Tannakian Reconstruction

The Analogy of the Distant Lights

Imagine two friends, Alice and Bob, separated by a wide river. Bob wants to identify which light on the far shore is Alice's. They agree on a protocol: every night at 10 p.m., Alice will illuminate her lights for exactly ten minutes.

Bob takes a long-exposure photograph each night. By the end of the year, he superimposes every single photo. While a single image might be ambiguous, the accumulation of images ensures that only Alice's window remains a bright, consistent spot.

This process is a form of encoding. While it might be ~~perfectly lossless~~ potentially lossy, it fundamentally preserves the structure of the source. If a relationship (a morphism) exists between two entities in the source, that relationship is mirrored in the target.

Generally, a single "snapshot" cannot reveal the full structure of the source category. However, if the target category possesses sufficient resolution and we aggregate all possible snapshots, the original morphisms can be recovered.

Understanding Fiber Functors

To achieve the "right resolution," we utilize the category of sets, denoted as Set. We focus on functors from a category $\mathcal{C}$ to Set (historically referred to as co-presheaves).

The mechanics are simple:

Objects in $\mathcal{C}$ $\rightarrow$ Sets in Set
Morphisms in $\mathcal{C}$ $\rightarrow$ Functions in Set

Usually, we keep the functor fixed and vary the objects. In Tannakian reconstruction, we do the opposite: we keep the objects constant and consider the totality of all functors.

The Probing Mapping

For any object $c \in \mathcal{C}$ , we define a mapping $\Phi_c$ that takes a functor and returns a set:

$\Phi_c : [\mathcal{C}, \text{Set}] \to \text{Set}$ $\Phi_c(\hat{F}) = \hat{F}(c)$

This mapping is itself functorial. If we have a natural transformation $\alpha$ between two functors $\hat{F}$ and $\hat{G}$ , it consists of a family of functions $\alpha_a : \hat{F}(a) \to \hat{G}(a)$ . The action of $\Phi_c$ on $\alpha$ is simply the component $\alpha_c$ .

Essentially, $\Phi_c$ "probes" an object and its immediate neighborhood via morphisms.

Analogy Component	Category Theory Equivalent
Long-exposure photo	A single Fiber Functor $\hat{F}$
Superimposing photos	The "End" ( $\int$ ) / Product of all functors
Alice's Window	The Object $c$
The Light Pattern	The Morphisms in $\mathcal{C}$

The Process of Tannakian Reconstruction

To reconstruct a hom-set, we examine the set of functions across all possible functors. This is equivalent to the set of natural transformations between two fiber functors $\hat{F}$ and $\hat{G}$ .

The "End" as a Superposition

A set of natural transformations can be expressed as an end, which functions like a massive product:

In our analogy, the end is the act of superimposing all photos. However, a product is fragile: if any single component is empty, the entire result is empty.

Wait, what stops us from picking a "bad" functor that is non-empty at $a$ but empty at $b$ ?

If such a "bad apple" existed, it would destroy the reconstruction because there are no functions from a non-empty set to an empty set. But because the Yoneda embedding is full and faithful, we are guaranteed a sufficient supply of functions to match the morphisms in $\mathcal{C}$ .

The Mathematical Proof

We can recover the original hom-set via this isomorphism:

$\text{Hom}(a, b) \cong \int_{\hat{F} \in [\mathcal{C}, \text{Set}]} \text{Hom}(\hat{F}(a), \hat{F}(b))$

The derivation follows these steps:

Expand: Use the Yoneda lemma to "explode" the functors inside the end.
Integrate: Apply the Yoneda reduction to "integrate" over $\hat{F}$ .
Simplify: The result simplifies back to $\text{Hom}(a, b)$ .

% Logic flow of the reconstruction
\text{End} \implies \int \text{Hom}(\hat{F}a, \hat{F}b) \implies \text{Hom}(a, b)

Example: The One-Object Category

Consider a category $\mathcal{C}$ with only one object. In this case, the hom-set is simply a monoid under composition.

Functors: A set-valued functor here is just a representation of that monoid.
Natural Transformations: These are known as equivariant maps.

Through Tannakian reconstruction, we can recover the entire monoid just by knowing the totality of its representations.

Summary Checklist

Define fiber functors $\mathcal{C} \to \text{Set}$ .
Establish the probing mapping $\Phi_c$ .
Aggregate all functors using an end.
Apply Yoneda Lemma/Reduction.
Recover the original morphisms.

Final Note: The requirement for naturality (or equivariance) is not an extra step—it is already baked into the very definition of the end.

$C, Set$ $Phi_c$