How the FTC Actually Works (And Why It Feels Like Magic Until It Doesn't)
The fundamental theorem of calculus says two things. The first part says if you have a continuous function on a closed interval and you build an area function out of it, that area function is differentiable and its derivative is the original function. The second part says you can evaluate a definite integral by finding any antiderivative and subtracting. That's it. Everything else is bookkeeping. I first ran into real trouble with this when I was checking homework for a remedial calc section. A student had integrated |x| from -2 to 2 by blindly applying Part 2 using x²/2 as the antiderivative and getting zero. The graph is a V shape sitting entirely above the axis, so the answer should be 4, not 0. The issue wasn't that the theorem was wrong, it was that the student was using an antiderivative that didn't actually work across the whole interval because the function wasn't smooth at zero. I showed them how to split the integral at the kink, which is the standard fix, but what they really needed was to understand when Part 2 is even allowed to apply.
Fundamental Theorem Of Calculus For Dummies
Let me put this in order without the textbook preamble. Part 1 is the constructibility guarantee. Given f continuous on [a,b], define F(x) = the integral from a to x of f(t) dt. Then F'(x) = f(x) for every x in (a,b). This is what makes the whole enterprise possible, because it tells us that differentiation and integration undo each other in at least one direction. Part 2 is the computation tool. If F is any antiderivative of f on [a,b], then the integral from a to b of f(x) dx equals F(b) minus F(a). You don't need to sum Riemann partitions. You find an antiderivative and plug in the endpoints. The thing nobody emphasizes enough is that Part 2 requires f to be integrable and F to be differentiable with F' = f everywhere on the interval. Students treat the second part like a universal key. It isn't. It fails in ways that look harmless until you get the wrong answer and can't figure out why. Here are the cases I see most often in practice. First, discontinuities inside the interval. If f has a jump at some point c between a and b, you can't just write down an antiderivative and evaluate endpoints. The function might still be integrable, but any antiderivative you write down needs to be checked at c. A simple example is the sign function on [-1,1]. The integral is zero because the positive and negative parts cancel, but if you use the Heaviside step function as your mental model for the antiderivative, you'll get confused about what value to assign at the discontinuity. The fix is always the same: split the integral at every discontinuity and handle each piece separately.
Second, infinite discontinuities. The improper integral of 1/sqrt(x) from 0 to 1 converges, but the integral of 1/x from 0 to 1 diverges. Both functions have the same type of bad behavior at zero. The difference is how badly. The power rule for integration only works when the exponent is not -1. When you see 1/x, your antiderivative is ln|x|, and that blows up at zero. Part 2 can't rescue you here without a limit process, and the limit process is where the convergence test lives, not the theorem itself. Third, functions that are integrable but don't have elementary antiderivatives. The integral of e^(-x²) from 0 to 1 is a perfectly well-defined number, roughly 0.7468. There is no closed-form antiderivative using standard functions. Part 2 still applies in principle, but you can't use it computationally without special functions like the error function or numerical quadrature. This is the case where the theorem is true but useless for hand calculation, and students get tripped up because they expect every integral to yield a nice formula. Here's a counter-intuitive point that shows up in exams. The antiderivative is not unique. If F is an antiderivative of f, then F(x) + C is also one for any constant C. Part 2 works regardless of which C you pick because the constants cancel in the subtraction F(b) - F(a). I tell my students to pick C = 0 and move on, but I've seen people waste ten minutes checking whether their arbitrary constant matters. It doesn't. The theorem is designed so that it can't matter.
Get the Full Details
Another thing that trips people up: Part 1 does not require f to be differentiable. It only requires f to be continuous. The resulting F is always differentiable, but f itself can be quite rough. Consider f(x) = |x| on [-1,1]. It's continuous everywhere, so Part 1 applies perfectly. The area function F(x) = integral from 0 to x of |t| dt is differentiable and its derivative is |x|. But f itself has a corner at zero. The theorem doesn't care. This is useful to remember because when you're trying to find an antiderivative by working backward from a derivative, the smoothness of the original function is not a requirement, only the continuity is. I once spent an afternoon debugging a student's program that computed definite integrals numerically and then compared the result against symbolic antiderivatives using Part 2. The mismatch came from a removable discontinuity. The function was defined differently at a single point than its limit, which meant the symbolic antiderivative assumed the continuous version while the numerical integrator picked up the actual value at that point. The difference was negligible for most purposes, but the symbolic and numerical answers disagreed by a tiny amount that looked like a bug. The fix was to make the function definition consistent with its continuous extension before applying either method. This is worth knowing because when you're doing real work with CAS tools, they'll give you an antiderivative that implicitly assumes continuity even if your input function has isolated discontinuities. The practical takeaway is simpler than the edge cases suggest. For the FTC to apply cleanly, make sure your function is continuous on the interval you're integrating over. If it's not, split the interval at every problem point and check whether each piece converges. If the function is continuous but has no elementary antiderivative, switch to numerical methods or special functions. If the function has an infinite discontinuity, test convergence with a p-test or comparison before attempting Part 2.
Most calculus courses spend two weeks on this theorem and then move on, but the amount of trouble people get into from misapplying it is disproportionate. The theorem itself is straightforward. The constraints around it are where the actual work lives.
