Quasiconvexity: The Mathematics of Stability in Materials and Optimization

The search for an optimal state—the configuration of lowest energy or least cost—is a fundamental pursuit across science and engineering. In this quest, mathematical convexity has long been the gold standard, guaranteeing that any locally best solution is also the globally best one. However, when we try to apply this simple idea to the complex behavior of real-world systems, such as elastic materials, it spectacularly fails. This gap between mathematical simplicity and physical reality necessitates a more subtle, powerful, and physically motivated generalization: quasiconvexity.

This article confronts a central paradox in continuum mechanics: the energy of a realistic material cannot be described by a convex function without violating the fundamental principle of frame indifference. This puzzle paves the way for a richer theory within the calculus of variations. We will uncover why quasiconvexity emerges as the "correct" condition for stability, fitting into a nuanced hierarchy of mathematical properties that govern material behavior.

Our exploration is divided into two main parts. In the "Principles and Mechanisms" chapter, we will unpack the definitions of quasiconvexity, polyconvexity, and rank-one convexity, dissecting the profound physical implications of the gaps between them. Subsequently, the "Applications and Interdisciplinary Connections" chapter will demonstrate how these seemingly abstract concepts are indispensable tools for predicting material behavior, ensuring numerical simulations are reliable, and solving complex problems in fields as diverse as finance and artificial intelligence.

In our journey to understand the world, we often seek to find the "best" way of doing things—the path of least resistance, the shape of lowest energy, the strategy of maximum return. In mathematics, this search is the province of optimization. At the heart of many optimization problems lies a beautifully simple idea: convexity. But as we shall see, the simplicity of a concept in one dimension can explode into a rich and complex hierarchy of ideas in higher dimensions, leading us to profound insights into the behavior of real-world materials.

Let's start with a familiar friend: a ​​convex function​​. In one dimension, you can think of it as a function whose graph is shaped like a bowl. For any two points on the graph, the straight line segment connecting them always lies on or above the graph. Formally, for a function $f$, any two points $x$ and $y$ in its domain, and any number $t$ between $0$ and $1$, the following holds:

This property is wonderfully powerful because it guarantees that any local minimum is also the global minimum—if you've found the bottom of a bowl, you've found the bottom.

But what if we relax this condition a little? What if we only require that the function's value along the line segment between $x$ and $y$ never exceeds the higher of the two endpoints, $f(x)$ and $f(y)$? This gives us the definition of a ​​quasiconvex function​​:

This seemingly small change has a wonderfully intuitive geometric meaning. It means that for any value $\alpha$, the set of all points where the function is less than or equal to $\alpha$ (what mathematicians call a ​​sublevel set​​) is a single, unbroken piece—a convex set. For a function of one variable, this just means an interval.

Let's look at an example. Consider the floor function $f(x) = \lfloor x \rfloor$, which rounds a number down to the nearest integer. Its graph is a series of steps, which is clearly not a simple bowl shape. Indeed, it fails the convexity test spectacularly. However, if you ask "for which values of $x$ is $\lfloor x \rfloor \le \alpha$?", the answer is always a simple interval like $(-\infty, k+1)$ for some integer $k$. Because all its sublevel sets are convex, the floor function is quasiconvex! Similarly, any function that is always increasing or always decreasing, like $f(x) = x^3$, is also quasiconvex because its sublevel sets are simple rays like $(-\infty, c]$. Quasiconvexity thus captures a much broader and more interesting class of "well-behaved" functions than convexity alone.

This is all well and good for functions of a single number, but in physics and engineering, we often deal with functions of more complex objects, like matrices. Imagine the energy stored in a block of rubber as you stretch and twist it. The state of deformation at each point is described by a matrix, the ​​deformation gradient​​ $F$. The energy of the material is a function of this matrix, $W(F)$.

Our first instinct might be to demand that this energy function $W(F)$ be convex. This seems like a natural requirement for material stability. If the energy function were not bowl-shaped, the material might spontaneously split into different states to lower its energy. But this simple and appealing idea leads directly to a physical paradox.

A fundamental property of any realistic material is ​​frame indifference​​ (also called objectivity): if you take a deformed body and simply rotate it in space, its stored energy shouldn't change. A rotated apple is still an apple, with the same internal stress and strain energy. Mathematically, this means $W(F) = W(QF)$ for any rotation matrix $Q$ (a member of the special orthogonal group, $\mathrm{SO}(3)$). We also know from experience that a material just sitting there, undeformed and unstressed, should have zero energy. This stress-free state can be represented by any rotation matrix $Q$, so we require $W(Q) = 0$ for all $Q \in \mathrm{SO}(3)$.

Now comes the twist. If we assume $W$ is convex, then by the definition of convexity, its value at any "average" of matrices must be less than or equal to the average of its values. Since $W(Q)=0$ for all rotations, it must be that $W$ is zero for any matrix in the convex hull of $\mathrm{SO}(3)$, the set of all possible weighted averages of rotation matrices. But what does this set contain? Consider the identity matrix $I$ (no rotation) and a 180-degree rotation $Q$ about an axis. Both have zero energy. Their average, the matrix $F_0 = \frac{1}{2}I + \frac{1}{2}Q$, would also have to have zero energy. But if you compute this matrix, you'll find its determinant is zero! A zero determinant means you've compressed a volume down to nothing. This should cost an enormous, if not infinite, amount of energy. Yet, the assumption of convexity demands it costs zero. This is a complete contradiction.

The conclusion is inescapable and profound: the energy functions of real materials cannot be convex in the simple sense. We need a more subtle and physically motivated notion of "convexity".

This physical paradox forced mathematicians to develop a whole hierarchy of weaker, more nuanced convexity conditions, tailored for functions of matrices. You can picture a pyramid of conditions, with the most restrictive at the top and the most general at the bottom:

​​Convexity $\implies$ Polyconvexity $\implies$ Quasiconvexity $\implies$ Rank-one Convexity​​

Let's meet the cast of characters:

• ​​Rank-one Convexity​​: This is the weakest and most fundamental condition. It requires the energy function to be convex only along certain directions in the space of matrices: those that connect two matrices differing by a ​​rank-one matrix​​. A rank-one matrix is the mathematical representation of a simple deformation, like a shear along a single plane or a stretch along a single axis. So, rank-one convexity is a bare-minimum stability requirement: the material must be energetically stable against these elementary deformations.

• ​​Quasiconvexity​​: This is the true star of our story, the "correct" notion of convexity for variational problems in mechanics. It has a less direct but far more profound definition, introduced by Charles Morrey. A function $W$ is quasiconvex if the energy of a body in a uniform state of deformation is always less than or equal to the average energy of a non-uniform state that, on average, looks the same. Imagine a block of material uniformly deformed by a gradient $F$. Its energy density is $W(F)$. Now, let's superimpose some microscopic wiggles on top of this deformation, represented by $\nabla\varphi(x)$, such that these wiggles average out to zero over the block. Quasiconvexity is the statement of stability: these wiggles cannot lower the average energy. Formally:

This is a sort of Jensen's inequality for gradients, and as we will see, it is the key to proving that a problem has a solution.

• ​​Polyconvexity​​: Introduced by John Ball, this is a "practical" condition that is stronger than quasiconvexity but much easier to verify. A function is polyconvex if its energy $W(F)$ can be written as a regular convex function (our familiar bowl shape) of a larger set of variables: the matrix $F$ itself, its ​​cofactor matrix​​ $\operatorname{cof} F$, and its ​​determinant​​ $\det F$. This has a beautiful physical interpretation: the energy is a "nice" function of how lengths (described by $F$), areas (described by $\operatorname{cof} F$), and volumes (described by $\det F$) are deformed.

The hierarchy is strict—each condition implies the one below it. But the reverse is not true, and the gaps between these definitions are where the physics gets fascinating and the mathematics gets deep.

• ​​Quasiconvexity does not imply Convexity​​: We saw this from the start. A simple example for matrix functions is $W(F) = (\det F)^2$. Since the function $g(z)=z^2$ is convex, $W(F)$ is polyconvex, which in turn means it is also quasiconvex. However, one can show it is not a convex function of the matrix $F$ itself. This allows the material's energy to depend on volume change in a non-trivial way, which is essential for realistic models.

• ​​Rank-one Convexity does not imply Quasiconvexity​​: This was a huge open problem, known as Morrey's Conjecture, for over fifty years. Are these two conditions the same? Is stability against simple shear deformations enough to guarantee stability against all possible micro-wiggles? The answer, shockingly, is no. In 1992, Vladimír Šverák constructed a mind-bending counterexample—a function that is rank-one convex but not quasiconvex. He discovered, in essence, a mathematical "material" that is stable if you apply any single, simple shear, but which becomes unstable and can lower its energy by forming a complex, coordinated pattern of oscillations. This was a monumental discovery, revealing a deep subtlety in the nature of material stability.

So why do we go through all this trouble to define this menagerie of convexities? The ultimate goal is to solve real-world problems, like finding the equilibrium shape of a bridge under load or the pattern formed by two liquids that don't mix. In the language of physics, this means finding the state that minimizes a total energy functional, often of the form $I(u) = \int_{\Omega} W(\nabla u) \,dx$.

To guarantee that such a minimum even exists, we often use what is called the ​​Direct Method in the Calculus of Variations​​. A crucial ingredient for this method is a property called ​​(Sequential) Weak Lower Semicontinuity (WLSC)​​. It's a technical name for a simple idea: if you have a sequence of states whose energies approach some minimum value, the energy of the final "limit" state cannot suddenly jump up. The energy can drop in the limit (think of a soap bubble bursting), but it cannot be higher than the limit of the energies.

And here is the punchline, the grand unification of all these ideas: for the energy functionals that describe so much of our physical world, ​​weak lower semicontinuity is equivalent to the quasiconvexity of the energy density $W$​​! Quasiconvexity is not just some abstract mathematical definition; it is exactly the right physical property that ensures our energy minimization problems are well-behaved and have stable solutions.

But what if the energy function $W$ for our material is not quasiconvex? This happens all the time, particularly in materials that can undergo phase transformations, like a liquid freezing into a solid, or a metal changing its crystal structure. These materials are often described by "multi-well" energy functions that are not quasiconvex. In this case, the minimization problem often has no classical solution! The material can always lower its energy by forming an ever-finer mixture of the different phases, creating what are called ​​microstructures​​.

This is not a failure of the theory; it's a triumph! The mathematics predicts the very real phenomenon of phase co-existence and the formation of intricate patterns in materials. To analyze this, we perform a procedure called ​​relaxation​​. We replace the badly-behaved, non-quasiconvex energy $W$ with its ​​quasiconvex envelope​​, $QW$. This new function, $QW$, is the largest possible quasiconvex function that you can fit underneath the graph of the original energy function $W$. The minimization problem for this new "relaxed" energy does have a solution, and that solution beautifully describes the average, macroscopic behavior of the material, automatically accounting for the energetic contributions of its microscopic wiggles. It's a way of finding the effective energy of the microstructure without having to model every single crystal, domain, or phase boundary, revealing a deep and elegant unity between abstract mathematics and the tangible structure of the world around us.

Now that we have some acquaintance with the mathematical machinery of quasiconvexity, you might be tempted to ask, "So what?" Is this just a clever game for mathematicians, a new kind of function to put in a zoo of curiosities? The wonderful answer is no! What we have stumbled upon is not a mere curiosity, but a deep principle of nature. It turns out that quasiconvexity, in its subtle way, governs the stability of the world around us—from the shape of a stretched piece of rubber to the very possibility of predicting material behavior with a computer. It is a key that unlocks a surprising unity across seemingly disparate fields of science and engineering. Let us go on a journey to see where this idea takes us.

Let's start with something you can hold in your hand: a block of rubber. If you push on it, or twist it, what shape does it take? We know from physics that it will settle into a state of equilibrium, a configuration that minimizes a quantity we call the total stored energy. To find this shape, we must solve a problem in the calculus of variations: find the deformation $y(x)$ that minimizes the integral of the energy density, $\mathcal{I}(y) = \int_{\Omega} W(\nabla y(x))\,dx$.

Now, you might think, "That's easy! We know how to minimize things. If the function $W$ is convex, everything is fine." But here, nature throws us a wonderful curveball. A simple, globally convex energy function $W$ is physically unrealistic for a solid material. Why? Because of a fundamental symmetry called frame-indifference. If you deform a piece of rubber and then simply rotate the whole thing in space, its stored energy shouldn't change. But this requirement is fundamentally at odds with strict convexity. If you try to build a model with a strictly convex energy, it won't be objective! So, what are we to do? Our most reliable tool for guaranteeing a unique, stable solution—convexity—is forbidden by basic physics.

This is where quasiconvexity makes its grand entrance. It is the precise, miraculous weakening of convexity that is exactly what's needed. If the energy density $W$ is quasiconvex, the direct method of the calculus of variations works, and we are guaranteed that a stable, energy-minimizing shape exists. We might lose uniqueness—several different shapes might have the same minimum energy, which is physically very interesting—but we don't lose existence. The problem has a solution.

Even better, there's a practical, verifiable condition called ​​polyconvexity​​ that ensures quasiconvexity. A function is polyconvex if it is a convex function of not just the deformation gradient $F$, but also of its higher-order relatives: its cofactor $\operatorname{cof}F$ and its determinant $\det F$. This isn't just a mathematical trick; it has a beautiful physical interpretation. The matrix $F$ tells you how tiny lines are stretched, $\operatorname{cof}F$ tells you how tiny areas are transformed, and $\det F$ tells you how tiny volumes change. By demanding convexity in this expanded space of geometric transformations, we build physically realistic models that are mathematically well-behaved.

What if we ignore this lesson? What if we try to work with a material whose energy function is not quasiconvex? Then something remarkable and troubling happens. The material discovers it can lower its energy by forming an infinitely fine mixture of different deformation states. Think of a crystal that can exist in two different lattice orientations. If the energy landscape has two wells (a hallmark of non-convexity), the material can achieve a lower average energy by creating a finely layered structure, alternating between the two states. This is called a ​​microstructure​​, and the effective energy of this composite is described by the quasiconvex envelope of the original energy function.

This phenomenon shows up dramatically when we turn to computers. Suppose you try to run a finite element simulation—the workhorse of modern engineering—to predict the shape of a material with a non-quasiconvex energy. The computer gets hopelessly confused. The solution it finds will be a chaotic, jagged mess of oscillations, and the pattern of this mess will depend entirely on the size of the mesh you chose for your calculation. Refine the mesh, and the oscillations just get finer, never converging to a smooth, physical shape. This is a disaster! It means your prediction is meaningless. Quasiconvexity is the property that tames this chaos. It acts as a selection principle, ruling out these energy-lowering oscillations and ensuring that our numerical models can converge to something real and predictable.

For a long time, scientists built these energy functions by hand, guided by theory and a few experiments. But in the age of big data and machine learning, we want to do better. We want to learn a material's behavior directly from vast amounts of experimental data. This leads to a fascinating challenge: how do you get a neural network, or some other flexible model, to learn a function from data while respecting the fundamental physical law of quasiconvexity?

Just fitting the data isn't enough; we might end up with a model that predicts numerical chaos. The solution is a beautiful fusion of classical mechanics and modern AI. We can design special "Input Convex Neural Networks" (ICNNs) that are, by their very architecture, guaranteed to be convex in their inputs. By feeding such a network not just the deformation $F$, but the full set of minors $(F, \operatorname{cof} F, \det F)$, we can train a model that is polyconvex by construction. The network learns the complex energy landscape from data, while we, the designers, have hard-coded the mathematical principle that ensures physical and numerical stability. It's a testament to how enduring and relevant these "abstract" mathematical concepts are.

You might be thinking this is all about rubber and crystals. But the reach of quasiconvexity is far broader. The theme of wrestling with non-convexity to find the best possible solution appears everywhere.

Consider ​​signal processing​​. An engineer wants to design an Infinite Impulse Response (IIR) filter—a fundamental building block of everything from audio equalizers to wireless communication systems. The goal is to find the filter's coefficients to best match a desired frequency response. This optimization problem is notoriously non-convex and hard to solve. However, it turns out that by cleverly reformulating the problem and sometimes accepting compromises (like fixing some parameters), one can arrive at a problem that is ​​quasiconvex​​. And for a quasiconvex problem, while we don't have a single magic formula, we have efficient algorithms (like bisection) that are guaranteed to find the global optimum. Here, quasiconvexity appears not as a law of nature, but as a hard-won structural property that makes an intractable engineering design problem solvable.

Or let's visit ​​finance​​. Imagine an "exotic" financial derivative, say a "capped call," whose payoff is zero up to a strike price $K$, rises linearly, and then flattens out at a maximum cap $C$. If you plot this payoff, you'll see it's not a convex function. But if you check its sublevel sets, you'll find they are all simple intervals. It is quasiconvex! This isn't just a label; it has consequences. If you want to find the cheapest portfolio of basic assets (like stock and cash) that will always be worth at least as much as this derivative's payoff, you have to solve an optimization problem. Because the payoff function is what it is, this "super-hedging" problem turns into a linear program—a type of convex optimization problem we know how to solve with extreme efficiency. The geometry of the payoff function dictates the complexity of hedging it.

For those who wish to venture deeper, the theory of quasiconvexity holds even more beautiful results. We've said it guarantees existence, but what about smoothness? Under stronger assumptions on the energy function—conditions on its growth and second derivatives—we get a stunning result known as ​​partial regularity​​. Any energy-minimizing configuration will be beautifully smooth (class $C^{1,\alpha}$, to be precise) almost everywhere. The set of points where it might be singular or rough is tiny; its Hausdorff dimension is small, and if the energy grows fast enough (for an exponent $p>n$), the singular set is empty!.

And what happens conceptually when quasiconvexity fails? The theory of ​​Young measures​​ gives us a powerful language. A minimizing sequence whose gradients are oscillating wildly doesn't just disappear. It converges in a weak sense to a new object: a spatially-varying probability measure, the Young measure. This measure tells us, at each point in the material, the probability of finding the gradient in a particular state. The system "relaxes" its energy by forming this probabilistic mixture. From this advanced perspective, quasiconvexity has an elegant alternative definition: a function $f$ is quasiconvex if its average over any gradient Young measure is always greater than or equal to the function evaluated at the average gradient. In other words, no probabilistic mixing can ever cheat the energy budget. This connects variational calculus to probability theory in a profound way.

So, we see that quasiconvexity is far from being a mere abstraction. It is a unifying thread, weaving through the fabric of the physical and engineered world. It is the subtle condition that distinguishes stable reality from unphysical chaos in materials. It is the dividing line between tractable and intractable design problems in engineering. It is even reflected in the structure of financial markets. By understanding this one concept, we gain a deeper appreciation for the mathematical principles that ensure order and predictability in a complex, non-convex universe.