Visual walkthrough — First-order linear ODEs — integrating factor method (derivation)

Everything starts from a single stubborn equation. Let us meet it.

---

Step 1 — Meet the equation, and see why it is "stuck"

WHAT. We have an unknown function $y(x)$ — a mystery curve. The symbol $\frac{dy}{dx}$ means "the slope of that curve at each point"; it is how fast $y$ rises as $x$ moves right. $P(x)$ and $Q(x)$ are known functions of $x$ (things we can compute at any $x$). We want to find the actual $y$.

WHY it is stuck. The left side is two separate pieces added together: a slope piece $\frac{dy}{dx}$ and a height piece $P(x)y$. To "undo" a slope we integrate — but you can only integrate a slope cleanly when the whole left side is the slope of one single thing. Right now it is the slope of one thing plus something else. We cannot integrate the sum directly.

PICTURE. Below, the two pieces are drawn as two separate arrows pulling in different directions. They refuse to combine into one motion — exactly the "two kids who won't hold hands" from the parent note.

---

Step 2 — The wish: what would make it un-stuck?

WHAT. Introduce a brand-new function $\mu (x)$ — read "mu of x", just a name for a helper multiplier we get to choose. Multiply every term of Step 1 by $\mu$: $\underset{\text{slope piece}\times \mu }{\underbrace{\mu \frac{dy}{dx}}}+\underset{\text{height piece}\times \mu }{\underbrace{\mu Py}}=\underset{\text{push}\times \mu }{\underbrace{\mu Q}}.$

WHY multiply? Multiplying does not change the solutions of an equation (as long as $\mu \ne 0$): if two things are equal, $\mu$ times each is still equal. So we are allowed to do it for free, and we will spend that freedom buying ourselves a nicer left side.

PICTURE. $\mu$ is drawn as an amber "coating" wrapped around both pieces — same coating on everyone, so they can finally merge.

---

Step 3 — The tool that merges products: the Product Rule

WHAT. The Product Rule says: the slope of "$\mu$ times $y$" is (first $\times$ slope of second) plus (slope of first $\times$ second). Here $\frac{d\mu }{dx}$ means "how fast our helper $\mu$ itself changes".

WHY line it up? Set our merged left side beside the Product Rule's right side and compare piece by piece — because if they match, our left side simply is $\frac{d}{dx}(\mu y)$.

our left side (Step 2)	Product Rule (Step 3)	verdict
$\mu \frac{dy}{dx}$	$\mu \frac{dy}{dx}$	already identical ✓
$\mu Py$	$\frac{d\mu }{dx}y$	must be forced to match

PICTURE. The first terms overlap perfectly (green tick); the second terms are only shaped the same — both "(something)$\times y$" — but the "somethings" differ. That mismatch is the one job left.

---

Step 4 — Force the match: a birth of a condition on $\mu$

WHAT. We are no longer hoping — we are choosing $\mu$ so that its growth rate $\frac{d\mu }{dx}$ equals itself times $P$. That single equation is the entire price of admission.

WHY this is progress. Look closely: this is an ODE, but a friendlier one. It contains only $\mu$ and $x$ — no $y$ at all. And it has the special shape "rate of change $=$ (something) $\times$ current value", which is a separable equation. We traded one hard problem (find $y$) for one easy problem (find $\mu$).

PICTURE. A dial: turning the requirement "second terms equal" collapses into the boxed condition $\frac{d\mu }{dx}=\mu P$. Notice $y$ has vanished from the picture entirely.

---

Step 5 — Solve for $\mu$ by separating variables

WHAT. Gather every $\mu$ on the left, every $x$ on the right: $\underset{\text{all the }\mu \text{ stuff}}{\underbrace{\frac{d\mu }{\mu }}}=\underset{\text{all the }x\text{ stuff}}{\underbrace{P(x)dx}}.$ Integrate both sides: $\int \frac{d\mu }{\mu }=\int P(x)dx⟹\underset{\text{from }\int d\mu /\mu }{\underbrace{\ln |\mu |}}=\int P(x)dx.$

WHY the logarithm? The integral of $\frac{1}{\mu }$ is $\ln |\mu |$ — that is the one antiderivative that "undoes" a reciprocal. To free $\mu$ from inside a logarithm we apply the exponential $e^{(\cdot )}$, because $e$ and $\ln$ are inverse operations: $|\mu |=e^{\int Pdx}.$

Where did $+C$ go? An integration constant would ride along as $e^C\cdot e^{\int Pdx}$ — a constant multiple of $\mu$. But multiplying $\mu$ by a constant multiplies every term of Step 2 by the same constant, which cancels out. So we keep the simplest choice $e^C=1$.

PICTURE. The separation splits the plane into a "$\mu$-column" and an "$x$-column"; the exponential curve rises out of the accumulated area $\int Pdx$.

---

Step 6 — Watch the left side collapse

WHAT. The two stubborn pieces have fused into the slope of a single object $\mu y$. The messy part of the problem is gone.

WHY it worked. Because we forced $\frac{d\mu }{dx}=\mu P$ in Step 4, the Product Rule's two terms now equal our two terms exactly — so their sum equals $\frac{d}{dx}(\mu y)$ by definition, not by luck.

PICTURE. The two arrows of Step 1, now both amber-coated, snap together into one bold arrow labelled $\frac{d}{dx}(\mu y)$.

---

Step 7 — Un-slope both sides and solve for $y$

WHAT. Integrate $\frac{d}{dx}(\mu y)=\mu Q$ with respect to $x$. Integrating a slope returns the thing itself: $\underset{\text{recovered product}}{\underbrace{\mu y}}=\int \mu (x)Q(x)dx+\underset{\text{the constant, here at last}}{\underbrace{C}}.$

WHY $+C$ now? This is the first genuine integration of the whole equation, so this is where the single family-labelling constant belongs. (Putting it earlier, inside $\mu$, would have cancelled — Step 5.)

Final divide. We have $\mu y$, not $y$. Divide by $\mu$ (legal because $\mu =e^{(\cdots )}$ is never zero — an exponential is always positive):

PICTURE. Integration lifts the single arrow into a family of curves stacked by the value of $C$; dividing by $\mu$ reshapes them into the actual solution curves $y(x)$.

---

Step 8 — Edge & degenerate cases (the ones that trip people)

Case A — $P(x)=0$. Then $\mu =e^{\int 0dx}=e^0=1$. The equation is already $\frac{dy}{dx}=Q$, so you just integrate directly. The method silently reduces to plain integration — nothing breaks.

Case B — $P,Q$ both constant. With $P=k$ (a number), $\mu =e^{kx}$. This is the constant-coefficient special case; the same formula produces the familiar $y=\frac{Q}{k}+C{e}^{-kx}$ shape.

Case C — the coefficient of $\frac{dy}{dx}$ is not $1$. If you meet $a(x)\frac{dy}{dx}+\dots$, you are not in standard form yet. Divide the whole equation by $a(x)$ first. If you skip this, the term you feed as $P$ is wrong and the collapse in Step 6 fails.

PICTURE. Three mini-panels: flat $\mu =1$, exponential $\mu =e^{kx}$, and a "STANDARDIZE FIRST" gate you must pass through.

---

The one-picture summary

One diagram compresses all eight steps: stuck equation → coat with $\mu$ → match the Product Rule → forced condition → separable solve → exponential $\mu$ → collapse → integrate → divide → solution.

Recall Feynman retelling — say it like a story

We started with an equation whose left side was a sum of two unfriendly pieces: a slope and a height. You can't integrate a sum like that in one move. So we invented a magic coating $\mu$ and painted the whole equation with it. Now we demanded that after coating, the left side become the slope of one single lump, $\mu y$. The only rule that turns a slope-of-a-product into two added terms is the Product Rule — so we lined our coated equation up against it and matched piece by piece. The first pieces matched for free; the second pieces matched only if $\mu$ grew at rate $\mu P$. That little demand was itself an easy separable equation whose answer — because "growth proportional to size" always means exponential — is $\mu =e^{\int P}$. With that coating on, the two pieces fused, we un-sloped (integrated) to get $\mu y$, dropped in the one constant $C$, and divided by the never-zero $\mu$ to reveal $y$. Every "special" case — $P=0$, constant coefficients, a non-1 leading coefficient — is just this same story with one knob turned.

Recall Rapid self-test

Why must we integrate rather than just add? ::: The left side is a slope; integration is the operation that recovers a function from its slope. Which single rule forces the whole construction? ::: The Product Rule — it is the only expansion of a product's derivative into two summed terms. Why is dividing by $\mu$ always legal? ::: $\mu =e^{\int P}$ is an exponential, which is strictly positive, hence never zero. What happens if $P=0$? ::: $\mu =e^0=1$; the method degenerates to plain direct integration.

---

Connections

• Product Rule — the engine of Step 3.
• Separable ODEs — solves the condition in Step 5.
• Exact ODEs — where integrating factors generalise.
• Bernoulli Equations — reduce to this linear form by substitution.
• Linear Constant-Coefficient ODEs — Case B of Step 8.