Where the last guide left you
Guide 3 showed the trick that makes the whole machine worthwhile: the transform of a derivative swaps differentiation for multiplication by s and — this is the gift — pulls the initial values y(0) and y'(0) straight into the algebra. So when you apply the Laplace transform to a linear constant-coefficient equation, the differential equation collapses into an ordinary equation in the single unknown Y(s), the transform of the answer. Solve that with grade-school algebra and you get Y(s) sitting alone on the left. The hard part is over — except you are now stranded in the s-world holding Y(s), not y(t).
That stranding is the problem this guide solves. The whole point of the transform was to escape calculus into algebra, do the easy work there, and then come home to a function of t. The trip home is the inverse Laplace transform: given Y(s), recover the unique y(t) whose transform it is. In principle the inverse is a contour integral in the complex plane (the Bromwich integral), which sounds terrifying. In practice you almost never touch that integral. Instead you reshape Y(s) until every piece matches an entry in the same transform table you already used going forward — and read each piece backwards.
Why partial fractions are the natural key
Here is the lucky structural fact that makes everything click. When the original equation is linear with constant coefficients, the Y(s) you solve for is always a rational function — a ratio N(s)/D(s) of two polynomials. The denominator D(s) is essentially the characteristic polynomial of the equation (plus whatever the forcing function contributed), and the numerator carries the initial data and the forcing. A whole rational function rarely matches a single table row. But its building blocks do.
Partial fraction decomposition is the algebra that smashes a complicated rational function into a sum of simple fragments — exactly the fragments the table speaks. A factor like (s - a) in the denominator contributes a term A/(s - a), whose inverse is the exponential A e^(at). A repeated factor (s - a)^2 contributes A/(s - a)^2, whose inverse brings in a t e^(at). An irreducible quadratic like (s - p)^2 + w^2 contributes the sine-and-cosine pieces of a damped oscillation. The denominator's factorization is the solution's anatomy written in advance.
denominator factor in D(s) partial-fraction piece inverse -> y(t) -------------------------------- ---------------------------- ------------------- (s - a) simple A / (s - a) A e^(at) (s - a)^2 repeated A/(s-a) + B/(s-a)^2 A e^(at) + B t e^(at) (s-p)^2 + w^2 complex pair (B s + C) / ((s-p)^2 + w^2) e^(pt)(... cos wt + ... sin wt)
Setting up and cracking the decomposition
The mechanics are worth doing slowly once so they become automatic forever. Suppose after transforming you reach Y(s) = (3s + 5) / ((s + 1)(s + 3)). The denominator is already factored into two distinct simple roots, so you write Y(s) = A/(s + 1) + B/(s + 3) and hunt for the numbers A and B. There are two honest ways to find them: clear the denominators and match coefficients, or use the cover-up shortcut for simple roots. Both give the same answer; pick whichever you trust.
- Factor the denominator completely, over the reals: distinct linear factors, repeated linear factors, and irreducible quadratics. This factoring IS finding the poles — the s-values where Y(s) blows up.
- Write the template: one unknown over each simple factor, and for a factor raised to power k, one term for each power 1 through k (so (s - a)^2 needs both A/(s - a) and B/(s - a)^2).
- Solve for the constants. Cover-up: to get A, cover the (s + 1) factor in Y(s) and evaluate the rest at s = -1, giving A = (3(-1) + 5)/(-1 + 3) = 1. Likewise B = (3(-3) + 5)/(-3 + 1) = 2.
- Read each fragment back through the table. A/(s + 1) = 1/(s + 1) inverts to e^(-t); B/(s + 3) = 2/(s + 3) inverts to 2 e^(-3t). Sum them: y(t) = e^(-t) + 2 e^(-3t).
Completing the square and the shifting theorem
Real equations love to produce complex roots — every underdamped oscillation does. When the denominator has an irreducible quadratic factor, do not force it over the complex numbers; instead complete the square to write it as (s - p)^2 + w^2. That single rewrite is what unlocks the oscillatory table rows, because the table knows s/(s^2 + w^2) inverts to cos(wt) and w/(s^2 + w^2) inverts to sin(wt). The p that pops out of the square is the decay (or growth) rate; the w is the ringing frequency. You can practically read the qualitative behaviour off the completed square before computing a thing.
The p inside (s - p)^2 + w^2 is handled by the first shifting theorem (the s-shift), and it is worth savouring how clean it is. The theorem says: replacing s by s - p in any transform corresponds to multiplying the time function by e^(pt). So once you can invert the centered piece — say w/(s^2 + w^2) gives sin(wt) — the shifted piece w/((s - p)^2 + w^2) simply gains a factor: e^(pt) sin(wt). All the messy-looking damping is just that one exponential envelope sliding in. Shifting in s is multiplying by an exponential in t; that pairing is the heartbeat of the whole table.
So the location of the poles in the s-plane tells the whole story before you invert. A pole on the negative real axis means a decaying exponential; a complex pair p ± iw means an oscillation e^(pt) with frequency w; the real part p decides whether that oscillation dies out (p < 0), holds steady (p = 0), or grows (p > 0). This is the engineer's habit of staring at the poles and predicting stability without ever returning to t. The partial fraction step is, quietly, a pole-by-pole reading of the system's destiny.
How it snaps into the full algorithm
Zoom out and see where this guide lives. The Laplace master algorithm has three moves: transform the equation (forward table plus the derivative rule with its built-in initial conditions), solve the resulting algebraic equation for Y(s), then invert. Everything in this guide is that third move. Guide 5, the finale, walks a complete initial value problem end to end and you will watch partial fractions do exactly this job in context — the inverse step is never the bottleneck once the decomposition is set up.
It is worth being honest about the boundaries. This clean partial-fraction route works precisely because the equation is linear with constant coefficients — that is what guarantees Y(s) is rational. The moment coefficients vary with t, or the equation goes nonlinear, Y(s) is no longer a tidy ratio of polynomials and table-and-partial-fractions stops being the answer; those problems lean on power series, numerical methods, or qualitative analysis from earlier rungs. And even in the linear world, a forcing function whose own transform is not rational (or one with switches and impulses) pushes you toward the second shifting theorem and convolution, which the next rung takes up.
One reassuring sanity check ties it back to everything before this rung. The terms partial fractions hand you — e^(at), t e^(at), e^(pt) cos(wt), e^(pt) sin(wt) — are exactly the building blocks you met when solving constant-coefficient equations the old way, by the characteristic equation. That is no coincidence: the denominator D(s) you are factoring is the characteristic polynomial in disguise. Laplace did not replace the old method; it repackaged it so the initial conditions ride along for free and the bookkeeping becomes pure algebra.