The left Riemann sum can either overestimate or underestimate the definite integral depending on the behavior of the function on the interval, making it a crucial concept for anyone studying numerical integration techniques.
Understanding Riemann Sums
A Riemann sum approximates the area under a curve by dividing the interval ([a, b]) into (n) subintervals of equal width (\Delta x = \frac{b-a}{n}). For each subinterval, a sample point is chosen, and the function value at that point is multiplied by (\Delta x). The sum of these products gives the approximation:
[ S = \sum_{i=1}^{n} f(x_i^*) ,\Delta x ]
The choice of sample point determines the type of sum: left endpoint ((x_i = a + (i-1)\Delta x)), right endpoint ((x_i = a + i\Delta x)), midpoint, or any other point within the subinterval.
Left Riemann Sum: Definition
When the sample point is taken as the left endpoint of each subinterval, the sum is called the left Riemann sum:
[ L_n = \sum_{i=1}^{n} f\bigl(a + (i-1)\Delta x\bigr),\Delta x ]
This rule is simple to compute and forms the basis for many algorithmic implementations Most people skip this — try not to..
Overestimate vs. Underestimate
1. Monotonic Functions
-
Increasing functions (e.g., (f(x)=x^2) on ([0,1])) have larger values at the right endpoint than at the left. So naturally, each rectangle’s height is underestimated, causing the left Riemann sum to underestimate the true integral.
-
Decreasing functions (e.g., (f(x)=\frac{1}{x}) on ([1,3])) have larger values at the left endpoint. The rectangles are overestimated, so the left Riemann sum overestimates the integral Nothing fancy..
2. Concavity
The concavity of the function also influences the direction of the error:
-
If the function is concave up (convex), the graph lies above the line segment connecting any two points. The left endpoint rectangle will sit below the curve, leading to an underestimate.
-
If the function is concave down, the graph lies below the line segment, and the left rectangle will sit above the curve, resulting in an overestimate Nothing fancy..
3. Visual Insight
Imagine drawing rectangles whose heights are determined by the left endpoint of each subinterval. If the curve rises quickly after the left point, the rectangle will be too short → underestimate. If the curve falls quickly after the left point, the rectangle will be too tall → overestimate It's one of those things that adds up. That alone is useful..
How to Determine the Direction for a Given Function
-
Check monotonicity on the entire interval.
- Increasing → underestimate.
- Decreasing → overestimate.
-
Examine concavity by computing the second derivative (f''(x)) Nothing fancy..
- (f''(x) > 0) (concave up) → underestimate.
- (f''(x) < 0) (concave down) → overestimate.
-
Combine both cues:
- For an increasing, concave‑up function, the left sum underestimates (both factors point the same way).
- For a decreasing, concave‑down function, the left sum overestimates (again, same direction).
Example Calculation
Consider (f(x)=\sin x) on ([0, \pi]) with (n=4) subintervals.
-
(\Delta x = \frac{\pi-0}{4}= \frac{\pi}{4}).
-
Left endpoints: (0,\ \frac{\pi}{4},\ \frac{\pi}{2},\ \frac{3\pi}{4}).
-
Function values:
- (f(0)=0)
- (f(\frac{\pi}{4})=\frac{\sqrt{2}}{2})
- (f(\frac{\pi}{2})=1)
- (f(\frac{3\pi}{4})=\frac{\sqrt{2}}{2})
-
Left Riemann sum:
[ L_4 = \left(0 + \frac{\sqrt{2}}{2} + 1 + \frac{\sqrt{2}}{2}\right)\frac{\pi}{4} = \left(1 + \sqrt{2}\right)\frac{\pi}{4} ]
- True integral:
[ \int_{0}^{\pi} \sin x ,dx = 2 ]
- Approximation:
[ L_4 = \frac{\pi}{4}(1+\sqrt{2}) \approx \frac{3.1416}{4}(2.4142) \approx 1.896 ]
Since (\sin x) is increasing on ([0,\pi]) and concave down (its second derivative (-\sin x) is negative), the left sum overestimates the integral, which is confirmed by the numeric result (1.Because of that, 896 > 2? actually 1.896 < 2, so it underestimates; let's re‑evaluate).
The function is increasing, so we expect underestimation. The second derivative is (-\sin x), which is negative on ((0,\pi)), indicating concave down, which would suggest overestimation. The conflicting cues show that monotonicity dominates for this particular interval because the function’s slope is positive throughout, and the rectangles are too short, leading to an underestimate.
This example illustrates that both monotonicity and concavity must be examined together; relying on a single factor can mislead the conclusion Simple, but easy to overlook..
Factors Influencing the Error Magnitude
-
Number of subintervals (n): As (n) increases, (\Delta x) becomes smaller, reducing the gap between the left endpoint value and the actual function value on the subinterval. The error therefore decreases as (n) grows, converging to the exact integral.
-
Interval length: Larger intervals amplify the discrepancy because each rectangle covers a broader region where the function may change more steeply.
-
Function smoothness: Highly oscillatory functions (e.g., (\sin(100x))) can cause erratic left‑endpoint values, leading to larger fluctuations in the sum. Smoother functions yield more predictable over/underestimation Easy to understand, harder to ignore. No workaround needed..
Common Misconceptions (FAQ)
Q1: Is the left Riemann sum always an underestimate?
A: No. For decreasing functions or concave‑down functions, the left sum overestimates the integral.
Q2: Does the left sum converge to the exact integral as n → ∞?
A: Yes. As the width (\Delta x) approaches zero, the left Riemann sum converges to the definite integral, regardless of whether it initially over‑ or underestimates And it works..
Q3: Can I improve the accuracy without changing the sample point?
A: Using midpoint or trapezoidal sums often reduces error because they incorporate information from both ends of each subinterval. Even so, increasing (n) alone also improves the left sum’s precision.
Q4: Does the shape of the interval (e.g., non‑uniform partitions) affect the result?
A: The standard left Riemann sum assumes equal subinterval widths. Non‑uniform partitions change (\Delta x_i) for each subinterval, and the left‑endpoint rule must be applied with the corresponding widths; the over/underestimate behavior may vary across subintervals And it works..
Conclusion
The left Riemann sum is not inherently an over‑ or underestimate; its tendency depends on the monotonicity and concavity of the function being integrated. By analyzing these properties before applying the left Riemann sum, you can predict the direction of the error and adjust (n) or choose a different Riemann sum variant to achieve the desired accuracy. For increasing functions, the left sum typically underestimates the integral, while for decreasing functions it overestimates it. Which means concave‑up functions push the sum under, and concave‑down functions push it over. Understanding this nuance is essential for anyone seeking reliable numerical integration in mathematics, physics, engineering, or data science Simple, but easy to overlook. Nothing fancy..
Worked Example: Estimating $\int_0^2 x^2 , dx$ with $n=4$
To solidify the concepts above, let’s walk through a concrete calculation. Consider $f(x) = x^2$ on the interval $[0, 2]$ using $n=4$ subintervals.
-
Calculate $\Delta x$: $ \Delta x = \frac{2 - 0}{4} = 0.5 $
-
Identify Left Endpoints ($x_i$): $x_0 = 0,\quad x_1 = 0.5,\quad x_2 = 1.0,\quad x_3 = 1.5$ (Note: $x_4 = 2.0$ is the right endpoint of the last subinterval and is not used in a left sum.)
-
Evaluate Function at Left Endpoints: $f(0) = 0^2 = 0$ $f(0.5) = 0.25$ $f(1.0) = 1.0$ $f(1.5) = 2.25$
-
Compute the Left Riemann Sum ($L_4$): $ L_4 = \Delta x \left[ f(0) + f(0.5) + f(1.0) + f(1.5) \right] $ $ L_4 = 0.5 \left[ 0 + 0.25 + 1.0 + 2.25 \right] = 0.5 \left[ 3.5 \right] = \mathbf{1.75} $
-
Analyze the Error: The exact integral is $\int_0^2 x^2 , dx = \left[ \frac{x^3}{3} \right]_0^2 = \frac{8}{3} \approx \mathbf{2.667}$ That alone is useful..
- Monotonicity: $f(x)=x^2$ is increasing on $[0, 2]$. The left sum uses the minimum value on each subinterval $\rightarrow$ Underestimate ($1.75 < 2.667$). ✅
- Concavity: $f''(x) = 2 > 0$ (Concave Up). The curve bends upward, lying above the left-endpoint rectangles $\rightarrow$ Underestimate. ✅
- Error Magnitude: $|2.667 - 1.75| \approx 0.917$.
-
Refining the Estimate ($n=8$): Doubling $n$ to $8$ ($\Delta x = 0.25$) yields $L_8 \approx 2.109$. The error shrinks to $\approx 0.558$, demonstrating the convergence discussed earlier.
Quick-Reference Decision Matrix
Before coding a numerical integration routine or solving a homework problem by hand, use this matrix to predict the left Riemann sum’s bias instantly.
| Function Behavior on $[a, b]$ | Left Sum Bias | Visual Intuition |
|---|---|---|
| Increasing & Concave Up (e.g.Here's the thing — , $e^x, x^2, x>0$) | Strong Underestimate | Rectangles sit "inside" the curve on both counts. |
| Increasing & Concave Down (e.g. |
Completing the Decision Matrix
The table you started is a handy quick‑look guide, but it is worth filling out the remaining cases so you can predict the left‑endpoint bias for any monotone‑concavity combination.
| Function Behavior on ([a, b]) | Left‑Sum Bias | Visual Intuition |
|---|---|---|
| Increasing & Concave Up (e. | ||
| Increasing & Concave Down (e.In practice, g. (\ln x, \sqrt{x})) | Ambiguous / Mixed | Monotonicity pulls Under; concavity pulls Over – the net error can be small, zero, or even change sign if one effect dominates. g. Now, (-e^{x}) on ([0,1])) |
| Constant (e.g. Because of that, g. | ||
| Decreasing & Concave Up (e.(e^{x}, x^{2}, x>0)) | Strong Underestimate | Rectangles sit “inside’’ the curve on both counts. On top of that, |
| Decreasing & Concave Down (e. g. (f(x)= -\sqrt{x}) on ([0,4])) | Strong Overestimate | The left‑hand rectangles now sit outside the curve because the function falls while curving upward. (f(x)=c)) |
Tip: When the bias is “Ambiguous / Mixed’’ it is often worthwhile to test a few values of (n) or to pair the left sum with a right‑sum (average gives the trapezoidal rule) to cancel the dominant error term.
Turning Theory into Code
Below is a compact Python snippet that puts the decision matrix to work automatically. It computes a left Riemann sum, estimates the sign of the error using monotonicity and concavity, and suggests a refinement strategy The details matter here. Took long enough..
import numpy as np
def left_riemann(f, a, b, n):
"""Return the left Riemann sum of f on [a,b] with n subintervals."""
dx = (b - a) / n
x = np.linspace(a,