You're staring at an integral. Which means product of two functions. One's a polynomial, the other's an exponential. Or maybe it's ln(x) times something algebraic. Because of that, your first instinct: u-substitution. But the derivative of the inside isn't sitting there waiting for you.
That's when integration by parts shows up. And if you've ever thought "I know the formula but I still get stuck picking u and dv" — you're not alone. The formula is simple. The judgment* is where everyone trips.
Let's walk through it together. Real examples. Real decisions. The stuff textbooks skip.
What Is Integration by Parts
It's the product rule in reverse. That's the whole idea.
You know d(uv) = u dv + v du*. Integrate both sides and rearrange: ∫ u dv = uv - ∫ v du.
That's it. Even so, one formula. But here's what they don't highlight enough: you're trading one integral for another. The goal isn't to apply the formula — it's to make the new integral easier* than the original.
Sometimes you apply it once and you're done. Sometimes twice. Sometimes three times. Sometimes you end up back where you started and solve algebraically. The pattern recognition comes from doing it wrong a few times first.
The LIATE Rule (And Why It's a Guideline, Not Law)
You've probably seen LIATE: Logarithmic, Inverse trig, Algebraic, Trigonometric, Exponential. Pick u from the leftmost category that appears.
It works most* of the time. But not always.
Example: ∫ x sin(x) dx. LIATE says Algebraic before Trigonometric → u = x, dv = sin(x) dx. Correct.
But ∫ x² e^x dx? LIATE says Algebraic before Exponential → u = x². Also correct.
What about ∫ ln(x) / x² dx? That's why lIATE says Logarithmic first → u = ln(x), dv = x⁻² dx. Works beautifully.
The trap: ∫ x √(x+1) dx. LIATE says Algebraic for both. Better: u = √(x+1), dv = x dx. Or just substitute u = x+1 first. But if you pick u = x, dv = √(x+1) dx, you get a messy v. LIATE didn't tell you that.
Use LIATE as a starting heuristic. Then think: which choice makes dv easy to integrate and du simpler?*
Why It Matters
Because a surprising number of integrals only* yield to parts. No substitution works. Here's the thing — no partial fractions. No trig identity.
The classics:
- ∫ xⁿ e^x dx (any polynomial times exponential)
- ∫ xⁿ sin(x) dx or ∫ xⁿ cos(x) dx
- ∫ ln(x) dx, ∫ arcsin(x) dx, ∫ arctan(x) dx
- ∫ e^x sin(x) dx — the one that loops back on itself
And in physics? Work integrals. Moment of inertia. Quantum mechanics expectation values. Laplace transforms. You'll see this formula everywhere.
More importantly: it teaches you how to think about integrals as transformations. Not "find the antiderivative" but "rewrite this into something I recognize." That mindset shift carries into every advanced technique.
How It Works — Step by Step
Let's do this properly. Not just "here's the answer." Here's the decision process*.
Example 1: ∫ x e^x dx
First question: what's u? What's dv?
Options:
- u = x, dv = e^x dx → du = dx, v = e^x
- u = e^x, dv = x dx → du = e^x dx, v = x²/2
Second option gives ∫ (x²/2) e^x dx. First option gives ∫ e^x dx. But harder*. Trivial*.
So: u = x, dv = e^x dx.
Apply formula: uv - ∫ v du = x e^x - ∫ e^x dx = x e^x - e^x + C.
Check: differentiate x e^x - e^x → e^x + x e^x - e^x = x e^x. ✓
Example 2: ∫ x² cos(x) dx
u = x² (algebraic), dv = cos(x) dx → du = 2x dx, v = sin(x)
First pass: x² sin(x) - ∫ 2x sin(x) dx
New integral: ∫ 2x sin(x) dx. Still a product. **Do parts again.
u = 2x, dv = sin(x) dx → du = 2 dx, v = -cos(x)
Second pass: 2x(-cos(x)) - ∫ -2 cos(x) dx = -2x cos(x) + 2 sin(x)
Plug back in: x² sin(x) - [-2x cos(x) + 2 sin(x)] + C = x² sin(x) + 2x cos(x) - 2 sin(x) + C
Notice the alternating signs. That pattern holds for any polynomial times sine or cosine.
Example 3: ∫ ln(x) dx
This one tricks people. "There's only one function!"
Rewrite: ∫ ln(x) · 1 dx
Now you have a product. LIATE: Logarithmic before Algebraic → u = ln(x), dv = dx
du = (1/x) dx, v = x
Formula: x ln(x) - ∫ x · (1/x) dx = x ln(x) - ∫ 1 dx = x ln(x) - x + C
Same trick works for ∫ arcsin(x) dx, ∫ arctan(x) dx, ∫ ln²(x) dx...
Example 4: ∫ e^x sin(x) dx — The Loop
u = e^x, dv = sin(x) dx → du = e^x dx, v = -cos(x)
First: -e^x cos(x) + ∫ e^x cos(x) dx
New integral: ∫ e^x cos(x) dx. Do parts again*.
u = e^x, dv = cos(x) dx → du = e^x dx, v = sin(x)
Second: e^x sin(x) - ∫ e^x sin(x) dx
Wait. That's the original integral* back.
Let I = ∫ e^x sin(x) dx.
Then I = -e^x cos(x) + [e^x sin(x) - I]
I = -e^x cos(x) + e^x sin(x) - I
2I = e^x (sin(x) - cos(x))
I = ½ e^x (sin(x) - cos(x)) + C
This "boomerang" technique appears in differential equations and Fourier analysis constantly. Recognize the pattern: exponential times sine/cosine → two rounds of parts → solve for I.
Example 5: ∫ x³ e^x dx — Tabular Method
When you have a polynomial that differentiates to zero eventually, use the tabular (DI) method. Saves writing.
| D (differentiate) | I (integrate) |
|---|---|
| x³ | e^x |
| 3x² | e^x |
| 6x | e^x |
| 6 | e^x |
| 0 | e^x |
Draw diagonal arrows with alternating signs: +,
+, −, +, −
Multiply diagonally and sum:
+x³·e^x − 3x²·e^x + 6x·e^x − 6·e^x + C
= e^x (x³ − 3x² + 6x − 6) + C
Differentiate to verify: e^x(x³−3x²+6x−6) + e^x(3x²−6x+6) = x³e^x. ✓
The tabular method isn't magic — it's just bookkeeping for repeated parts. If neither hits zero (e^x sin x), you'll loop. Use it when one column hits zero. That's your signal to switch to the algebraic "solve for I" approach.
Definite Integrals: Two Equivalent Paths
∫₀¹ x e^x dx
Path A: Find antiderivative first, then apply FTC. Worth keeping that in mind.
Antiderivative: x e^x − e^x (from Example 1)
Evaluate: (1·e¹ − e¹) − (0·e⁰ − e⁰) = 0 − (−1) = 1
Path B: Apply limits during* the parts formula.
∫ₐᵇ u dv = [uv]ₐᵇ − ∫ₐᵇ v du
u = x, dv = e^x dx → du = dx, v = e^x
[x e^x]₀¹ − ∫₀¹ e^x dx = (1·e − 0) − [e^x]₀¹ = e − (e − 1) = 1
Same result. Path B often avoids sign errors when substituting back into a messy antiderivative. **Default to Path B for definite integrals.
The "Add Zero" Maneuver
Some integrals look* like parts candidates but aren't.
Continue exploring with our guides on is buddhism a universal or ethnic religion and ap spanish language and culture calculator.
∫ x / (x+1)² dx
LIATE suggests u = x, dv = dx/(x+1)². That works but gets messy.
Instead: add zero in the numerator.
x = (x+1) − 1
∫ [(x+1) − 1] / (x+1)² dx = ∫ [1/(x+1) − 1/(x+1)²] dx
= ln|x+1| + 1/(x+1) + C
Algebraic simplification beats mechanical parts every time. Before committing to u and dv, ask: "Can I rewrite this to avoid parts entirely?"
When Parts Fails — And What to Do Instead
1. ∫ e^x² dx
No elementary antiderivative exists. Parts just shuffles the problem. This is the Gaussian integral — solved by polar coordinates trick, not parts.
2. ∫ sin(x)/x dx
The sine integral Si(x). Parts gives ∫ cos(x)/x dx — equally impossible.
3. ∫ x tan(x) dx
Parts: u = x, dv = tan x dx → v = −ln|cos x|. New integral ∫ ln|cos x| dx. No progress.
Recognition: If parts produces an integral harder* than the original, stop. You've hit a non-elementary integral or need a different technique (series, special functions, numerical methods).
The Hidden Pattern: Derivative of a Product
Integration by parts is the product rule integrated.
d(uv) = u dv + v du
∫ d(uv) = ∫ u dv + ∫ v du
uv = ∫ u dv + ∫ v du
∫ u dv = uv − ∫ v du
Every time you use parts, you're reversing a product rule differentiation. This is why it works for:
- Polynomial × exponential/trig (polynomial differentiates to zero)
- Log/inverse trig × polynomial (log/inverse trig differentiates to algebraic)
- Exponential × sine/cosine (derivatives cycle)
If the integrand isn't a derivative of some product, parts won't help.
Common Traps
| Trap | Symptom | Fix |
|---|---|---|
| Wrong u/dv choice | Integral gets harder | Apply LIATE; if still hard, swap |
| Forgetting dx | du or dv missing differential | Write differentials every time* |
| Sign errors in loops | Final answer off by factor of 2 or sign | Track signs in a table; solve for I carefully |
| Dropping +C | Indefinite integral without constant | Add C at the end, not middle |
| Parts on definite integral without limits | Messy back-substitution | Use [uv]ₐᵇ − ∫ₐᵇ v du directly |
Beyond Calculus II
This technique doesn't retire after the final
Beyond Calculus II
1. Integration by Parts in Vector and Integral Calculus
When the integrand lives on a manifold or in higher‑dimensional space, the product rule still governs the differential, but the notation becomes richer.
Line integrals.
For a scalar field (f) and a vector field (\mathbf{g}),
[ \int_{C} f, d\mathbf{g}= \big[ f\mathbf{g}\big]{C}-\int{C}\mathbf{g}, df . ]
If (\mathbf{g}= \nabla \phi) and (df = \nabla f\cdot d\mathbf{r}), the formula reduces to the familiar one‑dimensional version, yet it is indispensable for proving Green’s and Stokes’ theorems.
Surface and volume integrals.
The divergence theorem can be derived by applying the product rule to the triple product ( \mathbf{F}\cdot d\mathbf{S}). Likewise, the integration‑by‑parts identity for functions on a domain (\Omega) with boundary (\partial\Omega),
[ \int_{\Omega} u, dv = \int_{\partial\Omega} u v, \mathbf{n}, dS - \int_{\Omega} v, du , ]
underlies many weak‑solution formulations in partial differential equations.
2. Solving Differential Equations
Many linear ODEs and PDEs are tackled by repeatedly applying integration by parts to move derivatives from one factor to another.
-
Reduction of order.
If (y_1) solves a second‑order linear homogeneous ODE, one can find a second solution (y_2 = y_1 \int \frac{e^{-\int P,dx}}{y_1^2},dx) by integrating by parts on the resulting exact differential. -
Method of variation of parameters.
For a non‑homogeneous equation (y''+p(x)y'+q(x)y = r(x)), the particular solution is expressed as
[ y_p = -y_1 \int \frac{y_2, r}{W},dx + y_2 \int \frac{y_1, r}{W},dx , ]
where each integral is obtained after an integration‑by‑parts step that simplifies the integrand.
- Laplace and Fourier transforms.
The transform of a product (e^{at}f(t)) is (\mathcal{L}{e^{at}f(t)}=F(s-a)). When deriving transform properties, integration by parts moves the exponential factor across the integral sign, yielding the shift theorem and convolution identities.
3. Summation by Parts (Discrete Analogue)
In sequences and series, the discrete counterpart—summation by parts—mirrors the continuous formula:
[ \sum_{k=m}^{n} a_k (b_{k+1}-b_k)= a_{n+1}B_{n+1}-a_m B_m-\sum_{k=m}^{n} B_k (a_{k+1}-a_k), ]
where (B_k=\sum_{j=m}^{k} b_j). This technique is crucial for proving convergence of series, estimating error terms, and handling Abel summation.
4. Integration by Parts for Special Functions
Many special functions are defined through integrals that benefit from parts:
-
Bessel functions.
(J_n(x)=\frac{1}{\pi}\int_0^\pi \cos\bigl(n\tau - x\sin\tau\bigr),d\tau). Differentiating under the integral sign and integrating by parts yields recurrence relations such as (J_{n-1}+J_{n+1}= \frac{2n}{x}J_n). -
Hermite polynomials.
(H_n(x)=(-1)^n e^{x^2}\frac{d^n}{dx^n}e^{-x^2}). Repeated integration by parts connects them to orthogonality integrals (\int_{-\infty}^{\infty} H_m(x) H_n(x) e^{-x^2}dx = \sqrt{\pi}2^n n!\delta_{mn}). -
Beta and Gamma integrals.
The identity (\int_0^1 x^{a-1}(1-x)^{b-1}dx = B(a,b) = \frac{\Gamma(a)\Gamma(b)}{\Gamma(a+b)}) can be derived by integrating by parts, which also provides the functional equation (\Gamma(z+1)=z\Gamma(z)).
5. Numerical Integration and Adaptive Strategies
Even when an analytic antiderivative is unavailable, integration by parts can improve numerical stability. By rewriting an integral as
[ \int u,dv = uv - \int v,du,
Numerical Integration and Adaptive Strategies
Even when an analytic antiderivative is unavailable, integration by parts can improve numerical stability. By rewriting an integral as [ \int u,dv = uv - \int v,du, ] one can choose (u) and (dv) to minimize the growth of (v) or the oscillations of (du), which might otherwise amplify discretization errors. To give you an idea, integrating (\int_0^\infty e^{-x} \sin(x),dx) directly requires handling oscillatory decay, but integration by parts separates the oscillatory term into a manageable form.
In adaptive quadrature, integration by parts helps refine error estimates. But when integrating over a subinterval ([a, b]), the error term for a quadrature rule can be expressed using the remainder (R = \int_a^b f''(x) w(x),dx), where (w(x)) is a weight function. If (w(x)) has a singularity, integrating by parts transfers the singularity to a derivative of (f), potentially simplifying numerical integration if (f) is smooth. This technique is particularly useful in adaptive algorithms that dynamically adjust partitions based on error bounds.
Conclusion
Integration by parts is a versatile tool that bridges theoretical and applied mathematics. Its ability to transform intractable integrals into solvable forms underpins techniques in solving differential equations, analyzing special functions, and designing numerical methods. By shifting derivatives or leveraging discrete analogs like summation by parts, it reveals hidden structures in problems ranging from quantum mechanics to Fourier analysis. Whether simplifying recurrence relations for Bessel functions, deriving orthogonality of polynomials, or stabilizing numerical approximations, integration by parts remains indispensable for mathematicians and scientists seeking elegant solutions to complex problems.
