Accumulation and the Integral
How do local changes combine into a global outcome?
What structure must an accumulation process satisfy to be consistent with differentiation?
The derivative moves from global behavior to local behavior. Given a changing process, it asks what linear structure is visible at a point. It compresses a small neighborhood into first-order data: rate, direction, tangent behavior, local sensitivity.
The chain rule then explains how that local data travels through composition.
But local data is still local.
A speed at one instant does not tell us how far something traveled. A density at one point does not tell us the total mass of a region. A marginal cost at one level of production does not tell us the total cost of producing many units. Local change has to be assembled before it becomes a global quantity.
That assembly is the structural role of the integral.
The integral takes local contributions distributed across an interval or region and accumulates them into a total effect.
(local change)
d
│
│ integrate
▼
(global change)
This reverses the direction of attention from differentiation. Differentiation begins with a whole process and extracts its local rate. Integration begins with local rates or densities and asks for the total change they produce when combined across a domain.
Imagine a car moving along a road. At each moment, the speedometer gives a local rate: distance per unit time. The rate at a single instant is useful, but the journey has a global question: how much distance was covered between the start and the end?
If the speed were constant, the answer would be simple:
distance = speed * time
The structure becomes subtler when speed varies. Over a short interval of time, the speed may be treated as approximately constant. That gives a small contribution:
small distance ≈ speed at that moment * small time interval
The total distance is obtained by adding all those small contributions.
time interval: [a, b]
a ──|──|──|──|──|── b
│ │ │ │ │
local contributions
The integral is the operation that makes this accumulation precise.
At the calculus level, the picture is straightforward. Break the interval into small pieces. Estimate the contribution on each piece. Add the estimates. Refine the partition. Ask whether the accumulated total stabilizes.
For a function v(t) representing signed velocity, the accumulated change in position from a to b is written
∫_a^b v(t) dt
If v(t) is instead a nonnegative speed, the same form gives distance traveled rather than signed displacement. The formula is the same, but the interpretation depends on the local contribution being accumulated.
The notation records three structural pieces at once:
- the local contribution rule,
v(t) - the domain of accumulation, from
atob - the small scale of accumulation, indicated by
dt
The symbol dt should be read structurally before it is read technically. It marks the local unit over which contributions are being collected. The integral asks how all the pieces v(t) dt combine into a total.
In geometry, the same structure appears as area. If f(x) gives the height of a curve above the horizontal axis, then over a small interval of width dx, the local rectangular contribution is approximately
f(x) dx
Adding those contributions across an interval gives the area under the curve:
∫_a^b f(x) dx
The area picture is useful because it makes accumulation visible. Yet the structural idea is broader than area. The integral can accumulate velocity into displacement, density into mass, force into work, probability density into probability, and marginal change into total change.
The common pattern is local-to-global assembly.
local density / rate / contribution
│
│ sum coherently over a domain
▼
global amount / displacement / mass / area
This is why addition is built into integration. Local contributions must combine additively. If an interval is split into two adjacent pieces, the total over the whole interval should equal the total over the first piece plus the total over the second:
∫_a^c f(x) dx = ∫_a^b f(x) dx + ∫_b^c f(x) dx
That additivity is one of the defining structural constraints. Accumulation must respect decomposition of the domain. Otherwise the total would depend on how the interval was artificially cut into pieces.
Linearity is the second central constraint.
If two local contribution rules are added, their accumulated total should be the sum of their accumulated totals:
∫_a^b (f(x) + g(x)) dx = ∫_a^b f(x) dx + ∫_a^b g(x) dx
If every local contribution is scaled by a constant c, the total should scale by the same constant:
∫_a^b c f(x) dx = c ∫_a^b f(x) dx
These laws are familiar from computation, but structurally they say something deeper. The integral is a linear accumulation operator. It preserves the way local contributions combine.
This returns us to a theme from linear algebra: when behavior is additive and scalable, it becomes composable and predictable.
Integration also has to respect orientation. Accumulating from a to b and then reversing direction from b to a should change the sign:
∫_b^a f(x) dx = -∫_a^b f(x) dx
This sign is part of the structure. It distinguishes net change from unsigned size. If a quantity increases and then decreases, the integral of its rate records the net effect, allowing contributions with opposite orientation to cancel.
That cancellation is essential for change. A path can move forward and backward. A function can be positive and negative. A process can add and remove. The integral records the signed total unless the problem asks for a different structure, such as total distance or total variation.
The relationship with the derivative now becomes visible.
If F(t) records position, then its derivative F'(t) records velocity. The derivative moves from accumulated position to local rate. The integral moves from local rate back to accumulated change:
∫_a^b F'(t) dt = F(b) - F(a)
This is the fundamental structural link in calculus.
Differentiation extracts local change. Integration accumulates local change. When the hypotheses are right, doing one after the other recovers the net global change between endpoints.
global quantity F
│ differentiate
▼
local rate F'
│ integrate over [a,b]
▼
net change F(b) - F(a)
This relation is often called the Fundamental Theorem of Calculus. Its importance comes from the way it binds two different viewpoints into one coherent structure.
One viewpoint studies rates. The other studies totals. The theorem says that local rates and global net changes determine each other in a controlled way.
If we know an accumulated quantity F, differentiation gives its local rate F'.
If we know a local rate f, integration can construct an accumulated quantity
A(x) = ∫_a^x f(t) dt
Under suitable conditions, the derivative of this accumulation function is the original local contribution rule:
A'(x) = f(x)
This statement deserves attention. The variable upper limit turns integration into a function-building operation. Instead of asking only for one total over one interval, we ask for the accumulated total up to each point x.
a ───────────── x ───────── b
A(x) = accumulated contribution from a to x
As x moves, A(x) changes. The derivative of A at x recovers the local contribution being added at that point. In this sense, accumulation and local change are two faces of the same structure.
This also explains why antiderivatives are useful.
An antiderivative of f is a function F whose derivative is f. Once such an F is known, the accumulated total from a to b can be found by endpoint comparison:
∫_a^b f(x) dx = F(b) - F(a)
Antiderivatives are determined only up to an additive constant, and this endpoint difference cancels that constant. That is why the definite integral has a well-defined value even though the chosen antiderivative is not unique.
The integral is still the accumulation operation. The antiderivative is a way to compute that accumulation by finding a global quantity whose local change is the given integrand.
A potentially infinite accumulation across an interval becomes the difference of two endpoint values. Calculus works so well partly because the Fundamental Theorem turns local-to-global assembly into a computable endpoint operation.
The theorem also clarifies what integration preserves.
It does not preserve the pointwise shape of the local contribution rule. Many different functions can have the same total over a fixed interval. The invariant produced by a definite integral is the accumulated effect across the chosen domain.
Change the interval, and the total may change. Change the local contribution rule, and the total may change. Decompose the interval, and the totals of the pieces must add back to the total of the whole.
So the integral is attached to a domain of accumulation, just as the derivative is attached to a point.
derivative: point ──localize──▶ local linear behavior
integral: region ──accumulate──▶ global total
The two constructions answer opposite questions. The derivative asks: what is happening here, to first order? The integral asks: what is the total effect across this whole interval or region? Their compatibility is what makes calculus a single subject rather than two separate techniques.
There is also a compositional discipline inside integration. If a domain is assembled from pieces, accumulation over the domain is assembled from accumulation over the pieces. If local contribution rules are added, their accumulated totals add. If a contribution rule is scaled, its accumulated total scales. These are the structural laws that make integration coherent.
In higher dimensions, the same pattern expands. A density over a region accumulates into mass. A vector field along a curve accumulates into work or circulation. A differential form over a surface accumulates into flux. The representation changes, but the structural question remains:
What local data is assigned to each small piece, and how should those pieces compose into a global invariant?
This is the point where calculus begins to lean toward later mathematics. Integration over intervals leads to integration over regions, surfaces, manifolds, and more abstract spaces. The simple act of adding small contributions becomes a way to compare local structure with global structure.
Even at the elementary level, this local-to-global movement is the key.
Differentiation made curved change locally linear. The chain rule showed that local linearization respects composition. Integration now asks how the local pieces produced by calculus can be assembled into a total that is independent of arbitrary refinements.
That last phrase points to a boundary.
To define the integral carefully, one must say what it means for finer and finer sums to stabilize. One must decide which functions are integrable, which partitions count, and how infinite limiting processes are controlled. Those questions belong to analysis.
Calculus uses the idea of refinement. Analysis studies its legitimacy.
At this stage, the structural content is enough to complete the calculus layer. The objects are intervals, regions, and functions or fields assigning local contributions across them. The allowed operations are accumulation procedures that respect decomposition, scaling, and addition. What composes are decompositions of domains and additive combinations of local contribution rules. The invariants are total effect, net change, area, mass, work, probability, and other global quantities obtained by accumulation. The defining relation connecting this structure back to differentiation is
∫_a^b F'(x) dx = F(b) - F(a)
and, in the variable-limit form,
d/dx ∫_a^x f(t) dt = f(x)
Equality here means equality of accumulated effect, while formulas for computing an integral depend on representation, coordinates, and the available antiderivatives. Rigorous limits, convergence, completeness, and the classification of legitimate infinite processes remain outside the frame.
The integral reconstructs global behavior from local contributions.
Calculus now has its central loop.
Differentiate to pass from global change to local behavior. Use the chain rule to make local behavior respect composition. Integrate to accumulate local behavior back into global change.
But each step has relied on a quiet assumption: that refinement can stabilize, that infinite approximation can produce a well-defined object, and that the space being studied actually contains the limits these processes demand.
When does an infinite process converge?
And what structure guarantees that its limit has somewhere to land?