Differential Inclusion

The quest to predict the future has long been a cornerstone of science, perfectly encapsulated by the deterministic "clockwork universe" of ordinary differential equations (ODEs). In this classical view, knowing a system's precise state at one moment allows its entire future trajectory to be calculated. However, this elegant determinism breaks down when faced with uncertainty, physical constraints, or inherent choices, where the rate of change is not a single value but a collection of possibilities. This article addresses this gap by introducing the powerful framework of differential inclusions, which replaces the equals sign with a statement of belonging: $\dot{x} \in F(x)$.

This fundamental shift from a single path to a landscape of potential futures allows us to model a far richer set of phenomena. In the chapters that follow, you will discover the core concepts that govern this non-deterministic world. First, in "Principles and Mechanisms," we will explore the nature of solutions, the concept of a reachable set, and the surprising mathematical rules that emerge when uniqueness is abandoned. Then, in "Applications and Interdisciplinary Connections," we will witness how this single idea of differential constraint provides a profound and unifying language for seemingly unrelated fields, from the physics of solitons and the engineering of materials to the abstract foundations of mathematical logic.

For centuries, a central dream in science, articulated most clearly by thinkers like Pierre-Simon Laplace, has been the pursuit of a "clockwork universe." The idea is wonderfully elegant: if you know the precise position and velocity of every particle at one instant, and you know the laws of force that govern them, you can calculate the state of the universe for all of future time. This deterministic worldview is the soul of the ​​ordinary differential equation (ODE)​​, which states that the rate of change of a system, $\dot{x}$, is a unique function of its current state, $x$. In this world, every starting point has one and only one future, a single, inviolable trajectory laid out for all time.

But what if the world is a bit more... indecisive? What if our knowledge of the forces is incomplete? Or, more profoundly, what if the system itself has inherent choices? Imagine a boat in a choppy sea, buffeted by unpredictable gusts of wind, or a biological cell navigating a chemical gradient where its "decision" to move is not a single response but a range of possibilities. In these cases, the velocity vector is not a single, crisply defined arrow. Instead, it is a whole collection of possible arrows, a set of potential futures available at every instant. This is the world of the ​​differential inclusion​​.

The move from an ODE to a differential inclusion is a small, almost humble, change in notation, but it represents a monumental shift in philosophy. We replace the equals sign with the "element of" symbol:

This equation reads: "the velocity of the system at time $t$, $\dot{x}(t)$, is an element of the set of possible velocities, $F$, which depends on the current state $x(t)$." That little symbol, $\in$, is our gateway into a new reality. It tells us we are no longer on a single, pre-ordained railway track. We are in an open field, with a menu of possible directions at every step.

It is crucial to understand what this does, and does not, mean. This is not a ​​stochastic​​ system. A stochastic system involves randomness; its evolution is governed by explicit probabilistic rules, as if a cosmic die were being rolled at each step to pick a direction. A differential inclusion is simpler and, in a way, more fundamental. It makes no mention of probabilities. It is a ​​non-deterministic​​ model that simply lays out the complete set of all possible futures without assigning a likelihood to any of them. The system becomes deterministic only in the very special case where the set $F(x)$ shrinks to a single point for every $x$, and the resulting ODE happens to have unique solutions. Otherwise, we are left to grapple with a branching, diverging fan of possibilities.

If the future is no longer a single point on a line, what is it? It is a region, a domain of possibilities we call the ​​reachable set​​, $\mathcal{R}(T)$. This is the set of all possible states the system can be in at a future time $T$, having started from a given initial condition $x_0$. How do we begin to map the frontiers of this unknown territory?

Let's consider a simple one-dimensional case. Suppose the velocity $\dot{x}$ can be any value in an interval that changes with time, say $\dot{x} \in [-A, A e^{-t/\tau}]$. Or perhaps the interval of choices depends on the state itself, like $\dot{y} \in [y, y+1]$. To find the maximum possible position at time $T$, what must we do? The principle is wonderfully intuitive. To travel the farthest, you must, at every single moment, choose the fastest available velocity. Any hesitation, any choice of a slower speed for even an instant, means you will fall short of the absolute maximum distance. Likewise, to find the minimum possible position, you must consistently choose the most negative (or slowest) velocity at every instant.

This beautifully simple idea is the ​​principle of extremes​​. The boundary of the reachable set is traced out by the trajectories that never compromise, that always "push" or "pull" as hard as the rules allow. By solving the two ODEs corresponding to these extremal choices—$\dot{x}_{\max}(t) = \sup F(x(t), t)$ and $\dot{x}_{\min}(t) = \inf F(x(t), t)$—we can find the upper and lower bounds of the reachable interval. The length of this interval, $y_{\max}(T) - y_{\min}(T)$, gives us a quantitative measure of the system's uncertainty or freedom at time $T$.

The principle of extremes works perfectly in one dimension. But what happens in a plane, or in higher dimensions? Does the reachable set just become a simple box defined by the min/max velocities in each direction? The answer is a resounding "no," and it reveals a piece of mathematical magic.

Imagine you are controlling a particle in a plane, and your available velocity vectors, $\mathbf{v}$, are confined to two separate, disjoint disks. You can choose any velocity from the first disk, or any velocity from the second, but nothing in between. Where can you get to in one hour? Naively, you might expect the reachable set to be two disjoint regions, corresponding to the two sets of choices.

But what if you could switch between your choices infinitely fast? Suppose you spend half a second using a velocity $\mathbf{v}_1$ from the first disk, and the next half-second using a velocity $\mathbf{v}_2$ from the second disk. Over that one-second interval, your average velocity would be $\frac{1}{2}\mathbf{v}_1 + \frac{1}{2}\mathbf{v}_2$. By rapidly varying the proportion of time spent using each choice, you can effectively achieve any average velocity lying on the straight line segment connecting $\mathbf{v}_1$ and $\mathbf{v}_2$. This rapid-fire switching is known as ​​chattering​​.

The astonishing consequence, formalized in a result known as the ​​Filippov-Aumann Theorem​​, is that chattering allows the system to effectively "fill in" all the gaps in its velocity choices. The reachable set is not determined by the set of velocities $F$ itself, but by its ​​convex hull​​, denoted $\text{conv}(F)$. The convex hull is the shape you would get by stretching a rubber band around all the points in $F$. For the case of the two disjoint velocity disks, the convex hull is a single, connected "stadium" or "capsule" shape. The area of the reachable set $\mathcal{R}(T)$ is then simply the area of this convex hull, scaled up by a factor of $T^2$. Nature, through the dynamics of differential inclusions, always finds a way to take the average and convexify the set of possibilities.

This fundamental departure from uniqueness—the ability for paths to branch and for chattering to average out controls—forces us to rethink many classical results of dynamical systems. In the orderly world of ODEs with unique solutions, trajectories in a plane can never cross. If two paths were to meet, uniqueness would demand that they be the same path from that point forward and backward in time.

This non-crossing property is the absolute bedrock of profound theorems like the ​​Poincaré-Bendixson theorem​​, which states that if a trajectory in a plane is confined to a finite region without any equilibrium points, it must eventually approach a closed loop, a ​​periodic orbit​​. But in the world of differential inclusions, all bets are off. Trajectories can meet, cross, and go their separate ways. An orbit spiraling in a confined region might not settle into a simple loop. Its limit set could be a more complex, graph-like object, something unimaginable for a classical system.

Even more strangely, new types of behavior emerge. In systems with discontinuous rules—like a thermostat that switches on or off—a trajectory can hit the switching surface and get stuck, sliding along it in a way that is governed by a combination of the dynamics on either side. This ​​sliding mode​​ is a hallmark of Filippov systems (a major class of differential inclusions) and is a genuinely new dynamical phenomenon, one that is impossible in smooth ODEs but essential for modeling modern control systems, robotics, and electrical circuits with switches.

Thus, the differential inclusion is far more than a mathematical complication. It is a richer, more powerful language. It provides a framework to describe systems where uncertainty, constraints, or choices are not just minor nuisances but are at the very heart of the dynamics. By studying these principles, we learn to map the boundaries of a future that is not a single line, but a burgeoning landscape of possibilities, and we discover the surprising and beautiful rules that govern motion even when the path ahead is not unique.

Having grappled with the principles of differential inclusions, we now arrive at the most exciting part of our journey. We have seen that by replacing the strict command of an ordinary differential equation, $y' = f(x, y)$, with the more permissive condition $y' \in F(x, y)$, we open up a new world. This might seem like a step backward, a loss of precision. But in science, asking the right question is everything. It turns out that asking "what are the allowed directions of change?" instead of "what is the exact direction of change?" is an immensely powerful idea. It is the language of constraints, and it unifies a startling variety of phenomena, from the solitary dance of waves to the plastic flow of steel, and even to the very foundations of logical reasoning. Let us now explore some of these remarkable connections.

Imagine you are studying a complex physical system governed by a nonlinear partial differential equation (PDE). These equations, which describe everything from water waves to signals in optical fibers, are notoriously difficult to solve. Finding even one non-trivial solution can be a triumph. But what if there were a magical recipe, a kind of mathematical alchemy, that could take one known solution and transform it into an entirely new one?

This is precisely what a Bäcklund transformation does. It is not a single formula, but rather a system of differential relations that links a known solution, let's call it $u$, to a new, unknown solution, $v$. For instance, in the theory of solitons—those famously robust waves that travel without changing shape—the sine-Gordon equation $u_{xt} = \sin(u)$ is of central importance. A Bäcklund transformation for this equation is given by a pair of first-order constraints on the new solution $v$:

Notice what is happening here. These equations don't tell you what $v$ is. They tell you how its derivatives, its rates of change in space and time, are constrained by the known solution $u$ and by $v$ itself. The astonishing fact is that if you start with a solution $u$ and find a function $v$ that satisfies these two relations simultaneously, then this new function $v$ is guaranteed to be another solution to the sine-Gordon equation. This is because the system of relations has a built-in compatibility condition. The constraints are so perfectly intertwined that satisfying them forces the new function to obey the original, more complex, second-order PDE.

This is not an isolated trick. This powerful idea of using compatible differential constraints to generate solutions applies to a whole family of important "integrable" systems. Equations like the Tzitzéica equation, $u_{xy} = e^u - e^{-2u}$, and the Boussinesq equation for water waves also possess these remarkable transformations. By starting with a simple solution (even the "vacuum" solution, $u=0$), one can apply these transformations repeatedly to generate a whole hierarchy of complex, physically relevant solutions, like multi-soliton collisions. The differential relations act as a constructive algorithm, a set of instructions for building mathematical reality.

Let's turn from the ethereal world of waves to something you can hold in your hand: a piece of metal. When you bend a paperclip, it first deforms elastically, springing back to its original shape. But if you bend it too far, it deforms plastically—it stays bent. What is happening at the microscopic level, and how can we describe it mathematically?

The transition to plastic flow is governed by a yield criterion. The state of stress inside the material can be described by a set of numbers (the stress tensor). As long as the combination of stresses is below a certain threshold, the material is elastic. But once the stress hits that threshold—the "yield surface"—the material begins to flow. From that moment on, the state of stress is constrained to lie on this surface. It cannot go beyond it.

This is a perfect physical analogue of a differential inclusion. The state of the system (the stress field) is not determined by a single law of evolution, but is constrained to a set of allowed states. In the case of plane strain deformation (where the material is very long in one direction), this principle leads to a theory of extraordinary elegance and power: slip-line field theory.

When a rigid, perfectly plastic material yields, it deforms by shearing along two orthogonal families of curves, known as slip-lines or characteristics. The physics of force equilibrium, when combined with the Tresca yield criterion ($\tau_{\max} = k$, where $k$ is the material's constant shear yield stress), gives rise to a pair of beautiful differential relations. If we describe the stress state by the hydrostatic pressure $p$ and the angle $\theta$ of the principal stress direction, these relations, known as the Hencky relations, state that:

This is a profound result. The complex, seemingly chaotic flow of solid metal is governed by these simple, elegant rules. They tell us precisely how the pressure must change as the stress field rotates along these characteristic lines. Even better, these relations immediately imply the existence of conserved quantities. Along an $\alpha$-line, the quantity $p+2k\theta$ must be constant, and along a $\beta$-line, $p-2k\theta$ must be constant. This insight allows engineers to calculate the stress fields and forces required for practical manufacturing processes like forging, rolling, and extrusion, transforming an abstract mathematical constraint into a tool for designing the world around us.

Our final stop is the most abstract, and perhaps the most beautiful. It connects the world of differential constraints to the very foundations of mathematical logic. Consider the functions you learned about in school. Polynomials are "nice": a polynomial of degree $n$ can have at most $n$ roots. They don't wiggle infinitely often. The function $\sin(x)$, on the other hand, has infinitely many roots. Some functions are even more pathological. How can we describe this notion of "tameness" in a rigorous way?

Logicians developed the concept of o-minimal structures. An o-minimal structure is a collection of sets and functions where anything you can define in one dimension is just a finite collection of points and intervals. There are no infinitely oscillating curves, no fractal dust like the Cantor set. The sets definable from polynomials are o-minimal. But what about functions like the exponential function, $\exp(x)$? It seems "tame," but it's not a polynomial.

The key to expanding this world of "tame" geometry lies, once again, in differential constraints. A class of functions known as Pfaffian functions are built in a chain, where the derivative of each new function is constrained to be a polynomial of the functions that came before it. For example, $f_1(x) = \exp(x)$ satisfies the differential equation $f_1' = f_1$. If we then define $f_2(x) = \exp(\exp(x))$, its derivative is $f_2' = \exp(\exp(x)) \exp(x) = f_2 \cdot f_1$, again a polynomial in the previous functions.

This strict, triangular system of differential relations acts as a powerful brake on the complexity of the functions. A celebrated result in mathematics, stemming from the work of Askold Khovanskii, shows that this very constraint is what guarantees tameness. The Pfaffian conditions, coupled with the property of real analyticity, force the functions to have a finite number of zeros and a controlled geometric structure. This allows one to prove that the structure generated by these functions is o-minimal.

Think about the implications. The same fundamental idea—controlling a system's behavior by placing constraints on its derivatives—that allows us to generate soliton solutions and to model the flow of plastic solids, also provides the very foundation for a vast and beautiful area of mathematics that classifies which parts of the mathematical universe are simple and which are complex. It is a stunning example of the unity of scientific thought, where a single principle echoes through physics, engineering, and pure logic. It teaches us that sometimes, the deepest understanding comes not from dictating a single path, but from simply laying down the rules of the road.