Chemical Engineering Journal 456 (2023) 141035

Contents lists available at ScienceDirect

Chemical Engineering Journal

journal homepage: www.elsevier.com/locate/cej

Process-informed adsorbent design guidelines for direct air capture

John Young a, Fergus Mcilwaine a, Berend Smit b, Susana Garcia a, Mijndert van der Spek a
a
Research Centre for Carbon Solutions, Heriot-Watt University, Edinburgh EH14 4AS, United Kingdom
b
Laboratory of Molecular Simulation (LSMO), Institut des Sciences et Ingénierie Chimiques, Ecole Polytechnique Fédérale de Lausanne (EPFL), Rue de l’Industrie 17, CH-
1951 Sion, Switzerland

A R T I C L E I N F O A B S T R A C T

Keywords: Direct air capture using solid adsorbents is a proven technology critical to reducing our net greenhouse gas
Direct air capture emissions to zero and beyond. Currently, academic research into the technology mainly focuses on the devel­
Sorbent design opment of new adsorbents. However, there is a discord between the adsorbent design and process performance.
Adsorption
Many materials scientists focus on maximising metrics such as the CO2 capacity of their adsorbent. Here, we
Global sensitivity analysis
Machine learning
combine detailed process modelling, machine learning, and extensive global sensitivity analysis, which entails
Process modelling varying all of the model parameters together, on a direct air capture process to show that the dry CO2 adsorption
capacity does not influence process performance for an amine-functionalised adsorbent operating in a temper­
ature vacuum swing adsorption (TVSA) process, while it is important in a steam-assisted TVSA (S-TVSA) process.
In fact, adsorption kinetics, density, and thermal conductivity are all critical attributes to obtaining a low energy
penalty and reduced costs. The analysis also highlights the importance of heat transfer, directing process engi­
neers to (alternative) adsorber designs that maximise this. By an in-depth evaluation of how process performance
indicators are affected by materials properties and process operating parameters, this work provides guidance to
both material scientists and process engineers towards the design of a “unicorn adsorbent” and intensified DAC
processes. This will improve the performance of solid adsorbent direct air capture and help drive down the costs
of this vital technology to avert the worst impacts of climate change.

1. Introduction multi-component co-adsorption of CO2 and H2O on a benchmark amine-

The importance of limiting global warming to 1.5 ◦ C has been
highlighted by the Intergovernmental Panel on Climate Change (IPCC)
to prevent the worst impacts of climate change, including loss of species,
extreme weather events, and displacement of people [1,2]. Recently, the
IPCC has shown that to limit the earth’s warming to these maximum
temperature rise targets, we may have to reach net-zero CO2 emissions
as soon as 2040, and net negative CO2 emissions will be required beyond
this [3]. To reach net-zero and achieve net-negative emissions, it is
imperative that carbon dioxide removal (CDR) becomes available at
scale. In this work, we study solid adsorbent direct air capture (DAC)
employing amine-functionalised adsorbents in a temperature vacuum
swing adsorption (TVSA) process, with and without a steam purge to
regenerate the adsorbent [4].
Much solid adsorbent DAC research is focused on developing better
adsorbents, such as amine-grafted or impregnated silica, alumina, cel­
lulose, or metal–organic frameworks (MOFs), mostly optimising the CO2
uptake capacity [5–19]. In complement, we previously investigated the
functionalised adsorbent and proceeded to model and optimise its per­
formance in a cyclic TVSA process model, providing a benchmark of the
energy requirements of solid sorbent DAC [20,21].
The next critical step is to unravel how DAC performance can be
improved, leading to two key questions: i) “Which adsorbent parameters
critically impact process performance?” and ii) “What process parame­
ters and conditions critically impact process performance?”. The an­
swers to these questions are essential to guide future material scientists
and process engineers in developing improved adsorbents and an opti­
mised process.
To answer either question in full, we must go beyond existing process
modelling work [4,20,22,23], and investigate the whole material and
process design space simultaneously since many complex interactions
exist between all the different variables, making adsorption processes
highly non-linear. Here, we do this using global sensitivity analysis,
which varies all the parameters of interest within a process model at
once, maps the impact of variations in model inputs on model outputs,
and then determines which of these inputs contributes most to the

E-mail addresses: jpy1@hw.ac.uk (J. Young), f.mcilwaine@hw.ac.uk (F. Mcilwaine), berend.smit@epfl.ch (B. Smit), s.garcia@hw.ac.uk (S. Garcia), m.van_der_
spek@hw.ac.uk (M. van der Spek).

https://doi.org/10.1016/j.cej.2022.141035
Received 25 October 2022; Received in revised form 15 December 2022; Accepted 16 December 2022
Available online 17 December 2022
1385-8947/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Young et al.
Fig. 1. a) Process cycle diagram of a TVSA process using heating only via heat exchange pipes within the adsorbent bed. b) Process cycle diagram of an S-TVSA
process that uses a steam purge during the desorption step. Both figures have been adapted from Young et al. 2021 [20].
variance of the outputs (performance indicators) [24–27].
Performing global sensitivity analysis on an adsorption process is
non-trivial. Many input parameters exist, and many simulations (here,
up to 2.5 million) are required. Adsorption process models solve sets of
complex partial differential equations and simulate many cycles until
cyclic steady state is reached [28,29], and one simulation in our process
model may take up to 10 min on a desktop computer, making it infea­
sible to run a global sensitivity analysis of 2.5 million simulations. One
approach to reducing this computational time could be to leverage
shortcut models [30–32], but some descriptors from the more detailed
physical models (e.g., mass and heat transfer) are currently lost when
using these. It has been suggested that adsorbents with slow adsorption
kinetics, such as amine-functionalised adsorbents, should not be inves­
tigated using these models [33].
An alternative to shortcut process models is to use machine learning
surrogate models [34]. They can take the same inputs as the detailed
model and learn how to predict the outputs of the detailed model if
sufficient training data is provided. For example, ~30,000 detailed
model runs can be used to train a machine learning model that allows
you to perform millions of runs for global sensitivity analysis in just a
few seconds. Our synergistic approach to varying material and process
parameters together would ultimately not be possible without machine
learning. Meanwhile, simplifying the problem by fixing process pa­
rameters whilst varying material parameters could lead to false con­
clusions on the importance of each material parameter.
In this study, our main aim is to uncover which adsorbent parameters
contribute most to the technical performance of a DAC process
employing amine-functionalised adsorbents from a process perspective,
providing a set of process-informed adsorbent design guidelines to ma­
terial scientists. Additionally, our secondary objective is to analyse what
aspects of the process should be advanced to maximise the potential of
an optimised adsorbent. Our analysis will show that many adsorbent and
process parameters have key roles to play, but the CO2 adsorption ca­
pacity, for which newly developed adsorbents are commonly optimised,
is not among them when operating in a TVSA process, as the working
capacity of the adsorbent becomes limited by the increased CO2 partial
pressures during desorption. However, the CO2 adsorption capacity is
still key in an S-TVSA process, as the CO2 partial pressure is decreased to
2
Chemical Engineering Journal 456 (2023) 141035
virtually zero by the steam purge during desorption. Critical parameters
to either process include adsorption kinetics, water adsorption, thermal
conductivity, and density, based on which we provide the fingerprint of
“unicorn” DAC adsorbents and suggest a testing protocol for newly
developed sorbents to critically include these parameters.
2. Modelling approach
The modelling approach employed here combined detailed process
modelling with machine learning and global sensitivity analysis. This
approach has not been used for direct air capture adsorption processes
before and was needed to fulfil this study’s objectives: a global sensi­
tivity analysis is required because local sensitivity analyses may not do
justice to the complexities and non-linearity of adsorption processes
(confirmed by the global sensitivity analysis results in Section 3.2).
However, to vary many (here 15 or 16) input parameters of DAC pro­
cesses, a million or more simulations are needed, which would take
years when running a detailed adsorption model on a desktop computer.
Therefore, we trained machine learning models on a training set
generated with our detailed adsorption model and used the machine
learning models for the global sensitivity analysis.
2.1. Detailed adsorption model
The underlying detailed TVSA model used to train the machine
learning surrogate models was described in detail in previous work and
now in the Electronic Supplementary Information (ESI) Section 1 [20].
However, we applied two slight process variations within this model.
These variations are a) a TVSA process using indirect heating via heat
exchange pipes within the adsorbent bed (TVSA), shown in Fig. 1a, and
b) a TVSA process with a steam purge as the desorption step (S-TVSA),
shown in Fig. 1b.
Rather than using times to define the heating, desorption and cooling
steps, we terminated these steps when a predefined temperature level
was reached [20]. The S-TVSA process required alterations to the
boundary conditions and variables that defined the desorption step. For
example, the boundary conditions for the S-TVSA desorption step are
now the same as the adsorption step, with the inlet conditions defined as
J. Young et al. Chemical Engineering Journal 456 (2023) 141035

Table 1
Parameters varied as part of the global sensitivity analysis and the ranges of their values. Uniform distributions were used for all parameters apart from kCO2 where we
used a uniform distribution of its logarithm. The reasoning behind the ranges chosen are found in the ESI Table S8.
Parameter and units Symbol Low value High Process Type
value configuration

Maximum CO2 capacity [mol kg− 1] ns,0,CO2 3.6451 6.0752 Both Material
CO2 adsorption affinity parameter [J mol− 1] ΔHdry,CO2 − 123688 − 111908 Both Material
Monolayer H2O capacity [mol kg− 1] qm,H2O 0.90791 6.3554 Both Material
Co-adsorption wet CO2 adsorption affinity parameter [J mol− 1] ΔHwet − 136663 − 123647 Both Material
CO2 LDF constant [s− 1] kCO2 0.0003 0.03 Both Material
CO2 heat of adsorption [J mol− 1] ΔHCO2 − 50000 − 130000 Both Material
Adsorbent heat capacity [J kg− 1 K− 1] Cp,s 790 2370 Both Material
Adsorbent pellet density [kg m− 3] ρp 440 1320 Both Material
2
Heat transfer coefficient [W m− K− 1] HTC 3.5 24.5 Both Material and
process
Ratio of inlet velocity to minimum fluidisation velocity [-] v : vmf 0.5 0.9 Both Process
Heating fluid temperature [K] Theating 363 393 TVSA Process
Ratio of heating fluid and steam purge temperature to H2O saturation temperature at the vacuum pressure Tpurge : Tmin 1.05 1.15 S-TVSA Process
[–]
Vacuum pressure [Pa] Pvac 10,000 40,000 Both Process
Temperature difference between the column and heating fluid when the heating step is terminated [K] ΔTheating − 50 − 30 TVSA Process

Temperature difference between the column and the H2O saturation temperature when the heating step is ΔTheating 2 10 S-TVSA Process
terminated [K]
Temperature difference between the column and heating fluid when the desorption step is terminated [K] ΔTdes − 10 − 0.5 TVSA Process
Ratio of steam purge velocity to minimum fluidisation velocity [-] vpurge : vmf 0.01 0.20 S-TVSA Process
Saturation percentage at the column end (versus feed conditions) when the desorption step is terminated Satpurge,end 40 60 S-TVSA Process
[%]
Saturation percentage at the column end when the adsorption step is terminated [%] Satads,end 90 99 Both Process

the steam properties and the outlet pressure defined as the vacuum
pressure. The ESI Section 1 provides more details of the model equa­
tions, boundary conditions, performance indicator calculations, and
variables. Additionally, the work equivalent is calculated from the
specific thermal energy and the specific electricity consumption using
the Carnot efficiency [35]. Here, we assumed a turbine efficiency of
0.75. The Toth and Guggenheim-Anderson-de Boer (GAB) models were
used to model the CO2 and H2O adsorption equilibrium, respectively,
whilst the mechanistic co-adsorption model from Young et al. was used
to model the co-adsorption effect [20,36–38]. Meanwhile, it is assumed
that no nitrogen is adsorbed, as is true experimentally for Lewatit® VP
OC 1065 [20]. Additionally, the linear driving force (LDF) model was
used to model the adsorption mass transfer and kinetics [39].
The model for both processes was implemented in gPROMS, dis­
cretised using a central finite differences scheme, and solved using an
implicit Runge-Kutta method with a variable time step [40]. The pro­
cesses were run until cyclic steady state was achieved, defined as when
the composition, pressure, and temperature at five equidistant points
along the column at the end of a cycle, but also the step times, are within
0.5 % of their value at the end of the previous cycle.
The detailed model in gPROMS was called using gO:MATLAB, a tool
that allows gPROMS models to be called from MATLAB. The dataset was
collected by varying the chosen variables within their distribution of
interest, which we will define in Section 2.3, Table 1.1 Over 30,000
simulations were performed for each process. Five outputs are collected:
specific thermal energy, specific electricity, productivity, purity, and
working capacity. These are defined in the ESI Section 1.
2.2. Machine learning models
We have assessed that a multi-output artificial neural network (ANN) is
one of the simplest machine learning models to develop and most accurate
1
The exception is the S-TVSA parameter vpurge: vmf. The range used for
detailed model sampling was 0.01 to 0.75 due to the relatively unknown nature
of this parameter. However, we found that values above 0.2 are not relevant
during our preliminary investigation, so we limited the upper bound to 0.2 for
the global sensitivity analysis.
for this purpose. In our preliminary analysis, we also assessed single-
output ANNs, gaussian process regression, support vector machines,
gradient-boosted trees, and random forests. But, we found that these were
all outperformed by the multi-output ANN. Additionally, successful uses of
ANNs to predict adsorption process performance exist in literature to also
support this choice, as discussed in the ESI Section 2.1 [29,41–51].
The ANNs were implemented in Python using TensorFlow [52,53].
Firstly, the dataset was split randomly into a training and test set in a ratio.
The test set was defined as 10 % of the total dataset, whilst the training set
is the other 90 %. Then we normalised the training set using Z-score
normalisation, which involves subtracting the mean value from each input
and output before dividing by its standard deviation. The ANN applied was
a feedforward network with dense layers and no dropout [54]. The acti­
vation unit used for each node is the rectified linear unit (ReLU), and the
optimiser used for training is the Adam optimiser with a learning rate of
0.0005 [54]. Early stopping was applied as a regularisation strategy with a
stopping patience of 100 epochs and a validation dataset of 5 % of the
training dataset [54]. The total number of epochs used for training was
500. The loss function applied was mean absolute error (MAE), which was
chosen over mean squared error (MSE), so the model would not be biased
towards predicting regions of poor process performance particularly
accurately. An example is at low working capacities when specific thermal
energy and specific electrical energy are very high and sensitive to small
changes in input values. Hence, this region would be prone to large pre­
diction errors and greatly penalised by MSE. A hyperparameter optimi­
sation was performed using K-fold cross-validation using 7 folds and a grid
search, and the number of dense layers and the width of the dense layers
were varied as hyperparameters. A key point is that a singular ANN was
trained to predict all five outputs: specific thermal energy, specific elec­
tricity, productivity, purity, and working capacity. This was applied as a
regularisation and generalisation strategy as this should help the ANN find
fundamental features of the process as opposed to being a data-fitting
exercise [54]. The training and hyperparameter optimisation procedures
of the ANNs for both the TVSA and S-TVSA processes were identical.
2.3. Global sensitivity analysis
First, the variables of interest for each process and their

3
J. Young et al.
Fig. 2. Parity plots showing the predictions, for each performance indicator, of the two multi-output neural networks on the test set, along with the corresponding R2
value. The solid black lines are the parity lines (y = x), and are a guide for the eye. The dotted black lines represent a 10 % error in the ANN prediction value.
corresponding distributions were identified for the sensitivity analysis.
The selection aimed to include variables related to each important
adsorbent property or process parameter. For instance, parameters were
included that describe heat and mass transfer, uptake of CO2 and H2O,
adsorbent density and heat capacity, as well as pressure and temperature
settings and cycle step times. These variables and their ranges are shown
in Table 1, whilst the justification for our choices is found in the ESI
Table S8. Meanwhile, from experience, we know that some model inputs
are irrelevant to the research questions, so they were kept constant, and
the values chosen are in the ESI Tables S4–S7.
Uniform distributions were used for all parameters apart from the
CO2 LDF constant, kCO2 . This was a conservative choice, which was
implemented due to a lack of information to suggest otherwise,
following the maximum entropy principle [55]. For the CO2 LDF con­
stant, the logarithm was varied uniformly instead. The parameters were
varied around the properties of Lewatit® VP OC 1065 taken from our
previous work [20]. The ranges (minima and maxima) were assigned
based on a literature review to encompass the typical design space of
amine-functionalised adsorbents. We varied the isotherm parameters
enough to have visibly different isotherms that are still likely to apply to
amine-functionalised adsorbents based on a visual inspection and our
4
Chemical Engineering Journal 456 (2023) 141035
literature review. The range of isotherms can be observed in the ESI
Figs. S1-S2. The heat transfer coefficient is still highly uncertain, but we
varied our value around the range we used in previous work, which is
based on earlier process modelling work [20,30]. The process parame­
ters were varied based on preliminary parametric work and operational
limits that we imposed. Examples of these limits are:
• The inlet fluid velocity cannot exceed 90 % of the minimum fluid­
isation velocity.
• The temperature cannot exceed 120 ◦ C due to the stability limita­
tions of amine-functionalised adsorbents [56–60].
• The minimum vacuum pressure feasible at an industrial scale is
estimated to be around 0.1 bar.
• The minimum temperature of the column during a steam purge is 5
% above the H2O saturation temperature to avoid condensation.
As mentioned in Section 2.1, some steps are terminated when pre­
defined temperature levels are reached. Meanwhile, the CO2 saturation
level at the end of the column when the adsorption step is terminated is
used to determine the adsorption step time. Although we do extract the
associated step times, we found that the use of these proxies aids the
J. Young et al.
Fig. 3. Total order Sobol rankings of each parameter changing across the different performance indicators shown for both processes. The y-axis labels refer to the
lines beginning directly to the right of each label. Hence, they are in the order of the parameter rankings for productivity. Note here that we define productivity per
adsorbent bed volume.
learning of the ANNs as a form of feature engineering. As a result, there
are no step times included in Table 1. The proxies are Satads,end , ΔTheating ,
ΔTdes , and Satpurge,end . The parameters ΔTdes and Satpurge,end are for a TVSA
and S-TVSA process, respectively, whilst ΔTheating and Satads,end are in
both processes. Another important point to note is that we separated the
empirical Toth CO2 isotherm affinity parameter, ΔHdry,CO2 , and the heat
of adsorption used in the energy balance, ΔHCO2 , to unravel the impact
of each separately. In reality, the two parameters will be related to each
other.
A Sobol analysis was run using the ranges in Table 1 and Saltelli
sampling with the SALib package in Python to calculate a sensitivity
ranking [24,26,61–63]. Sobol analysis decomposes the variance
observed in output values into that explained by each (combination of)
input variable(s). It is a method especially suited for highly nonlinear
systems like adsorption processes [64]. Sobol indices represent how
much an input’s variance contributes to an output’s variance. Total
order Sobol indices report this, including all interactions between the
input and all other inputs being varied. In comparison, first-order Sobol
indices exclude all interactions with other inputs. Finally, second-order
Sobol indices consider just the interactions between the input parameter
and each other parameter alone.
The TVSA process has 15 parameters requiring 1,048,576 simula­
tions for the sensitivity analysis to calculate the total, first, and second-
order Sobol indices with error bars. The error bars are calculated using a
bootstrapping method [61]. The S-TVSA process has 16 parameters, so
we needed 2,228,224 simulations to do the same. Using the ANNs, this
sampling took just a few seconds for all performance indicators. For
comparison, the S-TVSA simulations would have taken around 42 years
using the detailed model on a normal desktop computer, showcasing the
sheer necessity of using surrogate models for this task.
2.4. Performance indicators
The ANNs predict the five outputs from the detailed model, which
are the specific thermal energy, the specific electricity, the productivity,
the purity, and the working capacity. The performance indicators we
will focus on are specific thermal energy, specific electricity, produc­
tivity, and purity. It has previously been shown that capital costs
dominate in the early stages of this technology’s development [65].
Productivity is the performance indicator with the greatest impact on
5
Chemical Engineering Journal 456 (2023) 141035
capital costs, as higher productivity will lead to fewer adsorption beds
required. However, as the technology develops, operating costs will
become more important in terms of cost [65]. Specific thermal energy
and specific electricity consumption will both be significant contributors
to these operating costs, so to minimise the long-term cost of the tech­
nology, these performance indicators are key. Purity is a performance
indicator that is a constraint depending on the final destination of the
CO2. For geological sequestration, it is essential to achieve high purities,
with the US Department of Energy - National Energy Technology Lab­
oratory (DOE/NETL) setting a lower limit of 95 % CO2 purity, for
example [66].
3. Results and discussion
3.1. Artificial neural network predictions
The parity plots in Fig. 2 show the predictions of the two optimised
ANN models on the test set.2 The predictions are all relatively accurate.
Most R2 values are above 0.96, with the exceptions being predictions of
specific thermal energy in a TVSA process, specific electricity in an S-
TVSA process, and purity for both processes. Working capacity and
productivity are predicted very accurately due to being the simplest two
performance indicators as they only involve the amount of CO2 recov­
ered, the mass or volume of the adsorbent, and the cycle time in the case
of productivity. In contrast, purity seems to be the most challenging
performance indicator to predict for the ANNs. The distributions of
purity of the processes are relatively narrow. As a result, there are fewer
examples of outliers in the dataset of detailed model simulations and
potentially more numerical noise from the detailed model. This is likely
having an impact on the ANN predictions. Additionally, only a very
small number of points exhibit greater than 10 % of error when
considering that there are ~3000 test samples, as shown in Fig. 2.
The TVSA ANN also appears to generalise very well. In the ESI
Figs. S20-S21, the global sensitivity samples plateau at a work
2
A visualisation of the ANNs’ hyperparameter optimisation can be found in
the ESI Figure S4. The optimal network parameters for the TVSA ANN model
are 14 layers and 1000 nodes per layer. Whereas, for the S-TVSA ANN model,
the optimal network parameters are 3 layers and 1000 nodes per layer.
J. Young et al. Chemical Engineering Journal 456 (2023) 141035

Fig. 4. Total order, first order, and second order Sobol indices showing the impact that each parameter varied has on each performance indicator for the TVSA
process. Note here that we define productivity per adsorbent bed volume.

equivalent of ~1.2 MJ kg− 1. When we constraint the purity to 98 % (i.e., cautiously.

Fig. 6), the global sensitivity samples plateau at a higher value of ~1.3
MJ kg− 1, which is above the thermodynamic minimum work of ~0.5
MJ kg− 1 for this purity constraint. In contrast, the S-TVSA model appears
to generalise more poorly and is likely simply regressing the data rather
than picking out real features from the process. In Fig. 6, the predicted
work equivalent of the S-TVSA ANN does not plateau, and the results
from the region of very low work equivalent (less than ~2.5 MJ kg− 1)
should be analysed cautiously. Although, the ESI Figs. S20–S21 show
that the global sensitivity samples from the S-TVSA model are extremely
sparse in this region, and hence all of the conclusions from this work are
unaffected. The ESI Figs. S19 and S20 show that the TVSA ANN always
predicts values of work equivalent within the area samples by the
detailed model data set, whilst the S-TVSA ANN does extrapolate. This is
another reason to view the low work equivalent S-TVSA results more
It is unlikely that collecting additional detailed process model data
would improve the ANN predictions. The ESI Figs. S5–S6 show learning
curves to assess the relationship between model performance and the
amount of training data. Learning saturates for both TVSA and S-TVSA
before 20,000 training samples. Any additional training data would not
substantially improve model performance.
3.2. Global sensitivity analysis
The rankings of importance of the input parameters and how these
rankings change across the different performance indicators are

6
J. Young et al.
Fig. 5. Total order, first order, and second order Sobol indices showing the impact that each parameter varied has on each performance indicator for the S-TVSA
process. Note here that we define productivity per adsorbent bed volume.
visualised in Fig. 3.3 Fig. 4 shows the TVSA Sobol indices for purity,
productivity, specific thermal energy, and specific electricity. Whereas
Fig. 5 shows the same for an S-TVSA process. Firstly, a key observation
from Figs. 3-5 is that DAC performance is dictated by both adsorbent and
process parameters. This confirms they need synergistic optimisation to
develop an optimal DAC process.
In a TVSA process, the process parameters that consistently rank high
are vacuum pressure and heating temperature. Meanwhile, the adsorbent
properties that consistently rank high are the CO2 LDF constant, mono­
layer H2O uptake, and the CO2 isotherm affinity parameter, whilst the heat
3
The scatter plot results of the global sensitivity analysis from every per­
formance indicator for both processes can be found in the ESI Figures S7-S16.
The results of the global sensitivity analysis for working capacity can be found
in the ESI Figures S17-S18.
7
Chemical Engineering Journal 456 (2023) 141035
transfer coefficient is also crucial for productivity. The material proper­
ties consistently ranked in the bottom half are the maximum CO2 ca­
pacity isotherm parameter, the density of the pellet, and in all but the
specific thermal energy, the heat capacity of the adsorbent.
The S-TVSA process shows a different ranking of parameter impor­
tance. For this process configuration, the superheating level of the steam
and steam purge flow rate are the key process variables to optimise. The
most important adsorbent parameter overall is the H2O-CO2 co-
adsorption parameter. However, the CO2 LDF constant is still among the
most important adsorbent parameters, alongside the CO2 isotherm af­
finity parameter, the heat transfer coefficient and, only for productivity,
the pellet density, which directly impacts the adsorbent bed density.
Meanwhile, the least important adsorbent properties are the heat ca­
pacity, the maximum CO2 capacity isotherm parameter, and the CO2
heat of adsorption. These first observations critically show that the
design of an optimal adsorbent is very much process-dependent and goes
J. Young et al. Chemical Engineering Journal 456 (2023) 141035

Table 2 should also be noted that we consider a type 1 CO2 isotherm model [67].
A qualitative analysis of the importance of each adsorbent parameter based on So, not all of our findings will apply to adsorbents with more exotic
the global sensitivity analysis in this study. We have added conductivity, as this isotherm shapes, such as “step-shaped” isotherms [10]. The optimisation
will impact the heat transfer coefficient and thermal stability, impacting the of an adsorbent property is occasionally linear. However, we need to
maximum temperature that can be utilised in the desorption step.9 accept that when improving one property, we may impair another, and an
Parameter TVSA process S-TVSA adsorbent that combines all optimal properties may not exist. Addition­
process ally, we can reason how to circumvent or avoid problems such as low
CO2 mass transfer/adsorption kinetics Critical Very thermal conductivity. For example, we can develop adsorbents that work
important with alternative faster heating methods such as microwave heating [68].
CO2 adsorption affinity parameter Very Very
important important
CO2 maximum capacity Unimportant Unimportant 3.3.1. Physical characteristics
Co-adsorption wet CO2 adsorption affinity Important Critical Fig. 6 shows scatter plots of work equivalent versus productivity with
parameter the points coloured by the heat transfer coefficient for values of work
H2O adsorption capacity Critical Important equivalent below the 2.5 MJ kg− 1 value found previously for Lewatit®
CO2 heat of adsorption Unimportant Unimportant
VP OC 1065 in a TVSA process [20]. The plot clearly shows that the
Heat capacity Important Unimportant
Pellet density Important Very highest productivities are only accessible with the highest heat transfer
important coefficient values in a TVSA process. The ESI Figs. S22 and S23 show the
Conductivity Very Important same plots for heat capacity and density, respectively. We observe that,
important
generally, the lowest energy consumption is only achievable with the
Thermal stability limit Important Important
9
The qualifiers are not based on a specific system and are purely qualitatively.
This is because some performance indicators have a greater impact on process
performance. For example, productivity is more important than specific elec­
tricity, as the specific electricity does not tend to vary as much across the whole
sensitivity analysis, and productivity is likely a greater contributor to cost
currently via the capital costs [65]. In addition, sometimes a parameter is not
one of the lowest ranked despite having a very small Sobol index, like ns,0,CO2 in a
S-TVSA process in its impact on specific thermal energy.

well beyond finding the adsorbent with the highest equilibrium uptake.
Finally, a key observation on all the Sobol indices is that, in Figs. 4
and 5, there are generally large differences between the first, second,
and total order values, leading to different importance rankings based on
each. This confirms that global sensitivity analysis is key to truly un­
derstanding the importance of adsorbent properties and adsorption
process parameters, which could not have been accomplished reliably
by local sensitivity methods alone. As previously discussed, this was only
possible using a machine learning surrogate model, showing the Fig. 7. Some properties of example solid supports normalised using Z-score
importance of combining traditional modelling methods with machine normalisation. Green is considered a positive property, and for density, where
learning to truly gain a deep understanding of adsorption processes. there is a trade-off between productivity and specific electricity, we chose a
high density to maximise productivity. The actual values can be found in the
ESI Table S9. Properties of amine-functionalised nano-fibrillated cellulose were
3.3. Implications for adsorbent design used to represent cellulose [69], whilst properties of γ-alumina were used to
represent alumina [70–72], and properties of SBA-15 were used to represent
Table 2 shows the assessed importance of each adsorbent parameter to silica [73–75]. Meanwhile, Lewatit is simply Lewatit® VP OC 1065, an example
of a functionalised resin [20,76,77]. Here, some of the properties are measured
use as a guide for adsorbent design. This work only encompassed the
after functionalisation. The exact values should be taken cautiously. (For
design space of amine-functionalised adsorbents, so many of the conclu­
interpretation of the references to colour in this figure legend, the reader is
sions will only hold for such adsorbents, and there is still much scope for
referred to the web version of this article.)
tuning and optimising the porous support and amine combination. It

Fig. 6. Scatter plots showing work equivalent versus productivity. Here, the points are coloured by the heat transfer coefficient. The upper bound of the plot is 2.5
MJ kg− 1, which is approximately the minimum previously found in a TVSA process using Lewatit® VP OC 1065 [20]. All points shown have a purity above 98 % and
the lower bound of the plot is set to be 0.5 MJ kg− 1, which is approximately the thermodynamic minimum (depending on capture rate and purity constraint).

8
J. Young et al.
Fig. 8. Scatter plots showing work equivalent versus productivity. Here, the points are coloured by the CO2 LDF constant. The upper bound of the plot is 2.5 MJ kg− 1,
which is approximately the minimum previously found in a TVSA process using Lewatit® VP OC 1065 [20]. All points shown have a purity above 98 % and the lower
bound of the plot is set to be 0.5 MJ kg− 1, which is approximately the thermodynamic minimum (depending on capture rate and purity constraint).
lowest heat capacities, particularly for the TVSA process. Additionally,
the highest productivities are only accessible to higher density adsor­
bents for both the TVSA and the S-TVSA processes.
Fig. 7 presents how we may start comparing the most common
supports used when designing amine-functionalised adsorbents. Here,
we have taken SBA-15 as an example of a silica support and γ-alumina as
an example of an alumina support. We have compared their heat ca­
pacity, conductivity, and density. We also added properties of func­
tionalised cellulose and Lewatit® VP OC 1065 for reference, although it
is important to note that functionalisation will change the properties of
alumina and silica. It is noticeable that alumina has a lower heat ca­
pacity and higher thermal conductivity when compared to silica.4
Another observation is that the thermal conductivity of Lewatit® VP OC
1065 is inferior compared to the rest of the adsorbents.
The ESI Figs. S7–S16 show we would like to apply higher desorption
temperatures in our processes than current adsorbent stability limits
allow in order to improve productivity and decrease energy consump­
tion. If we increase the desorption temperature, we will likely reach an
optimum when considering the trade-off between productivity and en­
ergy consumption. However, currently, optimisation efforts converge
onto the upper bound of temperature (~100 ◦ C) [20]. Some background
on amine-functionalised adsorbent stability is available in the ESI Sec­
tion 2.3 [60,78–82]. Based on this, primary amine-grafted adsorbents
appear to be the most promising route to preventing thermal and
oxidative degradation and accessing higher desorption temperatures.
3.3.2. CO2 adsorption kinetics
Figs. 3-5 show that CO2 adsorption kinetics or mass transfer is one of
the most critical factors in a TVSA process and also highly important in
an S-TVSA process, so we must focus efforts on optimising this. The
strong effect of CO2 adsorption kinetics on productivity is perhaps one of
the easiest to explain. Faster adsorption kinetics leads to a sharper
breakthrough curve with a shorter adsorption time and faster cycle time.
Meanwhile, depending on whether the desorption step is heat transfer
limited or mass transfer limited, faster adsorption kinetics can also
shorten the desorption step (which is only expected at very low values of
the LDF constant). The adsorption kinetics also affects the specific
electricity, again because of the LDF constant’s impact on the break­
through curve. Slower adsorption and a broader breakthrough curve
cause more CO2 to be lost from the end of the column during the
adsorption step, and hence more air is required to be forced through the
4
Interestingly, alumina has been identified as promising for other reasons
too. It has the lowest cradle-togate, and cradle-to-grave carbon footprint
compared to other amine-functionalised supports and is more stable than silica
in the presence of steam. [15,109].
9
Chemical Engineering Journal 456 (2023) 141035
column to recover the same amount of CO2. The impact of CO2
adsorption kinetics on the specific thermal energy in a TVSA process is
complex. There are two contributors to the specific thermal energy: the
total thermal energy input into the column per cycle, and the cycle CO2
working capacity. Firstly, although the LDF constant does impact the
CO2 working capacity, it ranks relatively low compared to other pa­
rameters, as shown in the ESI Fig. S17. So, the CO2 adsorption kinetics
must be impacting the total thermal energy input into the column. The
rate of thermal energy input is only dependent on the heating driving
force (the temperature difference between the column and heating
fluid), and the heat transfer coefficient. However, what this energy
contributes towards will depend on the rates of desorption of CO2 and
H2O, with any additional heat contributing towards raising the tem­
perature of the adsorbent and adsorbate. The rates of desorption depend
on the desorption driving force (the difference between the loading and
the equilibrium loading) and the LDF constants. If the CO2 LDF constant
is high, the desorption driving force can be lower whilst achieving the
same rates of desorption, so the heating temperature can be lower, and
hence less energy is required to heat the adsorbent bed. On top of this, a
lower heating temperature additionally affects the H2O working ca­
pacity, which will also be lower, reducing the heat required for
desorbing H2O. This is substantiated by the fact that the largest 2nd
order interaction for the impact of LDF constant on specific thermal
energy is with the heating temperature. A sharper CO2 breakthrough
curve can also lead to less H2O adsorption, depending on the H2O
adsorption capacity of the adsorbent, through a shorter adsorption step.
Less H2O adsorbed leads to lower thermal input required during
desorption, as less parasitic heat is necessary for H2O desorption. There
is also a large impact of the CO2 LDF constant on specific thermal energy
in an S-TVSA process. This is visualized in the scatter plots in the ESI
Fig. S15. The reason for this is that when the LDF constant, and hence
mass transfer, is too low, the desorption of CO2 occurs so slowly that the
desorption step becomes mass transfer limited. Consequently, much
more steam flows through the bed due to the longer desorption step
time, and more latent heat of steam evaporation is wasted. Again, this is
substantiated by the fact that the largest 2nd order interaction for the
impact of LDF constant on specific thermal energy is with the steam
purge flow rate.
Fig. 8 shows scatter plots of work equivalent versus productivity with
the points coloured by the CO2 LDF constant for values of work equiv­
alent below 2.5 MJ kg− 1 [20]. We observe that CO2 LDF constants of
0.003 s− 1 and above are requisite to reaching the lowest work equivalent
for both the TVSA and S-TVSA processes. This is equal to the value of the
CO2 LDF constant of Lewatit® VP OC 1065. This means that an improved
adsorbent needs the CO2 mass transfer and adsorption kinetics to be at
least as good as Lewatit® VP OC 1065. Beyond this, the highest
J. Young et al.
Fig. 9. Scatter plots showing work equivalent versus productivity. Here, the points are coloured by the CO2 adsorption affinity parameter. The upper bound of the
plot is 2.5 MJ kg− 1, which is approximately the minimum previously found in a TVSA process using Lewatit® VP OC 1065 [20]. All points shown have a purity above
98 % and the lower bound of the plot is set to be 0.5 MJ kg− 1, which is approximately the thermodynamic minimum (depending on capture rate and pu­
rity constraint).
productivities are only available if the LDF constant is 0.01 s− 1 and
above in an S-TVSA process and 0.003 s− 1 and above in a TVSA process.
Overloading an adsorbent with large amine molecules, such as pol­
yethylenimine (PEI), can create significant diffusional barriers for CO2
molecules to reach amine groups for adsorption [83–85]. This perhaps
favours smaller amine molecules, which need to be grafted rather than
impregnated for stability reasons, as detailed in the ESI Section 2.3.
Pores that are too small also impact diffusion, and hence the LDF con­
stant, negatively [86–88]. However, strategies are available to over­
come these limitations, such as pore expansion and, in the context of
polymer resins, cross-linker selection and optimisation [89–93]. More
details are available in the ESI Section 2.4. In Section 3.4, we shall
discuss how structuring adsorbents may also help overcome mass
transfer problems.
3.3.3. Equilibrium uptake
We have shown that the maximum, i.e., at infinite partial pressure,
CO2 capacity of the adsorbent does not significantly impact either pro­
cess configuration in the range studied, although the ESI Fig. S26 sug­
gests that high values of this parameter are preferred to reach the lowest
values of work equivalent in a TVSA process. This implies that we should
be willing to sacrifice amine loading and coverage if we can improve
other adsorbent properties, such as mass transfer.
However, the CO2 adsorption affinity parameter is important. The
CO2 adsorption affinity parameter, ΔHdry,CO2 , in the Toth isotherm
model used in this work, is equivalent to the heat of adsorption in the
Langmuir isotherm model [94]. This means it impacts the CO2 uptake at
low partial pressures. The CO2 isotherm affinity parameter, ΔHdry,CO2 ,
strongly impacts productivity and specific thermal energy in both pro­
cesses, whilst it also strongly impacts specific electricity and purity in an
S-TVSA process. The CO2 heat of adsorption, ΔHCO2 , will, in reality,
directly affect the affinity, ΔHdry,CO2 , such that an increase in ΔHCO2 will
lead to an increase in ΔHdry,CO2 . However, we separated these effects for
this exercise, as discussed in Section 2.3. The heat of adsorption, ΔHCO2 ,
has a minimal impact on the performance indicators, including the
specific thermal energy, but it does have a small effect on productivity in
a TVSA process.5
Fig. 9 shows scatter plots of work equivalent versus productivity with
the points coloured by the CO2 adsorption affinity parameter for values
of work equivalent below 2.5 MJ kg− 1 [20]. In a TVSA process, the
5
The isotherms from the three TVSA global sensitivity analysis points with
the lowest work equivalent and 98% CO2 purity are shown in the ESI Figure S3
as a demonstration of what the optimal isotherms in a TVSA process may look
like. Notably, they also have the maximum co-adsorption enhancement effect.
10
Chemical Engineering Journal 456 (2023) 141035
highest productivities and lowest work equivalents are only accessible
when the affinity parameter value is smaller or equal to that of Lewatit®
VP OC 1065, i.e., a less steep isotherm in the low-pressure region. This is
because the CO2 desorbed into the gas phase prevents further desorption
by limiting the reduction in the equilibrium capacity of CO2, and this is
especially prominent when the isotherm is steeper in the low-pressure
region due to a larger affinity. Meanwhile, in the S-TVSA process,
most of the points in this region of best process performance have the
largest affinity and hence the steepest isotherm in the low-pressure re­
gion. This means that for a TVSA process, we need to target more weakly
bonded chemisorbed species with a slightly less steep CO2 isotherm at
low pressures compared to Lewatit® VP OC 1065. In contrast, for an S-
TVSA process, we should target more strongly bonded chemisorbed
species with as steep an isotherm as possible at low pressures to maxi­
mise the capacity at adsorption conditions.6 In any case, our results
clearly show that adsorbent design should not, by default, focus on
lowering the heat of adsorption (the parameter driving affinity) to
minimise the thermal energy required during desorption, as it is a
relatively small contributor to the thermal energy requirement. For
example, in our previous work benchmarking a TVSA DAC process using
Lewatit® VP OC 0165, it accounted for ~10 % of the thermal energy
requirement [20].
Finally, Figs. 3-5 show that H2O adsorption equilibrium has a
considerable impact on process performance, especially in a TVSA
process. In the ESI Fig. S24, less water adsorption is preferred in a TVSA
process to reach the lowest work equivalent, but there is no obvious
trend within this region. Through CO2 and H2O co-adsorption, more
adsorbed H2O leads to higher CO2 adsorption capacities at ambient CO2
concentrations, increasing working capacity [20]. However, if enough
H2O is present during the desorption step, the same effect can also
significantly raise the CO2 adsorption capacity at the desorption con­
ditions, decreasing working capacity. This complicated effect, plus the
increased heat requirement and consequently longer desorption time
during the desorption step if more H2O is adsorbed, explains the sub­
stantial impact on productivity, specific thermal energy, and specific
electricity in a TVSA process. It is imperative to consider H2O adsorption
when evaluating process performance using process modelling.
3.3.4. The fingerprint of an optimal adsorbent and critical adsorbent
characterisation
Our results show that within the range studied (i.e., the design space
6
The ESI Figure S27 further confirms that higher CO2 adsorption capacities
under atmospheric concentrations lead to lower energy consumptions in an S-
TVSA process. However, the same is not true in a TVSA process.
J. Young et al.
of amine-functionalised adsorbents), the optimal adsorbent for either a
TVSA or an S-TVSA process should have the highest possible thermal
conductivity and density. Meanwhile, the CO2 LDF constant should be as
high as possible up to a value of 0.01 s− 1 (after which there is no
improvement in process performance). However, in reality, there may
be a trade-off between density and CO2 LDF constant. The optimal
adsorbent for a TVSA process should have the lowest heat capacity
possible in the range studied, if not lower. It should also have a CO2
isotherm with a gradient in the low-pressure region which is not steeper
than that of Lewatit® VP OC 1065. In contrast, for an S-TVSA process,
the CO2 isotherm should have as steep a gradient as possible in the low-
pressure region to maximise the CO2 adsorption capacity under ambient
conditions.
Grafting small primary amine molecules on porous structures or
creating porous polymer structures with small primary amine groups
appears to be the most promising route forward. Adsorbents impreg­
nated with PEI may have their performance restricted by mass transfer
and thermal stability. Sacrificing the CO2 capacity by reducing the
loading of amines may be desirable if it improves mass transfer. Alumina
is a particularly encouraging porous support due to its high conductiv­
ity, low specific heat capacity, and stability in the presence of steam. In
particular, the pores of whatever porous structure is used should be
expanded as much as possible without losing excessive mechanical
strength to maximise mass transfer. Although, the impact of pore
expansion on adsorbent density should also be investigated. We also
acknowledge that other key factors exist beyond those considered here,
such as synthesisability and cyclic stability, which will impact the
eventual feasibility of an adsorbent.
One key difficulty for process designers when selecting an adsorbent
to study more closely in a process is the lack of characterisation. This has
also been highlighted by a review from Low et al., which collated the
data availability from over 200 adsorbents studied for DAC, and found
that very few materials have the characterisation required for process
studies [95].7 There is an overwhelming number of amine-
functionalised adsorbents for DAC in literature. Yet, the only univer­
sally reported parameter is the uptake capacity at 0.4 mbar of CO2,
which, as discussed in this work, is not a critical parameter in isolation.
Wu et al. have attempted to develop a standard set of evaluations that
should be performed [96]. They identify-five key aspects that should be
reported: capacity, kinetics, ease of regeneration, stability, and avail­
ability. We agree that the evaluation of kinetics is critical. However,
applying a standard approach to the evaluation of kinetics is compli­
cated, and the result depends on the conditions and method used [97].
The adsorption half-time at 0.4 mbar of CO2 and a standard temperature
such as 25 ◦ C in a gravimetric set-up with a small sample, such as 50 mg,
to eliminate heat effects could be a simple and quick characterisation
method that would allow process designers to screen the literature for
adsorbents with fast kinetics quickly [98]. We agree with Wu et al. that
CO2 isotherms at a standard temperature such as 25 ◦ C should be
measured, and particular focus should be on regions of very low pressure
(the Henry regime) to capture the affinity. We also agree that stability
will be critical to the applicability of any amine-functionalised adsor­
bent. Conversely, we need to add that H2O adsorption is also vital, and
adsorbent designers should consistently start reporting H2O isotherms at
a standard temperature such as 25 ◦ C. We also see that the co-adsorption
effect is significant. It will need to be accounted for when designing the
process, but it may be too time-consuming to measure accurately at the
adsorbent development step. Perhaps, with thorough validation, models
can be developed to reliably predict this without having to measure
them experimentally [20]. The final two key parameters that should be
measured from our analysis are the adsorbent density and the thermal
7
The materials highlighted as having sufficient characterisation by Low et al.
are Lewatit® VP OC 1065 and amine-functionalised nanofibrillated cellulose.
[20,69,95]
11
Chemical Engineering Journal 456 (2023) 141035
conductivity.
To summarise, we propose six fast adsorbent characterisations that
should critically be measured and reported for any new amine-
functionalised DAC adsorbent to allow process designers to quickly
screen the literature for the most promising adsorbents to study more
closely. These are 1) the half-time of CO2 adsorption at 0.4 mbar of CO2
and 25 ◦ C, 2) a CO2 adsorption isotherm at 25 ◦ C with a focus on the
Henry regime, 3) a H2O adsorption isotherm at 25 ◦ C, 4) density, 5)
thermal conductivity, and 6) cyclic stability including the presence of
O2.
3.4. Implications for process and contactor optimisation
The global sensitivity analysis results and accompanying discussion
highlight the significance of optimising process design in terms of mass
transfer, heat transfer and pressure drop, which goes hand in hand with
contactor design. Due to the significance of the CO2 LDF constant for
both the TVSA and S-TVSA processes, it is vital to consider which con­
tactor design can maximise the mass transfer. Optimising the contactor
layout to minimise pressure drop and hence specific electricity via
blower work is also an important consideration [99]. Rezaei and Webley
have examined packed beds, monoliths, laminates, and foams in this
context [100]. They found that laminates with a small thickness and
spacing are a particularly promising option if challenges in their
manufacturing can be overcome. They exhibit a much shorter mass
transfer zone and a vastly improved overall adsorption kinetic constant
compared to a packed bed. However, this depends on the exact geometry
of the considered laminate, foam, monolith, or pellet being compared.
The drawback of structured adsorbents is a lower density of active
adsorbent. We found that the pellet density impacts productivity greatly
because it is directly related to the density of the active adsorbent.
Despite this, Rezaei and Webley showed that a PSA process still favours
laminates in terms of productivity [100]. Further investigation is
required to assess whether this is also true in a TVSA or S-TVSA DAC
process. In Section 3.3.4, we discuss the promising option of grafting
small primary amines onto alumina. We also propose forming this
adsorbent into thin laminates as an auspicious avenue for future work.
Heat transfer is one critical metric to optimise in both processes. In
the case of the TVSA process, it accelerates both the pre-heating and
desorption steps, and in the S-TVSA process, it accelerates the pre-
heating step. Zanco et al. showed how heat transfer in a packed bed
could be significantly improved by using open-cell foams with increased
productivity of a CO2 capture TSA process of 80 % [101]. Meanwhile,
alternative heating methods are important to consider, such as micro­
wave heating, Joule heating, and induction heating [68,102–106].
These can also help to electrify the process and reduce the requirement
for additional equipment to convert electricity into heat, like heat
pumps.
3.5. Process performance limits
We can also use this global sensitivity analysis to assess the potential
performance limits. The ranges of values used for each parameter are
plausible, but it should be noted that an increase in one may lead to a
decrease in another (e.g., an increase in the density via compaction of an
adsorbent pellet would likely negatively impact mass transfer). On top of
this, there are still huge uncertainties over the heat transfer coefficient
as this is yet to be measured experimentally in a real DAC process. For
this reason, it would not be fair to assess the potential limits of pro­
ductivity based on our study. However, we can gain some insights into
where energy consumption may go, while it should be noted that the
exact combination of parameters that lead to this energy consumption
may not be attainable. Conversely, an adsorbent and contactor design
breakthrough may occur that allows for even lower values.
To investigate the potential performance limits of direct air capture
using amine-functionalised adsorbents, we minimised the specific work
J. Young et al.
Table 3
A comparison of the minimum work equivalent case (achieving 98 % CO2 purity) in the TVSA global sensitivity analysis simulations to: current real-world pilot plant
performance, a current modelling benchmark, and future estimates. Second law thermodynamic efficiencies are calculated assuming a minimum thermodynamic work
of 0.5 MJ kg-1 [107].
Development Case Specific thermal Specific electrical
stage energy [MJ kg− 1] energy [MJ kg− 1]
Current Climeworks – Hinwil first-of-a 12 2.5
-kind plant data
Modelling benchmark 10 0.8
Future Climeworks – optimistic 5.4 1.8
estimates
Minimum work equivalent in 4.6 0.3
this study
equivalent of a TVSA process using the global sensitivity analysis and a
purity requirement of 98 %. The results, including a comparison with
pilot plant data, a modelling benchmark, and an optimistic industry
future estimate, are shown in Table 3. Firstly, note that the electricity
requirement from modelling studies appears to be much lower than
industry data and future estimates. This implies that additional model­
ling of the whole system (i.e., the whole gas flow path), rather than just
the adsorbent bed, may be required to capture the vacuum and blower
work accurately. Secondly, if a “unicorn” adsorbent is found, our results
suggest that the heat and electricity requirements can reduce by around
50 % compared to current modelling benchmarks through adsorbent
improvements alone. This agrees with optimistic energy consumption
estimates by Climeworks, who also suggest a possible 50 % reduction
compared to current values. So, whilst the second law thermodynamic
efficiencies of 45 % found in this study will likely be unattainable,
second law efficiencies of over 20 %, as predicted by Climeworks, may
be within reach if the right adsorbent is found. This contrasts with a
previous commentary from Herzog, who argued that second law effi­
ciencies of much above 10 % for DAC are unrealistic [107].8 However,
we believe this may not be very optimistic.
4. Conclusions
We varied all material and process parameters of interest together in
a global sensitivity analysis in order to investigate which properties of an
amine-functionalised adsorbent are critical to improved solid adsorbent
direct air capture performance. From this, we demonstrated the
fingerprint of optimal direct air capture adsorbents based on our set of
process-informed adsorbent design guidelines, whilst also showing
which process parameters impact performance the most. As part of our
guidelines, we made suggestions on which amine and support combi­
nation to pursue, as well as how adsorbents should be structured and
what alternative heating methods could be investigated further. To
allow us to perform the global sensitivity analysis, we developed two
artificial neural network models from a detailed process model for a
temperature vacuum swing adsorption process and a steam-assisted
temperature vacuum swing adsorption process, without which the
global sensitivity analysis would not have been possible. The key
contribution of this study was that it showed precisely which adsorbent
properties are critical for direct air capture processes and which are not
to focus further adsorbent design.
We found that mass transfer and adsorption kinetics are critical
within the design space of amine-functionalised adsorbents and can be
8 Department of Energy (US-DOE), USA, are gratefully acknowledged.
This argument is based on a study from 2011 which highlights an empirical
correlation between the concentration factor of the separation process and the
second law efficiency. [108] However, there are some flaws in this analysis: 1)
the number of data points in the empirical correlation is limited, 2) the cor­
relation is not very strong, 3) the concentration factor for DAC extrapolates
beyond any points in the data set, and 4) the analysis is agnostic of the
technology.
12
Chemical Engineering Journal 456 (2023) 141035
Specific work equivalent Approximate second law Reference
[MJ kg− 1] efficiency [%]
4.6 11 [109]
2.5 20 [20]
2.7 19 [109]
1.1 45 This work
optimised via adsorbent and contactor design. Meanwhile, the heat
transfer to the adsorbent also has a significant impact and should be
improved through better adsorbent thermal conductivity, alternative
contactor designs, or different heating methods. The maximum CO2
capacity of the adsorbent does not have a large impact on performance
within the range studied, and neither do the heat effects of the heat of
adsorption. However, high heats of adsorption are preferable in an S-
TVSA process, as they result from more strongly bound adsorbed species,
leading to a higher adsorption affinity and hence higher adsorption
capacities at the ultra-low concentrations of CO2. The converse is true
for a TVSA process within the scope of amine-functionalised adsorbents,
where high adsorption capacities for CO2 are not preferred. This implies
that the heat of adsorption should be optimised to control the isotherm
rather than minimised to reduce the heating requirement.
The new-found knowledge in this work also supports our guidelines
to material scientists on which fundamental properties should be
measured for any newly synthesised amine-functionalised adsorbent for
DAC. These are standardised CO2 adsorption kinetics, CO2 isotherms,
H2O isotherms, thermal conductivity, density, and stability. This data
will assist process engineers in screening the literature for the most
promising adsorbents. Additionally, by assessing this wide yet plausible
range of parameter values, we believe that optimistic future energy
consumption targets are surmountable if adsorbents can be developed
that combine the optimal critical properties unravelled here.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Data availability
Data will be made available on request.
Acknowledgements
The research in this article was supported by the PrISMa Project (No
299659), funded through the ACT programme (Accelerating CCS
Technologies, Horizon2020 Project No 294766). Financial contributions
made from: Department for Business, Energy & Industrial Strategy
(BEIS) together with extra funding from NERC and EPSRC Research
Councils, United Kingdom; The Research Council of Norway, (RCN),
Norway; Swiss Federal Office of Energy (SFOE), Switzerland; and US-
Additional financial support from TOTAL and Equinor is also gratefully
acknowledged.
The research in this article was supported by the USorb-DAC project,
which is supported by a grant from The Grantham Foundation for the
Protection of the Environment to RMI’s climate tech accelerator pro­
gram, Third Derivative.
J. Young et al.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.cej.2022.141035.
References
[1] V. Masson-Delmotte, P. Zhai, O. Pörtner, D. Roberts, J. Skea, P.R. Shukla, A.
Pirani, C. Moufouma-Okia, C. Péan, R. Pidcock, S. Connors, J.B.R. Matthews, Y.
Chen, X. Zhou, M.I. Gomis, E. Lonnoy, T. Maycokc, M. Tignor, T. Waterfield,
Global warming of 1.5◦ C. An IPCC Special Report on the impacts of global
warming of 1.5◦ C above pre-industrial levels and related global greenhouse gas
emission pathways, in the context of strengthening the global response to the
threat of climate change, 2018.
[2] H.-O. Pörtner, D.C. Roberts, H. Adams, I. Adelekan, C. Adler, R. Adrian, P.
Aldunce, E. Ali, R.A. Begum, B. Bednar-Friedl, R.B. Kerr, R. Biesbroek, J.
Birkmann, K. Bowen, M.A. Caretta, J. Carnicer, E. Castellanos, T.S. Cheong, W.
Chow, G. Cissé, S. Clayton, A. Constable, S. Cooley, M.J. Costello, M. Craig, W.
Cramer, R. Dawson, D. Dodman, J. Efitre, M. Garschagen, E.A. Gilmore, B.
Glavovic, D. Gutzler, M. Haasnoot, S. Harper, T. Hasegawa, B. Hayward, J.A.
Hicke, Y. Hirabayashi, C. Huang, K. Kalaba, W. Kiessling, A. Kitoh, R. Lasco, J.
Lawrence, M.F. Lemos, R. Lempert, C. Lennard, D. Ley, T. Lissner, Q. Liu, E.
Liwenga, S. Lluch-Cota, S. Löschke, S. Lucatello, Y. Luo, B. Mackey, K.
Mintenbeck, A. Mirzabaev, V. Möller, M. Moncassim Vale, M.D. Morecroft, L.
Mortsch, A. Mukherji, T. Mustonen, M. Mycoo, J. Nalau, A. Okem M. New, J.P.
Ometto, B. O’Neill, R. Pandey, C. Parmesan, M. Pelling, P.F. Pinho, J. Pinnegar, E.
S. Poloczanska, A. Prakash, B. Preston, M.-F. Racault, D. Reckien, A. Revi, S.K.
Rose, E.L.F. Schipper, D.N. Schmidt, D. Schoeman, R. Shaw, N.P. Simpson, C.
Singh, W. Solecki, L. Stringer, E. Totin, C.H. Trisos, Y. Trisurat, M. van Aalst, D.
Viner, M. Wairu, R. Warren, P. Wester, D. Wrathall, Z. Zaiton Ibrahim, Working
group II contribution to the IPCC sixth assessment report (AR6) - Summary for
policymakers, 2022.
[3] J. Skea, P. Skukla, A. Reisinger, R. Slade, M. Pathak, A. Al Khourdajie, R. van
Diemen, A. Abdulla, K. Akimoto, M. Babiker, Q. Bai, I. Bashmakov, C. Bataille, G.
Berndes, G. Blanco, K. Blok, M. Bustamante, E. Byers, L. Cabeza, K. Calvin, C.
Carraro, L. Clarke, A. Cowie, F. Creutzig, D. Dadi, D. Dasgupta, H. de Coninck, F.
Denton, S. Dhakal, N. Dubash, O. Geden, M. Grubb, C. Guivarch, S. Gupta, A.
Hahmann, K. Halsnaes, P. Jaramillo, K. Jiang, F. Jotzo, T. Jung, S. Ribeiro, S.
Khennas, S. Kilkis, S. Kreibiehl, V. Krey, E. Kriegler, W. Lamb, F. Lecocq, S.
Lwasa, N. Mahmoud, C. Mbow, D. McCollum, J. Minx, C. Mitchell, R. Mrabet, Y.
Mulugetta, G. Nabuurs, G. Nemet, P. Newman, L. Niamir, L. Nilsson, S. Nugroho,
C. Okereke, S. Pachauri, A. Patt, R. Pichs-Madruga, J. Pereira, L. Rajamani, K.
Riahi, J. Roy, Y. Saheb, R. Schaeffer, K. Seto, S. Some, L. Steg, F. Toth, D. Urge-
Vorsatz, D. van Vuuren, E. Verdolini, P. Vyas, Y. Wei, M. Williams, H. Winkler,
Working group III contribution to the IPCC sixth assessment report (AR6) -
Summary for policymakers, 2022.
[4] V. Stampi-Bombelli, M. van der Spek, M. Mazzotti, Analysis of direct capture of
CO2 from ambient air via steam-assisted temperature–vacuum swing adsorption,
Adsorption 26 (2020) 1183–1197, https://doi.org/10.1007/s10450-020-00249-
w.
[5] Y. Kuwahara, D.Y. Kang, J.R. Copeland, P. Bollini, C. Sievers, T. Kamegawa,
H. Yamashita, C.W. Jones, Enhanced CO2 adsorption over polymeric amines
supported on heteroatom-incorporated SBA-15 silica: Impact of heteroatom type
and loading on sorbent structure and adsorption performance, Chem. – A Eur. J.
18 (2012) 16649–16664, https://doi.org/10.1002/chem.201203144.
[6] S.A. Didas, A.R. Kulkarni, D.S. Sholl, C.W. Jones, Role of amine structure on
carbon dioxide adsorption from ultradilute gas streams such as ambient air,
ChemSusChem 5 (2012) 2058–2064, https://doi.org/10.1002/cssc.201200196.
[7] C. Gebald, J.A. Wurzbacher, A. Borgschulte, T. Zimmermann, A. Steinfeld, Single-
component and binary CO2 and H2O adsorption of amine-functionalized
cellulose, Environ. Sci. Tech. 48 (2014) 2497–2504, https://doi.org/10.1021/
es404430g.
[8] L.A. Darunte, A.D. Oetomo, K.S. Walton, D.S. Sholl, C.W. Jones, Direct Air
Capture of CO2 Using Amine Functionalized MIL-101(Cr), ACS Sustain. Chem.
Eng. 4 (2016) 5761–5768, https://doi.org/10.1021/acssuschemeng.6b01692.
[9] T.M. McDonald, W.R. Lee, J.A. Mason, B.M. Wiers, C.S. Hong, J.R. Long, Capture
of carbon dioxide from air and flue gas in the alkylamine-appended metal-organic
framework mmen-Mg 2 (dobpdc), J. Am. Chem. Soc. 134 (2012) 7056–7065,
https://doi.org/10.1021/ja300034j.
[10] T.M. McDonald, J.A. Mason, X. Kong, E.D. Bloch, D. Gygi, A. Dani, V. Crocellà,
F. Giordanino, S.O. Odoh, W.S. Drisdell, B. Vlaisavljevich, A.L. Dzubak, R. Poloni,
S.K. Schnell, N. Planas, K. Lee, T. Pascal, L.F. Wan, D. Prendergast, J.B. Neaton,
B. Smit, J.B. Kortright, L. Gagliardi, S. Bordiga, J.A. Reimer, J.R. Long,
Cooperative insertion of CO2 in diamine-appended metal-organic frameworks,
Nature 519 (2015) 303–308, https://doi.org/10.1038/nature14327.
[11] J. Yu, S.S.C. Chuang, The structure of adsorbed species on immobilized amines in
CO2 capture: An in situ IR study, Energy Fuel 30 (2016) 7579–7587, https://doi.
org/10.1021/acs.energyfuels.6b01423.
[12] W. Chaikittisilp, J.D. Lunn, D.F. Shantz, C.W. Jones, Poly(L-lysine) brush-
mesoporous silica hybrid material as a biomolecule-based adsorbent for CO 2
capture from simulated flue gas and air, Chem. – A Eur. J. 17 (2011)
10556–10561, https://doi.org/10.1002/chem.201101480.
13
Chemical Engineering Journal 456 (2023) 141035
[13] S. Choi, M.L. Gray, C.W. Jones, Amine-tethered solid adsorbents coupling high
adsorption capacity and regenerability for CO2 capture from ambient air,
ChemSusChem 4 (2011) 628–635, https://doi.org/10.1002/cssc.201000355.
[14] S.A. Didas, R. Zhu, N.A. Brunelli, D.S. Sholl, C.W. Jones, Thermal, oxidative and
CO2 induced degradation of primary amines used for CO2 capture: Effect of alkyl
linker on stability, J. Phys. Chem. C 118 (2014) 12302–12311, https://doi.org/
10.1021/jp5025137.
[15] W. Chaikittisilp, H.J. Kim, C.W. Jones, Mesoporous alumina-supported amines as
potential steam-stable adsorbents for capturing CO2 from simulated flue gas and
ambient air, Energy Fuel 25 (2011) 5528–5537, https://doi.org/10.1021/
ef201224v.
[16] M.A. Sakwa-Novak, C.W. Jones, Steam induced structural changes of a poly
(ethylenimine) impregnated γ-alumina sorbent for CO2 extraction from ambient
air, ACS Appl. Mater. Interfaces 6 (2014) 9245–9255, https://doi.org/10.1021/
am501500q.
[17] M.A. Sakwa-Novak, S. Tan, C.W. Jones, Role of Additives in Composite PEI/Oxide
CO2 Adsorbents: Enhancement in the Amine Efficiency of Supported PEI by PEG
in CO2 Capture from Simulated Ambient Air, ACS Appl. Mater. Interfaces 7
(2015) 24748–24759, https://doi.org/10.1021/acsami.5b07545.
[18] A.R. Sujan, S.H. Pang, G. Zhu, C.W. Jones, R.P. Lively, Direct CO2 Capture from
Air using Poly(ethylenimine)-Loaded Polymer/Silica Fiber Sorbents, ACS Sustain.
Chem. Eng. 7 (2019) 5264–5273, https://doi.org/10.1021/
acssuschemeng.8b06203.
[19] C. Gebald, J.A. Wurzbacher, P. Tingaut, A. Steinfeld, Stability of amine-
functionalized cellulose during temperature-vacuum-swing cycling for CO2
capture from air, Environ. Sci. Tech. 47 (2013) 10063–10070, https://doi.org/
10.1021/es401731p.
[20] J. Young, E. García-Díez, S. Garcia, M. van der Spek, The impact of binary water-
CO2 isotherm models on the optimal performance of sorbent-based direct air
capture processes, Energy Environ. Sci. 14 (2021) 5377–5394, https://doi.org/
10.1039/d1ee01272j.
[21] M. Parvazinia, S. Garcia, M. Maroto-Valer, CO2 capture by ion exchange resins as
amine functionalised adsorbents, Chem. Eng. J. 331 (2018) 335–342, https://doi.
org/10.1016/j.cej.2017.08.087.
[22] H.M. Schellevis, T.N. van Schagen, D.W.F. Brilman, Process optimization of a
fixed bed reactor system for direct air capture, Int. J. Greenh. Gas Control. 110
(2021), 103431, https://doi.org/10.1016/j.ijggc.2021.103431.
[23] S. Kim, G. Léonard, Process modelling of Direct Air Capture (DAC) of CO2 using
solid amine sorbents, Comput. Aided Chem. Eng. 51 (2022) 265–270, https://doi.
org/10.1016/B978-0-323-95879-0.50045-X.
[24] I. Sobol′ , Global sensitivity indices for nonlinear mathematical models and their
Monte Carlo estimates, Math. Comput. Simul 55 (2001) 271–280, https://doi.
org/10.1016/S0378-4754(00)00270-6.
[25] T. Strunge, P. Renforth, M. Van der Spek, Towards a business case for CO2
mineralisation in the cement industry, Commun. Earth Environ. 3 (2022) 59,
https://doi.org/10.1038/s43247-022-00390-0.
[26] A. Saltelli, M. Ratto, T. Andres, F. Campolongo, J. Cariboni, D. Gatelli, M. Saisana,
S. Tarantola, Global Sensitivity Analysis. The Primer, John Wiley & Sons,
Chichester, UK, 2008.
[27] T. Strunge, P. Renforth, M. van der Spek, Uncertainty analysis in ex-ante techno-
economic analysis of emission reduction technologies: a tutorial case study of
CO2 mineralisation, (2022) (submitted).
[28] N. Casas, J. Schell, R. Pini, M. Mazzotti, Fixed bed adsorption of CO2/H2
mixtures on activated carbon: Experiments and modeling, Adsorption 18 (2012)
143–161, https://doi.org/10.1007/s10450-012-9389-z.
[29] S.G. Subraveti, Z. Li, V. Prasad, A. Rajendran, Machine Learning-Based
Multiobjective Optimization of Pressure Swing Adsorption, Ind. Eng. Chem. Res.
58 (2019) 20412–20422, https://doi.org/10.1021/acs.iecr.9b04173.
[30] L. Joss, M. Gazzani, M. Hefti, D. Marx, M. Mazzotti, Temperature swing
adsorption for the recovery of the heavy component: An equilibrium-based
shortcut model, Ind. Eng. Chem. Res. 54 (2015) 3027–3038, https://doi.org/
10.1021/ie5048829.
[31] B.J. Maring, P.A. Webley, A new simplified pressure/vacuum swing adsorption
model for rapid adsorbent screening for CO2 capture applications, Int. J. Greenh.
Gas Control. 15 (2013) 16–31, https://doi.org/10.1016/J.IJGGC.2013.01.009.
[32] V.S. Balashankar, A.K. Rajagopalan, R. De Pauw, A.M. Avila, A. Rajendran,
Analysis of a Batch Adsorber Analogue for Rapid Screening of Adsorbents for
Postcombustion CO2 Capture, Ind. Eng. Chem. Res. 58 (2019) 3314–3328,
https://doi.org/10.1021/acs.iecr.8b05420.
[33] D. Danaci, M. Bui, N. Mac Dowell, C. Petit, Exploring the limits of adsorption-
based CO2 capture using MOFs with PVSA – from molecular design to process
economics, Mol. Syst. Des. Eng. 5 (2020) 212–231, https://doi.org/10.1039/
c9me00102f.
[34] A.H. Farmahini, S. Krishnamurthy, D. Friedrich, S. Brandani, L. Sarkisov,
Performance-Based Screening of Porous Materials for Carbon Capture, Chem.
Rev. 121 (2021) 10666–10741, https://doi.org/10.1021/acs.chemrev.0c01266.
[35] D. Danaci, P.A. Webley, C. Petit, Guidelines for Techno-Economic Analysis of
Adsorption Processes, Front Chem. Eng. 2 (2021) 1–11, https://doi.org/10.3389/
fceng.2020.602430.
[36] J. Toth, State equations of the solid gas interface layer, Acta Chim. Acad. Sci.
Hungaricae. 69 (1971) 311–317.
[37] R.B. Anderson, Modifications of the Brunauer, Emmett and Teller Equation,
J. Am. Chem. Soc. 68 (1946) 686–691, https://doi.org/10.1021/ja01208a049.
[38] R.B. Anderson, W. Keith Hall, Modifications of the Brunauer, Emmett and Teller
Equation II, J. Am. Chem. Soc. 70 (1948) 1727–1734, https://doi.org/10.1021/
ja01185a017.
J. Young et al.
[39] E. Glueckauf, Theory of chromatography. Part 10.—Formulæ for diffusion into
spheres and their application to chromatography, Trans. Faraday Soc. 51 (1955)
1540–1551, https://doi.org/10.1016/b978-0-08-057178-2.50005-4.
[40] Process Systems Enterprise, gPROMS, www.psenterprise.com/gproms 1998-
2022, (2022). www.psenterprise.com/gproms.
[41] J. Lewandowski, N.O. Lemcoff, S. Palosaari, Use of Neural Networks in the
Simulation and Optimization o Pressure Swing Adsorption Processes, Chem. Eng.
Technol. 21 (1998) 593–597, https://doi.org/10.1002/(SICI)1521-4125(199807)
21:7<593::AID-CEAT593>3.0.CO;2-U.
[42] C. Cleeton, A.H. Farmahini, L. Sarkisov, Performance-based ranking of porous
materials for PSA carbon capture under the uncertainty of experimental data,
Chem. Eng. J. 437 (2022), 135395, https://doi.org/10.1016/j.cej.2022.135395.
[43] A. Streb, M. Mazzotti, Performance limits of neural networks for optimizing an
adsorption process for hydrogen purification and CO2 capture, Comput. Chem.
Eng. (2022), 107974, https://doi.org/10.1016/j.compchemeng.2022.107974.
[44] N. Sundaram, Training neural networks for pressure swing adsorption processes,
Ind. Eng. Chem. Res. 38 (1999) 4449–4457, https://doi.org/10.1021/ie9901731.
[45] H.R. Sant Anna, A.G. Barreto, F.W. Tavares, M.B. de Souza, Machine learning
model and optimization of a PSA unit for methane-nitrogen separation, Comput.
Chem. Eng. 104 (2017) 377–391, https://doi.org/10.1016/j.
compchemeng.2017.05.006.
[46] K.N. Pai, V. Prasad, A. Rajendran, Generalized, Adsorbent-Agnostic, Artificial
Neural Network Framework for Rapid Simulation, Optimization, and Adsorbent
Screening of Adsorption Processes, Ind. Eng. Chem. Res. 59 (2020) 16730–16740,
https://doi.org/10.1021/acs.iecr.0c02339.
[47] K.T. Leperi, D. Yancy-Caballero, R.Q. Snurr, F. You, 110th Anniversary: Surrogate
Models Based on Artificial Neural Networks To Simulate and Optimize Pressure
Swing Adsorption Cycles for CO 2 Capture, Ind. Eng. Chem. Res. 58 (2019)
18241–18252, https://doi.org/10.1021/acs.iecr.9b02383.
[48] K.N. Pai, V. Prasad, A. Rajendran, Experimentally validated machine learning
frameworks for accelerated prediction of cyclic steady state and optimization of
pressure swing adsorption processes, Sep. Purif. Technol. 241 (2020), 116651,
https://doi.org/10.1016/j.seppur.2020.116651.
[49] A. Arora, S.S. Iyer, M.M.F. Hasan, Computational Material Screening Using
Artificial Neural Networks for Adsorption Gas Separation, J. Phys. Chem. C 124
(2020) 21446–21460, https://doi.org/10.1021/acs.jpcc.0c05900.
[50] G. Hüllen, J. Zhai, S.H. Kim, A. Sinha, M.J. Realff, F. Boukouvala, Managing
uncertainty in data-driven simulation-based optimization, Comput. Chem. Eng.
136 (2020), 106519, https://doi.org/10.1016/j.compchemeng.2019.106519.
[51] S.G. Subraveti, Z. Li, V. Prasad, A. Rajendran, Physics-Based Neural Networks for
Simulation and Synthesis of Cyclic Adsorption Processes, Ind. Eng. Chem. Res. 61
(2022) 4095–4113, https://doi.org/10.1021/acs.iecr.1c04731.
[52] M. Abadi, P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean, M. Devin, S. Ghemawat,
G. Irving, M. Isard, M. Kudlur, J. Levenberg, R. Monga, S. Moore, D. Murray, B.
Steiner, P. Tucker, V. Vasudevan, P. Warden, M. Wicke, Y. Yu, X. Zheng,
TensorFlow: A System for Large-Scale Machine Learning, in: 12th USENIX Symp.
Oper. Syst. Des. Implement, 2016: pp. 265–283.
[53] A. Géron, Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow,
second ed., O’Reilly Media, 2019.
[54] I. Goodfellow, Y. Bengio, A. Courville, Deep Learning, MIT Press, Cambridge, MA,
2016.
[55] T. Cover, J. Thomas, Maximum Entropy, in: Elem. Inf. Theory, 2nd editio, Wiley-
Interscience, 2006: pp. 409–426.
[56] Q. Yu, J.D.L.P. Delgado, R. Veneman, D.W.F. Brilman, Stability of a Benzyl Amine
Based CO2 Capture Adsorbent in View of Regeneration Strategies, Ind. Eng.
Chem. Res. 56 (2017) 3259–3269, https://doi.org/10.1021/acs.iecr.6b04645.
[57] M.J. Bos, V. Kroeze, S. Sutanto, D.W.F. Brilman, Evaluating Regeneration Options
of Solid Amine Sorbent for CO2 Removal, Ind. Eng. Chem. Res. 57 (2018)
11141–11153, https://doi.org/10.1021/acs.iecr.8b00768.
[58] W.R. Alesi, J.R. Kitchin, Evaluation of a primary amine-functionalized ion-
exchange resin for CO 2 capture, Ind. Eng. Chem. Res. 51 (2012) 6907–6915,
https://doi.org/10.1021/ie300452c.
[59] S.H.A.M. Leenders, M. Infantino, S. Van Paasen, G. Pankratova, J. Tsou, J.
Wijenberg, J. Yao, Validation of CO2 adsorbent properties at pilot plant scale and
their influence on CO2 capture costs, in: 14th Fundam. Adsorpt., Boulder, CO,
2022.
[60] M. Jahandar Lashaki, S. Khiavi, A. Sayari, Stability of amine-functionalized CO2
adsorbents: A multifaceted puzzle, Chem. Soc. Rev. 48 (2019) 3320–3405,
https://doi.org/10.1039/c8cs00877a.
[61] J. Herman, W. Usher, SALib: An open-source Python library for Sensitivity
Analysis, J. Open Source Softw. 2 (2017) 97. 10.21105/joss.00097.
[62] A. Saltelli, Making best use of model evaluations to compute sensitivity indices,
Comput. Phys. Commun. 145 (2002) 280–297, https://doi.org/10.1016/S0010-
4655(02)00280-1.
[63] A. Saltelli, P. Annoni, I. Azzini, F. Campolongo, M. Ratto, S. Tarantola, Variance
based sensitivity analysis of model output. Design and estimator for the total
sensitivity index, Comput. Phys. Commun. 181 (2010) 259–270, https://doi.org/
10.1016/j.cpc.2009.09.018.
[64] A. Ward, R. Pini, Integrated uncertainty quantification and sensitivity analysis of
single-component dynamic column breakthrough experiments, Adsorption 28
(2022) 161–183, https://doi.org/10.1007/s10450-022-00361-z.
[65] J. Young, N. McQueen, C. Charalambous, S. Foteinis, O. Hawrot, M. Ojeda, H.
Pilorgé, J. Andresen, P. Psarras, P. Renforth, S. Garcia, M. van der Spek, The cost
of direct air capture and storage: the impact of technological learning, regional
diversity, and policy., ChemRxiv. (2022). 10.26434/chemrxiv-2022-dp36t.
14
Chemical Engineering Journal 456 (2023) 141035
[66] DOE/NETL, Quality Guidelines for Energy System Studies – CO2 Impurity Design
Parameters, 2013.
[67] M. Thommes, K. Kaneko, A.V. Neimark, J.P. Olivier, F. Rodriguez-Reinoso,
J. Rouquerol, K.S.W. Sing, Physisorption of gases, with special reference to the
evaluation of surface area and pore size distribution (IUPAC Technical Report),
Pure Appl. Chem. 87 (2015) 1051–1069, https://doi.org/10.1515/pac-2014-
1117.
[68] T.N. van Schagen, P.J. van der Wal, D.W.F. Brilman, Development of a novel,
through-flow microwave-based regenerator for sorbent-based direct air capture,
Chem. Eng. J. Adv. 9 (2022), 100187, https://doi.org/10.1016/j.
ceja.2021.100187.
[69] J.A. Wurzbacher, C. Gebald, S. Brunner, A. Steinfeld, Heat and mass transfer of
temperature-vacuum swing desorption for CO2 capture from air, Chem. Eng. J.
283 (2016) 1329–1338, https://doi.org/10.1016/j.cej.2015.08.035.
[70] M.R. Karim, Synthesis of γ-Alumina Particles and Surface Characterization, Open
Colloid Sci. J. 4 (2011) 32–36, https://doi.org/10.2174/1876530001104010032.
[71] N. Huda, M.A. Whitney, M.H. Razmpoosh, A.P. Gerlich, J.Z. Wen, S.F. Corbin,
How phase (α and γ) and porosity affect specific heat capacity and thermal
conductivity of thermal storage alumina, J. Am. Ceram. Soc. 104 (2021)
1436–1447, https://doi.org/10.1111/jace.17528.
[72] J.J. Calvin, M. Asplund, Y. Zhang, B. Huang, B.F. Woodfield, Heat capacity and
thermodynamic functions of γ-Al2O3, J. Chem. Thermodyn. 112 (2017) 77–85,
https://doi.org/10.1016/j.jct.2017.04.011.
[73] Y. Liu, Z. Li, X. Yang, Y. Xing, C. Tsai, Q. Yang, Z. Wang, R.T. Yang, Performance
of mesoporous silicas (MCM-41 and SBA-15) and carbon (CMK-3) in the removal
of gas-phase naphthalene: adsorption capacity, rate and regenerability, RSC Adv.
6 (2016) 21193–21203, https://doi.org/10.1039/C5RA27289K.
[74] M. Ghedini, Manzoli, Di Michele, Puglia, Signoretto, Multifunctional and
Environmentally Friendly TiO2–SiO2 Mesoporous Materials for Sustainable
Green Buildings, Molecules 24 (2019) 4226, https://doi.org/10.3390/
molecules24234226.
[75] Y. Belmoujahid, M. Bonne, Y. Scudeller, D. Schleich, Y. Grohens, B. Lebeau,
Thermal conductivity of monolithic assemblies of SBA-15 ordered mesoporous
silica particles, Microporous Mesoporous Mater. 201 (2015) 124–133, https://
doi.org/10.1016/j.micromeso.2014.09.014.
[76] M.J. Bos, T. Kreuger, S.R.A. Kersten, D.W.F. Brilman, Study on transport
phenomena and intrinsic kinetics for CO2 adsorption in solid amine sorbent,
Chem. Eng. J. 377 (2019), 120374, https://doi.org/10.1016/j.cej.2018.11.072.
[77] E. Sonnleitner, G. Schöny, H. Hofbauer, Assessment of zeolite 13X and Lewatit®
VP OC 1065 for application in a continuous temperature swing adsorption process
for biogas upgrading, Biomass Convers. Biorefinery. 8 (2018) 379–395, https://
doi.org/10.1007/s13399-017-0293-3.
[78] A. Heydari-Gorji, A. Sayari, Thermal, Oxidative, and CO2-Induced Degradation of
Supported Polyethylenimine Adsorbents, Ind. Eng. Chem. Res. 51 (2012)
6887–6894, https://doi.org/10.1021/ie3003446.
[79] Z. Liu, Y. Teng, K. Zhang, Y. Cao, W. Pan, CO2 adsorption properties and thermal
stability of different amine-impregnated MCM-41 materials, J. Fuel Chem.
Technol. 41 (2013) 469–475, https://doi.org/10.1016/S1872-5813(13)60025-0.
[80] G. Calleja, R. Sanz, A. Arencibia, E.S. Sanz-Pérez, Influence of Drying Conditions
on Amine-Functionalized SBA-15 as Adsorbent of CO2, Top. Catal. 54 (2011)
135–145, https://doi.org/10.1007/s11244-011-9652-7.
[81] P. Bollini, S. Choi, J.H. Drese, C.W. Jones, Oxidative degradation of aminosilica
adsorbents relevant to postcombustion CO2 capture, Energy Fuel 25 (2011)
2416–2425, https://doi.org/10.1021/ef200140z.
[82] A. Sayari, Y. Belmabkhout, Stabilization of Amine-Containing CO 2 Adsorbents:
Dramatic Effect of Water Vapor, J. Am. Chem. Soc. 132 (2010) 6312–6314,
https://doi.org/10.1021/ja1013773.
[83] Z. Chen, S. Deng, H. Wei, B. Wang, J. Huang, G. Yu, Polyethylenimine-
impregnated resin for high CO2 adsorption: An efficient adsorbent for CO2
capture from simulated flue gas and ambient air, ACS Appl. Mater. Interfaces 5
(2013) 6937–6945, https://doi.org/10.1021/am400661b.
[84] A. Heydari-Gorji, A. Sayari, CO2 capture on polyethylenimine-impregnated
hydrophobic mesoporous silica: Experimental and kinetic modeling, Chem. Eng.
J. 173 (2011) 72–79, https://doi.org/10.1016/j.cej.2011.07.038.
[85] F. Liu, S. Chen, Y. Gao, Y. Xie, CO2 adsorption behavior and kinetics on
polyethylenimine modified porous phenolic resin, J. Porous Mater. 24 (2017)
1335–1342, https://doi.org/10.1007/s10934-017-0375-4.
[86] V. Zeleňák, M. Badaničová, D. Halamová, J. Čejka, A. Zukal, N. Murafa,
G. Goerigk, Amine-modified ordered mesoporous silica: Effect of pore size on
carbon dioxide capture, Chem. Eng. J. 144 (2008) 336–342, https://doi.org/
10.1016/j.cej.2008.07.025.
[87] L. Wang, M. Yao, X. Hu, G. Hu, J. Lu, M. Luo, M. Fan, Amine-modified ordered
mesoporous silica: The effect of pore size on CO2 capture performance, Appl.
Surf. Sci. 324 (2015) 286–292, https://doi.org/10.1016/j.apsusc.2014.10.135.
[88] Y. Belmabkhout, A. Sayari, Effect of pore expansion and amine functionalization
of mesoporous silica on CO2 adsorption over a wide range of conditions,
Adsorption 15 (2009) 318–328, https://doi.org/10.1007/s10450-009-9185-6.
[89] J.P. Thielemann, F. Girgsdies, R. Schlögl, C. Hess, Pore structure and surface area
of silica SBA-15: influence of washing and scale-up, Beilstein J. Nanotechnol. 2
(2011) 110–118, https://doi.org/10.3762/bjnano.2.13.
[90] B. Huang, C.H. Bartholomew, B.F. Woodfield, Facile synthesis of mesoporous
γ-alumina with tunable pore size: The effects of water to aluminum molar ratio in
hydrolysis of aluminum alkoxides, Microporous Mesoporous Mater. 183 (2014)
37–47, https://doi.org/10.1016/j.micromeso.2013.09.007.
[91] H. Deng, S. Grunder, K. Cordova, C. Valente, H. Furukawa, M. Hmadeh,
F. Gándara, A. Whalley, Z. Liu, S. Asahina, H. Kazumori, M. O’Keeffe, O. Terasaki,
J. Young et al.
J.F. Stoddart, O.M. Yaghi, Large-Pore Apertures in a Series of Metal-Organic
Frameworks, Science 336 (2012) 1018–1023, https://doi.org/10.7551/mitpress/
8876.003.0036.
[92] S. Yuan, L. Zou, J.-S. Qin, J. Li, L. Huang, L. Feng, X. Wang, M. Bosch, A. Alsalme,
T. Cagin, H.-C. Zhou, Construction of hierarchically porous metal–organic
frameworks through linker labilization, Nat. Commun. 8 (2017) 15356, https://
doi.org/10.1038/ncomms15356.
[93] L.B. Hamdy, A. Gougsa, W.Y. Chow, J.E. Russell, E. García-Díez, V. Kulakova,
S. Garcia, A.R. Barron, M. Taddei, E. Andreoli, Overcoming mass transfer
limitations in cross-linked polyethyleneimine-based adsorbents to enable
selective CO2 capture at ambient temperature, Mater. Adv. 3 (2022) 3174–3191,
https://doi.org/10.1039/D1MA01072G.
[94] D.M. Ruthven, Principles of Adsorption & Adsorption Processes, John Wiley &
Sons, 1984.
[95] M.-Y. (Ashlyn) Low, L. Barton, R. Pini, C. Petit, Analytical review of the current
state of knowledge of adsorption materials and processes for direct air capture,
Chem. Eng. Res. Des. (2022). 10.1016/j.cherd.2022.11.040.
[96] J. Wu, X. Zhu, Y. Chen, R. Wang, T. Ge, The analysis and evaluation of direct air
capture adsorbents on the material characterization level, Chem. Eng. J. 450
(2022), 137958, https://doi.org/10.1016/j.cej.2022.137958.
[97] J.-Y. Wang, E. Mangano, S. Brandani, D.M. Ruthven, A review of common
practices in gravimetric and volumetric adsorption kinetic experiments,
Adsorption 27 (2021) 295–318, https://doi.org/10.1007/s10450-020-00276-7.
[98] S. Choi, J.H. Drese, P.M. Eisenberger, C.W. Jones, Application of amine-tethered
solid sorbents for direct CO2 capture from the ambient air, Environ. Sci. Tech. 45
(2011) 2420–2427, https://doi.org/10.1021/es102797w.
[99] Q. Grossmann, V. Stampi-Bombelli, M. Mazzotti, Molecular to process scale: A
review of holistic direct air capture contactor design, in: 11th Trondheim Conf.
CO2 Capture, Transp. Storage, 2021: pp. 450–454.
[100] F. Rezaei, P. Webley, Optimum structured adsorbents for gas separation
processes, Chem. Eng. Sci. 64 (2009) 5182–5191, https://doi.org/10.1016/j.
ces.2009.08.029.
15
Chemical Engineering Journal 456 (2023) 141035
[101] S.E. Zanco, M. Ambrosetti, G. Groppi, E. Tronconi, M. Mazzotti, Heat transfer
intensification with packed open-cell foams in TSA processes for CO2 capture,
Chem. Eng. J. 430 (2022), 131000, https://doi.org/10.1016/j.cej.2021.131000.
[102] Y. Gomez-Rueda, B. Verougstraete, C. Ranga, E. Perez-Botella, F. Reniers, J.F.
M. Denayer, Rapid temperature swing adsorption using microwave regeneration
for carbon capture, Chem. Eng. J. 446 (2022), 137345, https://doi.org/10.1016/
j.cej.2022.137345.
[103] M. Gholami, B. Verougstraete, R. Vanoudenhoven, G.V. Baron, T. Van Assche, J.F.
M. Denayer, Induction heating as an alternative electrified heating method for
carbon capture process, Chem. Eng. J. 431 (2022), 133380, https://doi.org/
10.1016/j.cej.2021.133380.
[104] B. Verougstraete, A. Martín-Calvo, S. Van der Perre, G. Baron, V. Finsy, J.F.
M. Denayer, A new honeycomb carbon monolith for CO2 capture by rapid
temperature swing adsorption using steam regeneration, Chem. Eng. J. 383
(2020), 123075, https://doi.org/10.1016/j.cej.2019.123075.
[105] C. Grande, A. Rodrigues, Electric Swing Adsorption for CO2 removal from flue
gases, Int. J. Greenh. Gas Control. 2 (2008) 194–202, https://doi.org/10.1016/
S1750-5836(07)00116-8.
[106] T. Chronopoulos, Y. Fernandez-Diez, M.M. Maroto-Valer, R. Ocone, D.A. Reay,
CO2 desorption via microwave heating for post-combustion carbon capture,
Microporous Mesoporous Mater. 197 (2014) 288–290, https://doi.org/10.1016/j.
micromeso.2014.06.032.
[107] H.J. Herzog, Chapter 6 - Direct air capture, in: M. Bui, N. Mac Dowell (Eds.),
Greenh, Gas Remov. Technol., Royal Society of Chemistry, 2022.
[108] K.Z. House, A.C. Baclig, M. Ranjan, E.A. van Nierop, J. Wilcox, H.J. Herzog,
Economic and energetic analysis of capturing CO 2 from ambient air, Proc. Natl.
Acad. Sci. 108 (2011) 20428–20433, https://doi.org/10.1073/pnas.1012253108.
[109] S. Deutz, A. Bardow, Life-cycle assessment of an industrial direct air capture
process based on temperature–vacuum swing adsorption, Nat. Energy 6 (2021)
203–213, https://doi.org/10.1038/s41560-020-00771-9.