Why an Estimate Can Miss Important Behavior
A trapezoidal estimate uses function values at selected points to approximate accumulated change across an interval. Between each pair of sample points, it effectively treats the graph as a straight line. That is a useful approximation when the graph behaves smoothly and does not change dramatically between samples.
But a small set of values cannot describe every possible shape between them. A narrow peak may rise and fall between consecutive sample points. A graph may bend sharply even when its endpoint values appear ordinary. A discontinuity may make interpolation across an interval particularly misleading. The task in this tutorial is to diagnose these risks using the graph and the available data—not to claim that a visual inspection alone proves an estimate accurate.
What the Trapezoidal Estimate Assumes
Suppose the interval is divided at \(x_0,x_1,\ldots,x_n\), where \(x_0<x_1<\cdots<x_n\), and the sampled values are \(f(x_0),f(x_1),\ldots,f(x_n)\). On the subinterval from \(x_{i-1}\) to \(x_i\), the trapezoidal rule uses the area of a trapezoid whose parallel sides have lengths \(f(x_{i-1})\) and \(f(x_i)\). Its signed contribution is
Adding these contributions gives the composite estimate. This construction uses the sampled endpoint values of each subinterval, but it does not directly use what the graph does inside that subinterval.
The underlying approximation can be stated precisely. On each subinterval, replace the graph by the line segment joining the two sampled points. The trapezoidal estimate is the integral of those line segments. To see why, write the linear function joining \((a,f(a))\) and \((b,f(b))\) as
Integrating from \(a\) to \(b\), and using \(h=b-a\), gives
This is exactly the signed area used for one trapezoid. The approximation is therefore sensible when the line segments resemble the actual graph. If the graph bends sharply or behaves irregularly between the sampled points, the line segments may be poor stand-ins.
Warning Signs in a Graph or Data Table
Several observations should prompt caution. They are warning signs, not automatic proof that an estimate is inaccurate.
- Sharp curvature: the graph bends noticeably within a subinterval, so the straight segment may depart from it.
- Rapid changes in height: neighboring sampled values differ greatly over a short horizontal distance, or the graph appears steep between samples.
- Rapid changes in slope: successive secant slopes differ substantially. For adjacent samples, compare the quotients \(\frac{f(x_i)-f(x_{i-1})}{x_i-x_{i-1}}\).
- Localized peaks or dips: a feature that is narrow relative to the gap between sample points might be only partly sampled or missed entirely.
- Discontinuities or abrupt transitions: a line segment joining values on opposite sides of a jump does not represent the behavior of the function near the jump.
The graph and the table provide different kinds of evidence. A graph can reveal an apparent bend, peak, or jump between sampled points. A table can reveal large changes among the sampled values or changes in successive secant slopes. Neither can rule out every feature between the samples. Also check the horizontal scale: a graph that looks nearly straight on a compressed display may still have significant curvature on the interval being approximated.
Worked Example: A Curved Graph and a One-Trapezoid Estimate
Worked Example: A Curved Graph and a One-Trapezoid Estimate
Use one trapezoid to estimate \(\int_0^2 x^2\,dx\). Compare the estimate with the exact value, and identify the diagnostic warning.
The interval width is \(2-0=2\). The endpoint values are
Thus the trapezoidal estimate is
For a check, evaluate the integral exactly:
The difference between the estimate and the exact value is
The graph of \(x^2\) bends upward across the interval, so its shape is not well represented by the single endpoint-to-endpoint line segment. The exact comparison confirms that this one-trapezoid estimate is high by \(\frac{4}{3}\). The practical diagnosis is that a visibly curved graph over a wide subinterval deserves caution; the endpoint values alone do not show how much the graph bends between them.
Check: the antiderivative \(\frac{x^3}{3}\) differentiates to \(x^2\), and substituting the endpoints gives \(\frac{8}{3}\), as used above.
Why Sampled Values Cannot Guarantee Reliability
Even a table with identical endpoint values cannot show that the function stayed constant between those points. Here is a precise reason. Consider the zero function on \([0,2]\), and compare it with a triangular function that is zero at \(0\), \(1\), and \(2\), but rises to \(1\) at \(x=\frac12\) and falls back to zero at \(x=\frac35\). Define that second function to be zero elsewhere. The sample values at \(0,1,2\) are all zero for both functions, so a trapezoidal estimate using those three samples gives zero for each.
However, the triangular function has base width \(\frac35-\frac25=\frac15\) and height \(1\). Its integral is its triangular area:
The zero function has integral zero. Thus the same samples and the same trapezoidal estimate can correspond to different integral values. This example proves a general limitation of diagnosis from finite samples: sampled values alone cannot exclude an unsampled feature. A graph, contextual knowledge about the function, or additional samples can provide more evidence, but the conclusion must remain appropriately cautious.
Worked Example: A Narrow Feature Between Sample Points
Worked Example: A Narrow Feature Between Sample Points
Define \(f\) on \([0,2]\) by
Estimate \(\int_0^2 f(x)\,dx\) using the samples at \(x=0,1,2\), then compare with the exact integral.
The sampled values are \(f(0)=0\), \(f(1)=0\), and \(f(2)=0\), because all three sample points lie outside the nonzero part of the graph. The two subintervals have width \(1\), so the composite trapezoidal estimate is
The graph between \(0.4\) and \(0.6\) is a triangle: it rises from zero at \(x=0.4\) to \(f(0.5)=10(0.1)=1\), then falls to zero at \(x=0.6\). Its exact integral is therefore
The estimate is zero, although the integral is \(0.1\). The sampled values conceal the entire peak. This is a particularly serious diagnostic warning when a graph or the application suggests narrow spikes, short bursts, or other localized behavior. Additional samples placed inside the suspected feature would be needed to represent it.
Check: the triangle has base \(0.2\) and height \(1\), so its area is \(0.1\), confirming the exact integral.
Discontinuities Need Special Caution
A trapezoid joins sampled values with a line segment. If the function jumps inside a subinterval, that segment bridges the jump rather than representing the function on either side. If a sample lands exactly at a discontinuity, the function’s value at that single point may also influence the trapezoidal estimate, even though changing a function at one point does not change its integral when the integral exists.
Worked Example: A Step Discontinuity at a Sample Point
Worked Example: A Step Discontinuity at a Sample Point
Let \(f(x)=0\) for \(0\le x\le1\), and \(f(x)=2\) for \(1<x\le2\). Use the samples at \(0,1,2\) to form a trapezoidal estimate, then compare with the exact integral.
The sampled values are \(f(0)=0\), \(f(1)=0\), and \(f(2)=2\). Each subinterval has width \(1\), so
The function is zero on \([0,1]\) and equal to \(2\) on \((1,2]\). Its exact integral is
The trapezoidal estimate is \(1\), while the exact integral is \(2\). The line segment on \([1,2]\) rises from \(0\) to \(2\), but the function is \(2\) throughout \((1,2]\). This example shows why a detected jump should not be treated as ordinary smooth variation. A single value at the jump can affect the trapezoidal construction, while the integral reflects the values across intervals of positive width.
Check: the contribution from \((1,2]\) is a rectangle of height \(2\) and width \(1\), so the integral is \(2\), consistent with the calculation.
A Practical Diagnostic Process
Before trusting a trapezoidal estimate, inspect what is known about the graph and the sampling. The following sequence organizes that inspection:
Look for sharp bends, narrow extrema, abrupt transitions, or portions that are not shown clearly at the graph’s scale.
Calculate or estimate changes in height and successive secant slopes. Large changes indicate variation worth examining more closely.
Do the graph, the function’s context, or known features suggest behavior the sample points might miss?
If a jump or other discontinuity occurs, a line segment across it may be a poor model. Treat that interval separately when interpreting the estimate.
Describe the evidence for concern, but do not claim that a finite table proves accuracy or inaccuracy without further information.
Diagnostic Evidence Is Not an Error Guarantee
A graph with sharp curvature or rapidly changing sampled values signals a reason to question the estimate. A known discontinuity raises a more fundamental concern about connecting the sampled endpoints. Conversely, a smooth-looking graph and modest changes in the table are reassuring, but they are not a proof that no important behavior was missed.
Formal error bounds can quantify error when their assumptions are satisfied. In particular, a bound based on the second derivative requires that the function have the needed smoothness over the interval. A discontinuity or a sharp feature can invalidate those assumptions or make a bound based on global information too conservative to be useful. The diagnostic step is to check whether the function’s behavior and the sampling support the model being used before relying on a numerical result.
Check Your Understanding
Use the ideas in this tutorial to answer the following questions.
- Why can a sharply curved graph make the line-segment model behind a trapezoidal estimate questionable?
- What warning might be suggested by large changes between neighboring sampled values? Why is that warning not a proof that the estimate is inaccurate?
- How can two functions have the same sampled values but different integrals?
- Why can a discontinuity inside a subinterval make its trapezoid a poor representation of the function?
- What additional evidence would help investigate a suspected narrow peak between sample points?
