diff --git a/crystall.tex b/crystall.tex
index 404f8c4..e36c628 100644
--- a/crystall.tex
+++ b/crystall.tex
@@ -8,7 +8,7 @@ \section{New Stable Structure Prediction?} \label{crystall:sec:newstable}
 
 The ability to determine atomic structure of existing or hypothetical solids formed by chemical elements is one of the fundamental challanges in theoritical crystallography and in materials science as a whole. For the last several decades \cite{Hawthorne1994}, the two remaining challenges are predictions of (a) compositions forming compounds of interest and (b) candidate topologies or approximate atomic arrangements they can form. Given these, the exact structure can be solved experimentally, by matching X-ray diffraction profile to candidates using the method by \citet{LeBail1988} and interpreting the profile, or using deep learning \cite{Yue2024}. Similar analysis can be done, e.g., on the Raman spectrum \cite{Ferrari2013}. For hypothetical compounds in their pure states, the exact structure can also be solved using ab initio quantum mechanical methods, such as Kohn-Sham density functional theory (DFT) \cite{Kohn1965}, which are computationally expensive but can provide accurate results and continue to improve \cite{Kothakonda2023}.
 
-The latter problem of proposing the topologies has been traditionally approached by finding geometries satisfying constraints, like the principles systematized by [@Pauling1929], which included then-novel concepts like coördination. Over decades, this has evolved into automated computational tools like the \texttt{GRINSP} by \citet{LeBail2005} which can propose such structures. More recently, increasing computational power shifted focus towards testing many options, with some of the methods starting from random arrangements and collapsing them into local minima, like \texttt{AIRSS} by \citet{Pickard2011}, or by evolutionary techniques, like \texttt{USPEX} \cite{Glass2006}. 
+The latter problem of proposing the topologies has been traditionally approached by finding geometries satisfying constraints, like the principles systematized by \citet{Pauling1929TheCrystals}, which included then-novel concepts like coördination. Over decades, this has evolved into automated computational tools like the \texttt{GRINSP} by \citet{LeBail2005} which can propose such structures. More recently, increasing computational power shifted focus towards testing many options, with some of the methods starting from random arrangements and collapsing them into local minima, like \texttt{AIRSS} by \citet{Pickard2011}, or by evolutionary techniques, like \texttt{USPEX} \cite{Glass2006}. 
 
 Several efforts have been made to tackle both topological and compositional challenges simultaneously, such as high-throughput construction of the \texttt{AFLOW} \cite{Curtarolo2012} DFT Database \cite{Toher2018} (and others, described in Chapter \ref{chap:mpdd}) built by systematically populating the earlier, yet continuously growing, library of prototype structures \cite{Mehl2016} with new chemistries based on the expert knowledge. However, even for a limited set of elements, this problem becomes combinatorically challenging \cite{Krajewski2024Nimplex}. This prompted recent efforts into (1) systematic similarity-driven prediction of substitutions which would result in finding new low-energy structures \cite{Wang2021}; (2) into brute-force substitutions filtered by machine learning (ML) models \cite{Schmidt2023}; and (3) the combination thereof in \citet{Ye2022} and \citet{Merchant2023}, leading to rapid discovery of new structures stable at 0K temperature and 0Pa pressure against competing arrangements.