🔢 mathematics

Just Previsions

This paper investigates general previsions as positively homogeneous functionals, demonstrating that they can be represented as infima of sublinear previsions and suprema of superlinear previsions, thereby establishing homeomorphisms between spaces of previsions and specific hyperspaces via a double powerspace construction.

Original authors: Jean Goubault-Larrecq

Published 2026-05-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jean Goubault-Larrecq

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: What is a "Previon"?

Imagine you are a weather forecaster, but instead of predicting rain or sun, you are predicting the value of a "score" for any possible scenario. In math, these scenarios are functions (let's call them $h$ ).

A Previon is a rule that takes a scenario ( $h$ ) and gives you a number (a prediction). It has two special rules:

Scaling: If you double the scenario, your prediction doubles. (If the score is worth 10, and you double the stakes, the prediction is 20).
Continuity: If you slowly change the scenario, your prediction changes smoothly without jumping.

The paper is about understanding the "shape" of all possible rules (previsions) that follow these guidelines. The author wants to know: Can we build any complex prediction rule out of simpler, more rigid building blocks?

The Three Types of Predictors

To understand the main result, we need to meet three types of predictors:

Linear Predictors (The Fair Accountants): These are the "perfect" rules. They follow the rule: Prediction(A + B) = Prediction(A) + Prediction(B). They are like standard probability measures (like flipping a coin).
Sublinear Predictors (The Pessimists): These are cautious. They believe that Prediction(A + B) ≤ Prediction(A) + Prediction(B). They think combining two risky things might be less than the sum of their parts, or they simply add a "safety margin."
Superlinear Predictors (The Optimists): These are hopeful. They believe Prediction(A + B) ≥ Prediction(A) + Prediction(B). They think combining things creates extra value.

The paper focuses on "Just Previsions." These are rules that are not necessarily pessimistic or optimistic. They are the "wild cards" that don't strictly follow the pessimist or optimist rules. The question is: Can we describe these wild cards using only the pessimists and optimists?

The Main Discovery: Building Wild Cards from Simple Blocks

The author proves two amazing things:

1. The Pessimist's View (The Infimum)
Any "wild card" prediction rule can be built by taking the lowest possible value from a collection of "Pessimist" (sublinear) rules.

Analogy: Imagine you want to know the "true" height of a mountain, but you don't have a ruler. Instead, you ask 1,000 pessimistic surveyors. Each one draws a line above the mountain, claiming "The mountain is at least this high." If you take the lowest of all their lines, you get the exact outline of the mountain.
The Math: Every prevision is the pointwise infimum (the lowest point) of sublinear previsions.

2. The Optimist's View (The Supremum)
Conversely, any "wild card" rule can also be built by taking the highest possible value from a collection of "Optimist" (superlinear) rules.

Analogy: Now imagine 1,000 optimists drawing lines below the mountain, saying "The mountain is at least this high." If you take the highest of all their lines, you again get the exact outline of the mountain.
The Math: Every prevision is the pointwise supremum (the highest point) of superlinear previsions.

The "Double Shadow" Construction

The paper goes deeper, describing a geometric way to visualize this.

Imagine you have a room full of "Pessimists" (sublinear rules).

Step 1: You look at a specific "Wild Card" rule. You find all the Pessimists that are "above" it (rules that predict higher values).
Step 2: You treat this group of Pessimists as a single object (a shape in a higher-dimensional space).
Step 3: You look at the "shadow" this group casts. The paper shows that if you take the "shadow" of the "shadow" (a double orthogonality construction), you get your original Wild Card rule back.

This is called a Double Hyperspace Construction.

Metaphor: Think of a mirror. If you look at an object in a mirror, you see a reflection. If you put a second mirror behind the first one, you see the reflection of the reflection. The paper proves that for these prediction rules, looking at the "reflection of the reflection" brings you back to the original object perfectly.

Why Does This Matter? (According to the Paper)

The paper doesn't claim this will cure diseases or predict stock markets. Its value is purely structural and mathematical:

Simplification: It shows that even the most complex, unruly prediction rules are just combinations of simpler, well-behaved rules (the pessimists and optimists).
Topology: It proves that the "space" of all wild cards is mathematically identical (homeomorphic) to a specific, well-understood shape made of these simpler rules.
No Extra Conditions Needed: For the "Pessimist" view (building from sublinear rules), this works for any topological space (any kind of mathematical universe you can imagine). For the "Optimist" view, it works under mild, standard conditions.

Summary in One Sentence

The paper proves that any complex, non-linear prediction rule can be perfectly reconstructed by either taking the lowest limit of all "pessimistic" rules or the highest limit of all "optimistic" rules, effectively showing that the "wild" rules are just the shadows cast by these simpler, more rigid families.

Technical Summary of "Just Previsions" by Jean Goubault-Larrecq

Problem Statement
The paper investigates the structural properties of previsions, defined as positively homogeneous, Scott-continuous functionals mapping the space of lower semicontinuous functions ( $L_X$ ) to the extended non-negative reals ( $\mathbb{R}_+$ ). While previous work (specifically [8]) has extensively modeled mixtures of non-deterministic and probabilistic choice using sublinear or superlinear previsions, this paper focuses exclusively on general previsions without imposing sublinearity, superlinearity, or linearity constraints. The central question is whether general previsions can be characterized, decomposed, or represented in terms of these more restricted classes (sublinear and superlinear previsions) and how the spaces of these functionals relate to hyperspace constructions.

Methodology
The author employs tools from domain theory, topology, and the theory of cones. Key methodological components include:

Hyperspace Constructions: Utilizing Hoare ( $H$ ) and Smyth ( $Q$ ) hyperspaces over spaces of previsions. Specifically, the paper examines convex compact saturated subsets (Smyth) and convex closed subsets (Hoare) of spaces of sublinear and superlinear previsions.
Orthogonality Relations: The paper frames the relationship between sublinear and superlinear previsions through a Galois connection (polarity) defined by an orthogonality relation ( $\perp$ ). For $F^- \in P_{super}$ and $F^+ \in P_{sub}$ , $F^- \perp F^+$ is defined as $F^- \leq F^+$ . The study focuses on the "double orthogonals" ( $A^{\perp\perp}$ ) arising from this relation.
Cast Shadow Construction: A technical geometric construction (Section 4) is introduced to handle subnormalized and normalized previsions. This involves "casting shadows" of sets in a cone to preserve convexity and openness properties necessary for the representation theorems.
Retractions and Isomorphisms: The paper constructs continuous maps (retractions) between the space of general previsions ( $P^\star X$ ) and specific hyperspaces of sublinear/superlinear previsions, proving they are homeomorphisms under certain conditions.

Key Contributions and Results

Decomposition of Previsions:
- Sublinear Decomposition: The paper proves that every prevision (on any topological space $X$ ) can be expressed as the pointwise infimum of sublinear previsions.
- Superlinear Decomposition: Under mild conditions (specifically, that $L_X$ is a "convenient cone," which holds if $X$ is core-compact, and $X$ is compact and non-empty if the previsions are normalized), every prevision can be expressed as the pointwise supremum of superlinear previsions.
Isomorphism Theorems:
The paper establishes two primary homeomorphisms relating the space of general previsions $P^\star X$ to "double orthogonals" in hyperspaces:
- First Isomorphism: $P^\star X \cong Q^{\perp\perp}_v P^\star_{sub} X$ . The space of previsions is homeomorphic to the space of non-empty, convex, compact, saturated subsets of sublinear previsions that are double orthogonals with respect to the order relation. This holds for any topological space $X$ .
- Second Isomorphism: $P^\star X \cong H^{\perp\perp}_v P^\star_{super} X$ . The space of previsions is homeomorphic to the space of non-empty, convex, closed subsets of superlinear previsions that are double orthogonals. This requires $L_X$ to be a convenient cone (e.g., $X$ is core-compact).
Retraction Structures:
The paper demonstrates that $P^\star X$ is a retract of the larger convex hyperspaces $Q^{cvx}_v P^\star_{sub} X$ and $H^{cvx}_v P^\star_{super} X$ . The retraction maps are defined via pointwise infimum ( $min_X$ ) and pointwise supremum ( $sup_X$ ) over the respective sets of sublinear or superlinear previsions.
Algebraic Structure and Limitations:
While $P^\star X$ inherits a cone structure, the paper shows that the embedding of $P^\star X$ into the enclosing hyperspaces does not preserve algebraic operations (addition and convex combinations) in the standard way. Specifically, the infimum or supremum of two previsions in $P^\star X$ does not necessarily correspond to the convex hull of their images in the hyperspace. Examples are provided (using discrete spaces) to show that the double orthogonals form a proper subspace of the full convex hyperspaces and that the operations differ between the two spaces.

Significance and Claims
The author claims that these results extend the understanding of previsions beyond the linear, sublinear, or superlinear cases. By characterizing general previsions as double orthogonals, the paper connects the theory of previsions to the double hyperspace construction (specifically the de Brecht-Kawai construction relating Hoare and Smyth hyperspaces).

The paper argues that this construction is a variant of the double powerspace construction, elucidated through orthogonality relations. The significance lies in providing a unified topological and order-theoretic framework where general previsions are seen as "fixed points" (double orthogonals) within spaces of more restricted functionals. The paper explicitly notes that these characterizations hold without requiring $X$ to be a specific type of space (like a Polish space) for the sublinear case, though stronger conditions are needed for the superlinear case to ensure the existence of the required separating functionals.

The work does not propose new applications or experimental proposals but rather provides a foundational theoretical clarification of the structure of prevision spaces, linking them to established concepts in domain theory and hyperspace topology.