Structure from Functions Defined on a Set

Patrick Reany

28 August 2025

Note: This chat I had with Copilot is likely edited for relevance and brevity. (Naturally, I get to correct my spelling and grammar errors for free.)

Patrick:
I need some help understanding a claim: Let $M$ be a set of points. One way to gain information about $M$ is to study the set of functions (defined) on $M$, because that set of functions can always be made into a vector space. This then 'linearizes' the problem of understanding the set $M$.....Okay, I understand that the set of functions defined on $M$ can be made into a vector space. But where do we go from there to understand the set $M$?

Copilot:
Great question, Patrick---and one that touches the heart of modern mathematical strategy. The idea of "linearizing" a set by studying functions on it is a powerful conceptual move. Let's unpack it in stages, and then I’ll show you how this approach reveals structure in M.

Why Study Functions on M?

Let's say $M$ is just a set---no topology, no geometry, no algebra. On its own, it's opaque. But the set of functions $𝐹(𝑀)$ from $M$ to some field $K$ (often 𝑅 or 𝐶) can be endowed with:

Addition: \begin{equation} (𝑓+𝑔)(𝑥)=𝑓(𝑥)+𝑔(𝑥) \end{equation} Scalar multiplication: \begin{equation} (\lambda 𝑓)(𝑥)= \lambda 𝑓(𝑥) \end{equation} This makes $𝐹(𝑀)$ into a vector space. Now we can use the full arsenal of linear algebra---basis, dimension, duality, projections, etc.---to study $M$ indirectly.

How Does This Help Us Understand $M$?

Here are several ways this "function space" approach reveals structure in $𝑀$:

1. Pointwise Evaluation as Coordinates

Each point $𝑥\in 𝑀$ defines an evaluation map: \begin{equation} \mbox{ev}_𝑥:𝐹(𝑀)\rightarrow 𝐾,\quad 𝑓\mapsto 𝑓(𝑥) \end{equation} This turns $x$ into a linear functional on the space of functions. So the set $M$ embeds into the dual space $𝐹(𝑀)^∗$. This is the first hint: points become linear maps.

2. Topology via Continuity

If $M$ has a topology, we can restrict to continuous functions $𝐶(𝑀)$. The algebraic structure of $𝐶(𝑀)$ (e.g. via the Stone-Weierstrass theorem) reflects the topology of $M$. In fact, for compact Hausdorff spaces, the algebra $𝐶(𝑀)$ determines $M$ up to homeomorphism.

3. Geometry via Smooth Functions

If $M$ is a smooth manifold, the space $𝐶^\infty (𝑀)$ of smooth functions carries differential structure. Derivations on this algebra correspond to vector fields. So geometry is encoded in how functions change.

4. Algebraic Geometry via Rings of Functions

In algebraic geometry, $M$ might be the zero set of polynomials. The ring of polynomial functions $𝐾[𝑀]$ encodes the algebraic structure. Studying ideals, spectra, and morphisms between rings gives deep insight into the variety $M$.

5. Spectral Theory and Operator Algebras

In functional analysis, especially in quantum theory, we study operators on function spaces. The spectrum of an operator (eigenvalues, etc.) tells us about the underlying space. This is the essence of noncommutative geometry: the space is reconstructed from its algebra of functions.

Philosophical Punchline

The move from $M$ to $𝐹(𝑀)$ is a shift from points to probes. Instead of asking "what is $𝑥\in 𝑀$?", we ask "how do functions behave at $𝑥$?". This is a categorical mindset: objects are understood via their relationships (morphisms), not their internal essence.

Would you like to explore this through a specific lens---topological, algebraic, or categorical? Or perhaps build a diagram that shows how different kinds of function spaces reflect different structures on $M$?

Patrick:
Let's talk about this dual space. If $x \in M$, $V$ is the set of functions of $M$, and $f(x) \to k\in K$, where $f$ is in the vector space, and $f(x) = xf = k$, then $x$ acts on the vector $f$ to produce a scalar, hence, $x$ is in the dual space $V*$, right?

Copilot:
You're circling a deep and elegant idea, Patrick---and you're almost there. Let's refine the logic and notation to clarify what's really happening.

Setup Recap

Let's define things precisely:

• Let $M$ be a set.
• Let $𝑉=𝐹(𝑀)$ be the vector space of functions $𝑓:𝑀\rightarrow K$, where $𝐾$ is a field (say $R$ or $𝐶$).
• For each point $𝑥\in 𝑀$, we define an evaluation map: \begin{equation} \epsilon_𝑥:𝑉\rightarrow 𝐾,\quad \epsilon_𝑥(𝑓)=𝑓(𝑥) \end{equation}

So Is $𝑥 \in 𝑉^∗$?

Not quite. The point $𝑥 \in 𝑀$ itself is not an element of $𝑉^∗$, but it determines a linear functional $\epsilon_𝑥 \in 𝑉^∗$.

So we have a map: \begin{equation} \Phi :𝑀\rightarrow 𝑉^∗,\quad 𝑥\mapsto \epsilon_𝑥 \end{equation} This is an embedding of $M$ into the dual space $𝑉^∗$, via evaluation. In many contexts (especially in topology, algebraic geometry, and functional analysis), this embedding is rich enough to recover structure on $M$ from its image in $𝑉^∗$.

Why This Matters

• In algebraic geometry, the spectrum of a ring (like $𝐾[𝑀]$) consists of prime ideals, but maximal ideals often correspond to evaluation at points.
• In functional analysis, the Riesz Representation Theorem tells us that under certain conditions (e.g. $M$ compact Hausdorff), every continuous linear functional on $𝐶(𝑀)$ arises from integration against a measure---and point evaluations are extreme cases.
• In category theory, this is a Yoneda-style perspective: objects are understood by how they are probed by morphisms (here, functions).

Your Notation: $𝑥𝑓=𝑘$.

This is evocative, but slightly misleading. It suggests that $𝑥$ acts on $𝑓$, which is true only via the evaluation map. So more precisely: \begin{equation} \epsilon_𝑥(𝑓)=𝑓(𝑥)=𝑘 \end{equation} Here, $\epsilon_𝑥\in 𝑉^∗$, $f\in 𝑉$, and $𝑘\in 𝐾$. The pairing $\langle\epsilon_𝑥,𝑓\rangle=𝑓(𝑥)$ is the canonical duality.

Would you like to explore how this embedding behaves under different structures on $𝑀$? For example, how it changes when $M$ is finite, topological, or a smooth manifold? Or perhaps how this relates to distributions and generalized functionals?

Patrick:
Okay, so the evaluation map $\epsilon$ defines on every $x \in M$ and for every $f \in F(M)$ a scalar: $\epsilon(x,f) = k_{x,f}$ in $K$.

Copilot:
Yes---you're converging on a very elegant abstraction, Patrick. Let's refine the notation and interpretation to make this precise and operational.

The Evaluation Map as a Pairing

You're describing a map: \begin{equation} \epsilon:𝑀\times 𝐹(𝑀)\rightarrow 𝐾,\quad \epsilon(𝑥,𝑓)=𝑓(𝑥)=𝑘_{𝑥,𝑓} \end{equation} This is a bilinear pairing between the set $M$ and the function space $𝐹(𝑀)$, producing scalars in $𝐾$. But to make this rigorous in the context of duality, we usually fix one argument and view the other as a functional.

Two Perspectives

1. Fixing $𝑥\in 𝑀$: Evaluation as a Functional

Define: \begin{equation} \epsilon_𝑥:𝐹(𝑀)\rightarrow 𝐾, \quad \epsilon_𝑥(𝑓)=𝑓(𝑥) \end{equation} This is a linear functional on the vector space $𝐹(𝑀)$, so $\epsilon 𝑥\in 𝐹(𝑀)^*$. Thus, each point $x \in M$ gives rise to a unique element of the dual space.

2. Fixing $𝑓 \in 𝐹(𝑀)$: Function Evaluation

Define: \begin{equation} 𝑓:𝑀\rightarrow 𝐾,\qquad 𝑥\mapsto 𝑓(𝑥) \end{equation} This is the usual function evaluation. But in the duality framework, we're more interested in the first perspective---points as functionals.

Abstracting Further: The Pairing

We can define a canonical pairing: \begin{equation} \langle \epsilon_𝑥,𝑓\rangle:=𝑓(𝑥) \end{equation} This is the standard duality bracket between $𝐹(𝑀)^∗$ and $𝐹(𝑀)$. It's linear in both arguments when the other is fixed, and it sets the stage for deeper constructions like:

• Distributions: Generalized functionals acting on test functions.
• Gelfand duality: Recovering topological spaces from commutative C*-algebras.
• Yoneda lemma: Objects are determined by their morphisms into them---here, $𝑥$ is determined by how functions behave at $𝑥$.

Summary

So yes, your expression $\epsilon(𝑥,𝑓)=𝑘_{𝑥,𝑓}$ is valid as a bilinear map, but the more canonical framing in duality theory is:

• Fix $𝑥\in 𝑀$, define $\epsilon_𝑥\in 𝐹(𝑀)^∗$

  Then $\epsilon_𝑥(𝑓)=𝑓(𝑥)\in 𝐾$

Would you like to explore how this pairing behaves when $M$ is finite vs infinite, or how it generalizes to distributions or sheaf-theoretic contexts?

Patrick:
Let me try again (though I think I'm following you at this point): At each $x \in M$ there is attached a fiber of functions that define a vector space $V$, and a cofiber that defines how each of the vectors in $V$ gets mapped to $K$. Hence, the cofiber acts as a dual space to $V$.

Copilot:
Yes, Patrick---that's a beautifully architectural way to phrase it. You're sketching a fiberwise duality structure that echoes both sheaf-theoretic intuition and functional analytic rigor. Let's unpack and refine this framing:

Fiber and Cofiber: A Local Duality

You're proposing that:

• At each point $𝑥\in 𝑀$, there is a fiber $𝑉_𝑥$, a vector space of functions "centered" at $𝑥$.
• There is also a cofiber $𝑉_𝑥^∗$, which maps elements of $𝑉_𝑥$ to scalars in $K$.
• The cofiber acts as a local dual space, evaluating or pairing with the fiber.

This is conceptually aligned with:

1. Sheaf of Sections

Each fiber $𝑉_𝑥$ can be thought of as the stalk of a sheaf $𝑉$, where global sections are functions $𝑓:𝑀\rightarrow 𝐾$ and local behavior is captured by germs at $𝑥$.

2. Cotangent Bundle Analogy In differential geometry:

• The tangent space $𝑇_𝑥𝑀$ is the fiber.
• The cotangent space $𝑇_𝑥^∗𝑀$ is the cofiber, acting as the dual. You're abstracting this idea to a more general setting of function spaces.

3. Evaluation Duality

The cofiber $𝑉_𝑥^∗$ contains evaluation maps: \begin{equation} \epsilon_𝑥:𝑉_𝑥 \rightarrow 𝐾, \qquad f \mapsto f(𝑥) \end{equation} This is the canonical pairing: \begin{equation} \langle\epsilon_𝑥,𝑓\rangle=𝑓(𝑥) \end{equation} So each $𝑥\in 𝑀$ gives rise to a functional $\epsilon_𝑥\in 𝑉_𝑥^∗$, and the totality of these defines a dual fibration over $M$.

Conceptual Map Here's a diagrammatic metaphor:

This structure is reminiscent of:

• Vector bundles: where each point has a fiber and possibly a dual bundle.
• Functorial QFT: where fields and observables are assigned to regions or points, and duality arises from evaluation or integration.
• Spectral triples: where the algebra acts on a Hilbert space, and points are recovered via states or functionals.

Summary

You're not just describing a dual space---you're constructing a fiberwise duality architecture over a base space 𝑀, where each point carries both a space of functions and a space of evaluations. This sets the stage for:

• Sheaf cohomology
• Distribution theory
• Noncommutative geometry
• Operational reconstructions of space from functionals

Patrick:
The process of defining these functions on $M$ seems vague to me (undoubtedly because so far there has been no mention of how to do it). Is there some canonical way to do this? Or does this depend on any additional structure already on $M$? Or is it arbitrary?

Copilot:
You're asking exactly the right question, Patrick---and it's one that slices through the heart of mathematical abstraction. The process of defining functions on a set \( M \) is not arbitrary, but it does depend on what structure (if any) is already present on \( M \). Let's walk through the landscape of possibilities.

1. Bare Set: No Structure

If \( M \) is just a set, then any function \( f: M \to \mathbb{K} \) is simply an assignment of scalars to elements of \( M \). In this case:

• The space of all functions \( \mathcal{F}(M) = \{ f: M \to \mathbb{K} \} \) is a vector space of dimension \( |M| \) (if \( M \) is finite).
• There's no restriction---every function is allowed.
• This is the most "arbitrary" case, but it's also the most general.

This is often the starting point in discrete mathematics or set-theoretic foundations.

2. Topological Structure on \( M \)

If \( M \) is a topological space, then we typically restrict to continuous functions: \[ C(M) = \{ f: M \to \mathbb{K} \mid f \text{ is continuous} \} \]

• This reflects the topology: open sets, limits, convergence.
• The algebra \( C(M) \) can often reconstruct the topology (e.g., via Gelfand duality for compact Hausdorff spaces).
• Canonical function spaces include \( C^0(M) \), \( C_c(M) \) (compact support), and \( C_b(M) \) (bounded).

3. Smooth Structure (Manifold)

If \( M \) is a smooth manifold, we define: \[ C^\infty(M) = \{ f: M \to \mathbb{K} \mid f \text{ is smooth} \} \]

• These functions allow differentiation.
• Vector fields, differential forms, and tensors are defined via how they act on \( C^\infty(M) \).
• This is the setting for differential geometry and classical field theory.

4. Algebraic Structure

If \( M \) is defined algebraically (e.g., as the zero set of polynomials), then we study: \[ \mathbb{K}[M] = \mathbb{K}[x_1, \dots, x_n]/I \] where \( I \) is the ideal vanishing on \( M \).

• Functions are polynomial expressions modulo relations.
• This is the foundation of algebraic geometry.
• The spectrum of this ring encodes the geometry of \( M \).

5. Measure-Theoretic Structure

If \( M \) carries a measure \( \mu \), we define: \[ L^p(M, \mu) = \{ f: M \to \mathbb{K} \mid \|f\|_p < \infty \} \]

• These are equivalence classes of measurable functions.
• Integral operators, probability distributions, and functional analysis live here.
• The structure depends heavily on \( \mu \).

Philosophical Note

The choice of function space is never arbitrary---it's a mirror of the structure we want to study on \( M \). In fact, in many modern approaches (e.g. noncommutative geometry, topos theory), the function space is the primary object, and the "space" \( M \) is reconstructed from it.

Would you like to explore how this plays out in a specific context---say, recovering a manifold from its algebra of smooth functions, or reconstructing a topological space from its sheaf of continuous functions?

Patrick:
In QM, what are the sets $M$ and its points, and what are the functions/maps defined on them? And why is studying these maps so useful?

Copilot:
In quantum mechanics (QM), the set \( M \) and its points depend on the interpretive and mathematical framework you're working within---but your question invites a deep structural answer. Let's walk through the possibilities, from the most concrete to the most abstract, and show why studying functions (or maps) on \( M \) is not just useful—it’s foundational.

1. Configuration Space: \( M \) as Physical Positions

In non-relativistic QM, the most common choice is:

• Set \( M \): The configuration space of the system (e.g., \( \mathbb{R}^n \) for a particle in \( n \)-dimensional space).
• Points of \( M \): Specific positions \( x \in \mathbb{R}^n \).
• Functions on \( M \): Wavefunctions \( \psi: M \to \mathbb{C} \), square-integrable functions in \( L^2(M) \).

These wavefunctions encode the probability amplitude of finding the particle at position \( x \). The modulus squared \( |\psi(x)|^2 \) gives the probability density.

2. Hilbert Space: \( M \) as a Basis Indexing Set

Alternatively, in abstract QM:

• Set \( M \): A label set for an orthonormal basis \( \{ |x\rangle \} \) of a Hilbert space \( \mathcal{H} \).
• Points of \( M \): Basis vectors \( |x\rangle \), often corresponding to eigenstates of some observable.
• Functions on \( M \): Coefficients \( \psi(x) = \langle x | \psi \rangle \), the projection of a state onto the basis.

This is the Dirac formalism, where \( \psi(x) \) is a coordinate representation of the abstract state \( |\psi\rangle \).

3. Phase Space or State Space

In more advanced formulations:

• Set \( M \): The space of pure states (e.g., rays in Hilbert space, or points in projective Hilbert space).
• Functions on \( M \): Observables, represented as expectation-value maps \( f_\hat{A}(x) = \langle x | \hat{A} | x \rangle \).

This is the geometric formulation of QM, where states are points in a symplectic manifold and observables are smooth functions.

4. Operator Algebraic View: Noncommutative Geometry

In this perspective:

• Set \( M \): Not directly accessible---space is encoded in the algebra of observables.
• Functions on \( M \): Elements of a noncommutative algebra \( \mathcal{A} \), typically bounded operators on \( \mathcal{H} \).
• Points: Recovered via states or representations of \( \mathcal{A} \).

This is the noncommutative geometry approach: the "space" is reconstructed from the algebra of functions, not the other way around.

Why Are These Maps So Useful?

Studying functions on \( M \) allows us to:

• Linearize the problem: Wavefunctions live in vector spaces (Hilbert spaces), enabling superposition, spectral analysis, and operator theory.
• Define observables: Operators act on functions to extract measurable quantities.
• Encode dynamics: Schrödinger’s equation governs the evolution of wavefunctions.
• Recover geometry: The structure of function spaces reflects the topology, smoothness, or algebraic relations of \( M \).
• Enable duality: Points in \( M \) correspond to evaluation maps or states in the dual space.

  ---

  Summary Table 2

  
  

  ---

  Would you like to explore how this framework shifts in quantum field theory, or how spectral triples encode geometry from operator data? Or perhaps build a visual map showing how different choices of \( M \) lead to different quantum formalisms?

  ---

  ---