The Center Isn’t Always a Safe Haven: Star-Shapedness and Convexity on the Probability Simplex ✛ ✎
Context
In portfolio management, the advice is often to 'diversify to equilibrium' — move towards an equally-weighted portfolio to reduce risk. However, for a certain class of portfolio maps from stochastic portfolio theory, this strategy can backfire: edging towards the center can push the portfolio into short-position territory. Understanding the geometry of these regions is crucial for portfolio optimization theory and the analysis of dynamic strategies.
Methods
The authors employ a variational approach and differential relations to derive a normal form of constraints. The coordinate-aggregation decomposition paves the way for methods of nonlinear analysis, specifically the analysis of Jensen differences — a measure of the nonlinearity of the aggregation function. On the simplex graph, explicit examples are constructed with three vertices, using quadratic and power generating functions.
Results
A geometric mechanism for non-convexity is discovered: if the aggregation function has positive curvature at the barycenter, then there exist pairs of portfolios arbitrarily close to the center such that the midpoint of the segment violates the long-only constraint. For instance, in the three-dimensional case, the portfolios (3/8, 3/8, 1/4) and (1/4, 3/8, 3/8) are admissible, but their half-sum (5/16, 3/8, 5/16) is not. This is shown by direct computation — no heavy theory needed. Moreover, even star-shapedness with respect to the barycenter can break down: a configuration is found where a point lies in the long-position region, but its projection onto the line to the center leaves it. At the same time, sufficient conditions of concavity of the aggregation function or dominance of the second derivative restore stability. The entropic case is singled out as a limiting affine variant, where constraints reduce to simple coordinate thresholds.
Implications
These findings refine the limits of applicability of dynamic portfolio models and stochastic methods of capital management. They serve as a warning that the popular heuristic 'uniform diversification minimizes short-sale risk' is mathematically incorrect for many symmetric generators. This prompts a rethink of optimization algorithms in robo-advisory.
Future development
The immediate task is to develop a theory of the boundary of long-position regions, classifying entry and exit directions. Another direction is the search for similar convexity-breaking mechanisms beyond the symmetric separable class, where the structure of relationships between coordinates is more intricate.
Impact
The results will impact financial mathematics, computational geometry, and control theory — especially in problems with no-short-selling constraints.
Next steps
Next steps include analyzing asymmetric and non-separable generators, as well as testing the robustness of properties under small perturbations of the aggregation function.
Key open problems
Classifying long-position regions for arbitrary functionally generated portfolios remains open. This problem resonates with questions of convexity in the calculus of variations and with the geometry of differential constraints.
The knowledge machine's view
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The authors study the geometry of admissible portfolios on the probability simplex for symmetric separable functionally generated mappings. They show that the aggregating term can destroy convexity and star-shapedness with respect to the barycenter, providing concrete examples and sufficient conditions for preserving these properties.
Where its strength is
A strong point is the explicit analytical separation of the coordinate and aggregating terms, which reduces the complex geometry to properties of a single function, and the demonstration of the non-convexity mechanism on simple examples.
Where it could go next
- It might be interesting to see how similar violations of convexity arise in other systems with aggregation, for example in models of cell adhesion, where collective behavior is determined by nonlinear interactions.based on: How Cells Stick Together or Scatter arXiv:2604.26928
- Closely related is work on disorder-induced phase transitions; perhaps one could transfer methods from critical phenomena analysis to study the boundaries of the admissible region under random perturbations of the generating function.based on: When Disorder Gives Birth to Order arXiv:2605.08967
- The technique of decomposing into coordinate and aggregating parts could be useful for analyzing universality near singular points, as in work on why different systems behave the same way.based on: Why Different Systems Behave the Same arXiv:2607.09079
Worth trying
- Check whether non-convexity persists when small noise is added to the aggregating function.
- Investigate how the topology of the admissible region affects portfolio dynamics in a stochastic model.
- Construct numerical examples for a larger number of assets and study the scaling of the non-convexity measure.
Nearby works in our archive
- How Cells Stick Together or Scattersimilarity 0.59
- Where Does Life Draw the Line?similarity 0.58
- When Disorder Gives Birth to Ordersimilarity 0.58
- When Charge Loses Its Facesimilarity 0.58
- Why Quantum Resources Melt at the Horizonsimilarity 0.57
- Why Different Systems Behave the Samesimilarity 0.57
The archive is sparse around this work: the closest review is only 0.59, against a usual 0.64. It looks like few have drilled nearby.
In the quadratic example, the boundary of the admissible region for the second coordinate is a circle with radius √(47/80) ≈ 0.77 centered at a vertex of the simplex. So the long-only condition resembles landing inside a lunar crater: you can stand on its rim, but the crater’s center ejects you from the safety zone.
Key numbers
- dimension of counterexample: 3
- weight at barycenter: 1/3 ≈ 0.333
- radius of boundary circle: sqrt(47/80) ≈ 0.77
- depth of violation in second coordinate: -1/640 ≈ 0.00156