Technical overview
We employ a computer guided design approach to facilitate the
development of new CO2 absorbents with improved properties. Our method
is based on an evolutionary algorithm (EA), a global optimization
method ensuring that absorbents with multiple optimal properties are
reached. New structures are automatically assembled and evaluated by
one or several fitness functions, which are directly related to
the experimentally observed absorption of CO2 or alternatively to theoretically computed properties highly correlated with observations. Central to the method is the development of fitness functions based on predictive quantitative structure-property relationship (QSPR) models to make the iterative evolutionary algorithm computationally feasible.
Added value of project
The success of the proposed de novo method could be a game-changer for
carbon capture as it enables considerable speed-up of the development
process to produce cheaper and better absorbent molecules. The
economic and environmental impact could be significant.
Research challenges
There are three main areas of challenges to the project: 1) The
practical synthesis of proposed structures from modelling since
organic synthesis is in general difficult and time consuming, 2)
testing of new compounds for properties such as CO2 capture and
toxicity etc at low sample volumes and 3) the mathematical modelling
of relevant properties and theoretical search for new, synthesizable
structures.
Results so far
A series of 40 new imidazoles have been synthesized and tested for
properties such as CO2-absorbence, viscocity, density and equilibrium
data. In addition, several new structures have been developed for
ionic liquids that will also be tested in memebranes for increase CO2
selectivity.
The new apparatus developed in the project enables new option for
experimental screening and characterization of novel CO2 capture
solvents not currently available as bulk chemicals. The instrument
allows for quick automated screening of solvents on a milliliter
scale. The results from the measurements are comparable to results
obtained in larger apparatuses. The developed in-situ online
monitoring of solvent CO2-loading and gas phase concentration in the
apparatus may also be relevant for future online process
instrumentation of CO2 capture plants.
QSPR models for predicting properties of ionic liquids such as thermal
decomposition temperatures, degree of CO2 capture, melting points,
refractive indices, viscosities, heat capacities, densities and electric conductivity have been established. A total of 2098 cations and 336 anions have so far been mined from literature yielding over 700000 theoretical ionic liquid combinations of which only a small percentage (~1%) have been experimentally tested. On the basis of this we have initiated the construction of a database that will house the structures of the cations and anions, their predicted properties emerging from both machine learning and quantum chemical methods, along with experimental data where
available.
A combinatorial chemistry based approach was also established wherein over 200000 cations (spanning 7 different chemical groups) and 38 anions were used to assemble close to 8 million ionic liquids. Virtual screening based on machine learning models for various properties allowed for the ionic liquids to be tuned to the different application areas such as cellulose dissolution, gas capture, electrochemistry and biology.
Using a combination of Darwin inspired evolutionary design, machine learning and quantum chemical approaches, novel solvents for CO2 capture applications have been developed. The solvents span a range of chemistries, including amines, imidazoles and ionic liquids. The chemical and physical properties of these solvents with and without CO2 has been investigated, both theoretically and experimentally. The set-up allows for high-throughput combinatorial screening of molecules with desired properties.
We propose to employ a recently developed de novo design software to develop new absorbents, such as amines, amino acids and ionic liquids, for carbon capture. Our software has been successfully demonstrated to work in catalyst discovery and has the poten tial to boost development of new and more efficient absorbent molecules that are also cheap and easy to synthesize.
Our method is based on an evolutionary algorithm (EA), a global optimization method ensuring that absorbents with optimal properties are r eached. New structures are automatically assembled and evaluated by one or several fitness functions, which are here directly related to the observed absorption of CO2 or to theoretically computed properties highly correlated with observations. Central to the method is the development of fitness functions based on highly predictive quantitative structure-property relationship (QSPR) models to make the iterative evolutionary algorithm computationally tractable.
The success of the proposed de novo method could be a game-changer for carbon capture as it enables considerable speed-up of the development process and gives cheaper and better absorbent molecules. The economic and environmental impact could be significant.