Sunada Construction

In 1966, the mathematician Mark Kac posed a deceptively simple question: "Can one hear the shape of a drum?" This inquiry translates into a profound problem in spectral geometry: does the set of a shape's vibrational frequencies—its "sound" or spectrum—uniquely determine its geometric form? For decades, the answer remained elusive, as no one could find two differently shaped drums that produced the same sound. The deadlock was broken by Toshikazu Sunada, who introduced a systematic method for constructing entire families of such objects, revealing a stunning connection between geometry, wave physics, and the abstract mathematics of symmetry. This article delves into Sunada's remarkable construction, providing a blueprint for how different shapes can share the same spectral song.

The journey begins in the first chapter, ​​Principles and Mechanisms​​, which unpacks the secret recipe behind the construction. We will explore the crucial group-theoretic property of "almost conjugacy" and understand the two interconnected mechanisms—one based on a transplantation of wave patterns and the other rooted in the deeper language of representation theory—that explain why this algebraic condition leads to isospectrality. Following this, the chapter on ​​Applications and Interdisciplinary Connections​​ brings the theory to life. We will examine concrete examples of isospectral shapes, from tiled planar domains to curved hyperbolic surfaces, and explore the profound consequences of Sunada's work, revealing which geometric properties the spectrum can detect and which crucial details, like orientability, remain hidden from our "ear."

Sunada's method is a bit like geometric origami. You start with a large, highly symmetric manifold, let's call it the "master sheet" $X$. This sheet has a group of symmetries, $G$, which is a collection of transformations (like rotations or reflections) that leave $X$ looking unchanged. The key idea is to "fold" this master sheet in two different ways to create two smaller manifolds, $M_1$ and $M_2$. We do this by choosing two different sets of folding instructions, which correspond to two subgroups of symmetries, $H_1$ and $H_2$, from our main group $G$. The resulting shapes are the quotients, $M_1 = X/H_1$ and $M_2 = X/H_2$.

The crucial question is: what is the secret recipe that ensures $M_1$ and $M_2$ will sound the same, even if they look different? Sunada found that the answer lies in a subtle group-theoretic property called ​​almost conjugacy​​, or Gassmann equivalence.

Two subgroups, $H_1$ and $H_2$, are said to be ​​almost conjugate​​ in $G$ if they are indistinguishable from the perspective of the "types" of symmetries they contain. In any group, elements can be sorted into ​​conjugacy classes​​. You can think of a conjugacy class as a family of operations that are fundamentally the same type of action, just viewed from a different orientation. For example, in the symmetry group of a cube, all 90-degree rotations about an axis through the center of a face belong to the same conjugacy class. The almost conjugacy condition states that for every single conjugacy class $C$ in the parent group $G$, the number of elements $H_1$ contains from that class must be identical to the number of elements $H_2$ contains from it. In symbols, this is:

This condition is a delicate balancing act. It doesn't mean the specific symmetry elements in $H_1$ and $H_2$ are the same. It just means their "inventory" of symmetry types is perfectly matched. If the subgroups were conjugate (meaning one is just a "rotated" version of the other, $H_2 = gH_1g^{-1}$ for some $g \in G$), the resulting shapes $M_1$ and $M_2$ would be identical, which is uninteresting. The magic of Sunada's construction happens when we find subgroups that are almost conjugate but ​​not​​ conjugate. These are the pairs that yield drums with the same sound but different shapes.

Why does this abstract algebraic condition of almost conjugacy lead to two objects having the same vibrational frequencies? There are two beautiful ways to understand this. The first is more intuitive and can be pictured as a "transplantation" of wave patterns.

The vibrations of our drum, its eigenfunctions, are standing waves. On our small drums $M_1$ and $M_2$, these standing waves are simply the standing waves from the master sheet $X$ that respect the folding rules of $H_1$ and $H_2$, respectively. That is, an eigenfunction on $M_1$ is an $H_1$-invariant eigenfunction on $X$.

Sunada’s condition allows us to construct a remarkable mathematical machine, an operator $T$, that can take any standing wave on drum $M_1$ and perfectly transform it into a standing wave on drum $M_2$ that has the exact same frequency (eigenvalue). This transplantation works because the operator $T$ is built from the symmetries in $G$. Since symmetries preserve the geometry, the operator $T$ commutes with the Laplacian—the very operator that defines the wave equation. If something commutes with the wave operator, it doesn't change the frequencies of the waves it acts on.

The almost conjugacy condition is the guarantee that this transplantation is a perfect, one-to-one correspondence. For any given frequency, it ensures that the number of possible standing wave patterns on $M_1$ is exactly the same as the number of patterns on $M_2$. It's as if the two drums are built from the same set of fundamental tiles, just assembled in a different way. The transplantation operator shows how to rearrange the wave patterns on the tiles of the first drum to perfectly fit the second, without changing the energy of any of the waves.

The second explanation is more abstract but reveals a deeper unity between geometry and algebra, a perspective Feynman surely would have cherished. It involves the language of representation theory.

For any given frequency $\lambda$, the set of all possible standing waves (eigenfunctions) on the master sheet $X$ forms a vector space, $E_{\lambda}$. Because the group $G$ acts on $X$ by isometries, this space $E_{\lambda}$ is not just any vector space; it is a ​​representation​​ of the group $G$. It is a mathematical stage on which the symmetries of $G$ perform their dance.

The number of waves with frequency $\lambda$ that can exist on the small drum $M_1 = X/H_1$ is simply the dimension of the subspace of $E_{\lambda}$ that is left fixed by all the symmetries in $H_1$. Let's call this dimension $\dim(E_{\lambda}^{H_1})$. Our drums $M_1$ and $M_2$ will be isospectral if and only if $\dim(E_{\lambda}^{H_1}) = \dim(E_{\lambda}^{H_2})$ for every single possible frequency $\lambda$.

This seems like an impossible condition to satisfy! The eigenspaces $E_{\lambda}$ can be incredibly complex. How could we ensure this equality holds for all of them simultaneously? This is where Sunada's genius shines. He showed that the almost conjugacy condition is precisely the algebraic key that unlocks this geometric puzzle. A deep result in representation theory (related to Frobenius Reciprocity) states that if two subgroups $H_1$ and $H_2$ are almost conjugate, then for any representation $V$ of the group $G$, the dimension of the $H_1$-fixed subspace will equal the dimension of the $H_2$-fixed subspace.

By applying this powerful theorem to each eigenspace $E_{\lambda}$, we see that the multiplicities of every eigenvalue must match perfectly. The algebraic structure of the subgroups completely determines the spectral properties of the resulting geometric objects, regardless of the fine details of the master sheet $X$ or its spectrum. This is a profound echo of the unity of mathematics, where a question in geometry finds its answer in pure algebra.

So, we've built two drums, $M_1$ and $M_2$, that sound identical. But how can we be sure their shapes are truly different? We need to find a geometric invariant—a property of the shape—that is not determined by the sound and that differs between the two.

One of the most powerful such invariants is the ​​fundamental group​​, $\pi_1(M)$, which encodes information about the loops that can be drawn on a manifold. In Sunada's construction, if we choose our master sheet $X$ to be simply connected (meaning any loop can be shrunk to a point), then the fundamental group of the quotient drum $M_i = X/H_i$ is just the subgroup $H_i$ itself!.

This provides a brilliant strategy. If we can find a pair of subgroups $H_1$ and $H_2$ that are almost conjugate but are not even isomorphic as abstract groups (for example, one is the cyclic group of order 4 and the other is the Klein four-group), then the resulting drums will have non-isomorphic fundamental groups. Since any isometry (a shape-preserving transformation) between two manifolds must induce an isomorphism of their fundamental groups, the fact that their fundamental groups are different is a smoking gun: $M_1$ and $M_2$ cannot possibly be isometric. They are verifiably different shapes.

Sunada’s method is a triumph, providing a definitive "no" to Kac's original question. But like any powerful tool, it has its limitations, and studying them is just as illuminating.

For instance, what if we try to use this method to construct two non-identical flat tori that sound the same? A torus is a highly symmetric object. However, if we demand that our "folded" manifolds $M_1$ and $M_2$ also be tori, this geometric constraint forces our group of symmetries $G$ to be abelian (commutative). And here's the catch: in an abelian group, the only way for two subgroups to be almost conjugate is for them to be equal! So, the method collapses, only ever producing two identical tori. This doesn't mean non-isometric isospectral tori don't exist—John Milnor found the first example in 16 dimensions back in 1964—it just means Sunada's blueprint can't build them.

Furthermore, Sunada's construction leaves behind characteristic "fingerprints." For example, the resulting manifolds $M_1$ and $M_2$ will always share a common finite "unfolded" version (the master sheet $X$), making their fundamental groups mathematically related in a way called commensurable. They will also be isospectral not just for functions (sound waves) but for vibrations on higher-dimensional objects called differential forms. And the method cannot produce a pair of simply connected isospectral manifolds.

The discovery of these limitations shows us that the relationship between shape and sound is even more subtle and mysterious than we might have imagined. Sunada's construction opened a door, revealing a deep and elegant connection between symmetry and vibration. But beyond that door lies a landscape of geometric possibilities that we are still only beginning to explore.

After our journey through the principles and mechanisms of the Sunada construction, you might be left with a sense of algebraic elegance. But does this beautiful piece of mathematics have anything to say about the world we can measure and observe? Does it connect to other fields of science and thought? The answer is a resounding yes. In fact, the true power and beauty of the Sunada construction are revealed when we see it in action, for it provides a definitive, and quite surprising, answer to a famous question posed by the mathematician Mark Kac in 1966: "Can one hear the shape of a drum?"

The first concrete examples of isospectral yet non-isometric shapes were discovered by Carolyn Gordon, David Webb, and Scott Wolpert. Their construction can be understood as a beautiful application of Sunada's principles, bringing the abstract group theory down to a tangible, almost craft-like process of tiling the plane.

Imagine you have a large supply of identical triangular tiles. The game is to glue these tiles together edge-to-edge to form two different planar domains. Sunada's construction provides the blueprint. The specific blueprint used in one of the most famous examples is derived from the symmetries of the Fano plane—a finite geometric structure with seven points and seven lines. The group of symmetries is the finite simple group $G = \operatorname{PSL}(3,2)$. Within this group, one can find two special subgroups, $H_p$ and $H_\ell$, which are stabilizers of a point and a line, respectively. These two subgroups are not conjugate within $G$, but they satisfy the crucial "almost conjugate" condition: they have the same number of elements from every conjugacy class of $G$.

With these two subgroups in hand, the construction proceeds. One can devise a set of rules for gluing the triangular tiles, where the adjacencies are dictated by the group structure. Following the blueprint for $H_p$ results in one shape, and following the blueprint for $H_\ell$ results in another. Because the underlying group-theoretic data for $H_p$ and $H_\ell$ satisfies Sunada's condition, the two resulting domains—though visibly different in shape—are perfectly isospectral. They are two different drums that produce the exact same "sound." This demonstrates that the spectrum of the Laplacian does not capture the full geometry of even a simple domain in the plane.

The story gets even more interesting when we leave the flat Euclidean plane and venture into the curved world of hyperbolic geometry. Hyperbolic surfaces—surfaces with constant negative curvature, like the inside of a trumpet's bell—are a particularly rich playground for spectral geometry. Here, Sunada's method can be used to construct pairs of closed hyperbolic surfaces that sound the same but are shaped differently.

The construction follows a beautiful pattern. We start with a base hyperbolic surface $X$, whose fundamental group is $\pi_1(X)$. We can then find a homomorphism from this infinite group onto a finite group $G$. This creates a "common ancestor" manifold, $\tilde{X}$, which is a finite covering of $X$ and on which the group $G$ acts by isometries. Now, we just need to find a pair of subgroups, $H_1$ and $H_2$, in $G$ that are almost conjugate but not conjugate. Taking the quotients $\tilde{X}/H_1$ and $\tilde{X}/H_2$ gives us two new hyperbolic surfaces. Sunada's theorem guarantees they are isospectral because $H_1$ and $H_2$ are almost conjugate. And because they are not conjugate, the resulting surfaces are (typically) not isometric.

This result provides a stunning counterexample to what is known as "spectral rigidity." A class of manifolds is spectrally rigid if the spectrum is enough to uniquely determine the geometry. While it was proven by Mostow that hyperbolic manifolds in dimensions three and higher are remarkably rigid (their geometry is determined by their topology), this is spectacularly false for surfaces. Sunada's construction shows that the world of hyperbolic surfaces is flexible enough to allow for distinct shapes to share the same spectral song.

So, if the spectrum doesn't determine the shape, what does it determine? And what crucial information does it hide? The answers reveal the deep connections between the spectrum and other geometric invariants.

For any two isospectral manifolds, a set of properties known as "heat invariants" must be the same. These arise from the study of how heat would diffuse on the manifold. The first two such invariants are the total volume (or area, for a surface) and the total scalar curvature. So, any two drums that sound the same must have the same area and the same overall curvature integrated over the surface. This makes intuitive sense; a bigger drum ought to have a different fundamental tone.

The connections run deeper in the hyperbolic world. The Selberg trace formula acts like a Rosetta Stone, providing a direct and beautiful identity relating the spectrum of eigenvalues (the "sound") to the spectrum of lengths of all closed geodesics (the "shape" in a different sense). This formula implies that if two hyperbolic surfaces are isospectral, they must have the exact same unmarked length spectrum. That is, if you were to make a list of the lengths of all possible closed loops on each surface, the two lists would be identical, including multiplicities.

So, where is the difference hidden? The spectrum determines the set of lengths, but not which geodesic has which length. The "marked length spectrum," which associates lengths to specific homotopy classes, can be different. This opens the door to fascinating geometric differences between isospectral manifolds.

• ​​Injectivity Radius:​​ The injectivity radius is half the length of the shortest closed geodesic on a manifold. It's a measure of how "pinched" the manifold is at its tightest spot. Using a Sunada construction, one can build two isospectral surfaces where the shortest loop on one is, for instance, twice the length of a base loop, while on the other, it is three times that length. This means two drums that sound identical can have different-sized "thinnest necks". The spectrum hears all the loop lengths in aggregate, but it can't pick out the shortest one.

• ​​Orientability:​​ Perhaps the most astonishing hidden property is orientability. Can an orientable surface (like a sphere or a torus) sound the same as a non-orientable one (like a Klein bottle)? The answer, surprisingly, is yes. The construction relies on finding an almost-conjugate pair $(H_1, H_2)$ where one subgroup, $H_1$, contains only orientation-preserving isometries, while the other, $H_2$, contains at least one orientation-reversing isometry. The resulting quotient manifold $\tilde{X}/H_1$ will be orientable, while $\tilde{X}/H_2$ will be non-orientable. Yet, because the Laplacian on functions is insensitive to orientation, Sunada's theorem ensures they are isospectral. This is a profound result: the spectrum of the most fundamental operator on a manifold cannot distinguish a "two-sided" world from a "one-sided" one.

The power of the Sunada construction extends even further, showing its deep algebraic roots and its connections to other areas of mathematics.

• ​​From Manifolds to Orbifolds:​​ The requirement that the group $G$ acts freely on the covering space $\tilde{X}$ ensures that the quotients are smooth manifolds. But what if we relax this? What if the action has fixed points? The quotients become orbifolds, which are like manifolds with conical singularities. In a beautiful display of its robustness, the proof of Sunada's theorem still goes through. The isospectrality argument, which relies on a sum over conjugacy classes, elegantly handles the contributions from the singular points, showing that two almost-conjugate subgroups will produce isospectral orbifolds.

• ​​Strong Isospectrality and Hodge Theory:​​ So far, we have only talked about the spectrum of the Laplacian on functions (0-forms). But on a Riemannian manifold, one can define a Laplacian $\Delta_p$ for differential forms of any degree $p$. This leads to a richer question: if two manifolds are isospectral on functions, are they also isospectral on 1-forms, 2-forms, and so on? This is not guaranteed in general. However, the very same condition from Sunada's theorem—the equivalence of the permutation representations $\mathbb{C}[G/H_1]$ and $\mathbb{C}[G/H_2]$—is powerful enough to guarantee that the quotients $\tilde{X}/H_1$ and $\tilde{X}/H_2$ are isospectral for all degrees $p$. This property is called ​​strong isospectrality​​. This shows that the algebraic condition is not just fooling the Laplacian on functions; it's fooling the entire family of Hodge Laplacians, linking the construction to the very heart of Hodge theory and the topology of the manifold.

In the end, the journey through the applications of Sunada's construction is a tour of the surprising and deep unity of mathematics. It shows how a purely algebraic statement about the structure of finite groups can have profound consequences for the continuous world of geometry and analysis. The fact that we "cannot hear the shape of a drum" is not a failure of our hearing, but a revelation about the nature of sound itself. The spectrum of the Laplacian is a powerful geometric invariant, but it measures a kind of "averaged" or "symmetrized" version of the geometry. It can hear the area and the collection of all possible loop lengths, but it is deaf to the specific arrangement of those loops, the manifold's "handedness," and even its thinnest neck. Sunada's method gives us the spectacles to see precisely what the spectrum sees, and what it leaves hidden in the shadows.