diff --git a/implementation.tex b/implementation.tex
index 7777897454f839acf22d00fcc7e93a6d1dd440aa..79d3d8621fea6f0f253ac8c90d4ac82143beb5e0 100644
--- a/implementation.tex
+++ b/implementation.tex
@@ -56,6 +56,7 @@ The resource requirement of the frame applies to all tasks inside the frame.}
 \subsection{Genetic Algorithm}
 Our genetic algorithm is based on the implementation by \citet{Alexeev2014SimpleGeneticAlgorithm}.
 We ported his code from Haskell to Clean and extended it in a number of ways to make it fit our needs.
+The algorithm has four input parameters: the problem instance, the population size, the mutation probability, and the number of generations.
 
 \paragraph{Encoding}
 Our chromosome encoding scheme is based on the one by \citet{MyszkowskiLNS2018}, extended with support for tasks with multiple skill requirements.
@@ -77,13 +78,21 @@ Single-point crossover takes two chromosomes, randomly determines a cut-off poin
 The following table illustrates this scheme with two parents A and B.
 \begin{IEEEeqnarray*}{l"l}
 \textrm{Parents} & \textrm{Children}
-\\ \relax [A,A,A,A,A,A] & [A, A, B, B, B, B]
-\\ \relax [B,B,B,B,B,B] & [B, B, A, A, A, A]
+\\ \relax [A,A,A,A,A,A] & [A,A,B,B,B,B]
+\\ \relax [B,B,B,B,B,B] & [B,B,A,A,A,A]
 \end{IEEEeqnarray*}
 
+
 \paragraph{Mutation}
+Our mutation operation randomly determines a location in a given chromosome and changes it to a randomly selected entry in the corresponding domain.
 
 \paragraph{Selection}
+We use the selection method of \citet{Alexeev2014SimpleGeneticAlgorithm}, where all chromosomes in a population are paired with each other.
+This makes the algorithm run with quadratic complexity in the population size, so it is advised to use reasonably small populations.
+The children of crossover are subjected to mutation with the given mutation probability.
+Alexeev discards the old population and continues with the new one.
+We changed this so that the old and new populations are combined, sorted according to the fitness function, and then the best chromosomes are taken to be the new generation.
+This way the algorithm preserves the best individuals found so far.
 
 \paragraph{Constraints and invalid assignments}