Rule Of Partial Fraction

Partial fractions are just algebra in reverse

You take a rational expression and break it apart. That's it. The Rule Of Partial Fraction is a decomposition technique, not a theorem you need to memorize the proof of. Most people get stuck because they're trying to derive everything instead of learning the pattern-matching game. I've been doing these integrals since before Laplace transforms were the default answer to every engineering problem on a midterm. The basic idea: you have a fraction like (3x + 2)/(x^2 + x - 2) and you want to split it into simpler pieces. You factor the denominator, set up your unknown coefficients, and solve. The devil is always in the denominator.

The Rule Of Partial Fraction for Different Denominator Types

There are three cases you actually encounter in practice. Everything else is just a combination of these. Distinct linear factors go first - this is where you set up A/(x-a) + B/(x-b) + C/(x-c) and so on. Then you have repeated linear factors, which most textbooks handle fine but nobody warns you about the algebra mess that follows. And then irreducible quadratics, which is where things start eating your afternoon. The cover-up method works for distinct linear factors if you're quick. You cover up the factor you want the coefficient for, plug in the root, and what's left is your constant. It's fast until you need to show work for a professor who wants the full system of equations. Then you multiply everything through by the original denominator and equate coefficients. Both approaches give the same answer. The system-of-equations approach is less likely to make you miss a sign. I ran into a case last semester where someone gave me (2x^3 + 5x^2 - 4x - 3)/(x^2 - 4x - 5). The degree of the numerator was higher than the denominator, which means it's an improper fraction and partial fractions don't apply yet. Had to do polynomial long division first, got 2x + 13 plus a remainder term, and only then could I decompose the proper fraction part. Took about ten minutes that shouldn't have taken more than two because I kept second-guessing whether I needed to divide first. The rule is simple: if the numerator degree is greater than or equal to the denominator degree, divide first. End of story.

Setting Up the Decomposition

You always start by factoring the denominator completely. This is where most mistakes happen because people factor partially and then wonder why their system doesn't close. Take x^3 - x. That's x(x-1)(x+1), not x(x^2-1). The second form leaves a reducible quadratic sitting in there waiting to cause problems. For each linear factor, you get a constant over that factor. For each repeated linear factor to power n, you get n terms: A1/(x-r) + A2/(x-r)^2 + ... + An/(x-r)^n. For each irreducible quadratic factor, you get a linear numerator: (Bx + C) over that quadratic. The linear numerator over a quadratic part is the one beginners consistently botch. They write just a constant C when the numerator needs to be Bx + C because the quadratic itself could produce an x term when you multiply back through. I spent three weeks debugging a control systems lab report where my transfer function decomposition kept giving me wrong time-domain responses. The circuit was fine, the simulation was fine, and I'd written just a constant over an irreducible quadratic term instead of Bx + C. The arithmetic looked clean going through the bookwork but the inverse transform produced a shifted exponential that didn't match any physical behavior in the system. Changed it to a linear numerator, solved for both coefficients, and the response matched perfectly on the first try.

Solving for the Coefficients

You have two real options here. The first is substituting strategic values of x that zero out terms. Pick the roots of your linear factors and each one eliminates everything except the term you care about. This is fast and elegant when the numbers cooperate. The second is expanding everything and matching coefficients of like powers. This is slower but guaranteed to work regardless of how ugly the constants are. The hybrid approach is what I actually use. Cover up each linear factor to get those constants directly, then pick one or two easy x values to generate equations for the remaining unknowns. For a problem with distinct linear factors and one irreducible quadratic, you might get three constants from substitution and then use x = 0 and x = 1 to solve the last two. Saves you from setting up and solving a full 5x5 system when three of the five are immediate. When you have repeated factors, substitution still works for the highest-power term in that cluster. You cover up the highest power, plug in the root, and get that coefficient. The lower-power terms in the same cluster require either more substitution values or coefficient matching. There's a recursive formula for this but it's obscure enough that remembering it costs more mental overhead than it saves in computation time.

Common Pitfalls

The biggest mistake I see is skipping the improper fraction check. If you try to decompose (x^2 + 1)/(x - 1) directly, you'll get garbage because the method assumes a proper rational function. The second mistake is incomplete factorization - leaving (x^2 - 4) as a single quadratic factor instead of splitting it into (x-2)(x+2). The third is forgetting the x term in the numerator of irreducible quadratic pieces. You'll solve for C correctly but miss B entirely, and your final answer will look plausible until you integrate it and compare against a known result. Another edge case: when the denominator has no real roots at all, like x^4 + 1. This factors into two irreducible quadratics over the reals, but finding those factors requires either knowing the sum/difference of squares trick or accepting complex factors and then pairing them back up. For most coursework this is rare. For actual work, you'd typically use numerical methods or leave it in a different form entirely. This technique also breaks down when you need high numerical precision. The coefficient-matching system becomes ill-conditioned for high-degree polynomials with closely spaced roots. In those cases, residue-based methods or direct numerical decomposition give better results. I've seen this bite people working on filter design where the analytical answer was correct but the numerical implementation drifted because the partial fraction coefficients were orders of magnitude different from each other.

Quick Worked Example

Take (5x^2 - 2x - 19)/((x+3)(x+1)^2). The denominator has a distinct linear factor and a repeated linear factor. The decomposition form is A/(x+3) + B/(x+1) + C/(x+1)^2. Multiply through by the denominator: 5x^2 - 2x - 19 = A(x+1)^2 + B(x+3)(x+1) + C(x+3). Set x = -3: 45 + 6 - 19 = A(-2)^2, so 32 = 4A and A = 8. Set x = -1: 5 + 2 - 19 = C(2), so -12 = 2C and C = -6. For B, pick x = 0: -19 = A + 3B + 3C, so -19 = 8 + 3B - 18, which gives 3B = -9 and B = -3.

Check: 8/(x+3) - 3/(x+1) - 6/(x+1)^2. Multiply back through and you get the original numerator. The algebra is straightforward once you know which x values to plug in and which form to assume for the decomposition.

When the Rule Of Partial Fraction Doesn't Help

There are rational functions where partial fractions don't simplify your life. If you're doing numerical integration rather than symbolic work, the decomposition adds computation without removing complexity. If the denominator doesn't factor nicely over the reals and you need a numerical result, just use a quadrature rule. And if you're working with functions that have essential singularities or branch cuts, this method doesn't apply at all - you'd need Laurent series or contour integration instead. The real value shows up when you're integrating rational functions or finding inverse Laplace transforms. In both cases, the decomposed form maps directly to standard integral tables and transform pairs. That's why you learn it. It's not interesting as an algebra exercise. It's a preparation step for something that comes after.