Varifold

In the study of shapes and forms, few concepts are as intuitively appealing as that of a "minimal surface"—the shape a soap film assumes to minimize its area. While classical geometry provides tools to describe smooth surfaces, it struggles when confronted with the complex realities of nature, such as sharp junctions where multiple films meet or the bubbling and merging of shapes. This gap highlights the need for a more robust mathematical language, one that can handle singularities, layering, and limits with grace. The theory of varifolds, a cornerstone of geometric measure theory, was developed to meet this very challenge. This article provides a comprehensive introduction to this powerful concept. Across the following chapters, you will delve into the core principles of varifolds, learning how they generalize surfaces and how their "calculus" defines minimality, before exploring the vast applications of this theory, from modeling physical phenomena to proving fundamental theorems about the existence and smoothness of surfaces.

Geometric measure theory provides the setting for studying shapes, areas, and the quest for "minimality," as exemplified by a soap film snapping into a position of least area. To formalize these ideas, particularly for shapes with singularities, a central concept is required: the ​​varifold​​. While classical geometry uses points, lines, and surfaces, these concepts can be insufficient. When an existing mathematical language fails to describe natural phenomena, a new one must be invented. The varifold is such an invention.

Imagine trying to describe a cloud. You could try to list the position of every single water droplet, but that's a hopeless task. A much better way is to talk about the density of the cloud at different points in space. A varifold begins with a similar idea. Instead of thinking of a surface as an infinitely thin collection of points, we think of it as a distribution of mass or, more formally, a ​​Radon measure​​.

But a surface isn't just a cloud of dust; it has structure, it has direction at every point. A smooth piece of paper lying on a table has a clear two-dimensional plane associated with it at every point: the tangent plane. What if the surface is crumpled, or has singularities, like where three or more soap films meet? The idea of a single tangent plane might break down.

This is where the genius of the varifold concept, pioneered by Frederick J. Almgren, Jr., truly shines. A varifold not only keeps track of the location of the surface, but it also records the distribution of all possible tangent planes at each location. Instead of a measure just on our familiar space $\mathbb{R}^n$, a $k$-dimensional varifold is a measure on a much richer space: the product of positions and planes, $\mathbb{R}^n \times G(n,k)$. Here, $G(n,k)$, the ​​Grassmannian​​, is simply the collection of all possible $k$-dimensional planes that can exist in our $n$-dimensional space.

Think of it like this: imagine a vast field of tall prairie grass. A simple density measure tells you how much grass there is in any given acre. A varifold does more. At every spot, it also tells you which way the blades of grass are leaning—not just one direction, but a whole distribution of directions if the wind has been swirling. A varifold lets us say, "At this point $x$, there's this much 'surface-ness', and it's oriented mostly like this plane, but a little bit like that plane too."

Of course, we can always choose to ignore the orientation information. If we "sum up" or project all the tangent plane information at each point, we are left with a simple measure of how much surface area there is at each location. This is called the ​​weight measure​​ of the varifold, often denoted $|V|$. It’s our cloud density, the mass of the ghost in the machine.

When we look at two soap bubbles merging, their common wall is a single film. If a third bubble joins, they meet along a line, and we see three films coming together. It seems natural that surfaces in the real world can be layered. A varifold captures this beautifully through a ​​multiplicity function​​, $\theta$. For the most useful type of varifolds, called ​​integral varifolds​​, this multiplicity must be an integer: $1, 2, 3, \ldots$. It makes no physical sense to have one-and-a-half layers of a soap film. This integer-valuedness is deeply connected to the notion of ​​density​​, $\Theta^m(|V|, x)$. If you zoom in on a point $x$ on a varifold, the density tells you how many sheets of surface are passing through that point. For an integral varifold, this density will be an integer almost everywhere.

Now for a subtle but profound point: are varifolds oriented? Does a varifold know the difference between the "front" and "back" of a surface? The answer is a resounding ​​no​​. A varifold is inherently unoriented. The planes it records in the Grassmannian $G(n,k)$ are like infinite sheets of paper; they don't have a "top" or "bottom" side.

To see why this is so important, let’s contrast it with a different mathematical object called an ​​integral current​​, which is oriented. Imagine you have a current $T_M$ representing a surface $M$ with a chosen "up" direction. Now consider $-T_M$, the same surface but with the "down" direction chosen. If you add them, $T_M + (-T_M) = 0$. In the world of currents, they annihilate each other, like matter and antimatter!.

But what happens in the world of varifolds? The varifold $V_M$ associated with $M$ doesn't care about the "up" or "down" direction. So, the varifold for the "up" surface is the same as for the "down" surface. If you add them, you get $V_M + V_M = 2V_M$. You get a surface with multiplicity 2!.

This might seem like a technicality, but it's the very soul of the theory. When we take limits of surfaces—for instance, when we try to model a sequence of bubbling shapes that are converging—we don't want their area to just vanish into thin air because of orientation cancellation. Varifold convergence ensures that mass is preserved; surfaces can merge and double up, but they can't just disappear. This makes varifolds the perfect, robust tool for studying area-minimizing problems where shapes can merge and form singularities.

So we have our object. How do we find the one with the least area? In calculus, to find the minimum of a function, you take its derivative and set it to zero. We need to do the same for area. This is where the ​​first variation​​, $\delta V$, comes in.

Imagine you have your varifold $V$ and you "wiggle" it a tiny bit according to some smooth vector field $X$. The first variation $\delta V(X)$ tells you the initial rate of change of the area. It’s the derivative of the area functional. If a varifold is to be area-minimizing, it must at least be at a critical point, where the first-order change in area is zero for any small wiggle. Such a varifold is called ​​stationary​​. This is our generalized definition of a minimal surface.

What physical property of a surface tells it how to move to decrease its area? Curvature! Think of a stretched rubber band. Where it's curved, it has tension, and it tries to straighten out. For a surface, the analogous quantity is the ​​generalized mean curvature vector​​, $H$. It’s a vector at each point of the surface that measures how "bent" the surface is and points in the direction it would move to decrease its area most efficiently.

The first variation and the mean curvature are linked by a beautiful, fundamental formula: $\delta V(X) = - \int_{U} \langle H(x), X(x) \rangle\, d|V|(x)$ This equation is simply saying that the change in area ($\delta V$) when you deform the surface by a vector field $X$ is given by integrating how much the surface's "desire to move" (the vector $H$) aligns with the imposed wiggle $X$. The minus sign is a convention, ensuring that if you move the surface in the direction of its curvature vector ($X=H$), the area decreases.

From this, the conclusion is immediate: a varifold is stationary ($\delta V(X)=0$ for all $X$) if and only if its generalized mean curvature is zero ($H=0$) almost everywhere. Our search for minimal surfaces has become a search for surfaces with zero mean curvature!

With this machinery in place, we can now uncover some of the deep, almost magical, laws that govern these minimal surfaces.

First is the ​​Monotonicity Formula​​. It’s a breathtakingly simple yet powerful statement about stationary varifolds. It says that if you take a point $x_0$ on a stationary varifold and compute the ratio of the surface area inside a ball of radius $r$ to the area of a flat $m$-dimensional disk of radius $r$, this ratio, $\frac{|V|(B_r(x_0))}{\omega_m r^m}$, can only increase as your radius $r$ increases.

Think about that. It puts a rigid constraint on how area can be distributed. It forbids a minimal surface from being mostly empty near a point and then suddenly becoming very dense just a bit further out. This monotonicity has a profound consequence: when you zoom in on any point of a stationary varifold (a process called "blowing up"), the sequence of rescaled surfaces will always converge to a limiting shape. And this shape must be a ​​cone​​—a surface made of rays emanating from the origin. At a smooth point, this tangent cone is just a flat plane. At a singularity, like where two planes cross, the tangent cone is exactly that—two planes crossing. Monotonicity guarantees that these microscopic structures exist and are well-behaved.

This leads us to the grand finale: ​​Allard’s Regularity Theorem​​. We started this journey by allowing our surfaces to be very wild objects—just measures, really. The big question is, when are these generalized minimal surfaces ($H=0$) the beautiful, smooth soap films we see in the real world?

Allard's theorem provides a spectacular answer. It's a powerful $\varepsilon$-regularity result, a statement of the principle that "almost-perfection implies true-perfection." The theorem says, roughly, that if you look at a rectifiable varifold in a small ball, and you find that:

1. It is ​​almost flat​​ (its "tilt-excess," a measure of how much its tangent planes wobble, is very small).
2. It is ​​almost minimal​​ (the integral of its mean curvature is very small).
3. Its ​​density is close to one​​.

Then, the varifold is not just "almost" a smooth surface. It is a smooth surface inside a smaller ball! It is the graph of a $C^{1,\alpha}$ function, meaning it's differentiable and its derivative is even a bit better than just continuous.

This is the payoff. The abstract, powerful machinery of varifolds leads us right back to the concrete, beautiful world of smooth surfaces. It tells us that the singular, non-smooth parts of a minimal surface are the exception, not the rule. And it gives us a precise, quantitative way to distinguish them. Along the way, we've developed a richer understanding of what it even means to be a surface, preparing us to tackle problems in physics, materials science, and pure mathematics that were once far out of reach. We built a ghost in the machine, and in doing so, we learned how the machine truly works.

Having established the fundamental ideas of varifolds, it is natural to ask what this abstract machinery is used for. The theory of varifolds is not merely an exercise in generalization for its own sake. Rather, it is a powerful lens that brings a vast landscape of problems, from pure mathematics to the physics of materials, into sharp, unified focus. It allows mathematicians and scientists to ask—and answer—questions that were out of reach of the classical theory of surfaces. This section explores some of these applications to reveal how the language of varifolds illuminates a wide range of scientific problems.

The first duty of any new, more general theory is to successfully recapture the old, established truths. Varifolds do this beautifully. A classical, smooth minimal surface—a surface with zero mean curvature at every point, the aristocrat of differential geometry—is perfectly described in this new language as a ​​stationary varifold​​. The condition of being "stationary," meaning its area doesn't change to first order under any small deformation, is precisely the weak, integral formulation of the classical condition that the mean curvature vector $H$ is zero everywhere.

But varifolds immediately offer a bonus. The classical theory often relies on having a well-defined "inside" and "outside," a property known as orientability. A sphere is orientable; a Möbius strip is not. The varifold framework, by defining a surface as a measure on the space of unoriented tangent planes, doesn't care about orientability. A non-orientable minimal surface is just as welcome and just as easily described as a stationary varifold as its orientable cousin. This is the first hint of the theory's power: it strips away unnecessary assumptions to get at the heart of what it means to be "minimal."

Furthermore, this abstract notion connects directly to the concrete world of partial differential equations (PDEs). Consider a surface that is the graph of a function, $z = u(x,y)$. The condition that this surface be minimal translates into a fiendishly complex nonlinear PDE for the function $u$. The varifold framework provides a magnificent insight: the stationarity of the varifold corresponding to the graph is precisely equivalent to the function $u$ being a "weak solution" to this minimal surface equation. This allows mathematicians to use powerful tools from modern analysis to study surfaces that might not be perfectly smooth but are solutions in a more general, and often more physical, sense.

Perhaps the most compelling reason for the invention of varifolds is their ability to describe the world as it actually is, "warts and all." If you dip a wire frame into a soap solution, the resulting soap film is a beautiful embodiment of a minimal surface. But what happens when several films meet? You will often see three soap films meeting along a common line, forming angles of $120^\circ$. This Y-shaped junction is nature's way of minimizing area, but at the points along that line, the film is not a smooth surface. Classical differential geometry struggles here.

For varifold theory, this is no problem at all. A varifold can perfectly represent this union of three half-planes. The condition of stationarity, when analyzed at the junction, becomes a simple and elegant balancing act: the sum of the tangent vectors to the three sheets must be zero. The $120^\circ$ angle is exactly what is required for this balance,.

This leads to the beautiful concept of a ​​tangent cone​​. If you zoom in infinitely close to a point on a varifold, what do you see? For a smooth point, you just see its tangent plane—a flat, featureless expanse. But if you zoom in on a singular point, like the Y-junction, you see a cone: the same Y-shape stretching out to infinity. A wonderful theorem states that any tangent cone to a stationary varifold must itself be a stationary cone. In $\mathbb{R}^3$, this means its "link"—the curve it traces on the surface of a unit sphere—must be a network of geodesics (great circles) that are perfectly balanced at their junctions. So, the study of singularities in our space is transformed into a study of geodesic networks on a sphere!

This framework opened the door to a menagerie of new geometric objects, like the famous ​​Simons cone​​ in $\mathbb{R}^8$, a singular surface that, unlike the Y-junction, is genuinely area-minimizing. The existence of such objects showed that singularities are not just a nuisance, but an essential and deep feature of the theory of area minimization.

If the ability to model soap films seems a bit quaint, let's turn to a profound connection with modern physics and materials science. Many physical systems exhibit phase transitions, where a substance separates into distinct regions, like oil and water or different magnetic domains in a metal.

The ​​Allen-Cahn equation​​ is a celebrated PDE that models such phenomena. It describes a field $u(x)$ (which could represent, say, the concentration of one chemical) whose energy depends on two competing effects: a potential energy term $\frac{1}{\varepsilon}W(u)$ that prefers the system to be in one of two "pure" states (e.g., $u=-1$ for water, $u=1$ for oil), and a gradient energy term $\frac{\varepsilon}{2}|\nabla u|^2$ that penalizes sharp boundaries between these states. The parameter $\varepsilon$ controls the "fuzziness" of the interface.

What happens as you make the interface less and less fuzzy by letting $\varepsilon \to 0$? A miracle of modern analysis, known as $\Gamma$-convergence, takes place. The "diffuse" transition region, governed by the PDE, sharpens into an infinitesimally thin boundary. And the total energy of the system converges to the area of this boundary, multiplied by a constant that represents the surface tension. The crucial discovery is that this limiting boundary is not just any surface; it is a ​​stationary integral varifold​​.

This is a breathtaking result. The principle of area minimization, which we first saw in soap films, re-emerges as the governing principle for the boundaries between phases in a physical system. Nature, in seeking a low-energy state for the Allen-Cahn equation, organizes the interfaces into minimal surfaces. This shows that varifolds are not just a mathematical curiosity; they are the natural language for describing the energetic balance in a vast range of physical phenomena. Moreover, the integer multiplicities that varifolds allow correspond to the physical possibility of several transition layers collapsing onto the same location in the limit.

Beyond describing and modeling, the varifold framework is a creative force. It gives mathematicians the tools to prove the very existence of minimal surfaces where it was previously impossible. The celebrated ​​Almgren-Pitts min-max theory​​ is a prime example. In a spirit similar to finding a mountain pass, this theory considers continuous families of surfaces (or "cycles") that "sweep out" the ambient space. It then seeks the sweepout that minimizes the maximum area achieved along the way. The technical difficulty is immense, but the final result is pure gold: the min-max value of the area is realized by a stationary integral varifold. Varifolds provide the crucial analytic compactness needed to ensure that this intricate limiting process yields a concrete geometric object.

And what of the nature of these objects? We have seen that they can be singular. But when are they smooth? Here, geometric measure theory provides one of its most stunning results. For an area-minimizing hypersurface in an ambient space of dimension $n$, the singular set has a dimension of at most $n-8$. This means that in our familiar three-dimensional world ($n=3$), or even in a seven-dimensional one ($n=7$), the singular set has dimension $7-8 = -1$, which means it must be empty! Any area-minimizing "soap bubble" in a space of dimension 7 or less is guaranteed to be perfectly smooth. Singularities are a high-dimensional phenomenon. This is a fact of profound depth and strangeness.

Finally, what does a minimal surface look like from very far away? Just as we can zoom in to find a tangent cone, we can zoom out, or "blow down," to find its ​​tangent cone at infinity​​. This reveals the surface's asymptotic shape. This technique leads to another classic result with a modern twist: the ​​Bernstein Theorem​​. This theorem states that the only smooth minimal surface that can be written as a graph $z=u(x_1, \ldots, x_n)$ over the entirety of $\mathbb{R}^n$ is a flat plane, provided $n \le 7$. The modern proof involves showing that its tangent cone at infinity must be a plane, which then forces the surface itself to be a plane. It's a beautiful instance of using the modern machinery of varifolds to understand a clean, classical problem about the global nature of solutions.

From the junctions of soap films to the boundaries of phases, from proving existence to classifying shapes at infinity, varifolds provide a single, powerful, and elegant language. They have given us a canvas large enough to paint a unified picture of geometry, analysis, and physics. And at the heart of it all is a set of sophisticated analytical tools, such as the coarea formula, which essentially provides a "calculus for varifolds," allowing us to slice, integrate, and analyze these generalized surfaces with rigor and precision. This is the legacy of the theory: it took the simple, intuitive idea of minimizing area and gave us a universe.