The Lagrange spectrum maps how “irrational” numbers act

A centuries-old question—how closely fractions can mimic irrational numbers—led mathematicians to the Lagrange spectrum, a strange map of “irrationality.” From Diophantus to Hurwitz and Markov, the story ends with a fractal structure capped by the Freiman cons

Pi’s digits never settle into a repeating pattern. The square root of 2 does the same. And yet both turn up in the simplest geometry—circumferences and diagonals—like mathematical mysteries hidden in everyday shapes.

For thousands of years. scholars have chased the same kind of question: what exactly makes an irrational number hard to pin down with fractions?. We can always approximate an irrational number using fractions of integers, rational numbers. The trick is that the closeness depends on how large the denominator, q, is. The bigger the denominators you allow, the smaller the difference you can force.

But “approximate” is not one uniform game. Some irrational numbers yield to small, manageable fractions sooner than others. That difference is what mathematicians have been formalizing since the time of Diophantus of Alexandria. an ancient Greek mathematician who—more than a millennia ago—posed the idea of finding the smallest possible fraction that still differs by as little as possible from a given irrational number. It’s a question that still shapes research today.

Dirichlet’s contribution came in the 19th century. He considered the quantity you get when you subtract a fraction p⁄q from an irrational number α. He proved that this difference can be forced to be at most 1⁄q^2. In practical terms, it meant that for every irrational number α there are infinitely many fractions p⁄q. The accuracy scales with the square of the denominator: the larger q is for a carefully chosen fraction. the more accurately the value of the irrational number can be determined. Experts. naturally. tried to do better—not by abandoning q. but by making the part involving q in the right-hand side sharper. so the approximation improves faster than Dirichlet guaranteed.

That effort produced a major milestone. In 1891. mathematician Adolf Hurwitz found a strong candidate inequality: for every irrational number α there are infinitely many fractions p⁄q that satisfy the inequality above. Yet Hurwitz’s method hit a boundary. When α is the golden ratio, the inequality still works—but only if the constant involved is within a certain size. In other words, the golden ratio limited how far mathematicians could push the denominator-based improvement.

Lagrange numbers are built out of exactly that kind of obstruction.

In the late 19th century, Andrey Markov took a different pass. He tried to omit the golden ratio and focus on the remaining irrational values. The question was whether the denominator could be refined further for all other irrational numbers. The answer was yes: apart from numbers related to the golden ratio. infinitely many fractions p⁄q can be derived to satisfy the following inequality. But the improvement again stumbled—this time on a particular irrational number: √2. Setting α equal to √2 prevented a better approximation result, just as the golden ratio had earlier.

Markov excluded √2 as well, and the inequality could be improved again. Then another stubborn number appeared, forcing another exclusion. The process could be repeated many, many times. Each round produced a new constant that shows up in the denominator on the right-hand side of the inequality. First came √5 from Hurwitz’s work, then √2 from Markov’s initial effort, followed by √2 21⁄5, and so on.

These constants form an infinitely long sequence called “Lagrange numbers,” named after Joseph-Louis Lagrange. Markov demonstrated in 1880 that they gradually approach the limit of 3. For any specific irrational number. you can find the best possible inequality for approximating it and thereby identify its corresponding Lagrange number. In number theory. these values act like a scoreboard for approximation: the smaller the Lagrange number. the more “irrational” the number is—at least in the precise sense tied to how well it can be approximated by fractions.

There is a catch: the scoreboard doesn’t just climb smoothly.

Markov’s work allowed infinitely many Lagrange numbers between √5 and 3. but these correspond to a specific class of irrational numbers calculable using a quadratic equation. Beyond that, researchers found irrational numbers with Lagrange values larger than 3—puzzling in their own right. When the Lagrange values are written out from √5 to 3 and beyond, some patterns stand out sharply.

In the range below 3, the Lagrange numbers are discrete: they show up as individual values such as √5, 2√2 and √2 21⁄5. There are infinitely many of them, but they are not consecutive.

From the number 3 onward, the picture changes. The Lagrange spectrum becomes considerably more diverse. The numbers form what’s called a fractal structure—infinitely many continuous segments separated by gaps. It can be visualized as a barcode, with narrow stripes and thicker continuous stripes following one another. Even though broad behavior is known, some details remain unclear, including which gaps contain no Lagrange numbers at all.

Still, the fractal doesn’t last forever.

The structure ends at a point known as the Freiman constant, F. In 1968, Gregory Abelevich Freiman proved that every real number greater than or equal to F corresponds to a Lagrange number. In that sense, the spectrum has a unique limit for approximating an irrational number.

And yet the Freiman constant is also where the story feels most unfinished. Unlike constants such as pi or Euler’s number e. the Freiman constant has not appeared in any other context so far. There is also uncertainty about which irrational number corresponds to the Lagrange variable F. Freiman’s proof relied on complicated number-theoretic considerations rather than concrete calculations of the Lagrange variable of irrational numbers.

So even with centuries of progress from Diophantus’s starting question through Dirichlet. Hurwitz. Markov. and Freiman. the core feeling hasn’t faded: we can map parts of irrationality. but we still don’t grasp its full nature. The Lagrange spectrum shows how deep the rabbit hole goes—how “how irrational” can split into discrete points. fractal stretches. and a continuous line. and how the constant that caps it remains strangely isolated from the rest of mathematics.

irrational numbers Lagrange spectrum Diophantus Dirichlet Hurwitz Markov Lagrange numbers Freiman constant number theory Diophantine approximation fractal structure