The first time you really understood a proof, you probably weren’t reading it.
You were in the middle of something else, your own problem, your own mess of ideas, when the technique you’d read about three months earlier suddenly clicked into place. Not because you’d finally memorized it, but because you needed it. The stopping time argument, the corona decomposition, whatever it was: it stopped being a thing you’d seen and became a thing you could use. That moment has a texture you don’t forget.
There’s a framework floating around that tries to name what happened: the difference between a 2D lesson and a 3D lesson. The 2D lesson is reading the idea, being it encountering it in a paper, a textbook, a seminar. Flat. Valuable, but flat. The 3D lesson is implementing it, wrestling the idea into an actual proof, an actual project, where it either works or it doesn’t, and you have to figure out which. The metaphor isn’t perfect, but it captures something real about how mathematical knowledge is actually built.
The uncomfortable part is how much time we spend in two dimensions.
Graduate programs, by their nature, front-load 2D lessons. You read papers. You attend talks. You work through textbooks. This isn’t wrong, you have to encounter ideas before you can implement them, but the architecture of graduate training can create the illusion that comprehension equals capability. You followed the argument. You could reproduce the steps. And then you sit down to prove something yourself and find that the gap between “I understand this” and “I can do this” is wider than a canyon.
The failure stings in a particular way. It’s easy to read that gap as evidence of personal inadequacy, a sign you don’t belong, or didn’t really understand the paper as well as you thought. But the gap isn’t personal. It’s structural: the 2D lesson tells you what a technique does; the 3D lesson teaches you how it behaves, its load-bearing assumptions, where it breaks, what it costs inside a proof. You can’t learn the second without doing, and doing always means failing several times first.
What I find humbling, even now, is that this doesn’t stop. I’ve been a working mathematician long enough to have a reasonably established career, and I still notice myself stuck in two dimensions more often than I’d like to admit. I’ll read something beautiful, a new approach to a problem I care about, and feel the warm glow of comprehension. I followed it. I could explain the main idea at a seminar. And then I try to use it, really use it, and discover that I understood the map of the technique without having walked the terrain. The dimension shift from reading to doing is not a rite of passage you complete once in graduate school and never revisit. It is a recurring fact of mathematical life. Every new corner of mathematics you enter is a new 2D lesson waiting to become a 3D one.
What comes after the 3D lesson, though?
If we take the metaphor seriously, there’s a natural next question: is there a 4D lesson? I think the answer is yes, and it’s more unsettling than the first two.
The 4D lesson is teaching it. Not explaining the technique to yourself as you write up a proof, but standing in front of someone who doesn’t yet have it and finding that your explanation doesn’t work. That you can’t quite say why the stopping time happens at that moment, only that it does. That you’ve been using the technique correctly without ever fully articulating the logic that justifies it. Teaching forces a reckoning that neither reading nor doing requires: you have to make the implicit explicit. You have to find language for intuitions you’ve been operating on wordlessly for years.
Graduate students who TA courses report this experience constantly: they thought they understood something, tried to teach it, and discovered they understood a version of it, a version their students’ questions had no patience for. The knowledge was real, but incomplete in ways that only became visible when it had to travel across a room and into someone else’s head. Richard Feynman’s rule was “if you can’t explain it simply, you don’t understand it yet”, but even that understates what teaching actually reveals. It’s not just about simplicity. It’s about discovering the assumptions you didn’t know you were making, the steps you’d been skipping, the cases your mental model quietly excluded.
I’ve experienced this with my own mentees in a way I didn’t anticipate when I first started advising. I thought that having done the 3D work, having implemented a technique successfully in my own research, meant I could hand it across cleanly. What I found instead was that trying to engineer the dimension shift for someone else forced me through it again, differently. Watching a student struggle with a stopping time argument in a new context, asking questions I’d never had to answer out loud, I discovered corners of my own understanding that had never been fully lit. Mentoring, it turns out, is a 4D lesson you take alongside your student.
There’s a 5D lesson too, and it’s the one that makes a mathematician.
If teaching forces you to articulate what you know, transferring forces you to abstract it. The 5D lesson is recognizing a technique in a problem where it isn’t labeled, seeing the ghost of a stopping time argument inside something that looks nothing like harmonic analysis. This is mathematical maturity in its most precise sense: not just having a powerful tool, but understanding it abstractly enough that you can see its shadow in unfamiliar territory.
This is the difference between a mathematician who has mastered a technique and one who has made it their own. Mastery means you can execute reliably. Making it your own means it starts generating new ideas, suggesting new problems, whispering at you from the edge of something you haven’t yet proved. That doesn’t come from reading or implementing or even teaching alone, it comes from having been through all three, enough times, with enough different techniques, that you start to see the deep structural similarities between things that look different on the surface. The mathematician who has done this work has a different relationship to their toolkit. The ideas are no longer inert; they talk back.
This progression doesn’t only live in mathematics.
The surgeon’s training mantra, “see one, do one, teach one”, names three of the same dimensions. Surgeons don’t become capable in the operating room by watching; they become capable by doing, in the presence of someone more experienced. And they’re considered competent to train others only after they’ve taught, which is when the remaining gaps in their procedural knowledge become visible and fixable. The medical community has institutionalized the dimension shift. Academic mathematics has largely left it to chance.
That’s worth sitting with. Most mathematicians encounter 3D lessons almost by accident: a graduate student gets lucky with an advisor who assigns a problem that forces implementation, or they don’t get lucky and spend a year reading papers, understanding everything, and learning nothing that will carry them through the isolating work of original research. Good mentoring deliberately engineers the shift, identifies what a student has read but not done, and finds the problem, low-stakes enough to be safe and high-stakes enough to require genuine engagement, that forces the 3D lesson. And eventually it engineers the 4D lesson too, which is why the best advisors push students to give talks, to explain their work out loud, to encounter the productive humiliation of having an assumption challenged by a confused first-year. The goal isn’t to expose students to difficulty for its own sake. It’s to accelerate a transition that, left to chance, might never come.
What I’ve found in my own advising is that I can’t hand students the 3D lesson, I can only design conditions that make it likely. I can assign the right problem at the right moment. I can resist the urge to explain too much when a student is stuck at the edge of their understanding, because sometimes the stuck place is the lesson. And I can model the ongoing nature of it: let students see that their advisor is also, somewhere, in the middle of a 2D lesson that hasn’t yet become a 3D one. That the dimension shifts don’t end with a PhD, or a tenure decision, or any other milestone.
The brilliant paper you read last summer is still there for you, on the shelf. It will still be there next month. But if you’ve been reading without doing, without teaching, without struggling to transmit, if you’ve been confusing the map for the terrain, you might want to ask yourself honestly: in how many dimensions do you actually know this? And which dimension, if you committed to it, would change what you’re capable of next?