The Shape of the Thought

---

Einstein once joked that his greatest discovery in mathematics was suppressing the summation sign. The convention he introduced in 1916 — repeated indices implicitly sum — eliminated a symbol that appeared in nearly every equation of tensor calculus. The Σ was always there, always doing the same thing, always taking up space. Removing it didn’t change the math. It changed the thinking. With the summation sign gone, the structural relationships between tensors became visible. The notation got out of the way, and the physics stepped forward.

This is the central claim of mathematical notation design, usually left implicit: notation shapes cognition. The symbols aren’t labels stuck onto pre-existing thoughts. They are the medium through which the thoughts become possible. Choose badly, and you spend your cognitive budget parsing syntax. Choose well, and the notation does part of your reasoning for you.

The War That Shaped Calculus

The most consequential notation battle in history was fought between Newton and Leibniz in the late seventeenth century. Both invented calculus independently; both chose different symbols to express it.

Newton used dots: ẋ for the first derivative of x, ẍ for the second. Compact, clean, attached to the variable itself. Leibniz used fractions: dy/dx. Heavier, two symbols instead of one, but with a crucial property: the fraction looks like a ratio, and in many contexts it behaves like one. The chain rule in Leibniz’s notation is dy/dx = (dy/du)(du/dx) — you can “cancel” the du’s. In Newton’s notation, the same rule requires a separate formula with no visual mnemonic.

Leibniz won. Not immediately — Britain clung to Newton’s dots for a century, during which continental mathematics surged ahead. The reason wasn’t loyalty to a dead hero; it was that Leibniz’s notation scaled. When calculus moved to multiple variables, partial derivatives, and higher dimensions, dy/dx generalized naturally to ∂f/∂x. Newton’s dots didn’t. A notation designed for the problems of 1680 couldn’t carry the weight of 1780.

The lesson: notation is not just about the current problem. It’s about the space of problems the notation makes thinkable. Leibniz’s fraction invited manipulation, generalization, chain-rule reasoning. Newton’s dot was a label — correct, but inert.

What Makes Notation Good?

Charles Babbage — yes, the computer pioneer — wrote about notation design in the 1820s and identified principles that still hold: conciseness, univocity (one meaning per symbol), iconicity (the symbol should suggest its meaning), modularity, and symmetry. Modern frameworks add: cognitive load, compatibility with existing conventions, and what we might call transparency — whether the notation reveals or hides the structure of the object it represents.

These principles are in tension. Conciseness wants fewer symbols; transparency wants more. Einstein’s summation convention is maximally concise — it removes a symbol entirely — but it’s only transparent if you already know the convention. A student seeing aᵢbᵢ for the first time doesn’t know a sum is happening. Dirac’s bra-ket notation ⟨ψ|φ⟩ is beautifully concise for inner products and elegantly visual — the bracket literally closes around the contents — but it took quantum mechanics a generation to adopt it because it looked like nothing else in physics.

The deepest tension is between notation that’s compact (catches the eye in a glance, fits on a line) and notation that’s structural (shows you how the parts relate). Category theory’s commutative diagrams are maximally structural — you can trace paths through them, verify equalities by following arrows, prove theorems by “diagram chasing.” But they’re spatially expensive. A diagram that replaces three lines of equations might take half a page.

There is no resolution to this tension. There are only trade-offs, made in context, for particular communities and particular problems.

The Equal Sign Took a Century

We take = for granted, but Robert Recorde introduced it in 1557, and it didn’t become standard until the late 1600s. Descartes used a rotated Taurus symbol. Leibniz used a wedge. For over a hundred years, three notations for equality competed, and adoption was driven as much by publishing networks and national pride as by intrinsic quality.

This is the uncomfortable truth about notation: quality is necessary but not sufficient. Adoption depends on who publishes, who teaches, who builds the textbooks. Leibniz didn’t just invent better notation — he published it immediately, explained it to correspondents across Europe, and embedded it in a community of practice. Newton sat on his notation for decades. The network effect did the rest.

The same pattern repeats. The plus and minus signs (+, −) displaced the Italian abbreviations p and m over decades, driven by German printing houses and their distribution reach. The integral sign ∫ is a stylized S (for summa), chosen by Leibniz for its mnemonic quality — it looks like what it does. These symbols survived not because they were optimal in some abstract sense, but because they were good enough and available at the moment the community needed them.

Notation as Programming Language Design

Kenneth Iverson spent years in the 1950s frustrated by the inadequacy of conventional mathematical notation for expressing algorithms. His response was APL — a programming language that began as a notation system, only later becoming executable. Iverson’s Turing Award lecture was titled “Notation as a Tool of Thought,” and his central argument was that the properties of a notation — its ability to express common patterns concisely, to compose operations, to suggest generalizations — directly determine the quality of thinking it enables.

This is the thread that connects Leibniz to Julia. When Julia lets you write 2x instead of 2*x, or use Greek letters α and β as variable names, or define @constraint(model, 2x + 1 <= 4) as a macro that compiles to efficient optimization code — it’s making the same bet Leibniz made: that closing the gap between mathematical thought and computational expression is worth the cost in implementation complexity.

Haskell makes the bet differently: function application is f x, not f(x), mirroring how mathematicians write. Monads come from category theory. The language is designed so that the code is the proof, or at least resembles it enough that the translation is mechanical.

The cost is real. APL is famously unreadable to newcomers — a single line can replace a page of C, but understanding that line requires fluency in a symbol vocabulary that looks like a typographer’s nightmare. Julia’s macros generate code that’s harder to debug than the mathematical expression that inspired it. Haskell’s type system, derived from formal logic, produces error messages that read like dispatches from another dimension.

But the cost is always the same cost: the distance between the notation and the machine. Close the distance on the human side (make notation look like math), and it opens on the machine side (compilation, debugging, error messages). Close it on the machine side (explicit, verbose, step-by-step), and it opens on the human side (translation effort, cognitive load, transcription bugs).

The Notation Chooses You Back

There’s a subtler point that Babbage noticed and Iverson made explicit: notation is not a passive recording medium. It actively shapes what you think, what questions you ask, what generalizations you notice.

Write a problem in index notation, and you’ll think about components. Write it in coordinate-free notation, and you’ll think about geometric invariants. Write a quantum state as a column vector, and you’ll reach for matrix multiplication. Write it in bra-ket notation, and you’ll reach for inner products and projection operators. The answer is the same; the path to the answer is different; and the paths you don’t take — the questions that never occur to you because your notation doesn’t suggest them — are the invisible cost.

I once watched a mathematician I was working with rewrite half a thesis chapter because the notation had drifted. Not because the math was wrong — the math was fine — but because inconsistent notation forces the reader to maintain a translation table in working memory, and every byte of working memory spent on notation is a byte not spent on understanding the argument. The notation should be transparent enough to disappear.

The best notation is the notation you stop seeing. Like a well-designed tool, it becomes an extension of the hand. You don’t think about the hammer; you think about the nail. You don’t think about dy/dx; you think about the rate of change.

And that — the moment the symbol dissolves into the thought — is what good notation is for.

---

— V