The formula
There is no single formula here — there is a toolbox. Four tools cover most of what an engineer hits.
Substitution (u-sub): ∫ f(g(x))·g'(x) dx = ∫ f(u) du, u = g(x)
Reading: if the integrand carries its own derivative along for the ride, rename the inside as u and the mess collapses. This is the chain rule run backward.
Integration by parts: ∫ u dv = u·v − ∫ v du
Reading: trade the integral you have for a different one, hoping the new one is easier. This is the product rule run backward. The definite form is ∫[a,b] u v' dx = u(b)v(b) − u(a)v(a) − ∫[a,b] u' v dx.
Partial fractions: P(s)/Q(s) = A/(s−p₁) + B/(s−p₂) + ...
Reading: split one ugly rational function into a sum of simple pieces, one per root of the denominator. Each piece integrates (or inverse-transforms) on sight.
Trig substitution: √(a²−x²) → x = a·sin θ, √(a²+x²) → x = a·tan θ, √(x²−a²) → x = a·sec θ
Reading: a stubborn square root becomes a clean trig identity, because 1 − sin²θ = cos²θ and 1 + tan²θ = sec²θ erase the radical.
Where you meet it
Transfer function back to a time response. This is the daily one. You have
H(s) = (s+3)/((s+1)(s+2)), a Laplace-domain description of a filter, control loop, or circuit. To find how the thing actually behaves in time — the step response on the scope, the settling time in the spec — you run partial fractions to splitH(s)intoA/(s+1) + B/(s+2), then read each term straight off the Laplace table as a decaying exponential. Partial fractions is the bridge from the s-plane back to the real world.RMS and power on the bench. Computing the RMS value of a waveform means integrating its square over a period. For a rectified sine or a triangle wave, the integrand carries a
sin²or a product term, and you reach for a trig identity or integration by parts to get a closed number instead of a numerical estimate. See the crest-factor and rms-peak tools.Energy under a curve at a design review. Impulse from a thrust-time trace, charge from a current profile, area under a PSD. When the curve has a clean analytic form, a substitution turns the integral into a one-liner you can defend on a slide instead of a spreadsheet nobody trusts.
Antenna and field problems. Radiation integrals and inductance calculations routinely produce
1/√(a²+x²)forms. Trig substitution is what turns those into arctangents and logs you can actually evaluate.
How it works
The whole game is pattern-matching. Each technique is a lock, and the integrand tells you which key to grab.
Substitution is the first thing to try, always. Look for a chunk whose derivative also appears in the integrand, up to a constant. ∫ 2x·e^(x²) dx — the 2x is the derivative of x², so u = x² and it becomes ∫ e^u du = e^u. The mistake people make is forgetting to convert the limits on a definite integral, or dropping the du scaling factor. If your du doesn't account for every stray constant, the answer is off by that constant.
Integration by parts is for products where one factor gets simpler when you differentiate it. ∫ x·e^x dx: let u = x (differentiates to 1, simpler) and dv = e^x dx. You get x·e^x − ∫ e^x dx = (x−1)·e^x. Checked numerically, ∫[0,1] x·e^x dx = 1.000 exactly, and (x−1)e^x evaluated 0 to 1 gives 0 − (−1) = 1. The classic guide for picking u is LIATE — logs, inverse trig, algebraic, trig, exponential — differentiate whichever comes first. The trap is picking u and dv backward and landing on a harder integral than you started with. If parts makes it worse, swap your choice.
Partial fractions only works on a proper rational function — numerator degree strictly below denominator degree. If it isn't proper, do polynomial long division first; the quotient integrates trivially and the remainder is proper. Then factor the denominator. Distinct real roots each give one A/(s−p) term; a repeated root of multiplicity m needs terms for every power up to m; an irreducible quadratic gives a (Bs+C)/(quadratic) term that becomes a log plus an arctangent. Sanity check with 1/((s+1)(s+2)) = 1/(s+1) − 1/(s+2), whose inverse Laplace transform is e^(−t) − e^(−2t) — the exact time response you'd expect from two first-order poles. The gotcha engineers hit: complex-conjugate poles. Those are the oscillatory modes, and if you force them into real-only terms you lose the ringing. Keep the quadratic intact and you get the damped sinusoid.
Trig substitution is the specialist tool for radicals of the form √(a²±x²) or √(x²−a²). Pick the sub that turns the radical into a perfect square via a Pythagorean identity, integrate in θ, then draw the reference triangle to convert back to x. ∫ 1/√(1−x²) dx = arcsin(x) is the canonical result and the reason arcsin shows up in geometry-heavy integrals at all. The limit of validity: watch the domain. x = a·sin θ only makes sense for |x| ≤ a, and sec substitutions carry sign traps when x goes negative. Get the domain wrong and you get a real answer to an integral that was supposed to be undefined.
The meta-lesson: these are not four unrelated tricks. Substitution and parts are the chain and product rules read backward — every integration technique is a differentiation rule you already trust, running the other direction. That's why differentiation is mechanical and integration is an art. There is no algorithm that always terminates; some integrands, like e^(−x²), have no elementary antiderivative at all, and no amount of cleverness changes that. Knowing when to stop and switch to a numerical method is part of the skill.
History
The backward-running nature of these tools is why they arrived piecemeal, alongside calculus itself. Gottfried Leibniz worked out the substance of integration by parts in unpublished manuscripts of 1675, deriving it from a geometric argument — his notes record the equivalent of omn. xl = x omn. l − omn. omn. l — while he was still inventing the notation, writing "omn." where the integral sign would soon go [4]. The technique reached print through Brook Taylor, the English mathematician born in 1685, in his 1715 treatise Methodus Incrementorum Directa et Inversa — the same slim book that introduced the Taylor series and the calculus of finite differences [1][2]. Taylor is generally credited with inventing integration by parts as a stated method, packing it in among a change-of-variables formula and singular solutions of differential equations [2]. The book's importance went underappreciated for decades; Lagrange didn't call Taylor's theorem the foundational principle of differential calculus until 1772 [2].
Partial fractions has an even tidier origin story. The decomposition was worked out independently around 1702 by Leibniz and by Johann Bernoulli, both chasing the same prize — a general way to integrate rational functions [3]. That is exactly the problem it still solves: break P(x)/Q(x) into simple pieces and every piece integrates to a log, an arctangent, or a power [3]. Two centuries later, when Oliver Heaviside built his operational calculus for telegraph and circuit problems, partial fractions became the workhorse for inverting transforms — the ancestor of the s-plane-to-time-response move every controls engineer runs today.
Related tools
- /tools/rc-filter
- /tools/rc-charge-time
- /tools/q-factor-bandwidth
- /tools/lc-resonance
- /tools/rms-peak
- /tools/crest-factor
- /tools/capacitor-energy
- /tools/inductor-energy