Lévy-Steinitz Theorem: The Geometry of Infinite Sums

While the rearrangement of an infinite series of numbers can famously lead to any sum imaginable, a question naturally arises: what happens when we move from numbers on a line to vectors in space? Does the same chaos ensue, or does a new structure emerge? This is the central problem addressed by the Lévy-Steinitz theorem, a profound result that reveals a surprising and elegant order hidden within the infinite. Instead of a chaotic free-for-all, the set of all possible vector sums is always a clean geometric object—a point, a line, or a plane.

This article delves into the beautiful logic of this theorem. It is structured to first build your intuition and then explore its far-reaching consequences.

In the first chapter, ​​Principles and Mechanisms​​, we will dissect the "why" behind this geometric structure. You will learn about the crucial concept of "stable directions" and the elegant orthogonal relationship that constrains the freedom of rearrangement, ultimately dictating the shape of the possible outcomes.

Following this, the chapter on ​​Applications and Interdisciplinary Connections​​ will demonstrate the theorem's power in practice. We will see how it predicts the geometric fate of different series and explore its surprising relevance in fields beyond pure mathematics, including the abstract spaces of modern physics and computer graphics.

In our journey so far, we've encountered the surprising idea that by simply reshuffling the terms of certain infinite sums—the conditionally convergent ones—we can change the result. For numbers on a line, Bernhard Riemann discovered something astonishing: you can rearrange such a series to add up to any value you please, from negative infinity to positive infinity. It’s like having a bag of positive and negative steps of dwindling sizes, which you can arrange to land you anywhere on a path.

But what happens if our steps are not just forward and backward, but are vectors in a plane, or in three-dimensional space? If we have a series of vectors $\sum \vec{v}_n$, can we rearrange them to reach any point in the plane? Or are we constrained? Do we trace out a specific shape? The answer, unveiled by the beautiful ​​Lévy-Steinitz theorem​​, is one of the great examples of how abstract mathematical structure dictates tangible geometric outcomes. It turns out the set of all possible sums is not a chaotic mess; it is always a clean, simple geometric object: a point, a line, a plane, or a higher-dimensional equivalent (an ​​affine subspace​​). Our mission in this chapter is to understand the "why" behind this remarkable structure.

Let's imagine our vector series $\sum \vec{v}_n$ as a list of instructions for a walk in the park. Each vector is a step. A rearrangement is just taking those steps in a different order. The question is: what is the set of all possible endpoints of such a walk?

The secret lies in identifying directions of "stability" within the series. Imagine you have a compass that you can point in any direction, let's say a direction represented by a unit vector $\vec{u}$. Now, for each step $\vec{v}_n$ in our series, you measure how much progress it makes in your chosen direction $\vec{u}$. This is simply the dot product, $\vec{v}_n \cdot \vec{u}$. This gives you a new series of real numbers.

Now, we ask a crucial question: does this new real series of projected lengths converge absolutely? That is, does the sum of the absolute values, $\sum_{n=1}^\infty |\vec{v}_n \cdot \vec{u}|$, converge to a finite number?

If it does, then the direction $\vec{u}$ is a special, ​​stable direction​​ for our series. Why "stable"? Because for any direction where the projected series converges absolutely, Riemann's theorem no longer applies in its wild form. The sum of the components in that direction is fixed. No matter how you rearrange the original vector series, the sum of the projected components, $\sum (\vec{v}_{\sigma(n)} \cdot \vec{u})$, will always yield the same value. Rearranging the steps can't change the total distance you've traveled along that specific compass direction.

The set of all such stable directions, let's call it $U$, is not just a random collection of vectors. It forms a beautiful mathematical object in its own right: a ​​linear subspace​​. This means if you have two stable directions, any linear combination of them is also a stable direction. In a plane, this subspace $U$ can only be one of three things: the single point at the origin (if there are no stable directions), a line through the origin (if an entire line of directions is stable), or the entire plane itself (if all directions are stable).

So, if rearrangements can't change the sum in the stable directions of $U$, where can they work their magic? The answer is as elegant as it is surprising: they can only produce movement in the directions that are ​​orthogonal​​ (perpendicular) to all the stable directions.

This set of "unstable" or "wild" directions also forms a linear subspace, known as the ​​orthogonal complement​​ of $U$, denoted $U^\perp$. The Lévy-Steinitz theorem's grand conclusion is that the set of all possible rearrangement sums, let's call it $\mathcal{S}$, is simply the subspace $U^\perp$ shifted by some specific sum $\vec{S}_0$. In mathematical terms, $\mathcal{S} = \vec{S}_0 + U^\perp$.

This reveals a stunning duality: the geometry of the sum set is the orthogonal complement of the geometry of absolute convergence. The dimensions of these two spaces are linked by a simple formula in $k$-dimensional space: $\dim(\mathcal{S}) = \dim(U^\perp) = k - \dim(U)$ This single equation is our key to unlocking the geometry of rearrangements. Let's see it in action in the familiar 2D plane ($k=2$).

Based on our central principle, we can now see why a conditionally convergent series in the plane can only have one of three fates for its sum set.

​​1. The Entire Plane: Ultimate Freedom​​

What if there are no stable directions at all (apart from the trivial zero vector)? This means $U = \{\vec{0}\}$, so its dimension is $\dim(U)=0$. Our magic formula tells us the dimension of the sum set is $\dim(\mathcal{S}) = 2 - 0 = 2$. A 2-dimensional affine subspace of the plane is the plane itself! In this case, you can truly reach any point.

When does this happen? It happens when your vector terms $\vec{v}_n$ are sufficiently "un-aligned". Consider the classic series $\sum_{n=1}^\infty \frac{e^{in\theta}}{n}$. If $\theta/\pi$ is an irrational number, the directions of the terms $e^{in\theta}$ never repeat and are spread densely around the unit circle. There's no special direction $\vec{u}$ for which the projections converge absolutely. The series is "wild" in every direction, and the sums fill the entire complex plane. Similarly, for a hypothetical series in $\mathbb{R}^3$ like $\vec{v}_n = n^{-2/3}(\cos(\ln n), \sin(\ln n), (-1)^n)$, the terms spiral and flip in such a way that no non-zero functional can tame them; the sums can be rearranged to reach any point in 3D space.

​​2. A Single Point: Absolute Stability​​

What if every direction is stable? This means $U = \mathbb{R}^2$, so $\dim(U)=2$. The dimension of our sum set is then $\dim(\mathcal{S}) = 2 - 2 = 0$. A 0-dimensional space is a single point. This makes perfect sense: if the series converges absolutely in every direction, it means the series itself was ​​absolutely convergent​​ to begin with ($\sum|\vec{v}_n| \infty$). And for absolutely convergent series, all rearrangements yield the same sum.

​​3. A Line: Constrained Freedom​​

This is perhaps the most interesting case. The sum set forms a line when its dimension is 1. Our formula tells us this happens when $\dim(\mathcal{S}) = 1$, which requires $\dim(U) = 2 - 1 = 1$. In other words, ​​the set of sums is a line if and only if the set of stable directions is also a line​​.

This means there is one specific line's worth of directions (spanned by some vector $\vec{u}_0$) where the series is "tame", and the sum of projections is fixed. In the perpendicular direction (spanned by $\vec{u}_0^\perp$), the series is "wild", and we can rearrange it to achieve any value along that line.

We can actually hunt for this stable direction. Consider a hypothetical series with terms $\vec{v}_n = (\frac{(-1)^{n+1}}{n}, \frac{(-1)^{n+1}}{2n-1})$. The components are very similar, both behaving like $\frac{(-1)^{n+1}}{n}$ for large $n$. Let's look for a stable direction $\vec{u}=(a, b)$. The dot product is $\vec{v}_n \cdot \vec{u} = (-1)^{n+1} (\frac{a}{n} + \frac{b}{2n-1})$. For the series of absolute values $\sum |\vec{v}_n \cdot \vec{u}|$ to converge, we need the term in the parenthesis to decay faster than $1/n$. For large $n$, $\frac{1}{2n-1} \approx \frac{1}{2n}$. So we need $\frac{a}{n} + \frac{b}{2n} = \frac{1}{n}(a + \frac{b}{2})$ to be as small as possible. The only way to achieve faster convergence is if we have cancellation: $a + \frac{b}{2} = 0$. This condition, which is equivalent to $b=-2a$, defines a line of stable directions! For example, $\vec{u}=(1, -2)$ is a stable direction. The resulting line of sums must therefore be perpendicular to $(1, -2)$, meaning it points in the direction $(2, 1)$.

This same principle shines beautifully when components have different convergence rates. Imagine a series in $\mathbb{R}^3$ given by $\vec{v}_n = (\frac{(-1)^n}{n}, \frac{\sin n}{n^2}, \frac{\cos n}{n^3})$. The second and third components form absolutely convergent series on their own ($\sum 1/n^2$ and $\sum 1/n^3$ converge). The only conditional part is the first component. A direction $\vec{u}=(u_1, u_2, u_3)$ is stable if $\sum |\vec{u} \cdot \vec{v}_n|$ converges. The absolute convergence of the $y$ and $z$ parts can't be spoiled by $u_2$ and $u_3$. The only way to spoil the convergence is for the $\frac{(-1)^n}{n}$ term to survive. To tame the series, we must kill this term by choosing $u_1=0$. Thus, the subspace of stable directions $U$ is the entire $yz$-plane ($u_1=0$). The space of sums, $U^\perp$, is the line orthogonal to this plane: the $x$-axis. Rearrangements can only move the sum along the $x$-axis, forming a line.

The principle is universal: the geometry of what you can build is determined by the constraints you cannot escape. The wild, untamed nature of conditional convergence allows us to travel, but the directions of stability fence our journey into a specific, elegant affine subspace. The dance between the conditional and the absolute forges the geometry of the infinite.

Now that we have grappled with the principles behind the Lévy-Steinitz theorem, we can embark on a journey to see it in action. You might be tempted to think this is a niche result, a curiosity confined to the back pages of an analysis textbook. But nothing could be further from the truth. This theorem is a profound statement about structure, freedom, and predictability in spaces of any dimension. It surfaces in surprising and beautiful ways, from the familiar planes of Euclidean geometry to the abstract realms of quantum physics. Let us explore this hidden landscape.

In the one-dimensional world of real numbers, a rearranged conditionally convergent series is an agent of chaos; it can be bent to any value. But the moment we step into two or more dimensions, a remarkable order emerges. The set of all possible sums is no longer a free-for-all; it is always a clean, crisp geometric object called an affine subspace—a point, a line, a plane, or its higher-dimensional equivalent. What shape we get is not an accident. It is a direct consequence of the series' internal structure.

Imagine a series of vectors in a plane, $\mathbb{R}^2$. When can we rearrange it to create a sum anywhere on a specific ​​line​​? This happens when the "conditionally convergent character" of the series is constrained to a single direction. For instance, if the components of our vectors $\mathbf{v}_n = (x_n, y_n)$ are constructed such that their conditionally convergent parts are always moving along the line $y=x$, then no amount of shuffling can break that underlying correlation. The freedom to rearrange only allows us to slide the final sum up and down this line. The parts of the vectors that converge absolutely simply shift this entire line to a new position in the plane, without changing its direction.

What if we want more freedom? Let’s move to three dimensions, $\mathbb{R}^3$. Suppose we build a series $\sum \mathbf{v}_n$ where two components converge conditionally (say, with terms behaving like $\frac{(-1)^n}{n}$), but the third component converges absolutely (with terms like $\frac{1}{n^2}$). The absolutely convergent series adds up to a definite, unchangeable number regardless of the order of its terms. This acts as a "pin," locking the sum in that third dimension. If you rearrange the series, the sum's $z$-coordinate is fixed at the value $z_0 = \sum_{n=1}^\infty \frac{1}{n^2} = \frac{\pi^2}{6}$. However, in the other two dimensions, you retain the freedom to rearrange. This freedom is not total; it is confined to a plane. The set of all possible sums becomes a two-dimensional plane defined by the equation $z = z_0$. This abstract plane of sums can have tangible consequences. If we ask which of these possible sums lie within a unit sphere centered at the origin, the intersection forms a perfect circular disk whose area we can precisely calculate. The abstract theorem suddenly yields a concrete, measurable geometric feature.

This leads to a natural, exhilarating question: can we achieve total freedom? Can we construct a series that can be rearranged to sum to any point in the space? The Lévy-Steinitz theorem gives a resounding "yes" and tells us exactly how. The key is to ensure there are no "preferred directions" of absolute convergence. If, for any projection of the vector series onto any line, the resulting scalar series still fails to converge absolutely, then there are no constraints on our rearrangement power. We can steer the sum to any target we desire. This happens, for example, in series whose components involve terms like $\cos(n)$ and $\cos(n\sqrt{2})$, where the irrationality of the frequencies ensures that no linear combination can accidentally produce the rapid decay needed for absolute convergence. In these cases, the set of sums is the entire space—be it the complex plane $\mathbb{C}$ or the three-dimensional space $\mathbb{R}^3$. A fascinating example arises from the familiar power series for the logarithm, $\sum \frac{z^n}{n}$. This series can be rearranged to sum to any complex number precisely when $z$ lies on the unit circle, but is not equal to $1$—the very region where the series converges, but not absolutely.

The true power and unity of a mathematical idea are revealed when it transcends its original context. The "vectors" in the Lévy-Steinitz theorem need not be simple arrows in Euclidean space. They can be elements of any finite-dimensional vector space, including the exotic-sounding structures that form the bedrock of modern physics.

Consider the ​​quaternions​​, a four-dimensional extension of complex numbers used in everything from spacecraft orientation to 3D computer graphics. A quaternion series can be seen as a series in $\mathbb{R}^4$, and the theorem applies flawlessly. If we construct a series of quaternions where two components decay slowly (like $1/n$) and two decay quickly (like $1/n^2$), the theorem tells us that the rearrangement freedom will exist only in the two "slow" dimensions. The sum set will be a 2D plane embedded within the 4D space of quaternions, with its position and orientation determined entirely by the series's structure.

The connections become even more profound when we enter the world of quantum mechanics. The state of a quantum system, like the spin of an electron, is described not by a simple vector but by an element of a ​​Lie algebra​​. For example, the Lie algebra $\mathfrak{su}(2)$, which governs spin-1/2 particles, can be viewed as a 3D vector space where the basis vectors are related to the famous Pauli matrices ($\sigma_1, \sigma_2, \sigma_3$). A series of elements in $\mathfrak{su}(2)$ is just a vector series in disguise, and the Lévy-Steinitz theorem holds.

Imagine a series of "spin vectors" $\sum \mathbf{v}_n$, where the components along the $i\sigma_1$ and $i\sigma_2$ directions are conditionally convergent, but the component along the $i\sigma_3$ direction is absolutely convergent (e.g., its terms decay like $1/n^2$). The theorem predicts that while we can rearrange the series to get any desired values for the first two components, the third component is immutable. Its sum is fixed, a kind of "conserved quantity" under rearrangement. We can exploit this to perform a remarkable feat: we can find a specific rearrangement that makes the sums of the first two components exactly zero, thereby isolating the fixed, absolutely convergent part of the sum. This isn't just observation; it's a principle of control. We can even design series where the resulting line of sums passes through a special point, like the origin, by carefully choosing the parameters of the series.

From simple lines in a plane to the quantum mechanics of spin, the Lévy-Steinitz theorem provides a single, unifying framework. It reveals a deep truth about the interplay between order and chaos, freedom and constraint. It teaches us that even when dealing with the infinite, there is a hidden structure, a beautiful geometry waiting to be discovered. The sum may change, but the shape of possibility is written in the very terms of the series itself.