of 21
1
Hydrodynamics Change Tafel Slopes in Electrochemical CO
2
Reduction on Copper
Nicholas B. Watkins,
†a,b
Zachary J Schiffer,
†a,c
Yungchieh Lai,
a,c
Charles B. Musgrave III,
a,c
Harry A. Atwater,
a,c
William A. Goddard III,
a,b,c
Theodor Agapie,
a,b,
* Jonas Peters,
a,b,
* John M. Gregoire
a,c,
*
a
Liquid Sunlight Alliance, California Institute of Technology, Pasadena, CA, 91125, USA.
b
Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, CA, 91125, USA
c
Division of Engineering and Applied Science, California Institute of Technology, Pasadena, CA, 91125, USA
*Email:
agapie@caltech.edu,
jpeters@caltech.edu,
gregoire@caltech.edu
These authors contributed equally to this work.
Supplementary Information
Table of Contents
Materials and methods and synthetic procedures
2
Electrochemical measurements
3
COMSOL modeling information and LOWESS line fitting
4
Figure S1:
Cell designs
5
Figure S2:
Boundary layer determination
Table S1:
Boundary layer values
6
Figure S3:
CO
2
reduction partial current densities at different mass transport
7
Figure S4:
Potential-dependent selectivity data for all cells
8
Figure S5:
Methane Tafel plots and probability density curves
9
Figure S6-S9:
Ethylene full probability distributions
10-11
Figure S10-S13:
Methane full probability distributions
12-13
Figure S14:
0.25 M KHCO
3
Bayesian Statistics Summary for Ethylene and Methane
14
Figure S15-16:
Molecular dynamics discussion and calculations
15-16
Table S2:
CO
2
reduction selectivities and activities
17-20
References
21
2
Materials and Methods
All solvents and reagents were obtained from commercial sources and used as received, unless stated
otherwise. Cu foils (99.999% Cu, 25 mm × 25 mm × 0.5 mm), potassium carbonate (99.995%), phenanthroline
(≥99%),
and dibromoethane
(≥
99%) were purchased from Sigma Aldrich. Carbon rods (99.999% C) were purchased
from Strem Chemicals. Platinum foil (99.99% Pt, 25 mm × 25 mm × 0.05 mm), was purchased from Alfa Aesar. Natural
abundance CO
2
(Research grade) was purchased from Airgas. Water was purified by a Nanopure Analytical Ultrapure
Water System (Thermo Scientific) or a Milli-Q Advantage A10 Water Purification System (Millipore) with specific
resistance of 18.2
MΩ·cm
at 25 °C.
Prior to each use, copper foil was mechanically polished to a mirror-like finish using nanodiamond
suspension (first 3
μm
then 0.1
μm,
Buehler) followed by rinsing in water and drying under a stream of nitrogen gas.
The copper foil was then electropolished using a method similar to the one employed by Kuhl
et al
.: In a 85%
phosphoric acid bath, +2.1 V versus a carbon rod counter electrode was applied to the Cu foil for 5 minutes and the
foil was subsequently washed with copious amounts of ultra-pure water and dried under a stream of nitrogen gas.
Prior to each use, platinum foil was washed with water and flame-annealed using a butane torch for 10 s. CO
2
-
saturated potassium bicarbonate electrolyte (KHCO
3
, 0.1 M) was prepared by sparging an aqueous solution of
potassium carbonate (K
2
CO
3
, 0.05 M) with CO
2
for at least 1 h prior to electrolysis.
Cu thin film electrocatalysts were fabricated using DC magnetron sputtering of a 2” Cu metal target at 50 W
in 6 mTorr Ar onto a 100 mm-diameter Si wafer with an approximately 170 nm SiO2 diffusion barrier and 10 nm Ti
adhesion layer, using a previously described sputter system with 10-5 Pa base pressure.
1
After deposition, the films
were stored in a nitrogen purge box until the day of electrochemical testing, although no other catalyst treatment
was performed prior to electrocatalyst screening. The Cu-X (X: Co, Zn, Mn, In) thin film electrodes were deposited
under similar conditions from elemental metal targets with DC power adjusted to obtain designed composition in
the wafer center. All the metal targets were pre-cleaned in the presence of 6 mTorr Ar for 10 min to remove any
contaminants from the target surface. The non-confocal sputtering geometry provided a continuous composition
gradient across the Si wafer with the composition variation within each 5 mm diameter electrode being less than 1%
for the most Curich catalysts and about 2% for the most Cu-poor catalysts.
Synthetic Procedures
Synthesis of N,N’-ethylene-phenanthrolinium dibromide (
1-Br
2
) In a round bottom flask charged with a
magnetic stir bar, phenanthroline (500 mg, 2.8 mmol, 1 equiv.) was dissolved in dibromoethane (5 mL, 67.4 mmol, >
24 equiv.) and the final mixture was heated to 110 °C for 18 h. The precipitate formed was collected by filtration and
washed with hexane (3 x 10 mL) and acetone (3 x 10 mL) to afford the final product. Yield: 970 mg (94 %, 2.6 mmol).
1H and 13C NMR spectra were in accordance with reported values.
2
3
Electrochemical Measurements
All H-cell electrochemical experiments were carried out using a Biologic VMP3 multichannel potentiostat
with
copper foil as the working electrode and a platinum foil as the counter electrode. The cathode compartment was
separated from the anode compartment by a Selemion AMV anion-exchange membrane (AGC Engineering Co.). All
potentials were measured versus a leakless Ag/AgCl reference electrode (Innovative Instruments) with an outer
diameter of 5 mm that was inserted into the cathode compartment. The reference electrode was calibrated against
H
+
/H
2
on Pt in a 0.5 M sulfuric acid solution (0 V vs. standard hydrogen electrode).
Potentiostatic electrochemical impedance spectroscopy (PEIS) measurements were carried out prior to
each electrolysis experiment to determine the Ohmic resistance of the flow cell. The impedance measurements were
carried out at frequencies ranging from 100 Hz to 200 KHz to measure the solution resistance. A Nyquist plot was
plotted and in the high-frequency part a linear fit was performed, and the axis intersection was identified, the value
of which represents the Ohmic resistance of the cell. Typical values of the resistance range from 45 to 60
Ω.
All chronoamperometric experiments (unless stated otherwise) were performed for 30 min at 25 °C using CO
2
-
saturated 0.1 M KHCO
3
as electrolyte. The potentiostat was set to compensate for 85 % of the Ohmic drop, with the
remaining 15 % being compensated for after the measurements. The effluent gas stream coming from the flow cell
(10 mL/min) was flowed into the sample loops of a gas chromatograph (GC-FID/TCD, SRI 8610C, in Multi Gas 5
configuration) equipped with a HayeSep D column. Methane, ethylene and carbon monoxide were detected by a
methanizer-flame ionization detector (FID) and the hydrogen was detected by a thermal conductivity detector (TCD).
Every 10 minutes, 1 mL of gas was sampled to determine the concentration of gaseous products. Liquid products
were quantified by HPLC. Liquid products were only quantified in ANEC because the concentrations in the
recirculation H-cells were too low due to the requisite volume for recirculation.
4
COMSOL Modeling Information
The governing equations (material balances and fluid dynamics) were used within the Multiphysics Module and were
solved with the general solver in COMSOL Multiphysics 5.6 with default tolerances. The modeling domain was
discretized with a nonuniform physics-controlled mesh generated by COMSOL. To aid with convergence, all fluid
dynamics systems were first solved using the turbulence
model, then the result of that solution was used with
-
ε
the Low Re
turbulent model, and finally the results of the Low Re model were used as initial conditions for a
-
ε
laminar fluid dynamics model. Inlet velocity boundary conditions were set to fully developed flow with the flow rate
set to experimental values. Outlet boundary conditions were set to fully developed flow with an average pressure of
0 Pa. For the flux calculations, the concentration at the electrode was set to 1 mM and the inlet concentration was set
to 0 mM. The diffusion constant was set to 2.7E-9 m^2/s.
3
LOWESS Fitting Function
For Figures 3 and 5 in the main text, lines to guide the eye and provide a continuous function of the data are
plotted. These lines are calculated via a Locally Weighted Scatterplot Smoothing (LOWESS) model in Python.
Essentially, for each x-value in the dataset, a “smoothed” y-value is calculated by taking a weighted linear fit of
the nearest
n
data points. This value of
n
is, by default in the Python implementation,
of the total data points.
The weights for the linear fit are from a tricube function. Subsequent iterations of the refitting can be done with
altered
weights
according
to
the
residuals
of
the
previous
fitting.
See
Python’s
statsmodels.nonparametric.smoothers_lowess.lowess function for details on implementation and usage. In the
case of Figure 3, in addition to a fitting, a qualitative metric for error bars on the data is shown. This error was
calculated by taking a random sample of data points and then fitting the LOWESS to that sample. The average
of 500 such samples is taken to represent the “error” qualitatively of the fit. The following Python code was
used to generate the error from this fitting. Note that although this fitting procedure is “model-free”, there are
still parameters that we selected, including number of iterations, fraction of data points to fit, etc. Changing
these parameters will quantitatively change the fit (primarily the error bars), but will not qualitatively change
the results.
For Figure 5, the Faradaic efficiencies for the various gaseous products were fit using the LOWESS model as
implemented in Python. The raw data points are summed up and plotted, i.e., first the FE toward methane is
plotted, then the sum of FEs toward methane and ethylene is plotted as “ethylene”, etc. A similar procedure is
used to sum up the smoothed fits and label them on the plot.
5
Cell Design
Figure S1: a)
A schematic of ANalytical and ElectroChemistry (ANEC), which is an analytical electrochemistry
system previously published by our group that can efficiently detect a wide range of CO2R products.
4
b)
A
schematic of the sparged H-cell used by Kuhl
et al.
, whose schematic is reproduced herein.
5
c)
The H-cell used
for parallel flow experiments herein is a modification of that used
by Ager
et al.
6
d)
The angled H-cell is similar
to that of the cell designed by Ager
et al.
, but with the bottom inlet adjusted to face the electrode surface at a 20º
angle.
6
Boundary Layer Experiments
Figure S2:
These experiments were performed according to section SI-4 from Clark et al.
7
Sample data with ANEC,
where a) a CV is initially performed to determine the mass transport limited regime, and then a potential is chosen
(indicated by the dashed line) for subsequent constant potential experiments to determine the boundary layer at
different recirculation rates. c) The boundary layer was calculated using the equation below, where F is Faraday’s
constant,
￿[Fe(￿￿)
6
3−
] is the diffusivity of the ferricyanide ion ( 0.720×10-5 cm
2
s
-1
),
￿*[Fe(￿￿)
6
3−
] is the
concentration of ferricyanide ion in the bulk of the electrolyte (10 mM), and i
SS
is the steady-state current.
Table S1
: Cells and their associated boundary layer thicknesses.
Cell Geometry (flow rate)
Experimental Boundary
Layer Thickness (μm)
COMSOL Boundary Layer
Thickness (μm)
Ethylene Tafel slope
(mV/dec)
Ager et al. H-cell (20 sccm)
177.9 ± 21.6*
99.0
Parallel H-cell (280 μL/s)
106.9
242
94.9
Angled H-cell (280 μL/s)
33.4
128
66.1
ANEC (140 μL/s)**
36.2
57
39.2
* The average and associated error for 6 separate experiments
** 140 μL/s is the typical flow rate for ANEC experiments
7
CO
2
reduction partial current densities at different mass transport
a.
b.
Figure S3:
The partial current density towards CO
2
reduction products does not greatly change between a) cells at
different convective ratios or b) with the addition of
1-Br
2
. The parallel and angled cells have slightly lower current
densities as compared to ANEC or Kuhl, but are comparable to that of sparged H-cell Cu control data published
previously in our own group.
2
This change may be attributable to the type of polycrystalline copper used in each
experimentation.
8
8
Potential-dependent selectivity data for all cells
Figure S4:
The distribution of gaseous products are shown as stack plots for the a) sparged H-cell, b) parallel H-cell,
c) ANEC, and d) angled H-cell. The product distributions are shown with smooth trend lines calculated using LOWESS
(see page 4). Each data point corresponds to an individual experiment, with the exception of the Sparged H-cell data,
reproduced from [5], which is an average of three experiments.
9
Methane Bayesian Statistics Summary
Figure S5:
Tafel plots for a) ethylene from ANEC and Kuhl, et al.
5
A small plateau exists on the bottom of each ANEC
plot due to the noise floor of the instrument. All ANEC data points are single experiment values whereas Kuhl, et al.
are the averages of three independent experiments. b) Probability density curves for ethylene Tafel slope values
using Bayesian statistics for each cell geometry investigated.
Discussion Regarding Methane vs Ethylene Tafel
In Surendranath et al., their analysis indicates that ethylene will have a second order dependence at low CO coverage
and zeroth order at high CO coverage, and that methane will have a limiting-to-zero dependence at high CO coverage
and first order dependence at low CO coverage.
9
Based on their experimental results, their electrochemical
conditions of an ethanol-based electrolyte at -35
°C
provides a sufficiently high CO coverage at all pCO to realize the
zeroth-order relationship for ethylene and limiting-to-zero dependence for methane. With high convection, we
believe that we are operating in a regime with low surface coverage of CO due to the mass transport removing CO
from the surface; in this regime, we expect there to be a second order dependence on CO for ethylene and first order
for methane.
10
Ethylene full Bayesian probability distributions
Figure S6
. Bayesian analysis of the data from Kuhl et al. Data and mean
a posteriori
(MAP) fit are shown on left.
Probability distributions for the slope and limiting current are shown with MAP drawn.
Figure S7
. Bayesian analysis of the data from parallel flow H-Cell and mean
a posteriori
(MAP) fit are shown on left.
Probability distributions for the slope and limiting current are shown with MAP drawn.
11
Figure S8
. Bayesian analysis of the data from angled flow H-Cell and mean
a posteriori
(MAP) fit are shown on left.
Probability distributions for the slope and limiting current are shown with MAP drawn.
Figure S9
. Bayesian analysis of the data from ANEC cell and mean
a posteriori
(MAP) fit are shown on left.
Probability distributions for the slope and limiting current are shown with MAP drawn. Not shown is the
probability distribution for the plateau at the bottom. This plateau is a simple “kink” in the system representing the
lower detectability limit of the instruments.
12
Methane full Bayesian probability distributions
Figure S10
. Bayesian analysis of the data from Kuhl et al. Data and mean
a posteriori
(MAP) fit are shown on left.
Probability distributions for the slope and limiting current are shown with MAP drawn.
Figure S11
. Bayesian analysis of the data from parallel flow H-Cell and mean
a posteriori
(MAP) fit are shown on
left. Probability distributions for the slope and limiting current are shown with MAP drawn.
13
Figure S12
. Bayesian analysis of the data from angled flow H-Cell and mean
a posteriori
(MAP) fit are shown on
left. Probability distributions for the slope and limiting current are shown with MAP drawn.
Figure S13
. Bayesian analysis of the data from ANEC cell and mean
a posteriori
(MAP) fit are shown on left.
Probability distributions for the slope and limiting current are shown with MAP drawn. Not shown is the
probability distribution for the plateau at the bottom. This plateau is a simple “kink” in the system representing the
lower detectability limit of the instruments.
14
0.25 M KHCO
3
Bayesian Statistics Summary for Ethylene and Methane
Figure S14:
Tafel plots for a) ethylene and b) methane from ANEC with 0.25 M KHCO
3
with their mean
a posteriori
(MAP) fit shown on left. Probability distributions for the slope are shown with MAP drawn on the right.
15
Molecular Dynamics Discussion
All molecular dynamics calculations were performed using the Large-scale Atomic/Molecular Massively
Parallel Simulator (LAMMPS) software.
10
Valence (bond, angle, dihedral), electrostatic, and van-der-waals
potentials were modeled by the Universal Force Field (UFF).
11
We began all simulations by a steepest descent
minimization followed by a conjugate gradient minimization. The cell was then heated at constant volume (NVT
ensemble) from 1 K to the desired temperature over the period of 10 ps via the Nose-Hoover thermostat. The
cell was then maintained at the desired temperature (again NVT) for 2 ns to allow the system to reach
equilibrium.
After 2 ns of constant-temperature NVT dynamics, the 2-Phase Thermodynamics (2PT) method was
used to calculate CO diffusion coefficients (D
CO
).
12
In essence, 2PT calculates the velocity autocorrelation
function (VACF) and then integrates the VACF over time to yield D
CO
. The VACF was integrated over a period of
20 ps in order to achieve proper convergence. For all cases, D
CO
was averaged over 6 individual calculations to
ensure adequate sampling.
Pure water systems included 282 water molecules and a single CO molecule. The volume was chosen
to match the experimental density of pure water. Systems with additive featured 8 additive molecules, 200
waters, and a single CO molecule. Here the volume was kept the same as the system with no additive.
Figure S15:
Simulation boxes for pure water (left) and additive (right) systems. White atoms are hydrogen,
red are oxygen, grey are carbon, and blue are nitrogen. CO molecules are colored green for clarity.
16
Figure S16:
CO diffusion coefficients calculated at three temperatures (283, 293, 303) with and without the
N,N’-ethylene-phenanthrolinium dibromide
dimer (Add) present inside the box. The pure water calculations
match the experimental reference, whereas the additive/water mixture shows a significant decrease in
diffusion coefficient.
17
Table S2: All Faradaic efficiencies and currents
FE (%)
Cell
V
<I> (mA)
H
2
CO
CH
4
C
2
H
4
HCOOH
ACETATE
EtOH
1-PrOH
sum
-0.84
-2.33
66.2
0
0
0
26.7
0.0
0.0
0.0
92.9
-0.84
-2.40
63.1
0.6
0
0
28.6
0.0
0.0
0.0
92.3
-0.84
-2.48
62.1
0.7
0
0
18.4
0.0
0.0
0.0
81.2
-0.85
-2.26
59.4
2.4
0.1
0
15.4
0.0
0.0
0.0
77.3
-0.85
-2.38
63.5
3.9
0
0
8.6
0.0
0.0
0.0
76.0
-0.89
-2.89
92.3
5.7
0
0
20.5
0.0
0.0
0.0
118.5
-0.92
-2.76
79.6
6.3
0
0
31.7
0.0
0.0
0.0
117.6
-0.92
-3.47
59.6
2.2
0
0
30.1
0.0
0.0
0.0
91.9
-0.92
-3.38
69
3.7
0
0
27.3
0.0
0.0
0.0
100.0
-0.92
-3.48
56.3
3.2
0
0
31.1
0.0
0.0
0.0
90.6
-0.93
-3.08
51
6.1
0.1
0
26.6
0.0
0.0
0.0
83.8
-0.93
-3.04
47.3
7.4
0.1
0.1
31.3
0.0
0.0
0.0
86.2
-0.95
-3.56
49.6
4.5
0.1
0
24.5
0.0
0.0
0.0
78.7
-0.96
-3.09
64.3
8.2
0.2
0.2
29.0
0.0
0.0
0.0
101.9
-0.96
-2.96
61.8
10.2
0.3
0.3
30.8
0.0
0.0
0.0
103.4
-0.97
-3.22
60.1
9.1
0.3
0.3
37.5
0.0
0.0
0.0
107.3
-0.97
-3.53
44.3
5.6
0.3
0.2
35.1
0.0
0.0
0.0
85.5
-0.98
-3.10
47.4
10.3
0.6
0.6
36.5
0.0
0.0
0.0
95.4
-0.99
-3.72
39
7.2
1.3
0.8
32.0
0.0
0.0
0.0
80.3
-1.00
-4.34
33.4
12.2
4.1
3.3
39.8
0.0
1.4
0.7
94.9
-1.01
-4.13
42.1
6.5
2.7
1.8
32.3
0.0
0.0
0.0
85.4
-1.01
-3.94
42.7
6.4
4.5
2
34.0
0.0
0.0
0.9
90.5
-1.02
-3.81
35.8
7.9
2.7
1.6
36.1
0.0
0.0
0.0
84.1
-1.02
-3.72
37.2
11.9
3.9
1.9
18.3
0.0
5.3
1.8
80.2
-1.05
-6.81
20.7
3.1
21.1
12
22.1
0.0
3.9
6.6
89.5
-1.05
-6.50
26.6
3.8
20.2
6.7
28.6
0.0
0.7
1.2
87.8
-1.06
-6.42
26.8
10.5
12.4
6.7
29.7
0.0
0.0
2.5
88.6
-1.07
-6.03
28.3
6.8
12.9
5.1
26.3
0.0
0.9
2.1
82.4
-1.07
-5.64
28.4
10.8
16.5
4.8
13.7
0.0
6.0
1.6
81.8
-1.08
-9.94
15.1
1.9
28.7
14.9
14.2
0.0
4.4
5.8
85.0
-1.08
-9.81
22.3
2
31.3
8.9
14.6
0.0
2.8
1.0
83.0
-1.11
-8.76
23.2
8
19.7
7.7
20.7
0.0
0.0
3.0
82.3
-1.11
-8.76
24.4
4.5
22.5
6.5
22.5
0.0
1.4
0.5
82.3
ANEC
-1.11
-8.52
27
9
28.3
6.2
23.4
0.0
1.2
2.5
97.6
-0.78
-2.26
61.9
0.4
0
0
14.0
0.0
0.0
0.0
76.2
-0.86
-3.22
62.8
0.6
0
0
22.2
0.0
0.0
0.0
85.5
-0.91
-3.30
71.2
0.9
0
0
28.3
0.0
0.0
0.0
100.4
-0.96
-3.97
63
1.4
0.2
0
31.6
0.0
0.0
0.0
96.2
-0.99
-5.29
61.6
1.8
1.3
0.3
30.7
0.0
0.0
0.0
95.6
-1.02
-6.92
63.5
1.7
4.5
0.6
30.1
0.0
0.0
0.0
100.3
-1.05
-8.79
63.7
1.1
8.2
0.6
25.3
0.0
0.0
0.0
98.9
-1.07
-11.98
64.1
0.6
12.3
0.5
23.2
0.0
0.0
0.0
100.6
ANEC 0.25
M KHCO
3
-1.13
-15.71
74.4
0.4
14.7
0.3
17.7
0.0
0.0
0.0
107.5
ANEC 1-Br
2
-0.75
-0.38
52.4
0
0
0
19.3
0.0
0.0
0.0
71.7