🔢 mathematics

Grothendieck's Equality vs Voevodsky's Equality

This paper compares Homotopy Type Theory's treatment of equality with Grothendieck's approach by examining canonical constructions across algebraic structures and cohomology theories to improve the formalization of mathematics.

Original authors: Thomas Eckl

Published 2026-04-02

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Thomas Eckl

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

Imagine you are trying to build a massive, perfect library of all human knowledge. In the past, mathematicians built this library using a specific set of blueprints called Set Theory (like the ZFC axioms). It worked well for centuries, but recently, a new generation of builders has started using a different, more flexible set of blueprints called Type Theory, specifically a version called Homotopy Type Theory (HoTT).

This paper, written by Thomas Eckl, is like a guidebook for architects who are trying to translate old, classic mathematical designs into these new blueprints. It compares two different philosophies on how to handle "equality" (when two things are considered the same) and how to deal with "choices" (when there are multiple ways to build something).

Here is the breakdown of the paper's main ideas using everyday analogies:

1. The Problem: The "Universal Property" vs. The "Specific Blueprint"

In mathematics, many objects are defined by what they do rather than what they are. This is called a Universal Property.

The Analogy: Imagine you need a "Universal Key" that opens any door in a specific building. You don't care if the key is made of brass, steel, or plastic, or if it has a jagged edge or a smooth one. You only care that it opens the door.
The Old Way (Grothendieck): The great mathematician Alexander Grothendieck said, "If two keys open the door in the exact same way, they are the same key. Let's just treat them as identical." He called this "canonical."
The Problem: When you try to write a computer program (formalization) to prove things about these keys, the computer gets confused. If you tell the computer "Key A is the same as Key B," the computer might say, "Wait, I need to see the actual metal of the keys to prove they are the same!" The computer needs the specific blueprint (how the key was made), not just the promise that it works.

2. The Solution: Voevodsky's Equality (The "Univalence" Axiom)

Enter Vladimir Voevodsky and Homotopy Type Theory. They introduced a new rule called Univalence.

The Analogy: Univalence is like a magical translator. It says, "If two structures are equivalent (they work the same way), then they are equal."
The Catch: The paper argues that while this sounds like a perfect solution to Grothendieck's problem, it's actually more complicated. Univalence doesn't just say "they are the same"; it says "there is a path connecting them."
The Insight: The author shows that for many mathematical objects (like numbers or shapes), Univalence works great. But for complex algebraic structures, simply saying "they are equivalent" isn't always enough to make the computer's life easy. Sometimes, you still need to look at the specific construction (the blueprint) to get the job done efficiently.

3. The "Choice" Problem: Sign Flips and Boundary Maps

In advanced math (like Homological Algebra), you often have to make arbitrary choices.

The Analogy: Imagine you are building a bridge. You have to decide whether to paint the left support beam red and the right one blue, or vice versa. Mathematically, it doesn't matter which way you choose; the bridge will stand either way.
The Issue: In standard math, we often ignore these choices and say, "It's canonical." But in a computer proof, if you don't record which choice you made, the computer can't verify the bridge is safe.
The Paper's Advice: The author suggests a clever trick using Propositional Truncation.
- The Metaphor: Instead of forcing the computer to pick a specific color (Red/Blue), you tell it: "There exists a way to paint the beams such that the bridge stands."
- Why it works: If you are only trying to prove a statement (a "proposition") like "The bridge is safe," you don't need to know the exact color. You just need to know a valid coloring exists. This allows the computer to skip the tedious work of checking every possible choice, making the proof much faster and more efficient.

4. The "Working Mathematician" Perspective

The author writes this not as a pure logician, but as a "working mathematician."

The Goal: The paper isn't trying to reinvent the wheel of logic. It's asking: "How do we translate the way real mathematicians think into code that computers can understand?"
The Finding: The paper finds that while HoTT is powerful, we can't just blindly apply it. We have to be smart about how we define things. Sometimes, defining an object by its "Universal Property" (what it does) is better for theory, but defining it by its "Construction" (how it's built) is better for computer proofs.

5. The Big Picture: AI and the Future

The paper ends with a thought about Artificial Intelligence.

The Metaphor: Imagine a super-smart AI trying to learn math. If the AI only sees the "final answer" (the theorem), it might miss the "compression" of human thought—the way humans use shortcuts, analogies, and "canonical" identifications to skip steps.
The Warning: If we want AI to do real research, we need to understand how humans compress complex ideas into simple, efficient patterns. If we just feed the AI raw, unstructured data, it might get stuck in a "complexity barrier" (like an NP-hard problem) and never find the elegant solution a human would spot.

Summary

This paper is a bridge between old-school mathematical intuition (where we treat equivalent things as identical) and new-school computer formalization (where computers need strict, step-by-step instructions).

It teaches us that:

Equality is tricky: Just because two things work the same way doesn't always mean a computer sees them as equal without help.
Choices are okay: We don't always need to force a computer to make a specific choice; sometimes, just knowing a choice exists is enough to prove a theorem.
Efficiency matters: To make math work on computers (and eventually with AI), we need to find the most efficient way to describe mathematical objects, sometimes mixing the "what it does" with the "how it's built."

In short, it's a manual on how to teach a computer to think like a human mathematician without getting bogged down in the details of every single choice we make along the way.

Technical Summary: Grothendieck's Equality vs. Voevodsky's Equality

1. Problem Statement
The paper addresses a fundamental tension in the formalization of mathematics within the framework of Homotopy Type Theory (HoTT) and proof assistants like Lean. Specifically, it investigates how to handle universal properties and canonical isomorphisms efficiently.

The "Grothendieck" Approach: In classical mathematics (e.g., EGA), mathematicians often treat objects satisfying the same universal property as "identical" via a "canonical" isomorphism, effectively ignoring the specific construction details to streamline proofs.
The "Voevodsky" (HoTT) Approach: HoTT introduces the Univalence Axiom, which equates isomorphism with equality ( $A \simeq B \iff A = B$ ). However, the author argues that a naive application of univalence does not automatically solve the efficiency problems of formalization.
The Core Conflict: Formal proofs relying solely on universal properties are often inefficient or impossible without specific constructions. Conversely, transferring properties between different constructions of the same universal object requires cumbersome technical lemmas. Furthermore, "canonical" choices (like sign conventions in boundary maps) are often ambiguous in classical math, leading to potential errors if not tracked. The paper asks: How can HoTT reconcile the need for specific constructions with the desire for structural equality, and how does it handle non-canonical choices?

2. Methodology
The author, Thomas Eckl, adopts the perspective of a "working mathematician" rather than a foundational theorist. The methodology involves:

Comparative Analysis: Contrasting standard set-theoretic formalization practices (as seen in Lean's Mathlib) with the capabilities and constraints of HoTT.
Case Studies: The paper systematically analyzes a hierarchy of algebraic structures, moving from simple inductive types to complex cohomological theories:
- Inductive types (Natural numbers, Cartesian products).
- Higher Inductive Types (HITs) for Set Quotients, Monoids, Groups, and Rings.
- Universal constructions: Localization of rings, Tensor products, and Direct sums.
- Homological Algebra: $\delta$ -functors, derived functors, and cohomology.
Strickland's Characterization: The paper heavily utilizes a specific strategy proposed by Neil Strickland (and popularized by Kevin Buzzard): instead of defining objects by their universal property alone, one defines them by a characterization that implies the universal property but is computationally more tractable (e.g., defining localization via a kernel condition rather than just the universal mapping property).
Propositional Truncation: The author employs propositional truncation ( $\|A\|$ ) to handle "mere existence." This allows the mathematician to assert the existence of a choice (like a projective resolution or a boundary map) without committing to a specific construction, provided the final result is a proposition (a theorem) rather than a specific data object.

3. Key Contributions

Refining the Role of Univalence: The paper clarifies that Univalence equates types that are equivalent, but it does not automatically make different constructions "judgmentally equal" (computationally reducible to the same term). It argues that Univalence is more about attaching equivalences to equalities than about identifying types in a way that simplifies calculation without further work.
Characterization Strategies: The author proposes two main strategies for characterizing universal objects in HoTT:
1. Constructor-based: Defining objects by their constructors and free generation (e.g., lists for free monoids).
2. Property-based (Strickland-style): Defining objects by a set of conditions that imply the universal property (e.g., defining a localization $R[S^{-1}]$ by the property that elements are fractions and the kernel is the annihilator ideal). This approach is shown to be more efficient for formalization than relying purely on the universal mapping property.
Handling Non-Canonical Choices: The paper provides a rigorous framework for dealing with choices that are not unique (e.g., sign choices in boundary maps for $\delta$ $δ$ -functors).
- It demonstrates that while specific choices (like signs) affect the construction of boundary maps, the resulting universal $\delta$ -functor is unique up to canonical isomorphism (equality via univalence).
- It advocates for recording these choices as data when necessary for calculation but using propositional truncation to prove theorems where the specific choice is irrelevant.
Formalization of Homological Algebra: The paper successfully formalizes the construction of derived functors (like Tor and Ext) and the long exact sequence of cohomology within HoTT. It shows that the "universal" nature of these functors allows one to bypass the non-uniqueness of projective resolutions by proving that any two resolutions yield canonically equal results.

4. Key Results

Efficiency of Characterizations: The paper proves that characterizations like Strickland's for ring localizations (Prop 3.61) and the equational criterion for tensor products (Prop 3.67) are superior for formalization. They condense the description of equalities and kernels, avoiding the need for complex "yoga" of universal properties.
Uniqueness of Universal Objects: It is proven that in HoTT, any two objects satisfying the same universal property (e.g., two different constructions of a set quotient or a tensor product) are equal (not just isomorphic) via the Structure Identity Principle and Univalence, provided the objects are sets (0-types).
Cohomological Flatness Criterion: The paper culminates in the formalization of a flatness criterion for affine hypersurfaces (Prop 5.1). It demonstrates that one can prove deep commutative algebra results (e.g., $S/(f)$ is flat iff coefficients generate the unit ideal) using cohomological tools (Tor groups) without needing to explicitly construct specific projective resolutions for every step, relying instead on the mere existence of such resolutions.
Canonical vs. Non-Canonical: The results confirm that while the construction of boundary maps involves choices (signs, lifts), the functorial properties and the resulting cohomology theories are canonical. The "abuse of notation" common in classical math (treating different choices as the same) is justified in HoTT because the different choices lead to equal objects.

5. Significance

Bridging Practice and Theory: The paper serves as a "blueprint" for how working mathematicians can formalize advanced mathematics in HoTT. It validates that standard mathematical practices (like identifying canonical isomorphisms) can be faithfully translated into HoTT, provided one understands the distinction between construction and proposition.
Guidance for Formalization: It offers concrete guidelines for Lean/HoTT developers:
- Avoid relying solely on universal properties for computation; use explicit characterizations.
- Use propositional truncation to manage choices in existence proofs.
- Recognize that "canonical" in math often means "unique up to isomorphism," which HoTT treats as "unique up to equality," but this requires careful handling of the underlying structure.
AI and Formalization: The author argues that understanding the "human" compression of mathematical knowledge (how mathematicians ignore irrelevant construction details) is crucial for AI agents to generate research-level mathematics. If AI cannot navigate the complexity of NP-hard construction choices, it must rely on the "human" strategy of focusing on universal properties and canonical equivalences, which HoTT supports.
Foundational Insight: It clarifies the limitations of Univalence. While Univalence is powerful, it does not automatically solve the "efficiency" problem of formalization; it requires a shift in how objects are characterized (from pure universality to specific, tractable properties) to be truly effective.

In summary, the paper argues that HoTT does not replace the need for specific constructions but reframes how we interact with them. By distinguishing between the data of a construction (which may involve arbitrary choices) and the proposition of a theorem (which is independent of those choices), HoTT provides a rigorous foundation for the "canonical" reasoning used by mathematicians like Grothendieck, while offering tools (Univalence, HITs, Truncation) to manage the underlying complexity.