What Is the Average Value of a Function?
Averaging a list of numbers is familiar: add them up, divide by how many there are. But a function on an interval takes infinitely many values — you can't add them all. The integral solves exactly this problem: summing a continuum is what integration is, and dividing by the interval's length finishes the average:
favg = 1/(b−a) · ∫ab f(x) dx
Geometrically, favg is the height of the rectangle over [a, b] whose area equals the area under the curve. Imagine the region under the graph as water in a tank with a bumpy surface; let it settle flat, and the resulting level is the average value. That's why the graph above draws a horizontal line at favg — the area the curve holds above that line exactly fills the gaps below it.
How to Use This Calculator
Enter a function and an interval. The calculator evaluates the definite integral with high-precision Simpson's rule, divides by b − a, and reports the average. It also solves f(c) = favg to find where the function actually passes through its own average — the point guaranteed by the Mean Value Theorem for integrals — and marks the average as a dashed horizontal line against the shaded region so you can verify the equal-area picture by eye.
Worked Example
Find the average value of f(x) = x² on [0, 3]:
∫₀³ x² dx = x³/3 |₀³ = 9, so favg = 9 / (3 − 0) = 3
Note what the naive endpoint average would give: (f(0) + f(3))/2 = (0 + 9)/2 = 4.5 — wrong, because x² spends most of the interval at small values and only sprints up near the end. The integral weights every point equally along the x-axis, capturing that lopsidedness; two-point averaging cannot. The function crosses its true average where x² = 3, i.e. at c = √3 ≈ 1.732.
Average Value vs. Other "Averages"
| Quantity | Formula | What it averages |
|---|---|---|
| Average value of f | 1/(b−a)·∫f dx | The function's height over the interval |
| Average rate of change | [f(b)−f(a)]/(b−a) | The function's slope between endpoints |
| Endpoint average | [f(a)+f(b)]/2 | Just two sampled values — generally ≠ favg |
The first two are frequently confused: average value integrates f itself; average rate of change is the secant slope, connected to the derivative through the ordinary Mean Value Theorem. They coincide only by accident.
Common Mistakes to Avoid
- Averaging the endpoints. (f(a)+f(b))/2 uses two points out of infinitely many; as the worked example shows, it can be badly off for curved functions. (It's exact only for straight lines.)
- Forgetting to divide by (b − a). The bare integral is total accumulated area, not an average — the division converts "total" into "per unit length."
- Expecting the average to be positive. If f dips below the x-axis, signed area can make favg zero or negative — sin(x) over a full period averages exactly 0.
- Mixing up which theorem applies. "There's a c where f(c) = favg" is the MVT for integrals and needs f continuous; the derivative-based MVT is a different statement about slopes.
Real-World Applications
Average-of-a-function computations are everywhere measurements vary continuously. Your electricity bill is built on average power drawn over the billing period — the integral of instantaneous power divided by elapsed time — and "average daily temperature" in climate records is the integral of the continuous temperature trace over 24 hours divided by 24, not the average of two readings. In electronics, the RMS (root-mean-square) voltage stamped on every power supply is a close cousin: the square root of the average value of voltage squared, computed with exactly this page's formula applied to v(t)².
In economics and operations, average inventory over a season (for storage-cost accounting), average traffic flow over rush hour, and average concentration of a drug in the bloodstream over a dosing cycle are all integral averages — the drug case, "area under the curve" (AUC), is literally the pharmacokineticist's standard exposure metric, an average value multiplied back into a duration. Whenever a continuously varying quantity must be summarized by one fair number, this formula is the principled way to do it.
Frequently Asked Questions
Why does integrating give the "sum" of infinitely many values?
The definite integral is the limit of Riemann sums — adding up f at sample points weighted by tiny widths Δx. Dividing by (b−a) turns that weighted sum into a weighted mean, and in the limit the weighting becomes perfectly uniform along the axis. The Riemann Sum Calculator shows the finite version of exactly this process.
Must the function actually hit its average value somewhere?
If f is continuous on [a, b], yes — the Mean Value Theorem for integrals guarantees a c with f(c) = favg, and this calculator locates it. A discontinuous function can jump right over its average (a step function equal to 0 then 2 averages to 1 without ever equaling 1).
Can the average value be zero for a nonzero function?
Yes — symmetric positive and negative regions cancel. sin(x) over [0, 2π] averages 0 despite being nonzero almost everywhere. When cancellation isn't wanted (as in RMS calculations), the standard fix is averaging |f| or f².
How accurate is the numerical integration here?
Simpson's rule with thousands of subintervals is typically accurate to 6–8 significant figures for smooth functions — the same engine as the Definite Integral Calculator, which also cross-checks against symbolic antiderivatives when one exists.
What happens to the average as the interval shrinks?
As b → a, the average value converges to f(a) itself — averaging over a vanishing window recovers the function's value, one way to see that continuous functions are "locally constant-ish." It's the integral counterpart of the derivative's h → 0 limit.
How does a weighted average fit into this picture?
The plain average weights every x equally; a weighted average inserts a weight function, computing ∫f·w dx / ∫w dx. That generalization is everywhere: centers of mass weight position by density, expected values in probability weight outcomes by their probability density, and grade-point averages weight scores by credit hours. The formula on this page is the special case w(x) = 1 — mastering it makes every weighted variant a one-step upgrade.