These are the exact fragments the model serves — also available as an
ordered study guide.
Running the derivative backwards
A thin vertical strip of width delta-x and height f(x) sweeps rightward under a curve; each strip's area adds to the shaded region already swept, which is the accumulated total building up as x moves from 0 to 2.
A derivative asks how fast something is changing right now. An integral asks the reverse question: given how fast something has been changing, how much has piled up in total. Feed a derivative a position and it hands back a velocity; feed an integral a velocity and it hands back the distance traveled — the two operations undo each other, which is most of what makes calculus feel like one connected subject instead of two separate ones.
Picture a rate plotted as a curve — speed against time, say. At every instant, the height of the curve is how fast the quantity is growing. Multiply a tiny sliver of time by the height of the curve during that sliver, and the result is how much accumulated during that sliver; add up every sliver across the interval, and the total is the accumulated quantity, like an odometer accumulating miles from speed: it never measures distance directly, it just keeps adding up how far you traveled during each passing moment. Geometrically, that summed total turns out to be nothing more exotic than the area between the curve and the horizontal axis.
None of this needs anything beyond what a function, a slope, and a limit already are: a function still maps an input to an output, a derivative is still the slope of a curve at a point, and a limit is still what a quantity approaches as some other quantity is pushed toward an edge. Integration reuses all three ideas and points them in the opposite direction — backward, from rate to total, instead of forward, from position to rate.
Approximating area with rectangles
Four right-endpoint Riemann-sum rectangles under the curve y equals x squared on the interval from 0 to 2: each rectangle's height is the curve sampled at the right edge of its width-0.5 slice, giving heights 0.25, 1, 2.25 and 4, whose total of 3.75 approximates the curve's exact area of 8/3.
The cleanest way to estimate an area under a curve is also the crudest: chop the interval into equal-width pieces, stand a rectangle on each piece, and add up the rectangles' areas. Cut the interval from a to b into n equal pieces and each piece has width Δx = (b − a)/n; the height of each rectangle comes from sampling the function once somewhere inside its own piece — like estimating the area of an odd-shaped plot of land by laying down a row of thin, easy-to-measure rectangular strips and adding up their areas.
Where inside each piece you sample matters, at least until the pieces get thin. Sampling at the left edge of each piece gives a left-endpoint sum, sampling at the right edge gives a right-endpoint sum, and sampling in the middle gives a midpoint sum; on a curve that is rising the whole way, the left sum comes in low, the right sum comes in high, and the true area sits somewhere between them. For the curve f(x) = x² on the interval from 0 to 2, splitting into four equal pieces of width 0.5 and sampling at each piece's right edge gives heights of 0.25, 1, 2.25, and 4; multiplying each height by 0.5 and adding them up totals 3.75.
Formally, a Riemann sum adds up f(cᵢ) · Δxᵢ over every piece i, where cᵢ is whatever sample point the rule picked inside piece i. Bernhard Riemann gave this idea its rigorous form in 1854, defining the exact area as whatever the sum settles down to as the number of pieces grows without bound and each piece's width shrinks toward zero — the cruder the chop, the worse the estimate, and the finer the chop, the closer the rectangles hug the curve.
The definite integral as a limit
Push the Riemann-sum idea to its limit and the rectangles stop being an approximation at all. As the number of pieces n grows without bound and each piece's width shrinks toward zero, the sum of rectangle areas settles on one exact number — the definite integral of f from a to b, written ∫[a,b] f(x) dx. The ∫ symbol traces back to an unpublished 1675 manuscript of Gottfried Leibniz, who first wrote it privately before the notation appeared in print in 1686; it began as a stretched letter s, standing for summa, sum.
Every definite integral names three things: the function being accumulated, f, and the two endpoints of the interval, a and b, called the limits of integration. The dx marks which variable is sweeping across the interval — the same role Δx played among the rectangles, shrunk down to an infinitesimal width. A definite integral evaluates to a single number, not a family of functions, which is the detail that separates it most sharply from the antiderivative it will turn out to equal.
For the curve f(x) = x² on the interval from 0 to 2, the exact value is ∫[0,2] x² dx = 8/3, a little over 2.667 — squarely between a right-endpoint rectangle estimate of 3.75 and a left-endpoint estimate of 1.75, and closer still to a midpoint estimate of 2.625. Finer partitions bring every one of those estimates nearer to the same exact number.
When the curve dips below the axis
The line f(x) = x minus 1 crosses the x-axis at x=1 between x=0 and x=2: the triangular region above the axis from x=1 to x=2 has area 0.5 and counts as positive, the triangular region below the axis from x=0 to x=1 has area 0.5 and counts as negative, and the definite integral sums them to a net signed area of 0.
Every rectangle in a Riemann sum has a sign as well as a size. When the function is negative on some piece, its sampled height is a negative number, so that rectangle's area contributes a negative amount to the running total — like a running ledger where time spent above a baseline credits the total and time spent below it debits the total, so the final balance can land above zero, below zero, or exactly on it. A definite integral therefore does not measure the amount of ink between a curve and the axis; it measures net signed area, the area above the axis minus the area below it.
Take the line f(x) = x − 1 on the interval from 0 to 2. It sits below the axis from x = 0 to x = 1, tracing out a triangle of area 0.5 below the line y = 0, and above the axis from x = 1 to x = 2, tracing out a matching triangle of area 0.5 above it. The definite integral ∫[0,2] (x − 1) dx is not 1, the sum of the two triangle areas — it comes out to 0.5 − 0.5 = 0, because the below-axis triangle contributes negatively.
That cancellation is not an edge case to shrug off; it is the definition working as intended. A definite integral can come out positive, negative, or exactly zero depending on how much the curve spends above the axis versus below it over the interval in question, and reading a definite integral as area alone, with no attention to sign, silently discards exactly the information that made the below-axis region distinct from the above-axis one.
Undoing the derivative
Three curves that are vertical shifts of one another by a constant C: at every x they share the identical tangent slope, shown here at x=1, and the vertical gap between any two of them stays the same constant distance everywhere, illustrating that antiderivatives of the same function differ only by C.
An antiderivative of a function f is any function F whose derivative is f — differentiate F and f is what comes back out. Finding an antiderivative is running the derivative in reverse: instead of asking for the slope of a known curve, it asks which curve has a given slope everywhere.
That question never has just one answer. If F is an antiderivative of f, so is F plus any constant, because adding a constant to a function never changes its slope at any point. The indefinite integral ∫f(x) dx names this entire family at once and is written F(x) + C, where C stands for an unspecified constant — every member of the family is a vertical shift of every other member, all sharing the identical slope at each x.
The +C is not decoration; dropping it silently claims there is one specific antiderivative rather than an entire family of them, which is false for every function that has one. It is also the clearest line between the two kinds of integral: an indefinite integral is a family of functions with an unresolved constant, while a definite integral, bounded by two limits of integration, collapses to a single number with no constant left in it.
The bridge: the Fundamental Theorem of Calculus
The accumulation function F(x), the running area under f from 0 to x, has a tangent slope at each point equal to the height of f there: at x=1.5, F's tangent slope and f's height are both 1.5.
Riemann sums define the definite integral, and antiderivatives undo the derivative — the Fundamental Theorem of Calculus is the fact that these two separately built ideas are the same thing seen from different sides. It comes in two parts, and both carry the same condition: f must be continuous, an unbroken curve with no jumps or holes on the interval in question.
The first part says that if f is continuous, the accumulation function F(x) = ∫[a,x] f(t) dt — the running total of area swept out from a fixed starting point a up to a moving point x — is itself differentiable, and its derivative is f(x). In other words, differentiating an accumulated total hands back the original rate: the accumulation function's slope at any point equals the height of f at that same point.
The second part is the one that makes integrals computable without ever summing a rectangle: if f is continuous on [a, b] and F is any antiderivative of f, then ∫[a,b] f(x) dx = F(b) − F(a). Every constant C an antiderivative might carry cancels out in that subtraction, so the specific antiderivative chosen never matters — find one antiderivative, evaluate it at both endpoints, subtract, and the exact area drops out with no limit and no partition in sight.
A working toolkit
Two rules turn most antiderivative-finding from a search into arithmetic. Linearity says an antiderivative distributes over addition and pulls constants out front: the antiderivative of a sum is the sum of the antiderivatives, and the antiderivative of a constant times a function is that same constant times the antiderivative of the function. Neither rule extends to products or quotients of two functions — the antiderivative of f · g is not, in general, the product of their antiderivatives.
For powers, the antiderivative of xⁿ is xⁿ⁺¹/(n + 1) + C, for every exponent n except −1 — dividing by n + 1 is exactly what breaks at n = −1, since that would divide by zero. The antiderivative of x² is x³/3 + C, so on the interval from 0 to 2, ∫[0,2] x² dx evaluates to x³/3 at the two endpoints, 8/3 − 0, the same 8/3 a curve y = x² accumulates over that interval.
That one excluded exponent is not left uncovered; x⁻¹, better known as 1/x, has its own antiderivative, ln|x| + C. The absolute value is not optional: the natural logarithm is undefined for negative inputs, but 1/x is perfectly well defined there, so the antiderivative needs the absolute value to stay defined everywhere 1/x is.
Where accumulation shows up
Accumulation is not a purely mathematical curiosity; it is how a moving system turns one measured rate into the total that actually matters. A sensor reporting velocity does not report position directly — position is the integral of velocity over time, and velocity is itself the integral of acceleration over time, with each integration constant fixed by wherever the system started. A wheeled vehicle deciding where it is right now is running exactly this calculation, continuously, on whatever its sensors just measured.
Accumulation also shows up wherever a quantity builds up step by step rather than arriving all at once — a running cost totaled across many small steps, or a system whose future state depends on integrating how it is currently changing. Those are each their own subject with their own machinery, and integration is simply the piece of language they both borrow: the idea that many small rates, added up correctly, become one total.
None of that changes what integration fundamentally is: the derivative run backwards, turned into a total instead of a rate. Every accumulation problem, however it is dressed up, reduces to the same two questions this concept answered — how much has piled up, and can that pile be computed exactly rather than merely estimated.