system was only stable for a few minutes and
exhibited high cell potentials ( 12 ). Instabil-
ity of this process was already described by
Tsunetoet al.( 22 ), and recently mitigated by
Andersenet al. with a potential cycling strat-
egy ( 24 ), which enabled stability over several
days and increased the FE at 10-bar N 2 from
20% without cycling to 37% with cycling at an
energy efficiency (EE) of 7%. However, the
REFUEL program of the US Department of
Energy set a target of 90% FE at 300 mA/cm^2
and an EE of 60% ( 30 ). The current state of the
lithium-mediated process is clearly far from
this target. Especially the EE is currently a
major problem in the LiNR, because Li plating
requires largely negative potentials (−3.04 V
versus SHE). The overpotential losses of the
LiNR are portrayed and discussed in fig. S1. If
all overpotentials are minimized and hydro-
gen oxidation reaction (HOR) is utilized at the
counter electrode (CE), the resulting EE would
be 26%, assuming 80% FE.
Here we show that adding small amounts of
O 2 to the feed gas has a positive effect on both
the FE and the stability of the system. A FE of
up to 78.0 ± 1.3% at 20-bar N 2 can be achieved
by adding 0.5 to 0.8 mol % O 2 , resulting in an
EE of 11.7 ± 0.5% (calculation in supplemen-
tary materials). This EE calculation accounts
for neither EtOH as sacrificial proton donor
nor for the energy of pressurizing the system.
We used this framework as a basis for com-
parison to results from literature ( 24 , 25 ).
Using hydrogen oxidation at the CE in a GDE-
type system has previously been shown to
prevent solvent oxidation ( 12 ), and the use
of a phosphonium proton shuttle has been
experimentally verified to successfully shuttle
the newly created protons to the working elec-
trode while the phosphonium ylide becomes
reprotonated at the anode ( 11 ), thereby cir-
cumventing the sacrificial ethanol issue. Our
focus in this study lies on the finding that
oxygen increases the FE and not on the sac-
rificial proton donor issue, but we expect that
the previously reported solutions will also
apply to our system. The positive effect of
small amounts of oxygen is counterintuitive;
one would expect oxygen to contaminate the
active phase and result in loss of efficiency be-
cause of the oxygen reduction reaction (ORR).
Indeed, early experiments by Tsunetoet al.
showed that higher oxygen content in the gas
feed (synthetic air with 20% O 2 ) lowered the
FE considerably from ~50 to 0.1% at 50 bar ( 23 ).
We investigated the origin of the oxygen pro-
motion using a microkinetic model, suggest-
ing that the higher FE is related to slower Li+
diffusion through the SEI layer formed in the
presence of O 2 as observed by Wanget al. in
studies of Li–air battery materials ( 31 ). We
further tested this hypothesis by systemat-ically investigating the effect of O 2 on the SEI
layer through x-ray diffraction (XRD) and
x-ray photoelectron spectroscopy (XPS).Effect of oxygen on Faradaic efficiency
Systematic experiments with varying amounts
of O 2 added at 10- and 20-bar total gas pres-
sure were conducted. All experiments were
performed in an in-house–fabricated auto-
clave based on the design by Wiberget al.( 32 )
and carried out in a glass cell containing
~30 ml of electrolyte (fig. S2). The gases used
for the experiments were cleaned with purifiers
(NuPure) to levels as low as parts per trillion
by volume of nitrogen-containing impurities,
and the flow was controlled by mass flow con-
trollers (Brooks) to ensure proper gas mixing.
This was important, as improper gas mixing
affects the electrochemical reaction (figs. S3
and S4). The oxygen content was determined
by a mass spectrometer (Pfeiffer, omnistar GSD
320) attached to the autoclave with a 1-mm
orifice just above the electrochemical cell.
A representative mass spectrum is shown in1594 24 DECEMBER 2021•VOL 374 ISSUE 6575 science.orgSCIENCE
Fig. 2. Influence of oxygen content on the stability of the Li-mediated ammonia synthesis.
(A) Chronopotentiometry with−4 mA/cm^2 applied at 20 bar N 2 with variable O 2 content. The potentials
are normalized to the starting WE potential. A maximum of 50 C charge (black trace) is passed; however,
some measurements overloaded before 50 C was reached, as seen from the sudden decrease in WE
potential (blue traces) for low O 2 concentrations (all shown % O 2 are mol % O 2 ). A representative CE
potential (green trace) is shown, which is stable throughout the measurement, even for experiments
in which the WE overloaded. The dotted red line indicates the decrease of WE potential by 1 V. (B) The time
until the WE potential decreased by 1 V (tstable), as a function of O 2 content. All stable experiments were
included in 116+ min.Fig. 1. Influence of oxygen content on the FE of
the Li-mediated ammonia synthesis.Reactions
were conducted at 10 (green) and 20 (blue)
bar, where the added oxygen content inside the
autoclave was determined by mass spectrometry.
The reported pressure is the total pressure of
N 2 and O 2 combined. All experiments were operated
at a constant current of−4 mA/cm^2 until the
system overloaded or reached 50 C (116 min to
pass 50 C), unless stated otherwise.
RESEARCH | RESEARCH ARTICLES