ArticlePDF Available

Towards Predictive Synthesis of Inorganic Materials Using Network Science

Abstract and Figures

Accelerating materials discovery is the cornerstone of modern technological competitiveness. Yet, the inorganic synthesis of new compounds is often an important bottleneck in this quest. Well-established quantum chemistry and experimental synthesis methods combined with consolidated network science approaches might provide revolutionary knowledge to tackle this challenge. Recent pioneering studies in this direction have shown that the topological analysis of material networks hold great potential to effectively explore the synthesizability of inorganic compounds. In this Perspective we discuss the most exciting work in this area, in particular emerging new physicochemical insights and general concepts on how network science can significantly help reduce the timescales required to discover new materials and find synthetic routes for their fabrication. We also provide a perspective on outstanding problems, challenges and open questions.
Content may be subject to copyright.
Towards Predictive Synthesis of
Inorganic Materials Using Network
Alex Aziz and Javier Carrasco *
Centre for Cooperative Research on Alternative Energies (CIC energiGUNE), Basque Research and Technology Alliance (BRTA),
Vitoria-Gasteiz, Spain
Accelerating materials discovery is the cornerstone of modern technological
competitiveness. Yet, the inorganic synthesis of new compounds is often an important
bottleneck in this quest. Well-established quantum chemistry and experimental synthesis
methods combined with consolidated network science approaches might provide
revolutionary knowledge to tackle this challenge. Recent pioneering studies in this
direction have shown that the topological analysis of material networks hold great
potential to effectively explore the synthesizability of inorganic compounds. In this
Perspective we discuss the most exciting work in this area, in particular emerging new
physicochemical insights and general concepts on how network science can signicantly
help reduce the timescales required to discover new materials and nd synthetic routes for
their fabrication. We also provide a perspective on outstanding problems, challenges and
open questions.
Keywords: synthesis by design, materials synthesis networks, inorganic material design, network science,
Inorganic synthesis
Advanced materials are key enablers across many industries aimed at addressing the global
challenges of economic security, renewable and sustainable energy, and human welfare.
Innovationintheseelds often requires searching for new materials or optimizing existing
ones. The traditional materials discovery approach is to focus on archetypal compounds in which
a desirable property was rst observed. This approach involves trial-and-error chemical
exploration, which usually has high demands in terms of synthesis times and costs.
Therefore, accelerating the pace of discovery of new materials is essential to achieving global
competitiveness in the 21
century. Computational modelling has emerged as a powerful
complementary tool in accelerating the process of materials discovery. Thanks to the proven
predictive power of quantum chemistry methods, together with the spectacular growth of
computational resources, computer modelling is nowadays able to bring valuable insights in
understanding the structure, properties, and function of technological materials. In particular,
high-throughput screening of materials databases using rst-principles simulation approaches
have demonstrated a successful track record of guiding advances in materials science (Jain et al.,
2016), including areas as diverse as heterogeneous catalysis (Greeley et al., 2002),
thermoelectricity (Carrete et al., 2014), and energy storage (Van der Ven et al., 2020). With
an increase in computer resources and given computational modelling is progressively being
implemented in synergy with experiments, this trend is only likely to grow. However,
computational simulations in particular ab initio molecular dynamics are computationally
Edited by:
Malgorzata Biczysko,
Shanghai University, China
Reviewed by:
Joshua Schrier,
Fordham University, United States
Kun Yao,
Schrodinger United States
Javier Carrasco
Specialty section:
This article was submitted to
a section of the journal
Frontiers in Chemistry
Received: 20 October 2021
Accepted: 03 December 2021
Published: 21 December 2021
Aziz A and Carrasco J (2021) Towards
Predictive Synthesis of Inorganic
Materials Using Network Science.
Front. Chem. 9:798838.
doi: 10.3389/fchem.2021.798838
Frontiers in Chemistry | December 2021 | Volume 9 | Article 7988381
published: 21 December 2021
doi: 10.3389/fchem.2021.798838
demanding, energy intensive and risk being repeated multiple
times by various groups investigating similar materials.
An emerging alternative to traditional physical-based
approaches is data-driven modelling (Agrawal and Choudhary,
2016;Jennings et al., 2019;Noh et al., 2020;Lombardo et al.,
2021). As a matter of fact, recent trends in Big Data have raised
hopes for a new kind of paradigm to model complex systems with
a large number of strongly interacting elements. And
autonomous decision-making materials discovery schemes to
guide experimental campaigns are starting to emerge
(Montoya et al., 2020;Stach et al., 2021;Szymanski et al.,
2021). Data science may indeed help to answer many
fundamental research questions, especially as more and more
data becomes accessible (Hill et al., 2016). This is evident from the
rise in number and quality of computational materials databases
and related informatics such as the Materials Project
( and the Open Quantum Materials Database
(OQMD) ( that complement existing experimental
data sets such as the Inorganic Crystal Structure Database
(ICSD) (, NIST Materials Data
Repository (, or the Pauling File (
However, data science alone cannot develop fundamental
research questions by itself. Collecting data and then
identifying new patterns has the potential risk of ending up
with spurious correlations, without understanding the
underlying causal relationships. Indeed, this applies to all data
driven approaches, and care must be taken to benchmark and
verify datasets with experiment.
From a theoretical viewpoint, materials discovery faces a two-
fold major paradigm. On the one hand, the identication of
thermodynamically stable compounds, also referred to as a
structure prediction problem. And on the other,
synthesizability, which typically involves evaluating metastable
lifetimes and reaction energies. Thanks to a number of
methodological developments in the last 20 years, reliable
structure prediction can nowadays be efciently performed
without any prior knowledge or assumptions about the system
(Goedecker, 2004;Oganov et al., 2019;Tong et al., 2019). The
ability of these methods to predict not only the ground states, but
also low-energy metastable structures is indeed leading to the
identication of an increasing number of new virtual materials.
Thermodynamic considerations narrow down the chemical space
for where experimentalists should look (Szczypinski et al., 2021)
and indicate the synthesis probability of stable and metastable
structures in a rst rung approach (Aykol et al., 2018). Yet, the
problem of synthesizability remains. As a consequence, the
continuous proposition of new virtual materials with optimal
properties is often seen from experimentalists as a dreamland of
unachievable real materials. Without an efcient way to assess
actual synthetic routes towards novel stable compounds,
theoretical materials discovery is severely hindered. The
problem of synthesizability is exceptionally hard to solve
because as it needs to be addressed in a holistic manner. In
principle, predicting feasible synthetic routes for a new material
requires not only nding the lowest energy structures of
candidate reactants and products, but also proposing plausible
multi-step reaction mechanisms (including possible metastable
compounds) and computing transition state structures. Headway
is being made and new strategies are being proposed to
incorporate the dynamics of these complex chemical spaces.
One such strategy is the high-throughput analysis of possible
reaction pathways to target a specic inorganic crystal phase by
through reaction energies of reactants, the number of competing
phases and approximated nucleation barriers, at each step,
thereby identifying preferential synthesis routes (Aykol et al.,
2021). An alternative strategy employs the use of neural networks
to generate synthesis predictions for inorganic materials by
mining the scientic literature (Kim et al., 2020). This
approach would benet from the multitude of synthesis
data from unsuccessful experimental attempts, if such data
2021. However, the use of experimental synthesis data in data
driven approaches has been shown to have anthropogenic
biases in the choice of reagents and reaction conditions that
may ultimately lead to skewed networks (Jia et al., 2019). In a
computational approach the consideration of both
thermodynamics and kinetics along reaction pathways could
target the synthesis of any hypothetical material with
properties of interest, but leads to the exploration of large
chemical spaces and requires the use of sophisticated and
computationally demanding methods.
In this perspective we focus on a strategy based on the
application of network science (Barabási, 2016) that is starting
to gain momentum, using the power of network-based
representations and topological analysis to examine solid-state
chemical reactivity for materials discovery, specically a graph
based approach to mapping the thermodynamic relationships
between different materials. This bridge between the discovery of
new virtual materials and the simultaneous identication of likely
synthetic routes could guide experiments and accelerate materials
design and development.
Networks are very simple models, yet extremely useful to
represent complex systems, where the components of the
graph system are represented by nodes and their interactions
by links or edges. These links can be undirected (lines) or
directed (arrows), depending on the systems nature. For
example, a molecular chemical reaction network can be
represented as a directional connected graph. The reactants,
traverse a complex chemical space along reaction pathways
(links) that are governed by kinetics, through intermediates
(nodes) breaking and forming bonds, before nally reaching
the desired products. In contrast, in the crystalline network of
a solid, the nodes represent atoms and the links (bonds) are
undirected (Blatov et al., 2019,2021). What makes networks
useful is that their interaction structure (i.e., the networks
topology) accounts for their systemic properties and,
therefore, topological analysis can lead to applicable,
impactful outcomes. Topological characterization of
networks includes centrality analysis by computing average
Frontiers in Chemistry | December 2021 | Volume 9 | Article 7988382
Aziz and Carrasco Network Science for Inorganic Synthesis
degree and degree distributions (the degree is the number
of links a node has to other nodes) as well as other more
complex characteristics such as clustering coefcients,
betweenness, or hierarchy (Barabási, 2016). Figure 1
illustrates how some of these topological characterizations
can be useful, with a simple materials network example
built up from experimental thermochemistry data, to
analyse inorganic reactivity and identify common nodes in
large chemical spaces.
With the availability of computational materials databases and
the development of network theory we now have the underlying
data and technical know-how to utilize network science in
material discovery. To date there have been a few
representative studies modelling chemical spaces using
networks that have predominately been focused on fragment-
based drug discovery and ligand-based screening of organic
molecules (Tanaka et al., 2009;Kunimoto et al., 2017). In
deciphering reaction mechanisms a novel approach employs
the PageRank algorithm as a collective variable to graph the
possible molecular topologies along a specic reaction pathway
(Zhou et al., 2019). Taking a more general approach, the
pioneering work by Gothard et al., 2012 demonstrated that
the construction of a directional network from organic
reactions reported in the literature can predict sequential
synthesis steps using specicchemicallters including
functional groups and synthesis conditions in a one-pot
approach. Only recently, has this approach gained the
attention of the inorganic research community; from both a
pure crystal structure prediction perspective (Ahnert et al.,
2017) and in the consideration of synthesizability (Aykol et al.,
2019;Hegde et al., 2020;Blau et al., 2021). From a chemistry,
and materials science point of view network representations
are indeed a good approach to tackle synthesizability for the
following reasons:
(i) Chemical reaction spaces are generally very high
dimensional, the need to reduce this dimensionality often
results in a loss of information. Network representations
avoid this issue as there is no need for the construction of a
coordinate system or for any form of dimensionally
reduction. Networks are a natural representation of
chemical reactions (Choudhury et al., 2020).
(ii) Network science provides an intuitive conceptual
framework to statistically analyse many aspects of
reaction spaces and synthesis strategies, with many
meaningful descriptors (e.g., hubs, communities,
hierarchy, and betweenness, among many others)
(Barabási, 2016).
(iii) The rapidly expanding study of complex networks across a
wide range of disciplines has given rise to a large arsenal of
efcient algorithms and mathematical approaches to
quantify network properties and interpret their
characteristics. This development in network science
paves the way to apply these tools to synthesizability.
FIGURE 1 | A network of A + B C solid-state reactions (see Supplementary Table S1 in the Electronic Supplementary Information for details) is used here with
the clique percolation method implemented in CFinder (Adamcsek et al., 2006) to automatically identify only one common node (B
) between the two communities in
the Li-B-O-Na chemical space (A). Additionally, the dendrogram generated by the Girvan-Newman algorithm in (B) using Networkx helps to syste matically reproduce the
modules built into the network (Hagberg et al., 2008).
Frontiers in Chemistry | December 2021 | Volume 9 | Article 7988383
Aziz and Carrasco Network Science for Inorganic Synthesis
Hegde and coworkers have recently developed a unidirectional
materials network encoding the thermodynamic stability (at T
0 K) in the OQMD database (Hegde et al., 2020). The network
comprises of 21,300 nodes (inorganic compounds) with each
node able to connect to 3,850 edges, which represents the
number of two phase equilibria (thermodynamic equilibrium)
between phases, and highlights the dense nature of the network.
The comprehensive mapping of this materials network allows a
top down approach to tackling material stability, as a materials
nobility is measured as a function of the count (or number of
edges) of materials it has no reactivity against. As more data is
added this network has the scope to evolve and verify itself. Holes
in the network may identify materials yet to be discovered, and
subsequent topological analysis may offer an approach to realize
them starting with adjacent structures in the network. Their
discovery and synthesis will lead to the validation of the
network model and wide scale acceptance of network theory
as a strategy in materials discovery. In essence similar to the gaps
or holes in the periodic table predicted by Mendeleev in 1869,
with the rst such hole lled with the discovery of gallium in 1875
validating Mendeleevs periodic law.
The progressive development of network analysis may well
guide experimentalists to decipher which stable predicted
structures can indeed be synthesised. As an alternative to
determining synthesizability from thermodynamic
considerations, a novel time analysis approach combined with
machine learning has given a glimpse of how networks could be
utilized in this direction (Aykol et al., 2019). To reduce the size of
the network a subsample is taken, considering only materials that
share an edge with at least one physically stable material in the
same chemical space. An analysis of the network reveals some
interesting insights; the network is determined to be scale-free:
some nodes have a signicantly larger number of edges and are
thus referred to as hubs. This has two implications; materials
missing from the database will not hinder the discovery of others,
but missing hubs imply materials yet to be discovered and
identifying new hubs will accelerate the discovery in those
spaces. Using a machine-learning model based on certain
network properties of materials Aykol et al. (2019) determine
the likelihood of a predicted material in the network to actually be
synthesised but do not give an insight on their synthesis pathway.
In this respect the combination of a series of networks seems
natural. First, a directional network approach to determine the
probability of synthesis of a new material. Subsequently, a
directed reaction network approach to identify low-cost and
plausible reaction pathways for its fabrication. Ideally, such an
approach would employ optimized pathnding algorithms
similar to those in car navigation systems where one starts at
point A (the reactant) and nishes at point B (the product) whilst
choosing the quickest routes dependent on the trafc (kinetics),
but also considering intermediates, radicals, and ions, which will
have different stabilities dependent on their phase and synthesis
conditions, all whilst maintaining stoichiometric constraints.
This complexity is a signicant challenge that limits the size of
such a reaction network (Unsleber and Reiher, 2020). In this
regard neural networks have shown promise in navigating the
huge network space in organic molecular systems. Recently, a
three layered neural network has been able to uncover
retrosynthetic routes through the use of Monte Carlo tree
search algorithms (Segler et al., 2018) based on reactions
found in the Reaxys database, and we refer the reader to a
recent review on machine-learning methods for more
information (Meuwly, 2021). Compared to molecular
synthesis, inorganic synthesis prediction is more complex,
given the sheer number of elements, metastability and the
possibility of new unchartered materials. However, materials
networks have made progress, interdependencies between
materials have now been implemented in a directional
network that estimates the cost of going from reactant to
product ensuring stoichiometry is preserved along the path
(Blau et al., 2021). To ensure stoichiometry the network space
is continually expanded to ensure all the costs of producing or
removing the additional reactants required in the network are
accounted for. The network determines the cost solely on
thermodynamic considerations, but as databases expand, other
parameters such as kinetics, experimental reaction yields, or the
cost of precursors and their toxicity could also be included.
Indeed this has been demonstrated in subsequent work
expanding their network to include local chemical potential
(Todd et al., 2021). The success of the network is illustrated
by its ability to identify both proposed and novel-pathways in the
formation of lithium ethylene dicarbonate that forms at the solid
electrolyte interphase at the anode of lithium ion batteries.
Despite 6,000 species being needed with the analysis of over
4.5 million reactions the complete network was deduced on a
laptop in less than a day, highlighting the power of such a tool.
The 5 shortest pathwaysor most likely synthesis routes are
identied, two of which have previously been purported in the
literature (Blau et al., 2021). The omission of kinetics in the
network may lead to certain reaction pathways being omitted or
identied but unfeasible. One way to incorporate kinetics is their
subsequent manual consideration once a set of lowest cost
pathways is identied. This approach is employed to
determine whether lithium ethylene monocarbonate or
dicarbonate forms at the solid electrolyte interphase (Xie et al.,
2021). After construction of the graph reaction network and
elimination of duplicate pathways, the predominant pathways are
analysed, leading to the conclusion that paths without the
presence of water are kinetically unfeasible due to large energy
barriers. The requirement of water in the reaction pathway limits
the formation of lithium ethylene monocarbonate and also
suggests varying the water content at the interface could
control the ratio of formation of lithium ethylene
monocarbonate or dicarbonate. Such an insight is clearly
invaluable for experiments.
A somewhat simpler graph-based network that considers only
the thermodynamics of solid-state reactions built up from the
Material Project database and utilizes machine learning has
shown promise in predicting complex reaction pathways
(McDermott et al., 2021). Again, only taking into account
thermodynamic considerations both negative and positive free
Frontiers in Chemistry | December 2021 | Volume 9 | Article 7988384
Aziz and Carrasco Network Science for Inorganic Synthesis
energies are mapped as positive costs using the softplus function
(Dugas et al., 2001). This is a standard practise to ensure standard
pathnding algorithms can be used. Without kinetic
considerations this network is sufcient to predict the complex
reaction pathways reported in the literature for YMnO
and YBa
. Derived reaction routes
may well include hypothetical intermediates; in the case of
, the hypothetical compound Li
is identied
and ignored in the study. In the case of Fe
, a system of
only three elements less stringent constraints on metastability
above the hull of 0.5 eV per atom can be incorporated. This
highlights that even with relatively straightforward
thermodynamic network models trade-offs are still required.
Indeed, the maximum number of reaction pathways
(pathnding processes) and reaction combinations in reaching
the nal product are also of consideration and are set as
parameters in the network. The power of this network model
is demonstrated by the possibility of synthesis by design, with
the suggested synthesis routes for a hypothetical material
O that has been predicted to have superior
mobility (Rong et al., 2017). It is now also possible to
visualize certain available database online ( through
the MaterialNet interactive map (Choudhury et al., 2020). In
Figure 2 we take Na
, an undiscovered hypothetical
material reported in the literature (Gao et al., 2019) and use
the Materials Stability Network to identify other similar materials
in its chemical space and nd its expected synthesis probability to
be 99.4%. The next step in this top down approach would be to
identify possible synthesis pathways followed by experimental
validation. The identication of possible synthesis routes would
help experimentalists reduce the number of reactions pathways to
consider even if ultimately the network failed to predict the
optimal reaction pathway.
Advances in network models complemented with the recent
explosion of materials databases presents an opportunity to
develop a new pioneering research area in materials discovery
FIGURE 2 | Visualization of a local network for the hypothetical (undiscovered) material Na
(Gao et al., 2019) generated using the MaterialNet web application
(Choudhury et al., 2020) and expected to have a 99.4% probability of synthesis. To illustrate the local network environment Na and NaO derived materials are also added
to the chemical subspace. A reaction network (McDermott et al., 2021) could then be employed to identify the most likely synthesis pathways. In the structura lmodel Na
atoms are shown in yellow, Mn atoms in purple and O atoms in red.
Frontiers in Chemistry | December 2021 | Volume 9 | Article 7988385
Aziz and Carrasco Network Science for Inorganic Synthesis
and synthesis. Holes or gaps in networks may help identify
materials yet to be discovered and predictive synthesis routes
identied. To ensure the network representations are an accurate
representation of the chemical space, one must ensure the data is
complete, accurate and with no inherent bias. From a
computational perspective network models are highly
dependent on their original data and the difculty in standard
density functional theory approaches in dealing with correlated
systems raises questions on the validity of the f-block (and to a
lesser extent later d-block) thermodynamic data, and how to
accurately include them in the network. From an experimental
perspective anthropogenic biases in the choice of reagents and
reaction conditions in experimental synthesis may lead to skewed
networks (Jia et al., 2019). The immense chemical space; for
example, 10
combinations of possible materials for the
quaternary compounds formed from the rst 103 elements
are proposed (Davies et al., 2016), leads to a trade-off between
network size and detailed synthesis prediction. While, the
complete Materials Project network can be analysed to
predict the probability of a hypothetical structure being
synthesized, a much more detailed network is needed to
suggest a synthesis pathway, especially if molecular
precursor reactions are incorporated. Further development
of materials databases and/or machine learning approaches
will also be needed to incorporate kinetic costs or take into
account other considerations such as reaction yields, toxicity,
and congurational disorder or to predict the space group of a
material. Whilst the omission of kinetics and other
considerations, may lead to an incorrect hierarchy of
predicted pathways, the number of synthesis pathways
trialled could be dramatically reduced, maximising an
experimentalist`s research time. As this research area
evolves it will no doubt be an extremely powerful technique
to add to the arsenal available to the material-science
The original contributions presented in the study are included in
the article/Supplementary Material, further inquiries can be
directed to the corresponding author.
JC conceived the topic and supervised the work. AA reviewed the
current literature and prepared the manuscript. JC and AA jointly
prepared the nal submitted version.
This work is part of R and D and I project PID2019-
106519RB-I00 funded by MCIN/AEI/10.13039/501100011033.
The Supplementary Material for this article can be found online at:
Adamcsek, B., Palla, G., Farkas, I. J., Derényi, I., and Vicsek, T. (2006).
CFinder: Locating Cliques and Overlapping Modules in Biological
Networks. Bioinformatics 22, 10211023. doi:10.1093/bioinformatics/
Agrawal, A., and Choudhary, A. (2016). Perspective: Materials Informatics and Big
Data: Realization of the Fourth Paradigmof Science in Materials Science. APL
Mater. 4, 053208. doi:10.1063/1.4946894
Ahnert, S. E., Grant, W. P., and Pickard, C. J. (2017). Revealing and Exploiting
Hierarchical Material Structure through Complex Atomic Networks.
npj Comput. Mater. 3, 35. doi:10.1038/s41524-017-0035-x
Aykol, M., Dwaraknath, S. S., Sun, W., and Persson, K. A. (2018). Thermodynamic
Limit for Synthesis of Metastable Inorganic Materials. Sci. Adv. 4. eaaq0148.
Aykol, M., Hegde, V. I., Hung, L., Suram, S., Herring, P., Wolverton, C., et al.
(2019). Network Analysis of Synthesizable Materials Discovery. Nat. Commun.
10, 17. doi:10.1038/s41467-019-10030-5
Aykol, M., Montoya, J. H., and Hummelshøj, J. (2021). Rational Solid-State
Synthesis Routes for Inorganic Materials. J. Am. Chem. Soc. 143, 92449259.
Barabási, A.-L. (2016). Network Science. Northeastern University, Boston:
Cambridge University Press. Available at:
Blatov, V. A., Golov, A. A., Yang, C., Zeng, Q., and Kabanov, A. A. (2019). Network
Topological Model of Reconstructive Solid-State Transformations. Sci. Rep. 9,
6007. doi:10.1038/s41598-019-42483-5
Blatov, V. A., Yang, C., Tang, D., Zeng, Q., Golov, A. A., and Kabanov, A. A.
(2021). High-throughput Systematic Topological Generation of Low-
Energy Carbon Allotropes. npj Comput. Mater. 7, 15. doi:10.1038/
Blau, S. M., Patel, H. D., Spotte-Smith, E. W. C., Xie, X., Dwaraknath, S., and
Persson, K. A. (2021). A Chemically Consistent Graph Architecture for Massive
Reaction Networks Applied to Solid-Electrolyte Interphase Formation. Chem.
Sci. 12, 49314939. doi:10.1039/D0SC05647B
Carrete, J., Li, W., Mingo, N., Wang, S., and Curtarolo, S. (2014). Finding
Unprecedentedly Low-Thermal-Conductivity Half-Heusler Semiconductors
via High-Throughput Materials Modeling. Phys. Rev. X 4, 11019.
Choudhury, R., Aykol, M., Gratzl, S., Montoya, J., and Hummelshøj, J. (2020).
MaterialNet: A Web-Based Graph Explorer for Materials Science Data. Joss 5,
2105. doi:10.21105/joss.02105
Davies, D. W., Butler, K. T., Jackson, A. J., Morris, A., Frost, J. M., Skelton, J. M.,
et al. (2016). Computational Screening of All Stoichiometric Inorganic
Materials. Chem 1, 617627. doi:10.1016/j.chempr.2016.09.010
Dugas, C., Bengio, Y., Belisle, F., Nadeau, C., and Garcia, R. (2001). Incorporating
Second-Order Functional Knowledge for Better Option Pricing. In Advances in
Neural Information Processing Systems. Cambridge, MA: The MIT Press 13,
Gao, X., Wang, D., Lian, R., Kan, D., Mamoor, M., Wang, C., et al. (2019).
Electronic Properties, Phase Transformation, and Anionic Redox of Monoclinic
Na 2 MnO 3 Cathode Material for Sodium-Ion Batteries: First-Principle
Calculations. ChemElectroChem 6, 39873993. doi:10.1002/celc.201901019
Goedecker, S. (2004). Minima Hopping: An Efcient Search Method for the Global
Minimum of the Potential Energy Surface of Complex Molecular Systems.
J. Chem. Phys. 120, 99119917. doi:10.1063/1.1724816
Frontiers in Chemistry | December 2021 | Volume 9 | Article 7988386
Aziz and Carrasco Network Science for Inorganic Synthesis
Gothard, C. M., Soh, S., Gothard, N. A., Kowalczyk, B., Wei, Y., Baytekin, B., et al.
(2012). Rewiring Chemistry: Algorithmic Discovery and Experimental
Validation of One-Pot Reactions in the Network of Organic Chemistry.
Angew. Chem. Int. Ed. 51, 79227927. doi:10.1002/anie.201202155
Greeley, J., Nørskov, J. K., and Mavrikakis, M. (2002). Electronic Structure and
Catalysis on Metal Surfaces. Annu. Rev. Phys. Chem. 53, 319348. doi:10.1146/
Hagberg, A. A., Schult, D. A., and Swart, P. J. (2008). Exploring Network
Structure, Dynamics, and Function Using NetworkX,in Proceedings of the
7th Python in Science Conference. Editors G. Varoquaux, T. Vaught, and
J. Millman (Pasadena, CA USA), 1115. Available at: http://conference.scipy.
Hegde, V. I., Aykol, M., Kirklin, S., and Wolverton, C. (2020). The Phase Stability
Network of All Inorganic Materials. Sci. Adv. 6, eaay5606. doi:10.1126/
Hill, J., Mulholland, G., Persson, K., Seshadri, R., Wolverton, C., and Meredig, B.
(2016). Materials Science with Large-Scale Data and Informatics: Unlocking
New Opportunities. MRS Bull. 41, 399409. doi:10.1557/mrs.2016.93
Jain, A., Shin, Y., and Persson, K. A. (2016). Computational Predictions of Energy
Materials Using Density Functional Theory. Nat. Rev. Mater. 1, 15004.
Jennings, P. C., Lysgaard, S., Hummelshøj, J. S., Vegge, T., and Bligaard, T. (2019).
Genetic Algorithms for Computational Materials Discovery Accelerated by
Machine Learning. npj Comput. Mater. 5, 46. doi:10.1038/s41524-019-0181-4
Jia, X., Lynch, A., Huang, Y., Danielson, M., Langat, I., Milder, A., et al. (2019).
Anthropogenic Biases in Chemical Reaction Data Hinder Exploratory
Inorganic Synthesis. Nature 573, 251255. doi:10.1038/s41586-019-1540-5
Kim, E., Jensen, Z., van Grootel, A., Huang, K., Staib, M., Mysore, S., et al. (2020).
Inorganic Materials Synthesis Planning with Literature-Trained Neural
Networks. J. Chem. Inf. Model. 60, 11941201. doi:10.1021/acs.jcim.9b00995
Kovnir, K. (2021). Predictive Synthesis. Chem. Mater. 33, 48354841. doi:10.1021/
Kunimoto, R., Vogt, M., and Bajorath, J. (2017). Tracing Compound Pathways
Using Chemical Space Networks. Med. Chem. Commun. 8, 376384.
Lombardo, T., Duquesnoy, M., El-Bouysidy, H., Årén, F., Gallo-Bueno, A.,
Jørgensen, P. B., et al. (2021). Articial Intelligence Applied to Battery
Research: Hype or Reality. Chem. Rev. doi:10.1021/acs.chemrev.1c00108
McDermott, M. J., Dwaraknath, S. S., and Persson, K. A. (2021). A Graph-Based
Network for Predicting Chemical Reaction Pathways in Solid-State Materials
Synthesis. Nat. Commun. 12, 3097. doi:10.1038/s41467-021-23339-x
Meuwly, M. (2021). Machine Learning for Chemical Reactions. Chem. Rev. 121,
1021810239. doi:10.1021/acs.chemrev.1c00033
Montoya, J. H., Winther, K. T., Flores, R. A., Bligaard, T., Hummelshøj, J. S., and
Aykol, M. (2020). Autonomous Intelligent Agents for Accelerated Materials
Discovery. Chem. Sci. 11, 85178532. doi:10.1039/d0sc01101k
Noh, J., Gu, G. H., Kim, S., and Jung, Y. (2020). Machine-enabled Inverse Design of
Inorganic Solid Materials: Promises and Challenges. Chem. Sci. 11, 48714881.
Oganov, A. R., Pickard, C. J., Zhu, Q., and Needs, R. J. (2019). Structure Prediction
Drives Materials Discovery. Nat. Rev. Mater. 4, 331348. doi:10.1038/s41578-
Rong, Z., Xiao, P., Liu, M., Huang, W., Hannah, D. C., Scullin, W., et al. (2017). Fast
Mg2+ Diffusion in Mo3(PO4)3O for Mg Batteries. Chem. Commun. 53,
79988001. doi:10.1039/C7CC02903A
Segler, M. H. S., Preuss, M., and Waller, M. P. (2018). Planning Chemical Syntheses
with Deep Neural Networks and Symbolic AI. Nature 555, 604610.
Stach, E., DeCost, B., Kusne, A. G., Hattrick-Simpers, J., Brown, K. A., Reyes, K. G.,
et al. (2021). Autonomous Experimentation Systems for Materials
Development: A Community Perspective. Matter 4, 27022726. doi:10.1016/
Materials that Can Be Synthesised Chem. Sci. 12, 830840. doi:10.1039/
(2021). Toward Autonomous Design and Synthesis of Novel Inorganic
Materials. Mater. Horiz. 8, 21692198. doi:10.1039/D1MH00495F
Tanaka, N., Ohno, K., Niimi, T., Moritomo, A., Mori, K., and Orita, M. (2009).
Small-World Phenomena in Chemical Library Networks: Application to
Fragment-Based Drug Discovery. J. Chem. Inf. Model. 49, 26772686.
Todd, P. K., McDermott, M. J., Rom, C. L., Corrao, A. A., Denney, J. J.,
Dwaraknath, S. S., et al. (2021). Selectivity in Yttrium Manganese Oxide
Synthesis via Local Chemical Potentials in Hyperdimensional Phase Space.
J. Am. Chem. Soc. 143, 1518515194. doi:10.1021/jacs.1c06229
for Structure Prediction. Chin. Phys. B 28, 106105. doi:10.1088/1674-
Unsleber, J. P., and Reiher, M. (2020). The Exploration of Chemical Reaction
Networks. Annu. Rev. Phys. Chem. 71, 121142. doi:10.1146/annurev-
Van der Ven, A., Deng, Z., Banerjee, S., and Ong, S. P. (2020). Rechargeable Alkali-
Ion Battery Materials: Theory and Computation. Chem. Rev. 120, 69777019.
Xie, X., Clark Spotte-Smith, E. W., Wen, M., Patel, H. D., Blau, S. M., and Persson,
K. A. (2021). Data-Driven Prediction of Formation Mechanisms of Lithium
Ethylene Monocarbonate with an Automated Reaction Network. J. Am. Chem.
Soc. 143, 1324513258. doi:10.1021/jacs.1c05807
Zhou, T., Martinez-Baez, E., Schenter, G., and Clark, A. E. (2019). PageRank as
a Collective Variable to Study Complex Chemical Transformations and
Their Energy Landscapes. J. Chem. Phys. 150, 134102. doi:10.1063/
Conict of Interest: The authors declare that the research was conducted in the
absence of any commercial or nancial relationships that could be construed as a
potential conict of interest.
Publishers Note: All claims expressed in this article are solely those of the authors
and do not necessarily represent those of their afliated organizations, or those of
the publisher, the editors and the reviewers. Any product that may be evaluated in
this article, or claim that may be made by its manufacturer, is not guaranteed or
endorsed by the publisher.
Copyright © 2021 Aziz and Carrasco. This is an open-access article distributed under
the terms of the Creative Commons Attribution License (CC BY). The use,
distribution or reproduction in other forums is permitted, provided the original
author(s) and the copyright owner(s) are credited and that the original publication
in this journal is cited, in accordance with accepted academic practice. No use,
distribution or reproduction is permitted which does not comply with these terms.
Frontiers in Chemistry | December 2021 | Volume 9 | Article 7988387
Aziz and Carrasco Network Science for Inorganic Synthesis
Developing battery materials towards commercial use, from the early discovery through synthesis, processing, scaling up, and eventually to industrial production, may take decades. A notable example is Ni-based layered oxides, which were discovered as early as 1950s and intensively pursued as cathode active materials (CAMs) since the early 90s but have yet to realize their full commercial potential. Significant efforts have been devoted to materials development aiming at improving performance, far less to process development for large-scale synthesis and processing of CAMs. Herein, we present a rational design of calcination for scalable production of Ni-based CAMs. We start with an overview of the current understanding and knowledge gaps hindering rational process design and scaling-up of the calcination process. Then with specific examples, we demonstrate how to tackle those fundamental challenges through in situ characterization and multiscale modeling. We conclude by providing perspectives on the remaining challenges and emerging opportunities in commercial development of Ni-based cathodes, calling for more endeavors in this field.Graphical abstract
Full-text available
This is a critical review of artificial intelligence/machine learning (AI/ML) methods applied to battery research. It aims at providing a comprehensive, authoritative, and critical, yet easily understandable, review of general interest to the battery community. It addresses the concepts, approaches, tools, outcomes, and challenges of using AI/ML as an accelerator for the design and optimization of the next generation of batteries—a current hot topic. It intends to create both accessibility of these tools to the chemistry and electrochemical energy sciences communities and completeness in terms of the different battery R&D aspects covered.
Full-text available
Interfacial reactions are notoriously difficult to characterize, and robust prediction of the chemical evolution and associated functionality of the resulting surface film is one of the grand challenges of materials chemistry. The solid–electrolyte interphase (SEI), critical to Li-ion batteries (LIBs), exemplifies such a surface film, and despite decades of work, considerable controversy remains regarding the major components of the SEI as well as their formation mechanisms. Here we use a reaction network to investigate whether lithium ethylene monocarbonate (LEMC) or lithium ethylene dicarbonate (LEDC) is the major organic component of the LIB SEI. Our data-driven, automated methodology is based on a systematic generation of relevant species using a general fragmentation/recombination procedure which provides the basis for a vast thermodynamic reaction landscape, calculated with density functional theory. The shortest pathfinding algorithms are employed to explore the reaction landscape and obtain previously proposed formation mechanisms of LEMC as well as several new reaction pathways and intermediates. For example, we identify two novel LEMC formation mechanisms: one which involves LiH generation and another that involves breaking the (CH2)O–C(═O)OLi bond in LEDC. Most importantly, we find that all identified paths, which are also kinetically favorable under the explored conditions, require water as a reactant. This condition severely limits the amount of LEMC that can form, as compared with LEDC, a conclusion that has direct impact on the SEI formation in Li-ion energy storage systems. Finally, the data-driven framework presented here is generally applicable to any electrochemical system and expected to improve our understanding of surface passivation.
Full-text available
Accelerated inorganic synthesis remains a significant challenge in the search for novel, functional materials. Many of the principles which enable “synthesis by design” in synthetic organic chemistry do not exist in solid-state chemistry, despite the availability of extensive computed/experimental thermochemistry data. In this work, we present a chemical reaction network model for solid-state synthesis constructed from available thermochemistry data and devise a computationally tractable approach for suggesting likely reaction pathways via the application of pathfinding algorithms and linear combination of lowest-cost paths in the network. We demonstrate initial success of the network in predicting complex reaction pathways comparable to those reported in the literature for YMnO3, Y2Mn2O7, Fe2SiS4, and YBa2Cu3O6.5. The reaction network presents opportunities for enabling reaction pathway prediction, rapid iteration between experimental/theoretical results, and ultimately, control of the synthesis of solid-state materials. Predictive computational approaches are fundamental to accelerating solid-state inorganic synthesis. This work demonstrates a computational tractable approach constructed from available thermochemistry data and based on a graph-based network model for predicting solid-state inorganic reaction pathways.
Full-text available
Modeling reactivity with chemical reaction networks could yield fundamental mechanistic understanding that would expedite the development of processes and technologies for energy storage, medicine, catalysis, and more. Thus far, reaction networks have been limited in size by chemically inconsistent graph representations of multi-reactant reactions (e.g. A + B → C) that cannot enforce stoichiometric constraints, precluding the use of optimized shortest-path algorithms. Here, we report a chemically consistent graph architecture that overcomes these limitations using a novel multi-reactant representation and iterative cost-solving procedure. Our approach enables the identification of all low-cost pathways to desired products in massive reaction networks containing reactions of any stoichiometry, allowing for the investigation of vastly more complex systems than previously possible. Leveraging our architecture, we construct the first ever electrochemical reaction network from first-principles thermodynamic calculations to describe the formation of the Li-ion solid electrolyte interphase (SEI), which is critical for passivation of the negative electrode. Using this network comprised of nearly 6000 species and 4.5 million reactions, we interrogate the formation of a key SEI component, lithium ethylene dicarbonate. We automatically identify previously proposed mechanisms as well as multiple novel pathways containing counter-intuitive reactions that have not, to our knowledge, been reported in the literature. We envision that our framework and data-driven methodology will facilitate efforts to engineer the composition-related properties of the SEI - or of any complex chemical process - through selective control of reactivity.
In sharp contrast to molecular synthesis, materials synthesis is generally presumed to lack selectivity. The few known methods of designing selectivity in solid-state reactions have limited scope, such as topotactic reactions or strain stabilization. This contribution describes a general approach for searching large chemical spaces to identify selective reactions. This novel approach explains the ability of a nominally "innocent" Na2CO3 precursor to enable the metathesis synthesis of single-phase Y2Mn2O7: an outcome that was previously only accomplished at extreme pressures and which cannot be achieved with closely related precursors of Li2CO3 and K2CO3 under identical conditions. By calculating the required change in chemical potential across all possible reactant-product interfaces in an expanded chemical space including Y, Mn, O, alkali metals, and halogens, using thermodynamic parameters obtained from density functional theory calculations, we identify reactions that minimize the thermodynamic competition from intermediates. In this manner, only the Na-based intermediates minimize the distance in the hyperdimensional chemical potential space to Y2Mn2O7, thus providing selective access to a phase which was previously thought to be metastable. Experimental evidence validating this mechanism for pathway-dependent selectivity is provided by intermediates identified from in situ synchrotron-based crystallographic analysis. This approach of calculating chemical potential distances in hyperdimensional compositional spaces provides a general method for designing selective solid-state syntheses that will be useful for gaining access to metastable phases and for identifying reaction pathways that can reduce the synthesis temperature, and cost, of technological materials.
Solutions to many of the world's problems depend upon materials research and development. However, advanced materials can take decades to discover and decades more to fully deploy. Humans and robots have begun to partner to advance science and technology orders of magnitude faster than humans do today through the development and exploitation of closed-loop, autonomous experimentation systems. This review discusses the specific challenges and opportunities related to materials discovery and development that will emerge from this new paradigm. Our perspective incorporates input from stakeholders in academia, industry, government laboratories, and funding agencies. We outline the current status, barriers, and needed investments, culminating with a vision for the path forward. We intend the article to spark interest in this emerging research area and to motivate potential practitioners by illustrating early successes. We also aspire to encourage a creative reimagining of the next generation of materials science infrastructure. To this end, we frame future investments in materials science and technology, hardware and software infrastructure, artificial intelligence and autonomy methods, and critical workforce development for autonomous research.
Autonomous experimentation driven by artificial intelligence (AI) provides an exciting opportunity to revolutionize inorganic materials discovery and development. Herein, we review recent progress in the design of self-driving laboratories, including robotics to automate materials synthesis and characterization, in conjunction with AI to interpret experimental outcomes and propose new experimental procedures. We focus on efforts to automate inorganic synthesis through solution-based routes, solid-state reactions, and thin film deposition. In each case, connections are made to relevant work in organic chemistry, where automation is more common. Characterization techniques are primarily discussed in the context of phase identification, as this task is critical to understand what products have formed during synthesis. The application of deep learning to analyze multivariate characterization data and perform phase identification is examined. To achieve "closed-loop"materials synthesis and design, we further provide a detailed overview of optimization algorithms that use active learning to rationally guide experimental iterations. Finally, we highlight several key opportunities and challenges for the future development of self-driving inorganic materials synthesis platforms.