Why this matters
An ordinary equation usually answers with a number. "What number doubles to 4?" — the answer is just \(2\). "What number squares to 4?" — \(2\) and \(-2\), two of them. However many answers there are, they are a few points on the number line.
A differential equation, though, answers differently. A differential equation is an equation with a derivative (an instantaneous rate of change) in it. For example,
$$\frac{dy}{dx} = 2x$$Here \(\dfrac{dy}{dx}\) means "how fast \(y\) changes when \(x\) changes by a tiny amount," that is, the slope of the graph. So this equation is the command: "find the curve whose slope is always \(2x\)." Astonishingly, the answer to this command is not a number but a function, that is, a whole curve.
$$y = x^2 + C$$Here \(C\) is the arbitrary constant that always tags along with integration. It is also strange that the answer is not even a single curve but a family of curves, each shifted up or down by \(C\).
There is one more puzzle. When solving such an equation, we often do something odd. Although \(\dfrac{dy}{dx}\) is clearly a single chunk of a symbol meaning "slope," we peel the \(dx\) off and move it to the other side as if it were a fraction. Like this.
$$\frac{dy}{dx} = 2x \quad\Rightarrow\quad dy = 2x\,dx$$Are we allowed to handle something that is not a fraction as if it were one? This article threads these two puzzles — why the answer is a function, and why that shortcut is permitted — into one.
Why the answer is a function, not a number
The difference between an ordinary equation and a differential equation lies in what it asks.
- An ordinary equation asks "which number?" You look for a number to put in the unknown's place.
- A differential equation asks "which function?" What sits in the unknown's place is not a number but a whole function.
Look again at \(\dfrac{dy}{dx} = 2x\). This gives the information "there is an unknown function \(y\), and its slope at any \(x\) is \(2x\)." What we seek is the function \(y\) itself that meets this condition, not one particular number.
So how do we find that function? "The function whose slope is \(2x\)" means "the function whose derivative is \(2x\)," so we integrate — undo differentiation.
$$y = \int 2x \, dx = x^2 + C$$Here \(\int\) is the long S-shaped integral sign, meaning "add up all the finely sliced pieces." But why the result carries a \(C\) matters. Differentiation kills the constant term, sending it to \(0\) — differentiating \(x^2\) gives \(2x\), differentiating \(x^2 + 5\) gives \(2x\), and differentiating \(x^2 - 100\) gives the same \(2x\). So the condition "derivative is \(2x\)" alone cannot tell us the up-down position, and the answer comes out as a family of curves shifted by \(C\). This family is called the general solution. (This is covered more fully in why integrals carry a + C.)
One starting point picks out one curve
If the answer is a family of curves, then in an actual problem which one do we choose? This is where the starting point (initial condition) enters.
Suppose that among the family \(y = x^2 + C\) we want "the curve passing through the point \((1, 3)\)." Plugging that point into the equation,
$$3 = 1^2 + C \quad\Rightarrow\quad C = 2$$fixes \(C\) to exactly one value, and a single curve \(y = x^2 + 2\) is picked out. Had we wanted the starting point \((0, 0)\), then \(C = 0\) and it would be \(y = x^2\). This — one starting point plucking exactly one curve from the countless family — is the picture of solving a differential equation.
This scene is grasped at a glance through the direction field. A direction field is a picture with short arrows drawn all over the plane showing "the slope a curve must have at this point." For \(\dfrac{dy}{dx} = 2x\), the arrow slope at each point is set by that point's \(x\)-coordinate as \(2x\) (independent of \(y\)). Smoothly joining the flow of these arrows gives exactly a solution curve, and depending on where you start, a curve with a different \(C\) is drawn. In the first screen of the interactive below, click the field to set a starting point, and a single curve through that point is drawn riding the arrow flow.
Now the second puzzle: may we move dx?
In the example just now, \(\dfrac{dy}{dx}\) was an expression in \(x\) only, so simply integrating both sides finished it. But once the slope also depends on \(y\), the story changes. Look at a representative example.
$$\frac{dy}{dx} = y$$This is the command "find the curve whose slope always equals its own height." The higher it is, the more steeply it rises, so the answer looks like an explosively growing function. The method commonly used to solve it is separation of variables, and its first step is precisely that suspicious shortcut.
$$\frac{dy}{dx} = y \quad\Rightarrow\quad \frac{1}{y}\,dy = dx$$We treated \(\dfrac{dy}{dx}\) as a fraction, moving the \(y\) to the left and the \(dx\) to the right. Then we attach integral signs to both sides and integrate each.
$$\int \frac{1}{y}\,dy = \int 1 \, dx \quad\Rightarrow\quad \ln|y| = x + C$$Here \(\ln\) is the natural logarithm with base \(e\). Rearranging,
$$y = e^{x + C} = e^{C} e^{x} = C' e^{x}$$(since \(e^{C}\) is just another positive constant, we bundled it under a new name \(C'\)). The answer is again not a number but a function, and a family of curves stretched or shrunk vertically by \(C'\).
The computation is clean, but a nagging doubt remains. \(\dfrac{dy}{dx}\) is not a fraction but a single chunk of an operation called "the derivative," so may we really peel the \(dx\) inside it off and move it like an object?
The shortcut is really integration by substitution
To state the conclusion first: the peeling-and-moving move itself is a shortcut, but the result is honest. That is because the shortcut is nothing but a condensed way of writing integration by substitution, a sturdy rule. Let us proceed by the book, without moving \(dx\).
The start is the same. Divide both sides of \(\dfrac{dy}{dx} = y\) by \(y\) (this is dividing, a legitimate operation, not moving).
$$\frac{1}{y}\cdot\frac{dy}{dx} = 1$$Now integrate both sides with respect to \(x\). Instead of peeling the \(dx\) off, honestly keep it attached.
$$\int \frac{1}{y}\cdot\frac{dy}{dx}\,dx = \int 1 \, dx$$Here the left integral is the key. The rule of substitution says that setting \(u = y(x)\) gives \(du = \dfrac{dy}{dx}\,dx\). That is, the chunk \(\dfrac{dy}{dx}\,dx\) on the left turns wholesale into \(dy\).
$$\int \frac{1}{y}\,\underbrace{\frac{dy}{dx}\,dx}_{=\,dy} = \int \frac{1}{y}\,dy$$Then the left becomes \(\displaystyle\int \frac{1}{y}\,dy\) and the right \(\displaystyle\int 1\,dx\), which is exactly the same equation the shortcut arrived at earlier. In other words,
$$\text{the shortcut of writing } \frac{1}{y}\,dy = dx \;=\; \text{the substitution } u = y(x) \text{ run backward and abbreviated}$$The hand-motion of peeling the \(dx\) off and moving it is just shorthand — "it is tedious to write out substitution every time, so let us jot down only the result" — and beneath it lies integration by substitution, honestly reeling the chain rule backward.
Let us pin the intuition once more — \(dy\) and \(dx\) are respectively "a tiny change in \(y\)" and "a tiny change in \(x\)," and \(\dfrac{dy}{dx}\) is the ratio of the two. Because it has the form of a ratio, manipulating it like a fraction really does make the answer come out right, and that is no coincidence but something substitution guarantees from behind, which is why we can use it with confidence.
In one picture: the family of curves and the two paths
The interactive below has two screens.
- Direction field: the slope arrows of \(\dfrac{dy}{dx} = 2x\) are spread over the plane. Click anywhere on the field to set a starting point, and the solution curve \(y = x^2 + C\) through that point is drawn riding the arrow flow. Each time you move the starting point, \(C\) changes and a different line of the family is picked out.
- Two paths: solve \(\dfrac{dy}{dx} = y\) two ways side by side — "the shortcut of moving \(dx\)" and "substitution by the book." Stepping through, you can see the two paths meet at the very same integral in the middle and both arrive at the same answer \(y = C'e^{x}\).
The two screens say one thing. The answer to a differential equation is a whole curve, and the starting point picks out one of them, and the shortcut for solving it is an abbreviation of the honest rule that is integration by substitution.
The key summary
| Question | Answer |
|---|---|
| Why is the answer not a number? | What sits in the unknown's place is not a number but a function (a whole curve) |
| Why is the answer a family, not one curve? | Differentiation erased the constant term, so the shift \(C\) cannot be recovered |
| How do we pick one curve from the family? | Feed in one starting point (initial condition) to fix \(C\) |
| May we peel \(dx\) off and move it? | The move is a shortcut but the result is honest — substitution lies beneath |
| Why does handling it like a fraction work? | \(\dfrac{dy}{dx}\) is the ratio of two tiny changes, and substitution guarantees the move |
Threaded into one line — a differential equation asks "which function?", so its answer comes out as a family of curves with a starting point picking out one, and moving \(dx\) in separation of variables is an honest calculation that abbreviates integration by substitution.
When you study this with AI
If you miss the mental shift that "the answer is a function," differential equations become mere rules to memorize. When asking an AI, framing the question with the picture that "the answer is a family of curves and a starting point picks one" and the frame that "separation of variables is shorthand for substitution" threads the scattered rules into one.
Useful example questions:
- "Explain why the general solution of \(\dfrac{dy}{dx} = 2x\) is not a single curve but a family with a \(C\), in terms of the information differentiation erases."
- "Show step by step, using the substitution \(u = y(x)\), why moving \(dx\) in separation of variables is the same as integration by substitution."
- "In solving \(\dfrac{dy}{dx} = y\) to get \(y = C'e^{x}\), compute what \(C'\) becomes if the initial condition is \(y(0) = 3\)."
Closing
Two things feel strange when you first meet differential equations: that the answer is not a number but a family of curves, and the hand-motion of peeling the \(dx\) in \(\dfrac{dy}{dx}\) off and moving it like an object. But look at the roots and both are natural. A differential equation asks "which function?", so the answer cannot but be a function; because differentiation erased the up-down information the answer becomes a family; and one starting point plucks a single curve from it. And the shortcut of moving \(dx\) is no magic but an honest calculation that abbreviates integration by substitution, which reels the chain rule backward.
The picture of one starting point plucking a single curve from a family, and the scene of the shortcut and the by-the-book path meeting at the same integral — grasp these two pictures, and differential equations look not like a lump of new rules to memorize but like the next story flowing naturally from the differentiation and integration you already learned.
Related reading — good to read first: Why integration is the reverse of differentiation · Why integrals carry a + C → next: Why substitution works no matter what variable you choose · Why the chain rule is "multiplication" · draw it yourself: Desmos · Wolfram Alpha