In 1973, Andrey Kolmogorov proposed a non-probabilistic approach to statistics and model selection. Let each datum be a finite binary string and a model be a finite set of binary strings. Consider model classes consisting of models of given maximal Kolmogorov complexity. The Kolmogorov structure function of an individual data string expresses the relation between the complexity level constraint on a model class and the least log-cardinality of a model in the class containing the data. The structure function determines all stochastic properties of the individual data string: for every constrained model class it determines the individual best-fitting model in the class irrespective of whether the true model is in the model class considered or not. In the classical case we talk about a set of data with a probability distribution, and the properties are those of the expectations. In contrast, here we deal with individual data strings and the properties of the individual string focused on. In this setting, a property holds with certainty rather than with high probability as in the classical case. The Kolmogorov structure function precisely quantifies the goodness-of-fit of an individual model with respect to individual data.

The Kolmogorov structure function is used in the algorithmic information theory, also known as the theory of Kolmogorov complexity, for describing the structure of a string by use of models of increasing complexity.

The structure function was originally proposed by Kolmogorov in 1973 at a Soviet Information Theory symposium in Tallinn, but these results were not published p. 182. But the results were announced in in 1974, the only written record by Kolmogorov himself. One of his last scientific statements is (translated from the original Russian by L.A. Levin):

To each constructive object corresponds a function ${\displaystyle \Phi _{x}(k)}$ of a natural number k—the log of minimal cardinality of x-containing sets that allow definitions of complexity at most k. If the element x itself allows a simple definition, then the function ${\displaystyle \Phi }$ drops to 0 even for small k. Lacking such definition, the element is "random" in a negative sense. But it is positively "probabilistically random" only when function ${\displaystyle \Phi }$ having taken the value ${\displaystyle \Phi _{0}}$ at a relatively small ${\displaystyle k=k_{0}}$, then changes approximately as ${\displaystyle \Phi (k)=\Phi _{0}-(k-k_{0})}$.

— Kolmogorov, announcement cited above

It is discussed in Cover and Thomas. It is extensively studied in Vereshchagin and Vitányi where also the main properties are resolved. The Kolmogorov structure function can be written as $${\displaystyle h_{x}(\alpha )=\min _{S}\{\log |S|:x\in S,K(S)\leq \alpha \}}$$ where ${\displaystyle x}$ is a binary string of length ${\displaystyle n}$ with ${\displaystyle x\in S}$ where ${\displaystyle S}$ is a contemplated model (set of n-length strings) for ${\displaystyle x}$, ${\displaystyle K(S)}$ is the Kolmogorov complexity of ${\displaystyle S}$ and ${\displaystyle \alpha }$ is a nonnegative integer value bounding the complexity of the contemplated ${\displaystyle S}$'s. Clearly, this function is nonincreasing and reaches ${\displaystyle \log \left|\{x\}\right|=0}$ for ${\displaystyle \alpha =K(x)+c}$ where ${\displaystyle c}$ is the required number of bits to change ${\displaystyle x}$ into ${\displaystyle \{x\}}$ and ${\displaystyle K(x)}$ is the Kolmogorov complexity of ${\displaystyle x}$.

We define a set ${\displaystyle S}$ containing ${\displaystyle x}$ such that $${\displaystyle K(S)+K(x|S)=K(x)+O(1).}$$ The function ${\displaystyle h_{x}(\alpha )}$ never decreases more than a fixed independent constant below the diagonal called sufficiency line L defined by $${\displaystyle L(\alpha )+\alpha =K(x).}$$ It is approached to within a constant distance by the graph of ${\displaystyle h_{x}}$ for certain arguments (for instance, for ${\displaystyle \alpha =K(x)+c}$). For these ${\displaystyle \alpha }$'s we have ${\displaystyle \alpha +h_{x}(\alpha )=K(x)+O(1)}$ and the associated model ${\displaystyle S}$ (witness for ${\displaystyle h_{x}(\alpha )}$) is called an optimal set for ${\displaystyle x}$, and its description of ${\displaystyle K(S)\leq \alpha }$ bits is therefore an algorithmic sufficient statistic. We write `algorithmic' for `Kolmogorov complexity' by convention. The main properties of an algorithmic sufficient statistic are the following: If ${\displaystyle S}$ is an algorithmic sufficient statistic for ${\displaystyle x}$, then $${\displaystyle K(S)+\log |S|=K(x)+O(1).}$$ That is, the two-part description of ${\displaystyle x}$ using the model ${\displaystyle S}$ and as data-to-model code the index of ${\displaystyle x}$ in the enumeration of ${\displaystyle S}$ in ${\displaystyle \log |S|}$ bits, is as concise as the shortest one-part code of ${\displaystyle x}$ in ${\displaystyle K(x)}$ bits. This can be easily seen as follows: $${\displaystyle {\begin{aligned}K(x)&\leq K(x,S)+O(1)\\&\leq K(S)+K(x|S)+O(1)\\&\leq K(S)+\log |S|+O(1)\\&\leq K(x)+O(1),\end{aligned}}}$$

using straightforward inequalities and the sufficiency property, we find that ${\displaystyle K(x|S)=\log |S|+O(1)}$. (For example, given ${\displaystyle S\ni x}$, we can describe ${\displaystyle x}$ self-delimitingly (you can determine its end) in ${\displaystyle \log |S|+O(1)}$ bits.) Therefore, the randomness deficiency ${\displaystyle \log |S|-K(x|S)}$ of ${\displaystyle x}$ in ${\displaystyle S}$ is a constant, which means that ${\displaystyle x}$ is a typical (random) element of S. However, there can be models ${\displaystyle S}$ containing ${\displaystyle x}$ that are not sufficient statistics. An algorithmic sufficient statistic ${\displaystyle S}$ for ${\displaystyle x}$ has the additional property, apart from being a model of best fit, that ${\displaystyle K(x,S)=K(x)+O(1)}$ and therefore by the Kolmogorov complexity symmetry of information (the information about ${\displaystyle x}$ in ${\displaystyle S}$ is about the same as the information about ${\displaystyle S}$ in x) we have ${\displaystyle K(S|x^{*})=O(1)}$: the algorithmic sufficient statistic ${\displaystyle S}$ is a model of best fit that is almost completely determined by ${\displaystyle x}$. (${\displaystyle x^{*}}$ is a shortest program for ${\displaystyle x}$.) The algorithmic sufficient statistic associated with the least such ${\displaystyle \alpha }$ is called the algorithmic minimal sufficient statistic.