7,8).
One facet of the solution is the discovery and development of
zero-global-warming-potential alternative cooling technologies. In
principle, it is possible to take a heat engine, like a thermally regen-
erative electrochemical cycle (TREC)9, with sufficiently low internal
irreversibility and induce refrigeration via the reverse operation with
an external work input. While TRECs have been widely researched
for power generation10–12, the converse, a continuous electrochemi-
cally driven refrigerator, has not been previously demonstrated, to
the best of our knowledge. Electrochemically driven refrigeration has
been tenuously hypothesized and modelled in the past13–16. Duan et al.
developed a numerical model for a continuous electrochemical refrig-
erator based on the Stirling cycle, validated the open-circuit voltage
and overpotentials at various states of charge and current densities and
used the numerical model to predict cooling17. A Stirling-type cycle
inherently makes demonstration of electrochemical refrigeration
difficult as the entropic heat absorption is split between the moving
electrolyte, heat transfer across the electrochemical cell boundaries
and the thermal capacitance of the bespoke system. Recently, McKay
et al. demonstrated evidence of electrochemical cooling in a ther-
mogalvanics architecture18. However, the steady-state cooling load
reported was ~0.12 mW, and the electrolyte was heated to 50 °C to
reduce the activation overpotential. Furthermore, the maximum pos-
sible temperature span of this incarnation is limited by the trade-off
between the higher limiting current density and the thermal inertia
afforded by a higher electrolyte flow-rate.
In this study, we demonstrate a technology that is the refrigera-
tion analogue of a TREC10–12,19 relying on the thermogalvanic effect,
which utilizes the entropy change of redox reactions to generate
electrical work from a temperature difference. We develop the
Brayton Electrochemical Refrigerator (BECR), which achieves sus-
tained electrochemical refrigeration by employing three key fea-
tures. First, we identify two complementary half-cell reactions (that
is the triiodide/iodide and ferricyanide/ferrocyanide) that maximize
cooling and stability. Second, we develop modified flow battery
electrochemical cell architectures that facilitate continuous cooling
by increasing single-pass electrochemical conversion and reducing
electrochemical overpotential. Third, we base our technology on the
Brayton cycle as opposed to the Stirling cycle (Supplementary Note
1), which is frequently used by TREC systems. Additionally, unlike
the single-half-cell thermogalvanic architectures, our dual-half-cell
electrochemical system decouples ionic and heat transport, allowing
for access to higher current densities at lower Joule heat12,20. Finally,
we choose the Brayton cycle rather than the Stirling cycle because the
isothermal heat absorption process in a Stirling cycle competes with
the advecting electrolyte, which concomitantly exchanges heat with
the surroundings. This coupled heat and mass transfer problem for the
Stirling cycle is challenging; the Brayton cycle, which also has its own
challenges, circumvents this specific coupled problem by using a ther-
mally insulated electrochemical system and separate heat exchang-
ers. With the Brayton cycle, all of the electrochemical cooling lowers
the temperature of the electrolytes, and the heat absorption process
from the refrigerated space is accomplished with a conventional heat
exchanger, separately. Detailed analyses on the system dynamics of
both the BECR and the Stirling Electrochemical Refrigerator will be
provided in two forthcoming manuscripts (A. Rajan et al., manuscript
in preparation). Together, these features allow the BECR cycle devel-
oped in this work to achieve a peak cooling load of 0.934 W and a peak
coefficient of performance (COP) of 8.09. The maximum temperature
drop obtained is ~0.15 K at peak cooling load, while the theoretical
maximum temperature drop is ~2–7 K, depending on the electrolyte.
Concept of the BECR
The BECR scheme (Fig. 1b) employs two separated redox-active elec-
trolytes in a closed system comprising two electrochemical cells, two
circulation pumps and two heat exchangers. Refrigeration is achieved
by circulating the electrolytes through these components, where they

George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA, USA. 2Orca Sciences, Kirkland, WA, USA.
1

✉e-mail: shannon.yee@me.gatech.edu

Nature Energy | www.nature.com/natureenergy

Articles Nature Energy
a
b Hotter
TH
4 electrolytes

n 1 4
tio e– B A

Cation flux
ec

Isentropic reaction, ΔSrxn < 0

t rej e–
h ea
lar Hot side HX I
mo B′ A′
Iso
1 QH
TH Cold Hot
PIN
Temperature

cell cell
Isentropic reaction, ΔSrxn > 0

Cold side HX
e– A B

Cation flux
3
TC e– 2
3
QC
A′ B′
on
pti
absor TC
t
hea Colder
olar
Isom electrolytes
2

Entropy

Fig. 1 | Concept for the BECR. a, Temperature–entropy diagram for the BECR cycle showing the nature of the four constituent processes that convert the
electrolytes sequentially and cyclically from states 1 to 4. QH and QC are the rejected heat and the absorbed heat, respectively. The area enclosed within
the cycle is the electrical work input required to drive the cycle. b, Schematic showing the essential components operating in a continuous scheme. Two
redox-active electrolytes shown in purple (A + e− ⇌ A′) and yellow (B′ + e− ⇌ B) are pumped (shown using solid triangles) through four components, each
corresponding to one process shown in a and marked by the associated electrolyte states. An external power supply consumes power PIN and drives current
I through an adiabatic cold (electrochemical) cell, driving a net electrochemical reaction A + B → A′ + B′ with ∆Srxn > 0 isentropically. The entropic heat
required for the electrochemical reaction is absorbed from the lattice of the electrolytes, lowering the temperature of the electrolytes exiting the cold cell.
A cold side heat exchanger (HX) absorbs heat (represented by squiggly arrows) from a cold source TC devoid of any electrochemical reaction. The power
supply also drives the opposite electrochemical reaction, A′ + B′ → A + B with ∆Srxn < 0, in the adiabatic hot cell. In this cell, entropic heat liberation raises
the temperature of the electrolytes exiting the hot cell. The hot side HX rejects absorbed heat into a hot sink TH. Current continuity in the electrochemical
cells is preserved by an ion-exchange membrane that allows for an ionic current, shown to be cationic in this schematic.

are subject to the processes shown in the temperature–entropy dia-
gram in Fig. 1a. For tractability, Fig. 1a features the averaged state of
both electrolytes at a specific point in the scheme. The electrolytes
are driven from state 1 to state 2 in the cold electrochemical cell isen-
tropically using an electric current. The direction of this current is
such that the entropy change of the electrochemical reaction is posi-
tive (∆Srxn > 0), leading to a decrease in electrolyte temperature. The
reduction (A + e− → A′; e−, electron) and oxidation (B → B′ + e−) reac-
tions both require entropic heat, which is sourced from the electrolyte
lattice, since the cell is thermally insulated. A membrane keeps the
two electrolytes separate and shuttles ions in response to the electric
field created by the electron transfer between the half-cell reactions to
maintain charge neutrality. These ions must ideally each be a specta-
tor ion that is introduced into the electrolyte as a counterion or via a
supporting electrolyte. The electrolytes then absorb heat from the cold
source at a temperature TC in the cold side heat exchanger, thereby
heating the electrolytes from state 2 to state 3. In the hot electro-
chemical cell, the driven electrochemical reaction is opposite to that
of the cold electrochemical cell, that is, ∆Srxn < 0. In this case, both the
reduction (B + e− → B′) and oxidation (A → A′ + e−) reactions gener-
ate entropic heat, which is rejected into the electrolyte. Therefore, the
electrolytes heat up within the hot electrochemical cell, traversing
from state 3 to state 4. Finally, the electrolytes return to state 1 from
state 4 by rejecting heat to a hot sink at temperature TH in the hot side
exchanger. We detail the system dynamics of the BECR (A. Rajan et al.,
manuscript in preparation) and Stirling Electrochemical Refrigerator
systems, and their material level considerations, elsewhere21 and will
investigate them in future work.
as ∆Srxn/nF = αCELL, where αCELL is the cell temperature coefficient,
n is the number of electrons transferred, and F is the Faraday con-
stant. The cell temperature coefficient is the combination of con-
tributions from both electrolytes, αCELL ≈ αC – αA, where subscripts
C and A denote the cathode and anode, respectively11. To develop
the proof of concept, several half-cell reactions were first screened
(Supplementary Table 3) to find appropriate half-cell reactions,
using a non-isothermal temperature coefficient measurement
technique (Supplementary Note 2). We ultimately chose the aque-
ous I− 3 /I (αI > 
−
0) and ferricyanide/ferrocyanide (FCN3−/FCN4−,
αFCN < 0) half-cell reactions due to the relatively high αCELL ≈ αI – αFCN
value (~2 mV K–1) and the low crossover of both anionic
redox-active species when combined with a cation-exchange mem-
brane, such as Nafion, as employed in this work. The state of charge
(SOC)-dependent values of αI, αFCN and αCELL (measured using the
isothermal temperature coefficient measurement technique shown
in Supplementary Note 3) are shown in Fig. 2a. The SOC is defined
as the normalized electrons per unit volume of electrolyte that is
available for electron transfer in the direction of the cold cell elec-
trochemical reaction (ΔSrxn > 0). The maximum number of elec-
trons per volume is determined by the solubility of the limiting
reagent, in this case, potassium ferrocyanide at 0.6 M.
While the temperature coefficient represents the entropic driv-
ing force that leads to cooling, the heat capacity of the electrolyte
represents the thermal inertia. The volumetric heat capacities of the
electrolytes were measured at the 50% SOC configuration using a
calorimeter (Supplementary Note 4) to be 2.49 ± 0.38 MJ m−3 K−1
for aqueous I− 3 /I and 3.58 ± 0.78 MJ m  K for aqueous FCN /
− −3 −1 3−

FCN . We use these values as constants for our thermal calcula-

4−

Selection and characterization of electrolyte refrigerants tions, since we do not expect the heat capacities of the electrolytes
The BECR efficiency and cooling load are governed by the entropy to change substantially due to the small intra-electrolyte species
change of the combined electrochemical reaction, which is given conversion occurring with electron transfer. The SOC-dependent

Nature Energy | www.nature.com/natureenergy

Nature Energy Articles
a b
3 300

2
298

Final temperature (K)

296
α (mV K–1)

294
Direction of cold cell
–1 reaction
3I– + 2FCN3– I3– + 2FCN4–
∆Srxn > 0

292
–2

I3–/I–
I3–/I– FCN3–/FCN4– Cell FCN3–/FCN4–
–3 290
0 20 40 60 80 100 0 20 40 60 80 100
SOC (%) Final SOC (%)

Fig. 2 | Electrolyte characterization. a, Temperature coefficients of the individual electrolytes (non-isothermal measurement) and the full cell (isothermal
measurement) as a function of the SOC. We choose the direction of the reaction occurring in the cold electrochemical cell to be the SOC reference. The
dashed lines are linear fits that describe the temperature coefficients as a function of the SOC. b, The linear fit from a and the specific heat capacity of
the electrolyte are used to calculate the expected drop in electrolyte temperature as function of reaction extent along the isentropic path for a closed
electrochemical cell. The initial configuration starts at 90% SOC with the electrolytes in thermal equilibrium at 300 K. For this calculation, it is assumed
that the ion-exchange membrane has infinite thermal resistance. Details are provided in Supplementary Note 5.

temperature coefficients and heat capacities can be used to pre-
dict the maximum isentropic temperature difference as a function
of reaction extent, as depicted in Fig. 2b, achievable in the absence
of thermodynamic irreversibilities (that is, inter-electrolyte and
exo-cell heat transfer, and Joule heating), within a closed electro-
chemical cell. Supplementary Note 5 details how Fig. 2b was gen-
erated. In general, the temperature difference may be increased by
alternative half-cell reactions with larger temperature coefficients,
lower specific heat capacity and higher solubility of the limiting
reagent. To obtain the maximum isentropic drop in temperature
for a given flow-rate, electrode volume and operating current in an
open system, the electrochemical cell must adhere to the tenants of
a plug flow reactor (that is, without any heat or mass dispersion)22.
In such an architecture, for a given pair of electrolytes, the isentropic
drop in temperature increases with higher electrolyte residence time
and operating current. Additional thermodynamic cycle details,
the BECR figure of merit and the predicted performance will be
addressed in forthcoming theoretical work (A. Rajan et al., manu-
script in preparation). Additionally, we confirmed that the I− 3 /I //
−
Nafion//FCN /FCN system possessed the required lifetime for
3− 4−
the proof-of-concept experiment by cycling it at ~16 mA cm–2 within
a potential window far away from water-splitting regimes (between
–0.6 V and 0.6 V) for 100 cycles over 200 hours (Supplementary Note
6). For perspective, we operate the proof-of-concept refrigerator for
only 50 cycles and 20 hours, observing steady, continuous cooling.
BECR proof of concept
The proof-of-concept experiment (Fig. 3a) and the schematic in Fig.
1b are alike, save for the heat exchangers. In the proof-of-concept
prototype, one heat exchanger regulates the temperature of each
half-cell electrolyte at the inlets of the hot and cold cells to 27 °C
(that is, TH = TC = 27 °C), thereby, maintaining constant inlet tem-
peratures to clearly demonstrate that cooling is occurring from
only the electrochemical reactions and is not an artefact of ambi-
ent (room) temperature fluctuations. The electrochemical cells have
and a graphite serpentine flow field to increase the residence time,
and therefore the degree of electrochemical conversion of the elec-
trolyte. Electrolyte flow-rates were measured using optical tachom-
eters (Supplementary Note 7). Eight different electric current inputs
(up to 4 A) were sourced with a galvanostat to the electrochemi-
cal cells for 20 minutes to achieve a steady state (with 1 minute of
open-circuit data measurement on either side of the sourced cur-
rent domain; Fig. 3b). When the current is applied, the electrolyte
exiting the cold cell is driven to a lower SOC, decreasing the outlet
temperatures of both electrolytes using the entropy of the reactions.
Conversely, the electrochemical reaction in the hot cell increases the
SOC (the direction of current I is reverse in the hot cell relative to
the cold cell), in turn increasing the outlet temperature over time.
Since the electrochemical cells are in series and the flow-rates of
the electrolyte streams in both cells are equal, the SOC values of
the electrolytes entering both cells remain invariant after the steady
state has been achieved (when the voltages in the cells, VHOT and
VCOLD, saturate). The cells reach the thermal steady state (dTi/dt = 0;
Ti, temperature; t, time; subscript i, any of the electrolyte outlet tem-
peratures like C2 or H4, as seen in Fig. 3a) at ~7 minutes and elec-
tric steady state (dVi/dt = 0; Vi, voltage; subscript i, HOT or COLD)
at ~2 minutes. At the thermal steady state, the difference between
the electrolyte outlet and inlet temperatures is equal to the net heat
generation within the corresponding electrode. We used the differ-
ence of the values of this temperature differential at the steady state
(t = 21 minutes) and at the onset of the current (t = 1 minute) as the
steady-state temperature response of the electrolyte to the heat pro-
cess occurring in the electrode (Fig. 3c).
The steady-state temperature response was used to evaluate
the electrolyte-specific cooling load (Fig. 4a,b) and rejected heat
(Fig. 4c,d) using the heat capacity and flow-rate of the electrolytes.
The FCN3−/FCN4− electrolyte outperforms its counterpart because
|αFCN| > |αI| and because it has a lower overpotential at 4 A, where
the entropic cooling effects of the triiodide reduction are negated
by Joule heating. The larger Joule heating associated with the I− 3 /I
−

acrylic housing to increase thermal resistance to the surroundings reaction arises from its higher activation overpotential, attributed to

Nature Energy | www.nature.com/natureenergy

Articles Nature Energy

a b
TC1 Galvanostat
Peristaltic TC3 a
pump Peristaltic
pump

TC2 Cold cell

TC4 I

Cold cell
c
TH1 d
TH3
VCOLD = Va – Vb

VHOT = Vc – Vd

Hot cell
TH2
TH4 TE HX
TE HX

Hot cell

FCN3–/FCN4– electrolyte I3–/I– electrolyte

c d

0.2 0.1
TC1 TC2 TC3 TC4 ΔT21 ΔT43
0
0
Ti(t) – Ti(t = 0) (K)

–0.1
1.5 A 1A
–0.2 –0.2
ΔTij(t) – ΔTij(t = 0) (K)

0.5 0.1
ΔT21 ΔT43
0
0
–0.1
1.5 A TH1 TH2 TH3 TH4 2.5 A
–0.5 –0.2

0.3 0.1
ΔT21 ΔT43
Voltage (V)

0
0.1
–0.1
VCOLD VHOT
1.5 A 3.5 A
–0.1 –0.2
0 2 4 6 8 10 12 14 16 18 20 22 0 2 4 6 8 10 12 14 16 18 20 22
Time, t (min) Time, t (min)

Fig. 3 | BECR proof of concept. a,b, Schematics showing the thermo-hydraulic (a) and electrical (b) connections of the experimental set-up. Thermoelectric
(TE) modules regulate the temperature of the electrolytes at both electrochemical cells. The electrolyte temperatures are monitored at the points marked
by the variable resistor symbol. These temperatures, T, are labelled as C or H for cold or hot cell, and numbered 1 through 4 for I3−/I− electrolyte inlet,
I3−/I− electrolyte outlet, FCN3−/FCN4− electrolyte inlet, and FCN3−/FCN4− electrolyte outlet respectively. Voltage across the cold cell VCOLD (and hot cell
VHOT) is given by the potential difference at points a and b (and c and d) c, Representative data set for 1.5 A showing the time variation of the electrolyte
temperatures at the inlets and outlets of the electrochemical cells and the cell voltages. d, Time variation of the cold cell outlet and inlet temperature
differential for three representative current values. Here, ∆Tij is the difference between the cold cell electrolyte outlet and inlet temperatures and is defined
as TCi – TCj. The dark grey regions are when no current is sourced, and the indicated current is sourced in the light grey domain. In b and c, the lines and
associated shaded regions represent the mean value and standard deviation of the indicated parameter over six repetitions.

the large solvent reorganization energy of the covalent bond cleavage
that accompanies the triiodide reduction23. The electrolyte-specific
cooling loads were combined to obtain the total cooling load, Q̇C
(Fig. 4e). The COP, β, was evaluated using β = Q̇C /PIN (Fig. 4e),
where PIN = I(VHOT + VCOLD). We choose to neglect the circulating
pumping power; an estimate of the pumping power based on Darcy’s
law is provided in Supplementary Note 8. The BECR can achieve a
target cooling capacity by achieving a high degree of electrochemical
conversion at low electrolyte flow-rate. Therefore, unlike flow batter-
ies, the BECR is not bound by the trade-off between high pumping
power and high porous electrode-specific surface area. The precip-
itous drop in COP with increasing current (shown in Fig. 4e) has
thermodynamic and kinetic origins. Even though the cells operate
in the ohmic regime (Supplementary Fig. 16) at the higher current
values, the Joule heating scales with I2, increasing the irreversibility.
Furthermore, T2 and T4 (the average temperature of the electrolytes
exiting the cold cell and hot cell respectively, as shown in Fig. 1a)
decrease and increase respectively with increasing current. This leads
to heat transfer processes that are increasingly deviant from the iso-
thermal heat transfer characteristic of the ideal Carnot cycle. The
measured cooling load and COP (evaluated above) are in good agree-
ment with the predicted values, calculated using an energy balance.
Finally, we emphasize that while this proof-of-concept dem-
onstration of continuous electrochemical refrigeration is made

Nature Energy | www.nature.com/natureenergy

Nature Energy Articles
a c e
0.8 4
15 1.5
I3–/I–
0.6 3 Predicted β
Measured β
Predicted QC
0.4 2
Measured QC
12 1.2
0.2 1

Cooling

Electrolyte rejected heat (W)

Electrolyte cooling load (W)

Total cooling load, QC (W)

0 0
High Joule heat I3–/I–
9 0.9
–0.2 –1
b d

COP, β
1.2 8
FCN3–/FCN4–
6 0.6
0.9 6

0.6 4

0.3 2 3 0.3

Cooling
0 0
High Joule heat FCN3–/FCN4–
–0.3 –2 0 0
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Operating current, / (A) Operating current, / (A)
Operating current, / (A)

Fig. 4 | BECR performance. a,b, Electrolyte-specific cooling loads for the I−

3 /I (a) and FCN /FCN (b) electrolytes as a function of the operating
− 3− 4−

current. Blue regions indicate a positive cooling load where the entropic heat absorption dominates the Joule heat and the electrolytes are capable of
cooling. Red regions indicate a negative cooling load due to excessive Joule heating. c,d, Electrolyte-specific rejected heat for the I−
3 /I (c) and FCN /
− 3−

FCN4− (d) electrolytes as a function of the operating current. e, Measured and predicted values of the COP and total cooling load as a function of the
operating current. Markers and error bars (visible if bigger than the marker) represent the mean value and standard deviation of the quantity across six
repetitions. The lines and shaded bounds of the predicted COP (too small to be visible) and cooling load represent the same. Dashed lines serve as guides
for the measured values.

possible by appreciable electrolyte redox conversion, substantial the technologies (Fig. 5b). Through this lens, the entropic driving
improvement can still be made by increasing the residence time force for electrochemical refrigeration is more than double that of
of the electrolytes. The first Damköhler number, given by the ratio vapour compression.
of the reaction rate and the advective molar transport rate, for this Table 1 details the device COP and temperature drop achieved
work is ~0.08, which amounts to ~8% conversion for a plug flow by experimental works corresponding to the previously mentioned
reaction with first-order kinetics22. There is substantial scope for refrigeration technologies. We chose popular refrigerants used in
improvement by operating at higher currents (increased rate of other refrigeration technologies in Fig. 5 and Table 1 simply to pro-
reaction) or decreasing the volumetric flow-rate of the electrolytes vide context to our work. Research within each of these technolo-
(decreased molar influx). Of course, the latter must be done keep- gies may have led to other refrigerants that demonstrate superior
ing in mind that decreasing the electrolyte flow-rate to arbitrarily performance. For example, a rotary magnetocaloric refrigerator
low values will decrease the cooling load, and hence there exists an that employs LaFeSiH as the refrigerant has shown a temperature
optimum electrolyte flow-rate for a given current that provides the increase of 11 K and a COP of 1.9 (ref. 26). Table 1 shows that sub-
maximum electrolyte-specific cooling load. As future motivation, stantial work must be done to improve the temperature drop of
Kim et al. reported a twofold increase in the temperature coefficient our technology, while maintaining the COP, to improve its com-
of FCN3−/FCN4− by employing a 15 wt% methanol–water solution mercial viability. We believe that this can be done through addi-
as the solvent24. Yamada et al. increased the temperature coefficient tional research into high-temperature-coefficient half-cell reactions
of I−
3 /I to 2 mV K using a supramolecular scheme . These works
− –1 25
with high solubility, low-freezing-point solvents with low specific
could be immediately adopted into the BECR cycle to improve the heat capacities and electrochemical cell architectures that can drive
cooling load several fold. isentropic reactions to completion. Higher-temperature-coefficient
half-cell reactions with high solubility have larger entropy changes
Discussion and conclusions per unit volume of electrolyte, and thereby increase an electro-
We have compared the inverse gravimetric heat capacities and the lyte’s capacity for cooling, in addition to delaying the onset of
gravimetric maximum possible entropy change of existing refrig- temperature-induced precipitation. Low-vapour-pressure solvents
erants with those of our electrochemical refrigerants (Fig. 5a). (for example, ionic liquids) and low-freezing-point solvents (for
The maximum possible entropy change and the specific heat rep- example, alcohols) allow for the BECR cycle to operate in a larger
resent the entropic driving force and the thermal inertia towards parameter space (operating current and electrolyte flow-rate) with
cooling. With existing aqueous redox chemistries, we found that little threat of excessive internal pressurization and/or the electro-
electrochemical refrigeration compares well against vapour com- lyte freezing. Lower electrolyte-specific heat elicits a greater temper-
pression, suggesting that it could someday be an appreciable ature response per unit electrochemical reaction. Electrochemical
zero-global-warming-potential refrigeration technology. A more cell architectures that drive the reactions towards completion
compelling argument for electrochemical refrigeration can be made (higher first Damköhler number) isentropically (minimal over-
upon comparing the entropy change per unit carrier, ∆SCARRIER, for potential) can generate higher electrolyte temperature drops. The

Nature Energy | www.nature.com/natureenergy

Articles Nature Energy

a b
102 80

Solid to Solute to
gas solute
MC (Gd5Si2Ge2) Electron to
electron

60

101
Liquid solution
TE (Bi2Te3) to gas

ΔSCARRIER /R
Alternative
C–1 (kg K kJ–1)

solvents

40 Liquid to
MH (LaNi5)
gas
Solid to
VC (R140a) solid
100

20
Alternative
redox couples
EC
(Fe2+/3+//FCN3–/4–) Ab
(NH3 – water)

10–1 0
0 0.5 1.0 1.5 2.0 MC TE VC MH Ab EC
–1 –1
ΔS (kJ kg K )

Fig. 5 | Comparison of refrigeration technologies. a, Electrochemical refrigeration (EC) compared against popular refrigerants used in magnetocaloric
(MC), thermoelectric (TE), vapour compression (VC), metal hydride (MH) and absorption (Ab) refrigeration along the axes of the maximum possible
gravimetric entropy change (ΔS) and the inverse of the gravimetric specific heat capacity (C−1). b, Comparison of the entropy per unit carrier normalized
by the universal gas constant, R, for the same technologies along with the corresponding physical nature of the transformations. Data are compiled from
references listed herein33–38,27 and are summarized in Supplementary Table 4.

Table 1 | Comparison of device COP and temperature drop of some refrigeration technologies
Technology Refrigerant COP Temperature drop (K) Comment Reference
Electrochemical I3 /I //FCN /FCN
− − 3− 4−
8.09 0.07 Peak COP This work
3 /I //FCN /FCN
I− 0.93 0.15 Peak cooling load
− 3− 4−

Vapour compression R410a 3.1 6–8 Heating, ventilation and air conditioning 27
R134a 3.5 10–40 Refrigeration 28
Magnetocaloric Gadolinium 0.8 15.4 29
Thermoelectric Bismuth telluride 0.72 14 30
Absorption Ammonia–water 0.45 27 31
Metal hydride Lanthanum nickel 0.26 30 32

operating current and electrolyte flow-rate influence the COP and
cooling load, and they must be rationally optimized according to the
thermodynamic, kinetic and physical properties of the electrolytes.
In conclusion, we have demonstrated a continuous electro-
chemical refrigerator that is inspired by the reverse Brayton cycle. It
achieves a peak COP of 8.09 with a temperature drop of 0.07 K, and
a peak cooling load of 0.934 W (at 14.6 mW cm–2, normalized by the
cross-sectional area of the porous electrodes). We acknowledge that
only a modest temperature drop of 0.15 K at peak cooling load was
observed (Fig. 3c), but theoretically it could have achieved a tem-
perature drop of 2–7 K (Fig. 2b) if we were able to achieve a greater
degree of electrochemical conversion. The system demonstrates
cooling for a total of ~18 hours without any sign of deterioration of
the system performance.
Methods
Materials. All chemicals were purchased from commercial suppliers
(Sigma-Aldrich, VWR International, The Science Company), were of at least
reagent grade and were used without purification. MilliQ grade (19.7 MΩ cm)
ultrapure deionized water was used to prepare all the solutions. Non-aqueous
solvents were used without drying.
Electrochemical cells. The electrochemical cells were designed to maintain
an adiabatic environment and increase the residence time of the electrolytes
within the cell. Some 0.5-inch-thick acrylic endplates and a serpentine channel
made of resin-impregnated graphite (Graphite Store, MW001204) were used.
Supplementary Fig. 13 shows an exploded view of the cold and hot electrochemical
cells. The acrylic housing sandwiches the serpentine, porous graphite electrodes
(AvCarb, G600A) and cation-exchange membrane (Nafion 211). Eight M6 bolts
were used to hold the assembly together. Electrolytes were introduced into the
cells using polyvinylidene fluoride (PVDF) adaptors (McMaster-Carr, 5533K411).
The current collectors were thin copper strips that were placed in between the
acrylic housing and serpentine graphite channel. The size of the porous electrodes
(and therefore the rest of the assembly) was chosen using a conservative estimate.
A simple energy balance affords the following equation for small temperature
changes: V̇CV ΔT = IαT∞ − I2 R. Here, we assume that the majority of the
overpotential occurs due to the ohmic overpotential from the membrane
resistance. We use some conservative values for the flow-rate V̇ (1 ml s–1), heat
capacity CV (3.5 MJ m–3 K–1), temperature coefficient α (1.5 mV K–1) and membrane

Nature Energy | www.nature.com/natureenergy

Nature Energy
resistance R (1 Ω cm2). If we want to achieve a 0.5 K drop in temperature at 10 A at
an ambient temperature T∞ of 298 K, the cross-sectional area of the cell should be
~36 cm2. We exercise prudence and choose porous electrodes with an 8 cm × 8 cm
cross-sectional area to further decrease the effect of overpotentials.
Heat exchangers. We opted for an active heat exchanger (Supplementary Fig. 14)
rather than a passive heat exchanger to eliminate ambient temperature fluctuation
and maintain a constant electrochemical inlet cell temperature; this allows us
to unequivocally demonstrate electrochemical cooling decoupled from ambient
temperature fluctuations of the heat exchangers. The active heat exchange was
performed using a thermoelectric module attached to a cold plate, which had
a tube for the electrolyte to flow. Since the electrolytes must always be kept
from mixing with each other, we used two active heat exchanger modules (TE
Technology, CP-061HT). We embedded a 1/8-inch-outer-diameter stainless-steel
tube (chosen for its corrosion resistance to the triiodide/iodide and ferricyanide/
ferrocyanide redox couples) into two 1/4-inch-thick copper blocks, which were
mounted to the thermoelectric heat exchanger. A 1/4-inch-thick neoprene
rubber was use to insulate the thermoelectric module and copper block from the
environment. The parts of the stainless-steel tubing that were in contact with the
ambient air were liberally covered in a polymer as insulation. The temperature
was set and regulated by sourcing power to the thermoelectric modules using a
programmable temperature controller (SRS, PTC10) that employed two PTC440
TEC drivers. The temperature feedback to the closed proportional–integral–
derivative loop was provided by thermistors (OMEGA, item no. 44031) that
monitored the temperature of the electrolytes exiting the heat exchangers. The
programmable temperature controller used a PTC320 card to acquire inputs from
the thermistors located at the outlets.
Porous electrode and ion-exchange membrane preparation. Prior to all
experiments, the porous electrodes were immersed in an acid bath containing
three parts 68% nitric acid and one part 18 M sulfuric acid. They were heated to
80 °C and refluxed overnight. The electrodes were washed in deionized water until
pH neutral and then heated in an oven at 400 °C for three hours. The electrodes
were assembled into the electrochemical cells and flooded with electrolytes as
soon as possible. The as-purchased Nafion membrane was first heated in a 2 M
KOH solution at 60 °C overnight. It was then washed thoroughly with deionized
water and stored in a 4 M KCl solution. After each experiment, the membrane was
washed thoroughly with deionized water and stored in the KCl solution until the
next experiment.
Electrolytes for proof-of-concept experiment. Of all the redox-active species
employed in the proof-of-concept experiment, the species with the limiting
solubility was the ferrocyanide with a room temperature solubility of ~0.6 M.
Therefore, with reference to the cold electrochemical cell reaction, the cell is at
100% SOC when the ferrocyanide ion is at that value. Conversely it is at 0% SOC
when all of the ferrocyanide has been converted to ferricyanide at 0.6 M. The
quantity of the iodide ions and triiodide ions are determined using the solubility
constraint of the ferrocyanide species and by the requirement, albeit arbitrary, that
at 50% SOC, both the electrolytes have the same concentration of potassium ions.
Potassium ions were used for the iodide, ferricyanide and ferrocyanide species. The
triiodide ion is achieved by dissolving molecular iodine into an iodide solution,
which causes the iodine to abstract an iodide to form the triiodide. This reaction
has an equilibrium constant that strongly favours the triiodide ion, and therefore
it is assumed that no iodine is present in the solution. The proof-of-concept
experiment was primed with electrolyte at 80% SOC, and the details are provided
in Supplementary Table 2.
Proof-of-concept set-up. The tubing used to connect the components together
had a 1/4 inch outer diameter and 1/8 inch inner diameter. The tubing used to
contain the ferricyanide/ferrocyanide electrolyte was a flexible polyvinyl chloride
tubing, and rigid perfluoroalkoxy tubing was used to contain the triiodide/iodide
electrolyte due to the electrolyte’s tendency to engage in SN2 nucleophilic attack with
non-fluorinated polymers. All adaptors and fittings were made of PVDF. At the
inlet and outlet electrolyte ports of the electrochemical cells, resistive temperature
devices (RTDs) were introduced into the electrolyte streams using PVDF tee
adaptors. Our previous experiments showed that it was necessary for these RTDs
to have a metal shield to prevent capacitive coupling to the voltage applied by the
galvanostat. In addition to the components shown in Fig. 3a, we used two sealed
Erlenmeyer flasks for each electrolyte stream as a reservoir source (and sink) for
the inlets (and outlets) of the electrochemical cells, which served three purposes:
(1) they acted as gas traps that removed any air bubbles from the system, (2) they
allowed one pump to generate two flow streams and (3) by placing the outlet (or
inlet) tubing above (or below) the liquid level, they ensured the flow of electrolytes
in only one direction (eliminating any possibility of unintended back flow). K-type
thermocouples were placed close to the electrochemical cells to provide evidence
that the temperature changes caused by sourcing the electrochemical current are
not correlated to changes in environmental temperature. A photograph of the
completed set-up is shown in Supplementary Fig. 15. Peristaltic pumps (Control
Company, 3385) circulated the electrolyte. The hot and cold electrochemical cells
were connected electrically in series to the galvanostat (Biologic, VSP-300). Eight
Nature Energy | www.nature.com/natureenergy
Articles
stainless-steel shielded RTDs (Adafruit Industries, 3290) were introduced into
the electrolyte flow paths at the inlets and outlets of the electrochemical cells. The
set-up was assembled inside a fume hood whose temperature was monitored using
three thermocouples (Supplementary Fig. 17).
Calculation of equilibrium-time cell resistance. The equilibrium-time cell
resistance values during the proof-of-concept experiment were calculated using the
expression below, where i denotes the values corresponding to the hot or cold cell:
Vi (t = 21 minutes) − Vi (t = 1 min)
Ri =
I
Calculation of measured and predicted cooling loads. The measured total
cooling load was calculated by the sum of the electrolyte-specific products of the
volumetric specific heat (CV,e), volumetric flow-rate (V̇e) and equilibrium-time
drop in temperature (ΔTij):
∑
Q̇C,measured = V̇e CV,e ΔTij
e
Subscript e denotes electrolyte and subscripts i and j denote the RTDs located
at the electrolyte outlet and inlet on the cold cell, respectively. The predicted total
cooling load was calculated using an energy balance:
Q̇C,predicted = IαT − I2 RCOLD
T̄ is the volume-averaged temperature within the cold cell and RCOLD is the
resistance of the cold cell at steady-state. Since the maximum drop in temperature
in this work is ~0.2 K, we assume that the cell is isothermal at 27 °C.
Non-isothermal temperature coefficient measurements. Half-cell reactions
were screened using a non-isothermal temperature coefficient measurement.
To contain a variety of redox-active couples and solvent combinations, the cell
housing was made of a 7-cm-long polytetrafluoroethylene (PTFE) cylinder with
a 1.5 cm bore across its length, to house the electrolyte solution. The electrodes
were made of resin-impregnated graphite (Graphite Store, MW001204). After the
electrolyte was loaded into the cell housing, the PTFE annulus was sandwiched by
the graphite electrodes using neoprene gaskets and two M16 bolts that thread into
the bottom electrodes. Fibre glass washers were used to keep the graphite electrode
electrically isolated via zinc-plated bolts. The assembled cell was then sandwiched
between two thermoelectric modules (TE Technology, CP-061HT) using a lead
screw mechanism that minimizes interfacial thermal resistance by applying
pressure. When the experiment is started, the programmed thermoelectric
modules apply varying thermal gradients, as shown in Supplementary Fig. 3. The
top thermoelectric module is always maintained at a higher temperature than the
bottom one. The temperatures of the electrodes are continuously monitored using
K-type thermocouples, and the voltage across the two electrodes is measured using
a data acquisition unit (Agilent, 34970A). The hot electrode and cold electrode
are always connected to the positive and negative terminals, respectively. The
non-isothermal temperature coefficient is evaluated by the slope of the voltage
change and the difference in temperature between the two electrodes. For these
measurements, electrolytes were prepared with equal concentrations of the reduced
and oxidized species.
Isothermal temperature coefficient measurements. Once two promising half-cell
reactions with the appropriate temperature coefficients were shortlisted, they
were assembled in an electrochemical cell with an ion-exchange membrane of
choice (Supplementary Fig. 4). This was done for two reasons: (1) to confirm that
the temperature coefficients added up in a complete electrochemical cell set-up
and that there were no mechanisms that diminished the entropy of the half-cell
reactions, and (2) to evaluate the lifetime of the electrochemical system under
varying temperature under the open-circuit condition. To test a variety of redox
systems, the tenants of maximum inertness were maintained. The cell housing
was made of PTFE, and 2 cm through-holes were bored into the middle to house
the porous electrodes. Carbon felt (AvCarb, G600A) was used as the cathode and
anode material. The cathode and anode sandwiched the ion-exchange membrane
of choice. The current collectors were made of resin-impregnated graphite
(Graphite Store, MW001204) and zinc-plated screws were screwed into them
to facilitate robust electrical connections. The cell housing was sealed using M6
bolts and neoprene gaskets. The electrolytes were first introduced into the cell
using syringes and then air-tight sealed using redundant stop cocks and one-way
valves. This restricted any solvent motion due to osmotic pressure. After the
electrochemical cell was assembled and checked for any leakage, it was placed
in between the same two thermoelectric modules as described in the previous
section. This time, however, the modules were programmed to establish the same
temperature. The temperature feedbacks that informed the proportional–integral–
derivative controller were disconnected from the modules and firmly affixed on
the cathode and anode current collectors. Thermocouples were used to measure
the temperature of both current collectors, and the open-circuit voltage of the
cell was monitored continuously. Keeping in line with convention, the half-cell
reaction with the positive temperature coefficient was connected to the positive
Articles
terminal. When the experiment is started, the thermoelectric modules draw power
to drive the feedback temperatures (located on the current collectors) to the set
value. However, thermal equilibrium is achieved when the entire cell (specifically
the porous electrodes) attains the bespoke temperature. At this point, there should
ideally be no change in the open-circuit voltage with time. This was taken to be the
point of thermal equilibrium, and the open-circuit voltage and electrochemical cell
temperature were recorded for data processing. This was repeated for multiple set
values, and the rate of change of the voltage with respect to temperature afforded
the temperature coefficient of the half-cell reactions. Supplementary Fig. 5 shows
the generated data and processing.
Quantifying electrolyte-specific heat capacity. The measurement of
electrolyte-specific heats was mildly complicated by the tendency of the iodide/
triiodide electrolyte to liberate corrosive I2 vapour upon heating. This could be
detrimental to costly laboratory equipment (for example, automated differential
scanning calorimeters); therefore, we opted to perform simpler ‘Styrofoam cup’
calorimetry. The ‘Styrofoam cup’ calorimeter assembly, as shown in Supplementary
Fig. 6, is filled with some liquid whose specific heat capacity we want to measure.
A resistive heater is immersed into the liquid and generates heat at the rate PCAL.
A stir bar keeps the liquid well mixed so that the temperature is homogeneous.
Employing a simple energy balance for the liquid that begins at thermal
equilibrium (T = Tamb), the following expression is obtained under the assumption
of high thermal resistance of the Styrofoam cup:
dT
(VCV + CCAL ) ≈ PCAL
dt
In the above equation, CCAL is the lumped thermal capacitance of the cup
contents that is not the electrolyte of interest. Using the above equation, the specific
heat capacity of the liquid may be evaluated using a linear regression. For the
resistive heater, we chose a 1 Ω thin-film resistor (Riedon PF1262-1RF1) capable of
generating up to 20 W of heat. It was screwed into a 2 cm × 2 cm × 3 mm graphite
plate with some heat sink compound to improve its thermal conductance with the
liquid. The temperature of the liquid was measured using four RTDs (Adafruit
Industries, 3290) placed in random locations within the fluid (Supplementary
Fig. 7). Prior to the experiment, the resistive heater and RTDs were immersed
in the liquid of interest, contained in the Styrofoam cup, with fast strirring using
a stir bar (700 r.p.m.). During the experiment, a power source (Keithley 2230G-
30-3) was used to source 1.5 A through the heater, and the temperatures and
voltage across the resistor were continuously monitored using a data acquisition
unit (NI, CompactDAQ). The experimental time was short (<50 s) such that
the temperature–time slopes were linear, as shown in Supplementary Fig. 8. The
temperature of the liquid was taken to be the average of the four temperature traces,
and some of the data were visually truncated to ignore the noise floor regime. The
experiment was first validated using deionized water. We then used this set-up to
measure the specific heat capacities of the electrolytes (Supplementary Fig. 9). The
y intercept in the plot is different for the different liquids because of the polymer
adhesive that had to be applied to the resistive heater for corrosion resistance.
Electrolyte temperature as a function of reaction extent. For an arbitrary
electrochemical system, the first law of thermodynamics may be written as the
following expression:
( )
∑ ∑ ∑ ∑
dH = d Ni H̄i = H̄i dNi + Ni dH̄i = TdS + μi dNi
i i i i
In the above expression, H is the total enthalpy, N is the number of particles, H̄
is the partial molar enthalpy, T is temperature, S is entropy, μ is chemical potential
and subscript i(represents
) all species participating in the electrochemical system.
Recalling that ∂H̄i /∂T P = C̄P,i, where C̄P,i is the partial molar heat capacity at
constant pressure P, and employing the concept of an average heat capacity at
constant pressure P, CP, the above expression may be written as the following:
∑ ∑
TdS = CP dT + H̄i dNi − μi dNi
i i
Finally, we recognize that by fixing the initial and final states of the reaction, we
can define a reaction coordinate. One point on the reaction coordinate will then
map to relative concentrations of all species in the electrochemical system. We also
recognize that the changes in the partial molar enthalpies and chemical potential
are simply the enthalpy and Gibbs free energy of the reaction. The above equation
can now be written as the following for the special case of an isentropic reaction:
CP
dS = dT + ΔSrxn (ξ) dξ = 0
T subscript e represents the electrolyte. The uncertainty in the electrolyte-specific
The reaction coordinate ξ is interchangeable with the SOC. Finally, the linear
fit from the trends in Fig. 2a and specific heat capacities may be used to integrate
both sides of the above equation to find the final temperature as a function of
reaction extent.
Nature Energy
Quantifying electrolyte flow-rate. Electrolyte flow was generated using
two identical peristaltic pumps (Control Company, 3385). The peristaltic
pumps use a potentiometer to control their speed. Once connected to a
closed-flow loop, the speed of the pump decreases, and, therefore, the pump
flow-rate needs to be calibrated against the rotational speed that is set by the
potentiometer’s position when connected to the flow loop. To measure the
speed of the pump, we used an optical tachometry technique, as shown
in Supplementary Fig. 11. We first attached a reflective piece of foil to the
circumference of the pump head. Directly above the pump head, we mounted a
light-emitting diode (LED) and a photodiode (PD) within a recession cut into
a piece of foam. The assembly was arranged such that during the rotation of the
pump head, the LED and PD would be directly over the foil. At this instance, the
light emanating from the LED is reflected to the PD. The time-varying current
generated from the PD passes through a 909 kΩ resistor, and the time-varying
voltage is recorded by an oscilloscope (Tektronix, TBS1202B). This voltage trace
comprised moments of high intensity when the foil was directly underneath the
LED and PD, and moments of low intensity in the absence of the foil during the
rotation. The periodicity of this trace informed the pump speed. A calibration
experiment was performed to map the flow-rate of each pump to the pump
speed. For a specific setting on the pump, the speed was first measured using
the optical tachometry technique above. Then, the flow-rate for that setting
was measured using a graduated cylinder and a stop-watch. This was done for
three different settings, and a linear regression was performed to evaluate the
calibration (Supplementary Fig. 12). This was done for both pumps. During the
proof-of-concept experiment, the speeds of both pumps were recorded, and the
calibration curves were used to evaluate the flow-rate of the electrolytes. The raw
data acquired are shown in Supplementary Table 1.
Estimation of pumping power. The pressure drop ∆P experienced by a fluid as it
flows through a porous medium is given by Darcy’s law:
vλΔx
ΔP =
K
where ν is fluid velocity, λ is dynamic viscosity, ∆x is path length and K is
permeability. Converting the fluid velocity to volumetric flow-rate using the
cross-sectional area AC and multiplying both sides of the above equation with the
volumetric flow-rate V̇ to obtain the pumping power PPUMP gives the following:
V̇2 λΔx
PPUMP = ΔPV̇ =
AC K
The permeability for a porous electrode is given by the Kozeny–Carman
equation, given below, as a function of pore radius rp, electrode porosity ε and
Kozeny–Carman constant CKC:
4r2p ε2
K=
CKC (1 − ε)2
Using a pore radius of 10 μm, porosity of 0.7 and Kozeny–Carman constant
of 5.55, the above permeability is calculated to be 3.92 × 10–10 m2. This assumes
that the electrolyte takes the convoluted path of the serpentine graphite channels,
where each serpentine flow field has a total of 20 channels. The channels are 3 mm
wide and 3 mm deep. We conservatively estimate that the entire depth is occupied
by the porous electrode due to the compressive stress put into the electrochemical
cell assembly. In reality, there will be parallel pathways that reduce the pumping
demand, for example, the serpentine graphite channel area unobstructed by the
electrode or bypass electrolyte routes within the porous electrode that have very
short path lengths from the inlet to the outlet. The volumetric flow-rate of the
electrolyte is taken to be 1 ml s–1, and the dynamic viscosity of the electrolyte is
taken to be 8.9 × 10–4 Pa s. Using these parameters, the pumping power required to
drive the electrolyte through one porous electrode is estimated to be 0.4 W. Since
one pump drives the electrolyte through two porous electrodes in parallel, the total
power consumed by one pump is 0.2 W. We use two pumps to drive electrolyte
through two sets of porous electrodes, each in parallel; therefore, we estimate the
entire pumping power to be 0.4 W.
Uncertainty and error propagation. The electrolyte-specific cooling load QC,e at
thermal equilibrium (acquired at time teq) is given by the following expression:
QC,e = V̇e CV,e ΔTij (teq )
where V̇ is volumetric flow-rate, ∆Tij = Ti − Tj is the difference between the
electrolyte inlet and outlet temperatures of the cold electrochemical cell and
cooling load is then given by the following expression:
√(
)2 ( ) ( )
δQC,e δ V̇e δCV,e 2 δΔTij (teq ) 2
= + +
QC,e V̇e CV,e ΔTij (teq )
Nature Energy | www.nature.com/natureenergy
Nature Energy
The uncertainty in the equilibrium-time difference between the outlet and inlet
temperatures is given by the standard deviation of such differences acquired across
all repetitions and is given by the expression below:
� electrochemical heat engine and refrigeration efficiency. Trends Chem.
�  2
� N N https://doi.org/10.1016/j.trechm.2021.12.006 (2022).
� 1 � 1 �
δΔTij = � ΔTij,rep − ΔTij,rep 
N − 1 rep N rep
In the above equation, subscript rep represents each of N repetitions. The
uncertainty in the total cooling load, which is simply the sum of the electrolyte
contributions, is given by
√∑
δQC = δQ2C,e
e host–guest complexation and salt-induced crystallization. J. Am. Chem. Soc.
A similar approach is taken for rejected heat QH. The COP β is given by
the ratio of the total cooling load QC and the electrical work input WIN. The
uncertainty of the COP is then given by the corresponding formula for error
propagation:
√( ) ( ) 28. Whitman, B., Tomczyk, J., Johnson, B. & Silberstein, E. Refrigeration & Air
δβ δQC 2 δWIN 2
= +
β QC WIN
The uncertainty of the work input is given by the product of the current
(uncertainty of current regulated by the galvanostat is effectively zero) and the
uncertainty of the cell voltages.
Data availability
The data that support the results of this study are provided as Supplementary Data.
Source data are provided with this paper.
Received: 27 May 2021; Accepted: 17 December 2021;
Published: xx xx xxxx
References
1. Hansen, J. et al. Global temperature change. Proc. Natl Acad. Sci. USA 103,
14288–14293 (2006).
2. IPCC Climate Change 2014: Synthesis Report (eds Core Writing Team,
Pachauri, R. K. & Meyer, L. A.) (IPCC, 2014).
3. Ritchie, H. & Roser, M. Urbanization. Our World in Data https://
ourworldindata.org/urbanization (2018).
4. Efficient and Climate-Friendly Cooling (United Nations Environment
Programme, 2020); https://wedocs.unep.org/bitstream/handle/20.500.11822/
31587/ECFC.pdf?sequence=1&isAllowed=y
5. The Future of Cooling (International Energy Agency, 2018); https://www.iea.
org/futureofcooling/
6. IPCC. Safeguarding the Ozone Layer and the Global Climate System
(eds Metz, B. et al.) (Cambridge Univ. Press, 2005).
7. Velders, G. J. M., Fahey, D. W., Daniel, J. S., McFarland, M. & Andersen, S. O.
The large contribution of projected HFC emissions to future climate forcing.
Proc. Natl Acad. Sci. USA 106, 10949–10954 (2009).
8. Velders, G. J. M. et al. Preserving Montreal Protocol climate benefits by
limiting HFCs. Science 335, 922–923 (2012).
9. Chum, H. & Osteryoung, R. Review of Thermally Regenerative Electrochemical
Systems (Solar Energy Research Institute, 1981).
10. Hammond, R. H. & Risen, W. M. An electrochemical heat engine for direct
solar energy conversion. Sol. Energy 23, 443–449 (1979).
11. Lee, S. W. et al. An electrochemical system for efficiently harvesting
low-grade heat energy. Nat. Commun. 5, 3942 (2014).
12. Poletayev, A. D., McKay, I. S., Chueh, W. C. & Majumdar, A. Continuous
electrochemical heat engines. Energy Environ. Sci. 11, 2964–2971 (2018).
13. Kreysa, G. & Darbyshire, G. F. Theoretical consideration of electrochemical
heat pump systems. Electrochim. Acta 35, 1283–1289 (1990).
14. Dittmar, L., Jüttner, K. & Kreysa, G. in Electrochemical Engineering and
Energy (eds Lapicque, F., Storck, A. & Wragg, A. A.) 57–65 (Springer, 1995).
15. Gerlach, D. W. & Newell, T. A. Basic modelling of direct electrochemical
cooling. Int. J. Energy Res. 31, 439–454 (2007).
16. Newell, Ty. A. Thermodynamic analysis of an electrochemical refrigeration
cycle. Int. J. Energy Res. 24, 443–453 (2000).
17. Duan, Z. N., Qu, Z. G. & Zhang, J. F. Thermodynamic and electrochemical
performance analysis for an electrochemical refrigeration system based on
iron/vanadium redox couples. Electrochim. Acta 389, 138675 (2021).
18. McKay, I. S., Kunz, L. Y. & Majumdar, A. Electrochemical redox refrigeration.
Sci. Rep. 9, 13945 (2019).
19. Yang, Y. et al. Charging-free electrochemical system for harvesting low-grade
thermal energy. Proc. Natl Acad. Sci. USA 111, 17011–17016 (2014).
Nature Energy | www.nature.com/natureenergy
View publication stats
Articles
20. Henry, A. A new take on electrochemical heat engines. Joule 2,
1660–1661 (2018).
21. Rajan, A., McKay, I. S. & Yee, S. K. Electrolyte engineering can improve
22. Fogler, H. Essentials of Chemical Reaction Engineering 2nd edn (Pearson,
2017).
23. Marcus, R. A. & Sumi, H. Solvent dynamics and vibrational effects in
electron transfer reactions. J. Electroanal. Chem. Interfacial Electrochem. 204,
59–67 (1986).
24. Kim, T. et al. High thermopower of ferri/ferrocyanide redox couple in
organic-water solutions. Nano Energy 31, 160–167 (2017).
25. Zhou, H., Yamada, T. & Kimizuka, N. Supramolecular
thermo-electrochemical cells: enhanced thermoelectric performance by
138, 10502–10507 (2016).
26. Jacobs, S. et al. The performance of a large-scale rotary magnetic refrigerator.
N. Dev. Magn. Refrig. 37, 84–91 (2014).
27. Energy Savings Potential and RD&D Opportunities for Non-Vapor-Compression
HVAC Technologies (US DOE-BTO, 2014); https://doi.org/10.2172/1220817
Conditioning Technology (Cengage Learning, 2013).
29. Engelbrecht, K. et al. Experimental results for a novel rotary active magnetic
regenerator. Int. J. Refrig. 35, 1498–1505 (2012).
30. Atta, R. M. Solar thermoelectric cooling using closed loop heat exchangers
with macro channels. Heat Mass Trans. 53, 2241–2254 (2017).
31. Lima, A. A. S. et al. Absorption refrigeration systems based on ammonia as
refrigerant using different absorbents: review and applications. Energies 14,
48 (2021).
32. Feng, Q. et al. Development of a metal hydride refrigeration system as
an exhaust gas-driven automobile air conditioner. Renewable Energy 32,
2034–2052 (2007).
33. Thermodynamic Properties of DuPont Suva 410A Refrigerant (Chemours,
2019); https://www.freon.com/en/-/media/files/freon/freon-410a-si-
thermodynamic-properties.pdf?rev=6b72bfaa299142d697540982b88a56eb
34. Pecharsky, V. K. & Gschneidner, K. A. Jr Giant magnetocaloric effect in
Gd5(Si2Ge2). Phys. Rev. Lett. 78, 4494–4497 (1997).
35. Worswick, R. D., Dunn, A. G. & Staveley, L. A. K. The enthalpy of solution of
ammonia in water and in aqueous solutions of ammonium chloride and
ammonium bromide. J. Chem. Thermodyn. 6, 565–570 (1974).
36. Kim, J. H. et al. Iron (II/III) perchlorate electrolytes for electrochemically
harvesting low-grade thermal energy. Sci. Rep. 9, 8706 (2019).
37. Duan, J. et al. Aqueous thermogalvanic cells with a high Seebeck coefficient
for low-grade heat harvest. Nat. Commun. 9, 5146 (2018).
38. Makarov, D. M. & Egorov, G. I. Density and volumetric properties of the
aqueous solutions of urea at temperatures from T=(278 to 333) K and
pressures up to 100MPa. J. Chem. Thermodyn. 120, 164–173 (2018).
Acknowledgements
A.R. was partially supported by the Office of Naval Research (award no.
N00014-19-1-2162).
Author contributions
A.R. and S.K.Y. conceptualized the work. A.R. developed the methodology,
performed the experiments, analysed the data and wrote the original draught.
A.R., I.S.M. and S.K.Y. reviewed and edited the draught. S.K.Y. supervised the research
and acquired funding.
Competing interests
A.R. and S.K.Y. are inventors on a US patent (application no. 63/172,925, patent pending)
detailing the architecture, chemistry and operation of the BECR. I.S.M. declares no
competing interests.
Additional information
Supplementary information The online version contains supplementary material
available at https://doi.org/10.1038/s41560-021-00975-7.
Correspondence and requests for materials should be addressed to Shannon K. Yee.
Peer review information Nature Energy thanks Shien-Ping Feng, Neil Mathur,
Nini Pryds and the other, anonymous, reviewer(s) for their contribution to the peer
review of this work.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature Limited 2022