When Mathematics Looks Broken

What divergent series, imaginary numbers, and the Langlands program suggest about the limits of our mathematical language.

\[ 1 + 2 + 3 + 4 + \cdots = -\frac{1}{12} \]

The first time I encountered this identity, I found it almost offensive. Add positive integers forever and somehow get a negative fraction? A friend and former academic mentor sent me a Numberphile video about it and remarked that the derivation seemed buggy. I understood the objection. If we were doing ordinary arithmetic with an ordinary infinite sum, something had gone wrong. But I became interested in a different question: why did the supposedly wrong answer keep appearing in mathematics and theoretical physics?

The ordinary sum diverges. Its partial sums grow without bound. Yet \(-1/12\) belongs to a coherent mathematical story, with applications to quantum fields. What interests me is how an expression that looks errant can lead us toward structures that help describe nature.

A simpler discomfort comes from \(0.999\ldots = 1\). Every finite string of nines falls short of one, so it is tempting to imagine an infinitely small gap remaining. But the infinite decimal denotes the limit of those finite approximations. In the real numbers, that limit is exactly one: \(0.999\ldots - 1 = 0\). Two descriptions have encouraged us to imagine two numbers where there is only one. This requires no new arithmetic. It shows how easily the language itself can misdirect our intuition. The divergent sum takes us somewhere stranger: its ordinary interpretation really does fail, and understanding the finite value requires another framework.

The function beyond the series

Consider the Riemann zeta function. In the region where its defining series converges, we can write

\[ \zeta(s) = 1 + \frac{1}{2^s} + \frac{1}{3^s} + \cdots. \]

If we naively substitute \(s = -1\), we get our troublesome sum. But that substitution is outside the series’ domain of convergence.

Here is the beautiful part: the function has a unique meromorphic continuation. In plain language, it extends across the complex plane as an analytic function except for a pole at \(s = 1\). Once we require agreement with the original function and this analytic structure, the extension is forced. Nobody gets to choose a pleasing value at \(-1\). It is \(\zeta(-1) = -1/12\).

There is a second way to see why this number deserves attention. Instead of stopping the sum abruptly at \(N\), gradually fade out its large terms with a smooth cutoff \(\eta\), normalized so that \(\eta(0) = 1\). Under suitable smoothness and decay conditions, the result has the form

\[ \sum_{n} n\,\eta(n/N) = C_\eta N^2 - \frac{1}{12} + O(1/N). \]

The growing term depends on the cutoff. The constant \(-1/12\) survives across this class of cutoffs. Terence Tao explains this through the Euler–Maclaurin formula: a real-variable calculation recovers the value obtained through complex analysis. Agreement between the approaches gives us a reason to take the extension seriously, beyond the elegance of its definition.

One concrete physical example is a massless scalar field confined to a one-dimensional interval of length \(L\), with the field fixed to zero at its ends. Its standing-wave frequencies are integer multiples of \(\pi c/L\). Adding their quantum zero-point energies produces a divergent expression proportional to \(1 + 2 + 3 + \cdots\). After subtracting the unbounded-space contribution consistently, the finite Casimir energy is \(-\pi\hbar c/(24L)\), involving precisely the factor \(-1/12\). This is an idealized quantum-field model; the electromagnetic Casimir problem has additional geometry and modes. Zeta regularization offers one route to the finite answer, rather than being required for it. The surprise is that mathematics developed to understand an apparently absurd expression also participates in a physical calculation.

When impossible numbers became indispensable

There is a historical parallel that I find irresistible: imaginary numbers.

In the sixteenth century, Cardano’s formula for cubic equations confronted mathematicians with square roots of negative numbers even when the answers were real. Consider \(x^3 = 15x + 4\). The number 4 plainly works, but the formula produces \(\sqrt[3]{2 + 11i} + \sqrt[3]{2 - 11i}\). Bombelli recognized that \((2 + i)^3 = 2 + 11i\) and \((2 - i)^3 = 2 - 11i\). Choosing those cube roots, the mysterious expression becomes \((2 + i) + (2 - i) = 4\). The apparently impossible intermediate quantities had brought us back to a real answer.

The complex numbers gave these calculations a consistent home. Centuries later, they became central to the mathematics of waves, electrical circuits, and quantum mechanics. An extension motivated by algebra proved remarkably suited to describing the physical world. That later usefulness is the part I find hardest to shrug off.

Extending the numbers and continuing a function are different operations. Their shared philosophical interest lies in what happens to our sense of impossibility. A familiar restriction can look like a boundary of intelligibility until a richer framework makes the formerly strange object usable—and reveals relationships we could barely formulate before.

Euler, Ramanujan, and intuition before language

Euler and Ramanujan belong in this story for a very specific reason: both encountered this very value. In work presented in 1749 and published in 1768, Euler investigated relations between series of positive and reciprocal powers, anticipating the zeta functional equation. The association of \(1 + 2 + 3 + \cdots\) with \(-1/12\) predates Riemann’s mature complex-analytic treatment.

In his second letter to Hardy, dated 27 February 1913, Ramanujan explicitly associated the same series with \(-1/12\), adding the crucial words “under my theory.” He knew how outrageous it would sound outside that context. What fascinates me is the combination of audacity and specificity: he was asserting a mathematical relationship, with a particular meaning, that an ordinary reading of the notation concealed.

Ramanujan’s devotion to the goddess Namagiri is part of how he understood his own inspiration. I find that spiritual dimension moving, without treating it as mathematical evidence. A proof cannot establish the divine origin of an insight. But a complete account of mathematical discovery should leave room for how revelation-like discovery can feel: the experience of recognizing an order whose existence you had no reason to expect. The source of an intuition and the justification of its conclusions are separate questions.

The strange traffic between discrete and continuous

What makes this more than a curiosity about one divergent series is how often modern mathematics builds bridges between worlds that seem unrelated.

Consider an elliptic curve given by an equation with rational coefficients. Reduce the equation modulo a prime \(p\) of good reduction and count its points, including the point at infinity. Form the integer \(a_p = p + 1 - \#E(\mathbb{F}_p)\). The modularity theorem tells us that these integers are the prime-indexed coefficients of a corresponding weight-two modular form: a highly symmetric complex-analytic function. A collection of discrete counting problems is encoded in an analytic object. This is a concrete instance of the arithmetic–automorphic relationship developed far more broadly by the Langlands program.

The connection reaches beyond a resemblance in vocabulary. For example, for \(E: y^2 = x^3 - x\), the point counts at \(p = 3\), \(5\), and \(7\) are \(4\), \(8\), and \(8\). They give \(a_3 = 0\), \(a_5 = -2\), and \(a_7 = 0\), matching the coefficients of \(q^3\), \(q^5\), and \(q^7\) in its modular form, whose expansion begins \(q - 2q^5 - 3q^9 + \cdots\). The theorem establishes an entire correspondence, rather than a few numerical coincidences.

Langlands does not follow from regularizing the sum of the integers. It enlarges the question raised by that example: why do independently developed mathematical languages fit together so precisely? Arithmetic and analysis retain their distinctions, but their relationship can be far deeper than those distinctions initially suggest. I cannot look at correspondences like this without wondering how much of the underlying organization we still fail to see.

What are we beginning to see?

My question is less whether mathematics is invented or discovered than why expanding its language so often uncovers unexpected order. In the examples here, apparent nonsense gives way to precise structure. In some cases, that structure also helps us describe the physical world. Understanding the calculation settles a mathematical problem while leaving a philosophical astonishment intact: why should this work so well?

There is something almost spiritual about that encounter. One can understand why a mathematician might speak of discovery as revelation, or why mathematical intelligibility has invited religious interpretations. Neither beauty nor usefulness establishes a divine author. Yet calling the relationships consequences of definitions does not exhaust what is remarkable about them. We choose where to begin; we can still be startled by how much is waiting there, and by the places it leads.

I do not have a settled account of what this says about reality. My suspicion is that some of the categories that feel basic to us reflect the limits of our present understanding. When a larger framework connects them, we gain a glimpse of an organization that our earlier descriptions obscured. Whether that organization is intrinsic to nature, to mathematics, or to their relationship is the question I want to keep open.

And occasionally, what looks like broken mathematics is an invitation to discover a larger world.