Fully Nonlinear Partial Differential Equations

Many physical systems, from a simple lever to basic electrical circuits, can be described by linear equations where effects are directly proportional to their causes. However, the most profound and complex phenomena in nature—the crashing of a wave, the path of light through a complex medium, or the strategic decisions made in economics—are inherently nonlinear. This is the domain of nonlinear partial differential equations (PDEs), mathematical tools that capture the intricate feedback loops and disproportionate responses that define our world. A particularly challenging and powerful class within this domain is that of fully nonlinear PDEs.

This article tackles a central crisis in their study: the frequent breakdown of traditional calculus when solutions develop sharp corners, shocks, or other non-smooth features, rendering the very notion of a derivative meaningless. We will explore how mathematicians overcame this obstacle by redefining what it means to "solve" an equation. The discussion will proceed in two main parts. First, under "Principles and Mechanisms," we will delve into the nature of fully nonlinear PDEs, contrast them with simpler forms, and introduce the elegant and powerful theory of viscosity solutions. Second, in "Applications and Interdisciplinary Connections," we will witness these abstract concepts in action, discovering how they provide the essential language for describing problems in geometric optics, optimal control, and even the fundamental structure of spacetime in modern physics.

Imagine you are an engineer examining a machine. A simple machine, like a lever, is ​​linear​​: double the force you apply, and you double the output. The world of linear differential equations is much like this—predictable, proportional, and beautifully solvable with a toolkit of established methods. But nature is rarely so simple. Many of its most fascinating phenomena, from the flow of water to the propagation of light, are governed by machines of a far more intricate design: ​​nonlinear partial differential equations (PDEs)​​.

In this chapter, we’ll open the hood and explore the principles and mechanisms that make these equations tick. We will see that as the complexity grows, our old tools break, forcing us to invent clever new ways of thinking—a journey that leads from apparent chaos to a surprising and profound inner order.

Not all nonlinear equations are created equal. There's a hierarchy of complexity, and understanding it is the first step. Let's think of our PDE as a machine whose behavior depends on some function $u$ and its rates of change (its derivatives).

An equation is ​​quasi-linear​​ if it’s linear in its highest-order derivatives. This is like a machine whose most powerful gear operates smoothly, but its performance is constantly adjusted based on the machine’s current state. Consider the diffusion of a chemical through a special material where the diffusion rate depends on the chemical's concentration. This is modeled by an equation like $u_t = D(u) u_{xx}$. The highest derivative, $u_{xx}$, appears linearly. But its coefficient, the diffusion "constant" $D(u)$, isn't constant at all—it changes with the concentration $u$. Similarly, the inviscid Burgers' equation, $u_t + u u_x = 0$, a simple model for traffic flow or shockwaves in gas, is quasi-linear because the highest derivatives ($u_t$ and $u_x$) appear linearly, but the coefficient of $u_x$ is $u$ itself.

But what if the highest gear itself is nonlinear? What if the machine’s most fundamental operations are intrinsically complex? Then we enter the realm of ​​fully nonlinear PDEs​​. A classic example comes from optics: the ​​Eikonal equation​​, $(\nabla u)^2 = n^2(x, y)$. This equation describes how light propagates. The function $u(x,y)$ represents the arrival time of a light wave, and its gradient, $\nabla u$, is related to the direction of the light ray. The equation says that the magnitude of this gradient is determined by the material's refractive index $n(x,y)$. Notice the term $(\nabla u)^2$; the highest derivative (the gradient, a first derivative) is squared. The machine's response is no longer proportional, even at the highest level. This nonlinearity is the signature of a system that solves an optimization problem—in this case, light finding the path of least time.

This distinction between quasi-linear and fully nonlinear isn't just academic hair-splitting. It has profound consequences. Many fully nonlinear equations, and even some quasi-linear ones, refuse to yield the kind of elegantly smooth, "classical" solutions we are used to. Their solutions can develop sharp corners, jumps, or shocks. Think of a wave crashing on a beach, or a sonic boom from a supersonic jet. These are phenomena where physical quantities change so abruptly that their derivatives, in the classical sense, cease to exist.

Our traditional calculus, built on the idea of smooth, differentiable functions, hits a wall. If a solution isn't differentiable, how can we even plug it into an equation that contains its derivatives? This is a genuine crisis. It seems that the very language we use to write down the laws of nature is inadequate for describing their consequences.

To escape this crisis, mathematicians in the 1980s, led by Michael Crandall and Pierre-Louis Lions, developed a revolutionary idea: the ​​viscosity solution​​. The genius of this approach is to change the very meaning of what it means to be a "solution."

The idea is as elegant as it is powerful. If our candidate solution, $u$, is too rough and spiky to have its own derivatives, let’s not ask it for them. Instead, let's test it from the outside using perfectly smooth functions.

Imagine you have a spiky, mountainous terrain—this is our non-smooth function $u$. You want to know its "curvature" at a peak. You can't measure it directly. But what you can do is take a perfectly smooth, imaginary paraboloid, $\phi$, turn it upside down, and lower it until it just touches the peak of the mountain from above. At that single point of contact, two things are true: the height of the mountain and the paraboloid are the same, and the smooth paraboloid lies everywhere above the mountain nearby. Now, instead of asking the mountain about its curvature, we ask the paraboloid. We demand that the derivatives of the smooth function $\phi$ satisfy the PDE, but in a "one-sided" way.

This gives rise to two new concepts:

• A ​​viscosity subsolution​​ is a function that is "pushed down" by the PDE. To check it, we touch it from above with a smooth test function $\phi$. At the point of contact, we require that $\phi$ satisfy the inequality $F(D^2\phi, D\phi, \phi, x) \le 0$. The function is required to be ​​upper semicontinuous​​, which intuitively means it can jump down but not up, ensuring we can always find a test function to touch it from above.

• A ​​viscosity supersolution​​ is a function "pushed up" by the PDE. We test it by touching it from below with a smooth $\phi$, and at the point of contact, we demand $F(D^2\phi, D\phi, \phi, x) \ge 0$. This function must be ​​lower semicontinuous​​—it can jump up but not down.

A ​​viscosity solution​​ is then simply any function that is simultaneously a subsolution and a supersolution. Such a function must be continuous, but it doesn't need to be differentiable at all!

This might seem abstract, so let's make it concrete. Consider the 1D equation $-u'' + (u')^2 = 1$ with boundary values $u(0)=u(1)=0$. Let's try to find a simple shape that acts as a subsolution. A natural candidate that already satisfies the boundary conditions is a downward-opening parabola, $\phi(x) = \alpha x(1-x)$. By plugging this smooth function into the inequality $- \phi'' + (\phi')^2 \le 1$ and solving for the largest possible $\alpha$, we find that for any $\alpha \le \sqrt{2}-1$, our simple parabola is a valid subsolution. It's a "floor" that the true solution must lie above. This is a key step in ​​Perron's method​​, a powerful technique where the true solution is constructed by taking the "ceiling" of all possible subsolutions.

The viscosity method is incredibly flexible, but it doesn't work for just any equation. The PDE must obey certain structural rules. The most important of these is ​​ellipticity​​, which is a condition on how the operator $F$ responds to changes in the second derivative term, $D^2u$.

The most basic form is ​​degenerate ellipticity​​. In simple terms, this is a monotonicity condition. It says that if you make a function more "convex" (by adding a positive semidefinite matrix to its Hessian $D^2u$), the value of the operator $F$ should not increase. It should either decrease or stay the same. Many equations that arise from optimization problems, like the Hamilton-Jacobi-Bellman equations of optimal control, naturally satisfy this property. For example, operators of the form $F(X) = \sup_{\alpha} \{ -\operatorname{tr}(A_\alpha X) + \dots \}$, where the matrices $A_\alpha$ are positive semidefinite, are archetypal examples of degenerate elliptic operators.

A much stronger condition is ​​uniform ellipticity​​. Here, the operator's response is not just monotonic; it's quantitatively bounded. This is where the magnificent ​​Pucci extremal operators​​, $\mathcal{P}^+_{\lambda, \Lambda}$ and $\mathcal{P}^-_{\lambda, \Lambda}$, enter the stage. Think of them as universal referees defining the strictest possible bounds for a linear operator whose coefficients are constrained. The maximal operator $\mathcal{P}^+$ tells you the largest possible response to a change in convexity, while the minimal operator $\mathcal{P}^-$ tells you the smallest. A uniformly elliptic operator $F$ is one whose behavior is always "sandwiched" between these two Pucci operators:

This strict, uniform control is what buys us miracles.

Here is the most beautiful part of the story. We began by abandoning the assumption of smoothness to define solutions for very rough situations. But if we impose the strict rule of uniform ellipticity, smoothness magically reappears. The equation itself forces its own solutions to be well-behaved, a phenomenon known as ​​regularity​​.

One of the first signs of this is the celebrated ​​Krylov-Safonov Harnack inequality​​. It states that for any non-negative viscosity solution to a uniformly elliptic equation in a ball, its maximum value in a smaller interior ball is controlled by a constant times its minimum value in that same ball. This is profound. It means the solution cannot have arbitrarily sharp peaks right next to deep troughs. It is forced to be somewhat "flat." There is an inherent averaging property at work, a hidden smoothing mechanism embedded in the very structure of the equation.

The pinnacle of this line of thought is the ​​Evans-Krylov theorem​​. This theorem is like a mathematical rock tumbler. You put in a rough, merely continuous viscosity solution. The machinery of the theorem, powered by the equation's uniform ellipticity and a structural property like concavity (as is the case for the $\log \det$ operator in the complex Monge-Ampère equation), goes to work. When it's done, it hands you back a solution that is not only smooth but whose second derivatives are continuous. This "bootstrapping" of regularity is one of the deepest and most powerful results in modern analysis. It tells us that for the right class of problems, even if we start by imagining the worst-case, most jagged solution, the underlying mathematical law itself will polish it to a brilliant sheen.

This journey, from classifying equations to dealing with the breakdown of classical ideas and inventing a new notion of solution, culminates in the discovery that order can spontaneously emerge from the rigid logic of the equations themselves. It's a testament to the fact that even in the most nonlinear and complex corners of the mathematical universe, there is an inherent structure and a profound, unifying beauty waiting to be found.

Having grappled with the mechanisms of fully nonlinear partial differential equations and the subtle yet powerful concept of viscosity solutions, you might be wondering, "Where do these abstract ideas actually show up?" As it turns out, these equations are not just a playground for mathematicians. They are the language in which nature describes some of its most elegant and complex phenomena, from the path of a light ray to the very fabric of spacetime envisioned by modern physics. This is where the story gets truly exciting, as we see these mathematical tools leave the blackboard and reshape our understanding of the world.

Perhaps the most direct way to see a fully nonlinear PDE in action is to simply look. When you see a distorted image through a glass of water or the shimmering of a road on a hot day, you are witnessing the bending of light. Geometrical optics, the study of the paths of light rays, is governed by a beautifully simple-looking yet fully nonlinear equation: the ​​Eikonal equation​​. It takes the form $(\nabla u)^2 = n^2$, where the function $u(x)$ represents the arrival time of a light wave at a point $x$, and $n(x)$ is the refractive index of the medium. The nonlinearity here is in the term $(\nabla u)^2$, the squared magnitude of the gradient of $u$. The paths that light rays take are precisely the "characteristic curves" of this equation, paths that are always perpendicular to the wavefronts (the surfaces where $u$ is constant). Remarkably, solving this PDE is equivalent to finding the path of least time, a deep principle first articulated by Fermat. In this way, a fundamental physical principle is perfectly encoded in the structure of a nonlinear PDE. For example, a simple model for atmospheric layering can be described by a refractive index that changes with height, and solving the Eikonal equation reveals that light rays follow beautiful circular arcs through this medium.

These equations don't just describe static paths; they describe evolution and change. Imagine a soap bubble slowly shrinking, or the boundary between metallic grains in a cooling alloy. These are examples of ​​geometric flows​​, where a surface moves and changes its shape based on its own curvature. The speed of the surface at any point might depend on its mean curvature, its Gaussian curvature, or some other geometric quantity. This dependence is often nonlinear, and the evolution of the surface is described by a fully nonlinear PDE.

A classic example is the ​​Mean Curvature Flow​​, where the velocity of the surface is proportional to its mean curvature. This flow is governed by a parabolic nonlinear PDE. The "parabolicity" here isn't just a technical label; it's a guarantee that the flow is well-behaved, like a heat equation smoothing things out. One of the most important consequences is the ​​avoidance principle​​: two initially separate, closed surfaces evolving by such a flow will never touch or cross each other. This stability is a direct consequence of the mathematical property of ellipticity in the curvature function. It ensures that the evolution is predictable and doesn't spontaneously create wild singularities. This principle is not just mathematically beautiful; it's what makes these models useful for everything from computer graphics and image segmentation to modeling the physics of crystal growth.

Let's shift gears from the physical world to the world of decision-making. Imagine you are piloting a spacecraft to a target, trying to use the minimum amount of fuel, or you are an investor managing a portfolio to maximize returns while minimizing risk. In all such problems, you are seeking an optimal strategy or "control" over time. The mathematical framework for finding these best-case strategies is called optimal control theory, and its master equation is the ​​Hamilton-Jacobi-Bellman (HJB) equation​​.

The HJB equation is a PDE for a "value function," $V(t,x)$, which represents the best possible outcome you can achieve starting from state $x$ at time $t$. To find the best path forward, you must consider all possible immediate actions and choose the one that yields the best combination of immediate reward and future value. This act of choosing the "best" from a set of options introduces an operator like $\sup$ (supremum) or $\inf$ (infimum) into the equation, acting on the derivatives of $V$. This instantly makes the HJB equation fully nonlinear.

Here we encounter a wonderful difficulty. The value function $V$ is often not smooth! Consider the simple problem of finding the shortest path around an obstacle; the distance function has a "kink" or a "corner" along the line where it's equally fast to go left or right. It is not differentiable there. The same thing happens in more complex control problems; the value function develops kinks and corners where the optimal strategy abruptly switches. A classical PDE, which requires smooth, twice-differentiable solutions, simply has no answer.

This is where the genius of viscosity solutions comes to the fore. Instead of demanding that the solution itself be differentiable, the viscosity framework tests the function by "touching" it with smooth functions from above and below. It provides a way to make sense of the PDE even at the kinks. This brilliant idea turns out to be exactly what is needed. The value function from a control problem is the unique viscosity solution to its corresponding HJB equation. This framework ensures that the problem has a solution (existence), that there's only one (uniqueness), and that it correctly represents the control problem (consistency). The theory is so robust that it handles situations where the randomness in the system is "degenerate"—that is, not acting in all directions—a common scenario in engineering and finance. Other powerful probabilistic tools, like the theory of branching processes or backward stochastic differential equations, provide alternative nonlinear representations that have been instrumental in solving a wide array of semilinear problems arising in finance and population dynamics.

The journey of fully nonlinear PDEs culminates in some of the most profound questions in pure mathematics and theoretical physics. Here, the equations are used not just to model phenomena within space, but to understand the fundamental nature of space itself. The star of this story is the magnificent ​​Monge-Ampère equation​​.

In its simplest form, it looks like $\det(D^2u) = f(x)$, where $\det(D^2u)$ is the determinant of the Hessian matrix of second derivatives of a function $u$. This equation is the archetypal fully nonlinear elliptic PDE. What does it describe? For a convex function $u$, its Hessian is positive semi-definite, and $\det(D^2u)$ measures the "curvature" of its graph. For a classical solution, it's just a number. But for a non-smooth convex function, like the top of a cut diamond, we need a generalized solution. The theory of viscosity solutions once again provides the answer, and beautifully, it coincides with another notion from geometry: the ​​Alexandrov solution​​, where the operator $\det(D^2u)$ is reinterpreted as a measure that describes how the map of gradients of $u$ expands volumes.

The true magic of the Monge-Ampère equation appears when it emerges, seemingly out of nowhere, as the answer to a completely different problem: ​​optimal transport​​. Imagine you have a pile of sand with a certain density distribution, and you want to move it to form a new shape with a different density distribution. What is the most efficient way to move the sand, minimizing the total distance traveled? This problem, with applications from logistics and economics to image morphing, is solved by a "transport map." In his celebrated work, Brenier showed that this optimal map is the gradient of a convex function $u$. And the equation this function $u$ must satisfy is none other than the Monge-Ampère equation, where the right-hand side $f$ is determined by the two sand densities. It is one of the most stunning examples of the unity of mathematics.

Armed with this powerful equation, mathematicians and physicists have ventured into even deeper waters. In string theory, physicists postulate that our universe has extra, hidden dimensions that are curled up into a compact shape. The geometry of these hidden dimensions determines the laws of physics we observe. The most promising candidates for these shapes are ​​Calabi-Yau manifolds​​. A key feature of these spaces is that they are "Ricci-flat," meaning they have a special kind of geometric balance.

In a monumental achievement, Shing-Tung Yau proved the existence of these manifolds by solving a conjecture posed by Eugenio Calabi. His strategy was to translate the geometric problem into a single, complex version of the Monge-Ampère equation. He then solved this equation using the "continuity method," a heroic analytical feat that involves deforming a simple, solvable problem into the hard one and proving that a solution exists all along the path. The proof required establishing incredibly difficult a priori estimates to prevent the solution from "blowing up"—the "closedness" step—and showing local solvability via linearization—the "openness" step. Yau's solution of the Calabi conjecture, a triumph of the theory of fully nonlinear PDEs, provided the mathematical foundation for the geometric world of string theory.

This is not the only place where these equations sculpt the world of modern physics. Other special geometric shapes, like ​​special Lagrangian submanifolds​​—which are volume-minimizing surfaces inside Calabi-Yau manifolds and are thought to represent certain fundamental objects in string theory called D-branes—are also found by solving a related but distinct fully nonlinear PDE, the special Lagrangian equation.

From tracing light rays to optimizing fuel, from modeling soap bubbles to discovering the shape of hidden dimensions, fully nonlinear PDEs provide a unified and powerful language. They are a testament to how the pursuit of abstract mathematical structure can lead to profound insights into the workings of our universe, revealing a deep and often surprising beauty at the heart of reality.