Math Is Not Infrastructure

TL;DR

1. Most branches of math are just sets with different kinds of structure added on top. Once you see this, the whole landscape becomes readable.
2. The deepest results tend to come from connecting two areas that look like they have nothing to do with each other.

---

For the past five years, the math I use is Linear Algebra, Probability, Calculus, and Optimization — all in service of AI research. Math was infrastructure to me. I reached for it when I needed it, the same way I reach for a terminal.

That changed after a dinner with some old friends from high school. A few of them had gone into pure math research. At some point I asked, mostly out of politeness: "So what do you actually study?" They said something about Topology — structures, continuity, generalising what "space" means.

My brain looked for an application. Didn't find one. I moved on.

But I kept thinking about it. A few weeks later I started trying to map the territory — not to work through proofs, just to understand the shape of what these people were doing. What are the big ideas? Where are the open problems?

What I found was that pure mathematics has moved much further ahead of applied math than I had any sense of. Some of the gaps are measured in decades.

When the Framework Breaks

Mathematics tends to advance through crises — moments when the existing system can no longer account for something that clearly exists. Each one forced an expansion of what "number" or "space" even means.

• Negative numbers and $\mathbb{Z}$: If you have 3 apples and owe someone 5, you need something new.
• Rational numbers $\mathbb{Q}$: Division doesn't always resolve cleanly. $\mathbb{Q} = \left\{ \frac{p}{q} \mid p, q \in \mathbb{Z},\ q \neq 0 \right\}$ handles that.
• Irrational numbers and $\mathbb{R}$: The diagonal of a unit square is $\sqrt{2}$, and $\sqrt{2} \notin \mathbb{Q}$ — you can't write it as any fraction. This wasn't a minor inconvenience for the Pythagoreans. It broke their worldview. Hippasus, who proved it, was reportedly drowned at sea.
• Complex numbers $\mathbb{C}$: $x^2 + 1 = 0$ has no real solution. Rather than declare the equation meaningless, mathematicians defined $i^2 = -1$. The thing called "imaginary" became the foundation of quantum mechanics and signal processing.

Each time a wall appeared, getting past it created new mathematical structure. And every era has had walls.

We still have some. The Navier–Stokes equations describe viscous fluid flow:

$$\rho \left( \frac{\partial \mathbf{u}}{\partial t} + (\mathbf{u} \cdot \nabla)\mathbf{u} \right) = -\nabla p + \mu \nabla^2 \mathbf{u} + \mathbf{f}$$

Used constantly in aerodynamics and weather modelling. Whether smooth solutions always exist in three dimensions is still unproven — and carries a $1{,}000{,}000$ USD Millennium Prize.

Then there's the Collatz Conjecture. Take any positive integer $n$. If even, halve it. If odd, multiply by 3 and add 1. Does the sequence always reach 1?

$$f(n) = \begin{cases} n/2 & \text{if } n \equiv 0 \pmod{2} \\ 3n+1 & \text{if } n \equiv 1 \pmod{2} \end{cases}$$

In 2019, Terence Tao published a paper showing that almost all positive integers eventually reach a value close to 1 in a probabilistic sense. A serious result, and still not a complete proof. Tao has said something about problems like this that I keep coming back to: it's not that no one has climbed the cliff — it's that no one has invented the ladder yet.

The Foundational Crisis

Before getting into how the field is organised, I want to cover the moment it nearly fell apart.

Russell's Paradox (1901):

$$R = \{ x \mid x \notin x \} \implies R \in R \iff R \notin R$$

The set of all sets that don't contain themselves — does it contain itself? It sounds like wordplay. It isn't. Russell wasn't attacking Cantor's set theory — he was dismantling Gottlob Frege's Grundgesetze, an attempt to derive all of mathematics from pure logic. Frege received the letter while his second volume was at the printer. He wrote in the postscript: "A scientist can hardly meet with anything more undesirable than to have the foundation give way just as the work is finished."

The fallout raised a harder question: is mathematics itself consistent and complete? Kurt Gödel answered in 1931:

Incompleteness Theorems: In any sufficiently powerful, consistent axiomatic system, there exist true statements that cannot be proved within that system.

Not "we haven't found the proof yet." Provably unprovable.

After this, the field was rebuilt on ZFC — Zermelo–Fraenkel set theory with the Axiom of Choice, roughly nine axioms sufficient to construct essentially all of modern mathematics.

Sets With Rules

Everything starts with sets. A bare set has no structure — just a collection. The different branches of mathematics are what happens when you add different kinds of rules on top:

That last row is where I had a genuine "wait, what" moment: probability is just measure theory with one extra constraint — the total measure of the sample space equals 1. That's Kolmogorov's axioms from 1933. It's why modern probability is a branch of real analysis, not just counting outcomes:

$$P(\Omega) = 1, \quad P(A) \geq 0, \quad P\!\left(\bigsqcup_{i} A_i\right) = \sum_i P(A_i)$$

Combining sets with richer structure gives you the major spaces of modern mathematics:

• Banach spaces: A vector space with a norm and completeness — every Cauchy sequence converges. The $L^p$ spaces live here:

$$\|f\|_p = \left(\int |f(x)|^p\, dx\right)^{1/p}$$

• Hilbert spaces: A Banach space with an inner product — Euclidean geometry generalised to infinite dimensions. The quantum state of a particle is a vector in a Hilbert space:

$$\langle f, g \rangle = \int f(x)\overline{g(x)}\, dx$$

• Sobolev spaces $W^{k,p}(\Omega)$: Functions and their weak derivatives in $L^p$. You cannot seriously work on PDEs like Navier–Stokes without them.

Analysis

Analysis is the mathematics of change: limits, derivatives, integrals, convergence. But you need some notion of "closeness" to even define a limit — a topology or a metric.

• Real analysis: Functions on $\mathbb{R}$. Riemann and Lebesgue integration. This is roughly where most engineering curricula stop.
• Complex analysis: Functions on $\mathbb{C}$. Holomorphic functions are strangely well-behaved — knowing the values on a small contour determines the values everywhere inside it (Cauchy's integral theorem). The first time I read this I assumed I'd misunderstood something.
• Functional analysis: Instead of studying numbers or points, you study spaces of functions — each "point" is itself a function. Maps between these spaces are operators — the infinite-dimensional generalisation of matrices.

Connecting Discrete and Continuous

A rough cut through mathematics: discrete (integers, graphs, combinatorics) and continuous (smooth functions, geometry, analysis). The interesting thing is that mathematicians keep finding ways to connect these two worlds, and when they do, the results tend to be significant.

Fermat's Last Theorem

No positive integers $a, b, c$ and integer $n > 2$ satisfy:

$$a^n + b^n = c^n$$

Fermat wrote in a margin in 1637 that he had a proof, which the margin was too small to contain. For 350 years nobody found it. Andrew Wiles proved it in 1995, but not by attacking Fermat directly.

Wiles went through the Taniyama–Shimura conjecture: every elliptic curve over $\mathbb{Q}$ is modular. An elliptic curve ($y^2 = x^3 + ax + b$, a number-theoretic object) corresponds to a modular form (from complex analysis). Two things from opposite sides of the discrete/continuous divide are secretly the same object viewed differently.

In 1984, Gerhard Frey noticed that a solution to Fermat's equation would produce an elliptic curve too strange to be modular. Ken Ribet then proved that precisely. So: Taniyama–Shimura $\Rightarrow$ Fermat.

Wiles worked on this in secret for seven years, announced a proof in 1993, a gap was found, and fourteen months later he and Richard Taylor fixed it. The final proof is about 130 pages. Whatever Fermat had in mind in 1637 — if anything — it was not this.

The Langlands Programme

The Fermat story isn't isolated. It's one piece of the Langlands programme — a web of conjectured connections, sketched by Robert Langlands in 1967, between number theory, harmonic analysis, and representation theory. The goal is to build a dictionary where hard problems in one world become tractable using tools from another.

One instance: the Green–Tao theorem (2004), which shows the primes contain arbitrarily long arithmetic progressions. Green and Tao approached the primes — discrete, seemingly chaotic — through higher-order Fourier analysis:

$$\hat{f}(\xi) = \int_{\mathbb{R}} f(x)\, e^{-2\pi i x \xi}\, dx$$

Same move as Wiles. When direct attack is too hard, translate the problem into a setting where better tools exist.

Category Theory

Category theory studies not mathematical objects themselves but the relationships between structures — morphisms, functors — at the most general level. It doesn't care what objects are, only how they relate through structure-preserving maps.

The most useful application I came across: Algebraic Topology. Comparing two topological spaces directly is hard. So instead, you map them to algebraic groups, where computation is feasible. A geometric problem becomes an algebra problem. Same pattern as above.

A note on physics

String theory depends on differential geometry and Calabi–Yau manifolds. The debt runs the other direction too.

The Poincaré Conjecture was proved by Grigori Perelman in the early 2000s using Ricci Flow — an equation describing how a Riemannian metric evolves:

$$\frac{\partial g_{ij}}{\partial t} = -2 R_{ij}$$

Richard Hamilton proposed this in 1982 and saw its potential, but got stuck when the flow produced singularities — points where the geometry blows up. Perelman's contribution was Ricci Flow with surgery: cutting out and patching singularities so the flow could continue. That was the actual invention.

Perelman declined the Fields Medal and the $1{,}000{,}000$ USD prize.

What This Has to Do With Engineering

More than I expected, though not in the ways I was looking for.

Filter stability in digital signal processing requires all poles of the transfer function to lie inside the unit circle $|z| < 1$ in the complex plane:

$$H(z) = \frac{B(z)}{A(z)}, \quad z \in \mathbb{C}$$

That's a direct application of the topology of $\mathbb{C}$. If you understand the structure of that space, stability analysis becomes geometric reasoning.

More surprising: Harmonic Analysis — the extension of Fourier series — is simultaneously what Tao used to find structure in the primes and the mathematical foundation of audio compression and digital communications. I did not expect those to be the same tools.

The Vocabulary Problem

The vague answer my friends gave — "Topology studies structure, continuity, generalisation of space" — wasn't vague. I just didn't have enough vocabulary to parse it.

I'm not going to become a pure mathematician. But spending a few months on this changed how I think about the field. There's a lot that doesn't show up in engineering curricula, and some of it turns out to connect to things I already do.

This is one engineer's attempt to map unfamiliar territory — not an academic reference. Happy to be corrected.