|
<!DOCTYPE html><html> |
|
<head> |
|
<title></title> |
|
<style type="text/css"> |
|
<!-- |
|
.xflip { |
|
-moz-transform: scaleX(-1); |
|
-webkit-transform: scaleX(-1); |
|
-o-transform: scaleX(-1); |
|
transform: scaleX(-1); |
|
filter: fliph; |
|
} |
|
.yflip { |
|
-moz-transform: scaleY(-1); |
|
-webkit-transform: scaleY(-1); |
|
-o-transform: scaleY(-1); |
|
transform: scaleY(-1); |
|
filter: flipv; |
|
} |
|
.xyflip { |
|
-moz-transform: scaleX(-1) scaleY(-1); |
|
-webkit-transform: scaleX(-1) scaleY(-1); |
|
-o-transform: scaleX(-1) scaleY(-1); |
|
transform: scaleX(-1) scaleY(-1); |
|
filter: fliph + flipv; |
|
} |
|
--> |
|
</style> |
|
</head> |
|
<body> |
|
<a name=1></a>HLF 2016, Sep. 22, 2016, Heidelberg.<br/> |
|
UniMath<br/> |
|
by Vladimir Voevodsky  <br/> |
|
from the Institute for Advanced Study in Princeton, NJ. <br/> |
|
<hr/> |
|
<a name=2></a>Part 1. Univalent foundations<br/> |
|
2<br/> |
|
<hr/> |
|
<a name=3></a><img src="Unimath-3_1.png"/><br/> |
|
<img src="Unimath-3_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
Today we face a problem that involves two difficult to satisfy conditions. <br/> |
|
On the one hand we have to find a way for computer assisted verification of <br/>mathematical proofs.<br/> |
|
This is necessary, first of all, because we have to stop the dissolution of the <br/>concept of proof in mathematics.<br/> |
|
On the other hand  we have to preserve the intimate connection between <br/>mathematics and the world of human intuition.<br/> |
|
This connection is what moves mathematics forward and what we often <br/>experience as the beauty of mathematics. <br/> |
|
3<br/> |
|
<hr/> |
|
<a name=4></a><img src="Unimath-4_1.png"/><br/> |
|
<img src="Unimath-4_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
The Univalent Foundations (UF) is, a yet imperfect, solution to this problem.<br/> |
|
In their original form,  the UF combined three components:<br/> |
|
• the view of mathematics as the study of structures on sets and their higher <br/> |
|
analogs, <br/> |
|
• the idea that the higher analogs of sets are reflected in the set-based <br/> |
|
mathematics as homotopy types, <br/> |
|
• the idea that one can formalize our intuition about structures on these higher <br/> |
|
analogs using the Martin-Lof  Type Theory (MLTT) extended with the Law of <br/>Excluded Middle for propositions (LEM) , the Axiom of Choice for sets (AC), <br/>the Univalence Axiom (UA) and the Resizing Rules (RR).<br/> |
|
4<br/> |
|
<hr/> |
|
<a name=5></a><img src="Unimath-5_1.png"/><br/> |
|
<img src="Unimath-5_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
The main new concepts that were since added to these are the following: <br/> |
|
• the understanding that a lot of mathematics can be formalized in the MLTT <br/> |
|
without the LEM and the AC and that excluding these two axioms one <br/>obtains foundations for a <i>new form of constructive mathematics</i>,<br/> |
|
• the understanding that classical mathematics appears as a subset of this new <br/> |
|
constructive mathematics,<br/> |
|
• the understanding that the MLTT extended with the UA is an imperfect <br/> |
|
formalization system for this constructive mathematics and that it should be <br/>possible  to integrate the UA into the MLTT obtaining a new type theory <br/>with better computational properties.<br/> |
|
5<br/> |
|
<hr/> |
|
<a name=6></a><img src="Unimath-6_1.png"/><br/> |
|
<img src="Unimath-6_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
What does it mean for a formalization system to be constructive?<br/> |
|
Some expressions in type theory are said to be in normal form. Any <br/>expression can be automatically and deterministically “normalized”, that is, an <br/>equivalent expression in normal form can be computed. <br/> |
|
In type theory there are type expressions and element expressions. If “T” is a <br/>type (expression) and “o” is an element (expression) one writes “o:T” if the <br/>type of “o” is “T”. <br/> |
|
6<br/> |
|
<hr/> |
|
<a name=7></a><img src="Unimath-7_1.png"/><br/> |
|
<img src="Unimath-7_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
In most type systems there is the type of natural numbers. In the UniMath it is <br/>written as “nat”.<br/> |
|
There is the zero element “O:nat” and the successor function “S” from “nat” to <br/>“nat” that intuitively corresponds to the function that takes “n” to “1+n”. <br/> |
|
A constructive system satisfies the <i>canonicity property</i> for “nat”, which asserts <br/>that the normal form of any expression “o:nat” has the form “S(S(….(SO)..))”.<br/> |
|
By counting how many “S” there is in the normal form one obtains an actual <br/>natural number from any element expression of type “nat”. <br/> |
|
7<br/> |
|
<hr/> |
|
<a name=8></a><img src="Unimath-8_1.png"/><br/> |
|
<img src="Unimath-8_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
This is a tremendously strong property. <br/> |
|
Consider the example: a set “X:hSet” is defined to be finite if there exists an <br/>isomorphism between it and the standard finite set “stn n”. Here “n” is an <br/>expression of type “nat”. It is well defined and one obtains a function “fincard”  <br/>from finite sets to “nat” called the cardinality - the number of elements of the <br/>set.<br/> |
|
Now suppose that I have proved, constructively, that “X” is finite. Then <br/> |
|
“(fincard X):nat” <br/> |
|
is defined. By normalizing “fincard X” I obtain an actual natural number.<br/> |
|
If I had a constructive proof of <i>Faltings’s Theorem, </i>stating that the number of <br/>rational points on a curve of genus >1 is finite, I could find the actual number <br/>of points on any curve of genus >1.  <br/> |
|
8<br/> |
|
<hr/> |
|
<a name=9></a><img src="Unimath-9_1.png"/><br/> |
|
<img src="Unimath-9_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
We don’t know whether such a proof exists. It is a very interesting and hard <br/>problem. <br/> |
|
The reason that the MLTT+UA is an imperfect system for constructive  <br/>formalization is that while MLTT itself has the canonicity property MLTT+UA <br/>does not.<br/> |
|
Therefore, formalizing the proof of Faltings’s Theorem in the UniMath, which is <br/>based on MLTT+UA, would not immediately give us an algorithm to compute <br/>the number of rational points on a curve of genus >1.<br/> |
|
This is where a new type theory that integrates the UA into the MLTT in such <br/>a way as to preserve the canonicity would help. <br/> |
|
9<br/> |
|
<hr/> |
|
<a name=10></a><img src="Unimath-10_1.png"/><br/> |
|
<img src="Unimath-10_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
The search for such a type theory became one of the main driving forces in <br/>the development of the UF.<br/> |
|
Today several groups are working on the construction and  implementation in <br/>a proof assistant of candidate type theories.  <br/> |
|
The <i>cubical type theory</i> and the prototype proof assistant <i>cubicaltt</i> created by <br/>the group of Thierry Coquand with the help of many researchers from <br/>different parts of the world is at the most advanced stage of development <br/>today. <br/> |
|
A proof in the UniMath easily translates into a proof in the cubilatt.<br/> |
|
10<br/> |
|
<hr/> |
|
<a name=11></a><img src="Unimath-11_1.png"/><br/> |
|
<img src="Unimath-11_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
The new form of the UF that emerges can be seen as combining the following <br/>components:<br/> |
|
• the view of mathematics as the study of structures on sets and their higher <br/> |
|
analogs, <br/> |
|
• the view of mathematics as constructive with the classical mathematics being <br/> |
|
a subset consisting of the results that require LEM and/or AC among their <br/>assumptions,<br/> |
|
• the idea that the higher analogs of sets are reflected in the set-based <br/> |
|
mathematics as constructive homotopy types - objects of the new <br/>constructive homotopy theory that can so far be formulated only in terms of <br/>cubical sets,<br/> |
|
• the idea that one can formalize our intuition about structures on these higher <br/> |
|
analogs using Cubical Type Theory (CTT).<br/> |
|
11<br/> |
|
<hr/> |
|
<a name=12></a><img src="Unimath-12_1.png"/><br/> |
|
<img src="Unimath-12_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
In addition to the understanding that to obtain a formal system for the new <br/>constructive mathematics the UA needs to be integrated into the MLTT <br/>constructively, several more things are felt as lacking in the MLTT+UA:<br/> |
|
• higher inductive types, <br/> |
|
• resizing rules,<br/> |
|
• a possible strict extensional equality combined with the “fibrancy discipline”,<br/> |
|
• as yet unknown mechanism to construct the types of structures that involve <br/> |
|
infinite hierarchies of coherence conditions.  <br/> |
|
Surprisingly,  it might be easier to add these features to the CTT than to the <br/>MLTT. The work in these directions is ongoing. <br/> |
|
12<br/> |
|
<hr/> |
|
<a name=13></a>Part 2. The UniMath library<br/> |
|
13<br/> |
|
<hr/> |
|
<a name=14></a><img src="Unimath-14_1.png"/><br/> |
|
<img src="Unimath-14_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
In the development of the UniMath library we attempt to do something that <br/>might be compared with the effort by the Bourbaki group to write a <br/>systematic exposition of mathematics based on the set theory and the view of <br/>mathematics as studying structures on sets.<br/> |
|
The effort by Bourbaki stalled at some point around the middle of the 20th <br/>century, in part, because it was very complicated to describe the emerging <br/>category-theoretic constructions in set-theoretic terms.<br/> |
|
14<br/> |
|
<hr/> |
|
<a name=15></a><img src="Unimath-15_1.png"/><br/> |
|
<img src="Unimath-15_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
One may however ask, is there any mathematical innovation in what we are <br/>doing? Is there a discovery of the unknown in the work on the UniMath?<br/> |
|
We have already seen how well-known problems in fields such as arithmetic <br/>algebraic geometry can be related to the search for a new foundation of <br/>constructive mathematics and for building proofs in the UniMath.<br/> |
|
Here is a different example.<br/> |
|
15<br/> |
|
<hr/> |
|
<a name=16></a><img src="Unimath-16_1.png"/><br/> |
|
<img src="Unimath-16_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
Some years ago, at the IAS, I had a conversation at lunch with Armand Borel. I <br/>mentioned how I like Bourbaki “Algebra” and how it helped me to become a <br/>mathematician. <br/> |
|
I then mentioned that some places there were really dense. For example, said I, <br/>the description of the tensor product was hard to follow. <br/> |
|
Of course, said Borel, <i>we have invented tensor product to get a systematic <br/>exposition of multi-linear maps</i>. <br/> |
|
It was new research, this is why it was not very smoothly written. <br/> |
|
16<br/> |
|
<hr/> |
|
<a name=17></a><img src="Unimath-17_1.png"/><br/> |
|
<img src="Unimath-17_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
I was amazed.<br/> |
|
It is hard to imagine today’s mathematics without the concept of the tensor <br/>product. It would never occurred to me that it was invented by Bourbaki with <br/>the only purpose to obtain a more systematic exposition of multi-linear maps <br/>of vector spaces!<br/> |
|
This example shows how a major innovation can emerge from the work on <br/>systematization of knowledge. <br/> |
|
17<br/> |
|
<hr/> |
|
<a name=18></a><img src="Unimath-18_1.png"/><br/> |
|
<img src="Unimath-18_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
Finally, a few words to those mathematicians who will decide to understand <br/>UniMath and maybe to contribute to it. <br/> |
|
The UniMath library is being created using the proof assistant Coq. It is freely <br/>available on GitHub.<br/> |
|
The language of Coq is a very substantial extension of the MLTT and UniMath <br/>uses a very small subset of the full Coq language that approximately <br/>corresponds to the original MLTT.<br/> |
|
18<br/> |
|
<hr/> |
|
<a name=19></a><img src="Unimath-19_1.png"/><br/> |
|
<img src="Unimath-19_2.png"/><br/> |
|
<img src="Unimath-19_3.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
The first file in the UniMath after the <i>Basics/preamble.v</i> is <i>Basics/PartA</i>.v.<br/> |
|
The first line in <i>Basics/PartA.v</i> after the preamble section is as follows:<br/> |
|
    <br/> |
|
It should be understood as a declaration of intent to define a constant called <br/><i>fromempty </i>whose type is described by the expression that is written to the <br/>right of the colon. <br/> |
|
Following this line there is a paragraph that starts with <i>Proof.</i> and ends with <br/><i>Defined. </i>where the constant is actually defined using the little sub-programs of <br/>Coq called tactics which help to build complex expressions of the underlying <br/>type theory language in simple steps.  <br/> |
|
19<br/> |
|
<hr/> |
|
<a name=20></a><img src="Unimath-20_1.png"/><br/> |
|
<img src="Unimath-20_2.png"/><br/> |
|
Univalent Foundations<br/> |
|
UniMath library<br/> |
|
A mathematician who wants to understand UniMath should expect a very <br/>non-linear learning curve:<br/> |
|
• In the lectures that I gave in Oxford and in the similar lectures in the Hebrew <br/> |
|
University it took me the whole first lecture to explain what that first line <br/>and the following it paragraph really mean.<br/> |
|
• In the next lecture I was able to explain the next few hundred lines of PartA.<br/> |
|
• By the fourth lecture in Oxford, the video of which can be found on my <br/> |
|
website, I was explaining the invariant formalization of fibration sequences.<br/> |
|
20<br/> |
|
<hr/> |
|
<a name=21></a>I hope that was able to show how important Univalent Foundations are and <br/> |
|
how important is the work on libraries such as UniMath.<br/> |
|
Thank you!<br/> |
|
21<br/> |
|
<hr/> |
|
</body> |
|
</html> |
|
|