What Is a Differential Equation?

From one term to a whole subject

In Volume I you met the differential equation almost in passing — an equation that relates a function to its own derivatives. You probably saw exactly one example, something like dy/dx = k y for exponential growth, solved it by guessing y = C e^{k x}, and moved on. That single line was a doorway. Walk through it and you find one of the largest and most useful subjects in all of applied mathematics. The reason is simple: nature almost never tells you a quantity directly. It tells you the rate at which the quantity changes, and leaves you to reconstruct the quantity itself.

Newton's second law is a differential equation: force equals mass times acceleration, and acceleration is d^2x/dt^2, so F = m d^2x/dt^2 is a law about the second derivative of position. Cooling, radioactive decay, the charge on a capacitor, the population of a species, the temperature along a rod, the shape of a hanging cable — each is governed by a relation between a quantity and its rate of change. A differential equation is the mathematical sentence that encodes such a law of change. Solving it means recovering the quantity from the law its derivatives obey.

Three labels that tell you what you are facing

Before you try to solve anything, you classify it — because the class decides which methods even have a chance. The first label is the order: the highest derivative that appears. dy/dx = k y is first order; F = m d^2x/dt^2 is second order. Order roughly counts how much freedom the solution carries — a first-order equation needs one piece of starting information, a second-order one needs two (think position and velocity). The whole point of starting this rung with first-order equations is that one is the simplest place to learn the trade.

The second label is the degree: once the equation is written as a polynomial in its derivatives (no roots, no fractions of derivatives), the degree is the power on the highest-order derivative. Most equations you meet are first degree — dy/dx appears to the first power — but something like (dy/dx)^2 = 1 + y is second degree. Be honest here: degree is only defined after you have cleared roots and made it polynomial in the derivatives, and for some equations that is impossible, so they simply have no degree. Order and degree are different questions; do not let the words blur together.

The third label is the one that matters most: linear versus nonlinear. An equation is linear if the unknown y and all its derivatives appear only to the first power, never multiplied together and never inside a sine, a square, or an exponential — so it has the shape a_n(x) y^{(n)} + ... + a_1(x) y' + a_0(x) y = g(x), where the coefficients may depend on x but not on y. The instant you see y^2, or y times y', or sin(y), the equation is nonlinear. This single distinction is the great watershed of the whole subject. Linear equations enjoy the superposition that makes them tractable; nonlinear ones can do wild things — multiple equilibria, blow-up in finite time, chaos — and rarely yield a clean formula.

A family of solutions, then one specific answer

Solve dy/dx = k y and you do not get a single curve — you get y = C e^{k x} for *every* constant C, an infinite family of curves stacked on top of one another. That family, with its arbitrary constants, is the general solution, and the count of constants matches the order: one for first order, two for second, and so on. The constants are not a nuisance; they are exactly the freedom nature left open until you supply the missing facts.

Those facts come as initial conditions. Pin down where the curve passes at a starting instant — say y = 3 when x = 0 — and the family collapses to one member: a particular solution. A differential equation together with enough initial conditions to fix the constants is an initial-value problem, or IVP, and it is the form almost every real model takes. The equation is the law; the initial condition is the present state; the solution is the future and past the law unrolls from that state. For dy/dx = k y with y(0) = 3, the constant is forced to C = 3 and the one true answer is y = 3 e^{k x}.

Does an IVP always have exactly one answer? Usually, but not automatically — and this is where honesty matters. The existence-uniqueness theorem (Picard-Lindelof) guarantees one and only one solution through a given starting point, but only locally — on some interval around that point, not forever — and only when the right-hand side is well behaved (continuous, with a bounded sensitivity to y, the so-called Lipschitz condition). Break the hypotheses and strange things happen: dy/dx = sqrt(y) with y(0) = 0 has both y = 0 and y = x^2/4 as solutions, because sqrt(y) has an infinite slope at y = 0. The theorem is a promise with fine print; read the fine print.

What it really means to "solve" one

Here is the idea that reorganizes the whole subject in your head. "Solving" a differential equation is not one thing — it is at least four genuinely different goals, and a good applied mathematician keeps all four on the workbench, reaching for whichever the problem allows. The honest truth is that the first one, the kind you were taught to expect, is the rarest.

1. Closed form. Find an exact formula in elementary functions, like y = 3 e^{k x}. This is the dream, and the next guides give you the tactics that win it — separable, linear, exact equations. But most differential equations have no closed-form solution at all, just as most integrals are non-elementary; this is the exception, not the rule.
2. Series. When no formula exists, build the solution as a power series, y = a_0 + a_1 x + a_2 x^2 + ..., and grind out the coefficients one by one from the equation. This trades a formula for an infinite recipe — closely kin to the Taylor series you already know — and it is how the famous special functions (Bessel, Legendre) are born.
3. Numerical. Give up exactness for arithmetic: start at the known point and step forward in tiny increments, using the equation itself to predict each next value, as in Euler's method. This always works, to any accuracy you are willing to pay for in computing time, and it is what a computer actually does when it simulates a rocket or the weather.
4. Qualitative. Often you do not need the curve, only its character: Does it grow, decay, or settle? Toward what value? Is that resting state stable? You can read all of this straight off the equation without solving it — from its slope field and equilibria — which is the subject of the next guide on slope fields.

The picture: every equation is already a slope field

Here is the single most clarifying picture for first-order equations, and the bridge to the rest of this rung. Write the equation in the form dy/dx = f(x, y). Now read it geometrically. At every point (x, y) in the plane, the right-hand side f(x, y) hands you a number, and that number is the slope a solution curve must have if it passes through that point. So the equation does not describe one curve — it paints a whole field of tiny slope arrows across the plane, one at every point, before you have solved anything at all.

That field of arrows is the direction field (or slope field), and solving the equation is nothing more than threading a curve through the plane so that it stays tangent to the little arrow at every point it crosses. An initial condition just nails down one point the thread must pass through; the existence-uniqueness theorem says that — under its hypotheses — exactly one thread runs through that point. You can literally see why a solution exists and why it is unique: arrows fill the plane, and through a given point there is one way to follow the flow.

dy/dx = f(x, y)              the law, read as a slope at every point

         y                   tiny arrows = the direction field
         |   /   /   /   /
         |  /   /   /   /     a solution = a curve kept tangent
         | /   /   _____      to the arrow at each point it meets
         |/   _---      ---   (here the field flattens out:
         +----------------- x  an equilibrium, dy/dx -> 0)

Where this rung is taking you

With the language in place, the path is clear. The next guide stays purely geometric — reading a slope field, spotting equilibria, and judging which steady states a solution drifts toward — so you can understand any first-order equation, solvable or not, at a glance. After that we go hunting for closed forms: separable equations you split and integrate, exact equations you recognize as a total differential, and linear equations you tame with an integrating factor. The classification you just learned is your map: order one, first degree, linear or not — those three labels tell you which method to even try.

Keep the big shape in mind as you go. A differential equation is a law of change; its general solution is a family with as many free constants as its order; an initial-value problem picks out one member by fixing the present state; and "solving" can mean a formula, a series, a numerical march, or a qualitative read — whichever the equation, and your purpose, actually call for. That four-way flexibility is exactly why this subject reaches so far into physics, engineering, biology, and finance. One Volume I term has become a whole craft.