Use the Tangent Line as a Local Approximation
A worked explanation of use the tangent line as a local approximation, including the key limitation and a checkable example.
Use the Tangent Line as a Local Approximation
A tangent-line estimate drifts as you move away from the contact point, and the question is how far you can go before the drift becomes misleading. A linear approximation calculator lets you compare that estimate with the function's actual value once you have chosen the contact point. For a differentiable function, the error grows as the square of the distance from the point of tangency. The practical answer depends on the function's curvature, not on a fixed distance. To use the tangent line as a local approximation, you need a concrete bound that tells you when the estimate is still trustworthy.
A Worked Example
Apply the error bound to f(x) = √x at a = 4. The tangent-line slope is f′(4) = 1/(2√4) = 0.25, so the linearization is L(x) = 2 + 0.25(x − 4). At x = 4.1, the estimate is 2.025, and the actual √4.1 ≈ 2.02485. The error is about 0.00015. At x = 5, L(5) = 2.25, but √5 ≈ 2.23607, producing an error of 0.01393. The error at x = 5 is roughly ninety-three times larger than at x = 4.1, even though the distance from the contact point only grew from 0.1 to 1.
The error formula explains why. The Lagrange remainder states |f(x) − L(x)| ≤ (M/2)(x − a)², where M is the maximum of |f″| on the interval between a and x. For √x, f″(x) = −1/(4x³⁄²). On the interval [4, 5], the largest value of |f″| occurs at x = 4, where it equals 1/32 = 0.03125. At x = 5, the error bound becomes (0.03125/2)(1)² ≈ 0.0156, which does contain the actual error of 0.01393. At x = 4.1, the bound is (0.03125/2)(0.1)² ≈ 0.000156, containing the actual 0.00015.
Calculation Steps
To decide how close you need to stay, compute the error bound for your function before you trust the estimate. The general formula L(x) = f(a) + f′(a)(x − a) is the tangent line, and the error formula gives the maximum possible deviation. For functions where the second derivative is known, you can find the largest tolerable |x − a| by solving (M/2)(x − a)² ≤ acceptable error.
For √x at a = 4, suppose you need the estimate to be within 0.01. Set (0.03125/2)(x − 4)² ≤ 0.01. This gives (x − 4)² ≤ 0.64, so |x − 4| ≤ 0.8. The tangent-line estimate is trustworthy within 0.8 units of the contact point for this error tolerance. Change the tolerance to 0.001, and the bound shrinks to |x − 4| ≤ 0.253.
At the point where the next calculation becomes necessary, open the linked tool to test your hand-computed bound against a numerical check. A calculator that computes L(x) and the actual f(x) simultaneously will confirm whether your bound matches the real error.
Why the Method Works
The error bound depends on the maximum of |f″| on the interval between a and x. For functions with high curvature, M is large, and the tolerable distance shrinks. Consider f(x) = 1/x at a = 0.1. The derivative f′(x) = −1/x², so f′(0.1) = −100. The linearization is L(x) = 10 + (−100)(x − 0.1). At x = 0.2, the estimate gives L(0.2) = 10 + (−100)(0.1) = 0, but the actual 1/0.2 = 5. The error is 5, which is the same magnitude as the function value itself. The second derivative f″(x) = 2/x³, and at x = 0.1, |f″| = 2000. The bound (M/2)(0.1)² with M = 2000 gives 10, which does contain the error of 5, but the estimate is useless.
This is the failure case. The tangent-line estimate cannot be trusted at x = 0.2 for this function because the curvature near the contact point is extreme. The distance of 0.1 from a = 0.1 is too large. To get a reliable estimate, you would need to stay much closer, within roughly 0.01 units, or choose a different contact point.
The honest definition of the tangent line as the limit of secant lines means the line only guarantees local behavior. The line is the best straight-line fit at exactly that point, but the quality of the fit degrades as you leave. The most common error is assuming the line approximates well over a fixed distance, like 0.1, for every function. It does not. The distance that counts is relative to the curvature, not absolute.
| Function | Contact Point a | Test Point x | Distance |x−a| | M (max |f″|) | Error Bound (M/2)(x−a)² | Actual Error | Trustworthy? |
|---|---|---|---|---|---|---|---|
| √x | 4 | 4.1 | 0.1 | 0.03125 | 0.000156 | 0.00015 | Yes |
| √x | 4 | 5 | 1 | 0.03125 | 0.0156 | 0.01393 | Yes, within 0.02 |
| 1/x | 0.1 | 0.2 | 0.1 | 2000 | 10 | 5 | No, error equals value |
| 1/x | 0.1 | 0.11 | 0.01 | 2000 | 0.1 | 0.0909 | Marginal |
The Single Thing That Most Often Goes Wrong
The most common error is assuming a fixed distance, like 0.1 or 1, works for every function. The tangent-line estimate at a point of tangency only guarantees accuracy within a range that depends on the second derivative. For a function with low curvature, you can move farther. For a function with high curvature, you must stay very close. Always compute the error bound before acting on the estimate. The failure case of 1/x at a=0.1 shows that a distance of 0.1 can produce an error that makes the estimate worse than guessing.
Common Questions
How close to the contact point can a tangent-line estimate be trusted?
It depends on the function's second derivative and your acceptable error. Solve (M/2)(x−a)² ≤ error tolerance, where M is the maximum of |f″| on the interval. For √x at a=4 with a tolerance of 0.01, the estimate is trustworthy within 0.8 units. For 1/x at a=0.1, the same tolerance requires staying within 0.01 units.
What is the error bound formula for linearization?
The Lagrange remainder gives |f(x)−L(x)| ≤ (M/2)(x−a)², where M is the maximum of |f″| on the interval between a and x. The error is in the same units as f(x).
Why does the tangent-line estimate fail for 1/x at a=0.1?
The second derivative is 2/x³, which equals 2000 at x=0.1. This large curvature means the error bound grows quickly. At x=0.2, the actual error is 5, and the bound predicts up to 10. The estimate is unreliable because the function changes slope rapidly near the contact point.
How do I compute M for my function?
Find f″(x) and determine its maximum absolute value on the closed interval between a and x. For a function like √x on [4,5], f″(x)=−1/(4x³/²), and the largest value occurs at x=4, giving M=0.03125. For a function like 1/x on [0.1,0.2], the largest value occurs at x=0.1, giving M=2000.
Can I trust the tangent-line estimate if the function has an inflection point at the contact point?
Yes, the estimate is still valid, but the tangent line will cross the curve at the inflection point. The error bound still applies because f″ changes sign. The linearization remains the best local straight-line fit, but the line crosses rather than staying on one side of the curve.
What happens if the function is not differentiable at the contact point?
No tangent line exists in the usual sense. A corner or cusp means the limit of secant lines does not exist, so you cannot form a linearization. The calculator's nDeriv( may still return a value, but that value is meaningless. Check both left-hand and right-hand limits of the difference quotient before trusting any estimate.
Does the error bound work for functions with vertical tangents?
No. A vertical tangent occurs where the derivative is undefined (the limit of secant slopes goes to infinity). The linearization formula L(x)=f(a)+f′(a)(x−a) assumes a finite derivative. For a vertical tangent, you cannot write the line in slope-intercept form, and the error bound does not apply.