Integration by Substitution Fails at the Differential, Not the Algebra

If an integral means "run the antiderivative machine on this function", then $dx$ is the full stop at the end of a sentence, with nothing to convert. If it means "add up many thin products of a height and a width", then $dx$ is the width: one of the two factors. Change the variable and the width changes with it.

When 150 calculus students at two US colleges were asked in writing what a definite integral means, explanations about the area under a curve and about undoing a derivative came up often, while explanations about adding up many small pieces came up much less often^[1]. If the summing picture is rare, treating $dx$ as a factor is rare too.

Take $\int 2x\cos(x^2)\,dx$ and set $u = x^2$. Substituting inside the cosine and nothing else produces

$$\int 2x\cos(u)\,dx$$

Before reading on, say in your own words what that line changed.

The symbol $u$ is not a new free quantity; it is $x^2$ wearing a hat, so the line says what the original said. Worse, it invites you to treat $u$ as a constant and integrate $2x$ to get $x^2\cos(u)$, which is $x^2\cos(x^2)$. Differentiate that back and you get $2x\cos(x^2) - 2x^3\sin(x^2)$. The second term was never in the problem.

The complete version converts both halves. From $u = x^2$ comes $du = 2x\,dx$, already in the integrand, so the integral becomes $\int\cos(u)\,du = \sin(u) + C = \sin(x^2) + C$. Check by differentiating: $\frac{d}{dx}\sin(x^2) = 2x\cos(x^2)$.

Explaining the broken line is what does the work. In one experiment with about two hundred middle school algebra students, those shown a worked-out wrong solution and asked to explain what had gone wrong got better at solving equations, and those who understood the least at the start benefited more than the others^[2]. That was algebra, not integration. So write the correct line directly under the wrong one.

Now take $\int x\cos(x^2)\,dx$, with two candidates on offer. With $u = x^2$ we get $du = 2x\,dx$, so $x\,dx = \frac{1}{2}\,du$, and the integral becomes $\frac{1}{2}\int\cos(u)\,du = \frac{1}{2}\sin(x^2) + C$. Check it: $\frac{d}{dx}\left[\frac{1}{2}\sin(x^2)\right] = x\cos(x^2)$.

With $u = \cos(x^2)$, the biggest-looking piece, we get $du = -2x\sin(x^2)\,dx$. There is no $\sin(x^2)$ in the integrand, so it has nothing to attach to.

Reaching for the largest piece is not carelessness. In a set of experiments in physics, people still learning the subject grouped problems by how they looked on the surface, while experienced solvers sorted the same problems by the principle that would solve them^[3]. That study was about physics problems, not integrals. Size is visible on the page; "its derivative is also present" is not.

The drill that trains this has no integration in it. For each integrand, name one candidate $u$, decide whether $du$ is already there up to a constant multiple, and stop.

• $\int \frac{\ln x}{x}\,dx$: try $u = \ln x$; $du = \frac{1}{x}\,dx$ is there. Yes.
• $\int e^{x^2}\,dx$: try $u = x^2$; $du = 2x\,dx$ needs an $x$ the integrand lacks. No.
• $\int \frac{x}{1+x^2}\,dx$: try $u = 1+x^2$; $du = 2x\,dx$ is there up to $\frac{1}{2}$. Yes.
• $\int x\ln x\,dx$: try $u = \ln x$; $\frac{1}{x}\,dx$ is absent. No; this wants parts.

All four take under a minute, and none ends in an antiderivative. In studies with middle and high school students, practice at seeing the structure of equations across transformations, without ever solving one, was followed by large gains in how quickly they could then solve equations^[4]. That was school algebra, not calculus, and spotting a usable substitution asks more than spotting an equation's shape. What carries over is narrower: the decision gets practiced on its own. When we build practice sets at Learn4Less, a recognition round comes first.

1. Is any $x$ left? If so the substitution is unfinished: you cannot integrate with respect to $u$ while $x$ remains.
2. Did the differential change? You need $du = g'(x)\,dx$ because $dx$ is the other factor in the product being summed, not a label naming the variable.
3. Did you return to $x$? An indefinite integral answers a question about $x$, so $\sin(u) + C$ is not yet an answer; $\sin(x^2) + C$ is.

A definite integral adds a fourth check: the original bounds were positions on the $x$-axis, and they stop meaning anything once $u$ takes over. On $\int_0^1 2x\cos(x^2)\,dx$ the bounds travel with the substitution: $x=0$ gives $u=0$, $x=1$ gives $u=1$, and the value is $\sin(1)$. Keep the old bounds and you are feeding $x$-values to a function of $u$; that mistake has its own post on substitution bounds.

Attaching a reason is not decoration. A review of research on mathematics learning found that knowing how to carry out a procedure and knowing why it works each build the other up^[5]. The same review found that hardly anyone has tested whether teaching the idea first beats teaching the steps first, so there is little evidence either way^[5].

Why did this work all week and collapse today? A substitution chapter is a pile of substitution problems. You never had to ask whether $du$ was present; the heading had promised it.

A review of the research on how practice sets are ordered makes the same point generally: when every question uses one method, students never work out which method a question needs^[6]. Experiments comparing mixed sets against one-method sets have favored mixing, though that review's author says the evidence was limited and little of it came from real classrooms^[6]. One study did run in a school: 126 seventh-graders given the very same problems over three months scored higher on a surprise test a month later when the problems had been mixed up rather than grouped by kind^[7].

So build the mixed set yourself: ten integrals from across the chapter with the headings hidden, some falling to substitution, some needing parts, some partial fractions, some only a rewrite. That trial was seventh-grade mathematics, not calculus, so treat it as a reason to try mixing rather than a promise about your next test^[7].

None of the seven sources looked at integration by substitution^{[3][1][6][7][4][2][5]}. The nearest one asked those 150 US calculus students to explain in writing what a definite integral means, not to carry one out^[1]. The rest came from physics undergraduates, seventh-graders, and middle and high school algebra students^[3][7][4][2]. Carrying those rooms over to your calculus page is my argument, not a measured result.

Two of the seven sources are reviews rather than experiments, and one says outright that its evidence was limited and classroom studies were still needed^[6][5]. Nothing here tells you how often the differential goes missing, whether recognition drills built from integrands would help the way equation drills did, or whether explaining a wrong substitution beats studying a correct one^[6][4][2].

The mathematics sits elsewhere. If $u = g(x)$ then $du = g'(x)\,dx$, and the whole rule is

$$\int f(g(x))g'(x)\,dx = \int f(u)\,du$$

Your integrand has to match that left-hand side up to a constant multiple, with anything left over rewritable in $u$; otherwise the substitution does not close. Every algebraic claim above can be checked by differentiating the answer, which needs no outside evidence.