The Legendre transform, the Laplace transform and valuations

This paper characterizes the Legendre, Laplace, and identity transforms as the unique continuous, $\mathrm{SL}(n)$ contravariant valuations with specific conjugation or duality properties on spaces of convex and log-concave functions.

The Gist

Imagine you are standing in a vast, multidimensional landscape. In this world, shapes aren't just solid blocks; they are also functions—mathematical recipes that tell you how high a hill is at any given point.

This paper is like a detective story. The author, Jin Li, is trying to identify specific "magic tools" (mathematical transforms) that can reshape this landscape. The goal is to figure out: What makes a tool the "Legendre Transform" or the "Laplace Transform"? Is there a unique set of rules that only these specific tools follow?

Here is the breakdown of the paper using simple analogies.

1. The Rules of the Game: "Valuations"

First, the paper defines what a "Valuation" is. Imagine you have a set of Lego blocks.

• If you take two blocks and snap them together (Union), and then take the overlapping part (Intersection), a Valuation is a rule that says:

  "The value of the big combined block + the value of the tiny overlap = The value of block A + The value of block B."

In math terms, this rule must hold true for functions too. The author is looking for tools that respect this "Lego logic."

2. The First Mystery: The Legendre Transform

The Legendre Transform is a famous tool in physics and economics. It's like a mirror that flips a convex hill upside down and swaps its coordinates. If you have a hill defined by position, the Legendre transform redefines it by "slope."

The Detective Work:
The author asks: "If I give you a tool that acts like a valuation, and it behaves nicely when you stretch or rotate the landscape (SL(n) contravariance), and it has a special relationship with translations (sliding things around), is it the Legendre Transform?"

• The Translation Trick: Imagine sliding a hill to the right.
  • The Legendre Transform has a magical property: If you slide the original hill, the transformed hill doesn't just slide; it gets a "tilt" added to it.
  • Conversely, if you tilt the original hill, the transformed hill just slides.
  • It's like a conjugate dance: One partner slides, the other tilts. They swap roles perfectly.

The Discovery (Theorem 1.1):
The author proves that if a tool is continuous, respects the "Lego logic," respects rotations, and performs this perfect "slide-tilt dance," it must be the Legendre Transform (plus maybe a constant number added to it). No other tool can do this.

3. The Second Mystery: Log-Concave Functions and the Laplace Transform

Next, the author looks at a different type of landscape: Log-Concave functions.

• Think of these as "bell curves" or probability distributions (like the Gaussian curve). They are the mathematical cousins of the hills we just looked at.
• The Laplace Transform is another famous tool used to turn complex differential equations into simple algebra. It's like a translator that turns a difficult story into a simple summary.

The Twist:
When the author tries to apply the same "slide-tilt dance" rules to these bell curves, something surprising happens.

• The Legendre Transform (or its cousin, the "Duality Transform") still works perfectly.
• BUT, the Laplace Transform also fits the rules!

The Discovery (Theorem 1.3):
If you relax the rules just a tiny bit for these bell curves, the "magic tool" isn't just one thing anymore. It turns out to be a mixture.

The tool could be a blend of the Duality Transform (the mirror) and the Laplace Transform (the translator).

It's like saying, "If you have a machine that sorts mail by sliding and tilting, it could be a standard sorter, OR it could be a machine that also scans the letters with a barcode reader." Both fit the description.

4. The "Dual" Perspective

Finally, the author uses a concept called Dual Valuations.

• Imagine you have a map of a city. The "Dual Map" is a map of the roads instead of the buildings.
• The author shows that if you understand the rules for the "hills" (Convex functions), you automatically know the rules for the "flat ground" (Finite convex functions) and the "bell curves" (Log-concave functions) just by flipping the map over.
• This allows them to prove that the Identity Transform (the tool that does nothing but say "Hello, I am the same") is the only tool that behaves a certain way on flat ground.

Summary: What's the Big Deal?

In the world of mathematics, there are many ways to describe shapes and functions. This paper is significant because it uniquely identifies the most important tools in the toolbox.

• Before: We knew what the Legendre and Laplace transforms did.
• Now: We know exactly what they are based on their fundamental behavior (how they handle sliding, tilting, and combining shapes).

The Takeaway:
Just as you can identify a specific type of wrench by how it fits a bolt and how it feels in your hand, this paper proves that the Legendre and Laplace transforms are the only mathematical tools that fit the specific "grip" of being continuous, rotation-friendly, and translation-conjugate. It's a beautiful piece of mathematical detective work that defines the identity of these famous transforms.

Technical Summary

Here is a detailed technical summary of the paper "The Legendre transform, the Laplace transform and valuations" by Jin Li.

1. Problem Statement

The paper addresses the problem of characterizing fundamental integral transforms (specifically the Legendre transform and the Laplace transform) within the framework of valuation theory.

While valuation theory has a long history in characterizing geometric operators on convex bodies (e.g., intrinsic volumes, polar bodies), its application to function spaces is a newer field. Previous characterizations of transforms like the Legendre transform (e.g., by Artstein-Avidan and Milman) relied heavily on strong assumptions such as bijectivity and order-reversing isomorphisms (conjugation of min/max).

The Core Challenge:
The author seeks to characterize these transforms using weaker, more natural assumptions suitable for valuation theory:

1. Valuation Property: The map must satisfy the inclusion-exclusion principle ($Z(f \vee g) + Z(f \wedge g) = Zf + Zg$).
2. Geometric Invariance: The map must be $SL(n)$ contravariant (or covariant).
3. Continuity: The map must be continuous with respect to specific topologies (epi-convergence for convex functions, hypo-convergence for log-concave functions).
4. Translation Behavior: The map must interact with translations in a specific "conjugation" manner.

The paper specifically investigates whether these conditions uniquely identify the Legendre transform on super-coercive convex functions and the Laplace transform on log-concave functions, without assuming bijectivity.

2. Methodology

The author employs a multi-step methodology combining functional analysis, convex geometry, and measure theory:

• Function Spaces:
  • Convex Functions: Focuses on $\text{Conv}_{sc}(\mathbb{R}^n)$ (proper, super-coercive, lower semi-continuous convex functions) and $\text{Conv}(\mathbb{R}^n; \mathbb{R})$ (finite convex functions).
  • Log-Concave Functions: Focuses on $\text{LC}_{sc}(\mathbb{R}^n) = \{e^{-u} : u \in \text{Conv}_{sc}(\mathbb{R}^n)\}$.
• Valuation Definition: Defines a valuation $Z$ on a lattice of functions where the lattice operations are pointwise maximum ($\vee$) and minimum ($\wedge$).
• Translation Conjugation: Introduces the concept of a "translation conjugation." For convex functions, the Legendre transform maps the standard translation $\tau_y u(x) = u(x-y)$ to the dual translation $u + \ell_y$ (where $\ell_y(x) = x \cdot y$), and vice versa.
• Dual Valuations: Utilizes the concept of dual valuations. If $Z$ is a valuation on $\text{Conv}_{sc}$, its dual $Z^*$ is defined on $\text{Conv}(\mathbb{R}^n; \mathbb{R})$ via the Legendre transform ($Z^*(u) = Z(u^*)$). This allows transferring classification results between the two spaces.
• Reduction to Convex Polytopes: The proof strategy involves restricting the valuation to indicator functions of convex polytopes ($I_P$). This reduces the functional problem to a problem on convex bodies, where existing results (e.g., by Ludwig, Blaschke) can be applied.
• Functional Equations: Solves specific functional equations (Cauchy-type and exponential-type) derived from the translation and $SL(n)$ properties to determine the exact form of the coefficients in the valuation.
• Dimensional Constraints: The main theorems are established for dimensions $n \geq 3$, relying on specific rigidity results for $SL(n)$ valuations in higher dimensions.

3. Key Contributions and Results

A. Characterization of the Legendre Transform

Theorem 1.1: For $n \geq 3$, a transform $Z: \text{Conv}_{sc}(\mathbb{R}^n) \to F(\mathbb{R}^n; \mathbb{R})$ is a continuous, $SL(n)$ contravariant valuation that acts as a translation conjugation (mapping $\tau_y u \to Z(u) + \ell_y$ and $u + \ell_y \to \tau_y Z(u)$) if and only if:
$$Zu = u^* + c$$
where $u^*$ is the Legendre transform and $c$ is a constant.

• Significance: This removes the need for the bijectivity assumption found in previous works (Artstein-Avidan & Milman). It proves that the Legendre transform is the unique valuation with these specific symmetry and translation properties.

B. Characterization of the Laplace Transform on Log-Concave Functions

The paper distinguishes between two types of translation behaviors on log-concave functions ($f = e^{-u}$):

1. Strict Translation Conjugation: If $Z$ satisfies the conjugation for both standard and dual translations, it characterizes the duality transform ($f^\circ = e^{-u^*}$).
   • Theorem 1.2: $Zf = c f^\circ$.
2. Modified Translation Conjugation: If the sign in the translation relation is altered (specifically, $Z(\tau_y f) = e^{\ell_y} Z(f)$), the Laplace transform appears.
   • Theorem 1.3: $Z$ is a continuous, $SL(n)$ contravariant valuation satisfying the modified translation relations if and only if:
     $$Zf = c_1 f^\circ + c_2 \mathcal{L}f$$
     where $\mathcal{L}f(x) = \int_{\mathbb{R}^n} e^{x \cdot y} f(y) dy$ is the Laplace transform.

C. Characterization of the Identity Transform

Using the concept of dual valuations, the paper derives characterizations for the identity transform on finite convex functions and positive log-concave functions.

• Theorem 6.1: On finite convex functions, the only continuous, $SL(n)$ covariant valuation that is a translation homomorphism is $Zu = u + c$.
• Theorem 6.2: On positive log-concave functions, the only such transform is $Zf = cf$.

D. Generalizations

The paper provides generalized theorems (Theorems 5.2, 5.3, 5.7) that cover cases where the translation conjugation involves scaling factors ($\varepsilon$) or specific linear maps ($A(y) = \sigma y$), showing that the Legendre and Laplace transforms remain the fundamental building blocks even under these generalized conditions.

4. Significance

• Unification of Geometry and Analysis: The paper bridges the gap between classical valuation theory on convex bodies (geometric) and valuation theory on function spaces (analytic). It demonstrates that the structural rigidity of the Legendre and Laplace transforms is deeply rooted in their interaction with the group $SL(n)$ and translation operations.
• Relaxation of Assumptions: By proving these characterizations without assuming bijectivity or order-reversing isomorphism, the results are more robust and applicable to a wider class of operators.
• Novelty of Laplace Characterization: While the Legendre transform has been studied extensively in this context, the characterization of the Laplace transform as a valuation on log-concave functions is a significant new contribution. It highlights a fundamental difference between the behavior of convex functions and log-concave functions under valuation theory.
• Foundation for Future Research: The introduction of "dual valuations" provides a powerful tool for transferring classification results between different function spaces (e.g., from super-coercive to finite convex functions), opening avenues for further classification of other integral transforms.

In summary, Jin Li establishes that the Legendre and Laplace transforms are not just arbitrary integral operators but are uniquely determined by their fundamental algebraic (valuation), geometric ($SL(n)$), and translational properties within the space of convex and log-concave functions.