Comparing Concentration and Separation

In the study of how resources are distributed, economists and statisticians have long relied on tools to measure concentration. Imagine a pile of gold coins scattered across a table. If all the coins sit in one small heap, the distribution is highly concentrated. If they are spread out evenly across the entire surface, the distribution is uniform. For decades, researchers have used specific numbers to describe how "clumped" a single pile of resources is, asking questions like: How much of the world's copper is produced by just a few countries? How unequal is the distribution of wealth in a single nation? These questions usually treat each resource or group in isolation, measuring how far its distribution strays from a perfect average.

However, the world rarely presents us with isolated piles. We often face multiple distributions at once: the production of lithium, cobalt, and nickel, for instance, which are all critical for modern batteries. A traditional analysis might tell us that lithium is highly concentrated in one region and cobalt is highly concentrated in another. But this approach misses a crucial piece of the puzzle: are those two regions the same, or are they completely different? If the countries producing lithium are entirely different from those producing cobalt, the risks to a supply chain are spread out. If they are the same, the risk is doubled. Until now, there has been no standard, unified way to measure not just how concentrated each resource is, but how easily these different distributions can be separated from one another.

A new study by Arkadiusz Maciuk at the Wroclaw University of Economics and Business proposes a fresh way to look at this problem. Instead of treating concentration and separation as two different ideas, the author suggests viewing concentration as a special case of separation. The core idea is simple: if you can easily sort a group of people or countries into distinct teams so that each team captures a large share of a specific resource, then those resources are well separated. If the resources are all mixed together in the same places, they are hard to separate. The paper builds a mathematical framework that allows researchers to compare any number of distributions at once, determining whether one set of distributions is "more separable" than another.

The researchers developed a method that treats the set of agents—such as countries or individuals—as a single group that can be divided in various ways. They imagined a process where every country is split into fractional parts, with each part assigned to a different resource. For example, a country might give 60% of its "attention" to lithium and 40% to cobalt. By calculating all the possible ways to make these splits, the researchers created a map of what is achievable. If one set of resource distributions allows for a wider range of successful splits than another, it is considered more separated. This approach unifies the measurement of a single resource's concentration with the measurement of how distinct multiple resources are from each other.

One of the most significant findings is that this new way of comparing distributions connects directly to how well a system can identify the source of a signal. In the language of decision theory, the ability to separate distributions is equivalent to the ability to correctly guess which resource is being observed, no matter what the initial probabilities are. If a set of distributions is highly separable, an observer can almost always tell which resource they are looking at. If the distributions are not separable, the observer will frequently make mistakes. This link means the new method is not just a geometric exercise; it is a practical tool for understanding how much information is contained in the geographical layout of resources.

The study also clarifies how this new method relates to older, well-known mathematical tools. When looking at just one resource compared to a perfect average, the new method produces the exact same results as the standard measure of inequality used in economics. When looking at two resources, it aligns with established theories about how experiments and information structures compare. However, the paper reveals that when three or more resources are involved, the situation becomes more complex. The new method does not simply reduce to the old rules; it captures a unique kind of relationship that previous tools missed. Specifically, it shows that two sets of resources can look similar under old rules but be fundamentally different in how they can be separated into distinct groups.

To demonstrate the value of this approach, the author applied it to real-world data on the global production of copper, lithium, cobalt, and nickel. They analyzed how these production patterns have changed from 2005 to 2025. The results showed that while the concentration of each metal individually remained high and relatively stable, the separation between them changed dramatically. For instance, the production structures of lithium and cobalt became increasingly distinct over time, with the separation index rising from 0.8686 in 2005 to 0.9781 in 2025. This means that the production structures of lithium and cobalt became increasingly easy to separate into two groups of countries capturing large shares of the respective allocations. In 2025, this interpretation has a direct geographical counterpart: the dominant center of cobalt production is the Democratic Republic of the Congo, whereas nickel production is highly concentrated in Indonesia. This shift means that a supply chain relying on both metals is less vulnerable to a single regional disruption than it would have been two decades ago.

The researchers found that looking at these metals individually would have missed this crucial development. Standard concentration indices showed that lithium remained concentrated and cobalt remained concentrated, but they did not reveal that the locations of that concentration were moving apart. By using the new separation index, the study uncovered a hidden layer of safety in the global supply chain: the risks are becoming geographically diversified even as the individual markets remain tight. This distinction is vital for policymakers and industry leaders who need to understand not just where resources are, but how the locations of different resources relate to one another.

The paper also introduces a family of new indices to measure this separation, ranging from a strict "worst-case" guarantee to an average performance measure. The strictest version asks: what is the minimum amount of each resource that can be guaranteed to a distinct group of countries? The average version asks: what is the total amount of resources that can be captured by the best possible grouping? The study shows that these different perspectives can lead to different conclusions about the same data, offering a more nuanced view of risk and opportunity. For example, while the average separation between lithium, cobalt, and nickel production increased significantly, the strict guarantee for all three simultaneously was slightly lower, indicating that while the overall picture is improving, there are still some overlaps that prevent a perfect separation.

Ultimately, this work provides a new language for describing the geometry of distribution. It moves beyond asking how unequal a single system is to asking how distinct multiple systems are from one another. By treating the problem as one of separation rather than just concentration, the study offers a more complete picture of how resources, populations, and economic activities are organized in space. The findings suggest that in a world of interconnected supply chains, understanding the relative positions of different distributions is just as important as understanding their individual sizes. This new framework allows us to see that while the world may still be full of concentrated resources, the way those concentrations are arranged relative to each other is shifting, creating new patterns of risk and resilience that were previously invisible to standard analysis.