Linear First-Order Equations and the Integrating Factor

When separation runs out

You have already climbed past the first solvable family: the separable equation, where every x lives in one bag and every y in another. The trouble is that most equations refuse to come apart like that. Try dy/dx = x - y: the x and the y are stuck together by a plain subtraction, and no amount of rearranging will let you write all-y on one side and all-x on the other. Separation, your first real tool, simply does not apply here.

But notice something hopeful about dy/dx = x - y: the unknown y shows up only by itself, never squared, never multiplied by y', never inside a sine. Rewrite it as y' + y = x and the shape jumps out. This is the first-order linear equation, the second great solvable family — and unlike separation, which works only in lucky cases, this one comes with a guaranteed method that never asks you to be clever or lucky.

What 'linear' actually means

A first-order linear equation can always be massaged into one standard shape: y' + p(x) y = q(x). Here p(x) and q(x) are known functions of x alone — they may be constants, or genuine functions like e^x or 1/x. The whole point of 'linear' is a restriction on how y and y' are allowed to enter: each appears only to the first power, never multiplied together, never squared, never fed into a function. So y, y', and a coefficient that depends on x — that is all you are permitted.

The right-hand side q(x) earns a special name: the forcing or input. When q(x) = 0 the equation is called homogeneous, and a happy coincidence appears — a homogeneous linear equation y' + p(x) y = 0 is also separable, solvable by the tool you already own. The genuinely new territory is the nonhomogeneous case, q(x) not zero, where the forcing pushes the solution around and separation gives up. That is exactly the gap the integrating factor was invented to fill.

The trick: collapse the left side into one derivative

Here is the idea that makes everything click. Stare at the left side of y' + p(x) y = q(x). It looks almost — but not quite — like the result of a product rule. Recall that the product rule turns (mu y)' into mu y' + mu' y. If only the messy left side mu y' + p mu y were secretly a product rule in disguise, we could rewrite it as (mu y)' and the equation would become trivial. The integrating factor is precisely the multiplier mu(x) engineered to make that disguise real.

Multiply every term of the equation by some not-yet-known function mu(x). The left side becomes mu y' + mu p y. Compare that with the genuine product-rule expansion (mu y)' = mu y' + mu' y. The two match exactly the moment mu' = mu p — that is, when mu's own derivative equals p times mu. So we just need a function whose derivative is p times itself, and exponentials do exactly that. Choosing mu(x) = e^(integral of p(x) dx) guarantees mu' = mu p, because the derivative of e^(integral p) is p times e^(integral p). The factor is not magic; it is reverse-engineered from the one condition that makes the product rule fire.

y' + p(x) y = q(x)            (standard form)
mu(x) = e^( integral p(x) dx )   (the integrating factor)
mu y' + mu p y = mu q
( mu y )'        = mu q        (left side is now ONE derivative)
mu y = integral( mu q ) dx + C
y = ( integral( mu q ) dx + C ) / mu

Why this never gets stuck

Once the left side is the single derivative (mu y)', the equation reads (mu y)' = mu(x) q(x), and you finish by integrating both sides and dividing back by mu. The deep comfort here is unconditional reliability: as long as p and q are continuous on an interval, this integrating-factor recipe returns the general solution on that interval — no lucky structure required, no guessing. This stands in sharp contrast to separation, which only works when the variables happen to come apart. The next guide walks the four steps slowly; for now, just feel the certainty that the method always exists.

Take the small example y' + (1/x) y = x. Here p = 1/x, so mu = e^(integral of (1/x) dx) = e^(ln x) = x. Multiply through by x to get x y' + y = x^2, whose left side is exactly (x y)'. Now (x y)' = x^2, integrate to x y = x^3/3 + C, and divide by x to land on y = x^2/3 + C/x. Notice the coefficient p was a genuine function of x, not a constant — the method handled it identically.

Reading the answer: two pieces with two jobs

Look back at y = x^2/3 + C/x and you can already see a pattern that runs through all linear theory. Every general solution splits cleanly into two parts: a homogeneous part plus a particular part, written y = y_h + y_p. The homogeneous part y_h carries the one arbitrary constant C and solves the forcing-off equation y' + p(x) y = 0; the particular part y_p is any single function that satisfies the full equation with q(x) present. Above, C/x is y_h and x^2/3 is y_p.

This split exists because linear equations obey superposition: responses to combined inputs are just combined responses, so you can build a full solution by adding a free homogeneous wobble onto one fixed particular answer. The arbitrary constant C is then pinned down by an initial condition. For a constant-coefficient case like y' + 2 y = 6, the pieces become y_h = C e^(-2x) (a part that decays as time runs on) and y_p = 3 (a part that just sits there) — the seed of the transient-plus-steady-state picture the final guide of this rung will develop.

One reassurance about that split: there is no single 'correct' particular solution y_p — any function that satisfies the full equation will do. Two valid choices differ only by a homogeneous solution, and the free constant C absorbs that difference, so the general solution is exactly the same either way. You never have to worry that you picked the 'wrong' y_p.

Honest edges and where you are headed

Two honest cautions keep you out of trouble. First, the integrating factor only works once the equation is in standard form, with the coefficient of y' equal to 1. If you start from a(x) y' + b(x) y = c(x), divide by a(x) first — otherwise you compute mu from the wrong p and the whole thing fails silently. Second, that division can introduce excluded points: dividing by x earlier means the solution C/x is only valid away from x = 0, so the answer lives on an interval of validity that may not be the whole line.

Unlike separable equations, the linear recipe has no lost-constant-solution pitfall — there is no hidden division by something involving y that could throw away an equilibrium, because the only division is by mu(x), which never depends on y. That tidiness is a structural gift of linearity, the same gift that, by the way, prevents linear first-order solutions from blowing up in finite time the way a nonlinear y' = y^2 can.

1. Put the equation in standard form y' + p(x) y = q(x) — divide through if the coefficient of y' is not already 1.
2. Build the integrating factor mu(x) = e^(integral of p(x) dx); any one antiderivative works, no constant needed.
3. Multiply through by mu and recognize the left side as the single derivative (mu y)', so the equation reads (mu y)' = mu q.
4. Integrate both sides (keeping the constant C here), then divide by mu to isolate y — and use any initial condition to pin C.