How this instrument works
A Taylor polynomial of order n rebuilds a function near a center point a using its derivatives at that single point, and the further x sits from a, the more the polynomial can drift from the real curve. The Lagrange error bound puts a hard ceiling on that drift: bound = M·|x−a|ⁿ⁺¹ ⁄ (n+1)!, where M is the largest size the (n+1)th derivative reaches anywhere between a and x. You never need the exact remainder or an unknown intermediate point — only an honest estimate of how large that next derivative could possibly get.
The reasoning behind the formula traces back to Taylor's theorem, which states the true remainder equals f⁽ⁿ⁺¹⁾(c)⁄(n+1)! times (x−a)ⁿ⁺¹ for some c sitting between a and x. Nobody can pin down that c in advance, so the practical move is to swap it for M, the worst case the derivative could reach across that whole stretch. Swapping an unknown exact value for a known worst case is what turns an abstract theorem into a number you can actually compute and trust as a guarantee.
Two patterns are worth internalizing. First, moving x closer to the center a shrinks |x−a| toward zero, so the bound collapses toward zero too — evaluate right at the center and the guaranteed error is exactly zero, since there's nothing left to approximate across. Second, raising n grows the factorial in the denominator far faster than the numerator typically grows, which is exactly why adding more terms to a Taylor polynomial usually tightens the guarantee, provided M itself doesn't explode as the derivative order climbs.
- Enter M, the largest size the (n+1)th derivative reaches anywhere between the center and your evaluation point.
- Set a, the center the Taylor polynomial is built around.
- Set x, the point where you want to know the approximation's worst-case error.
- Set n, the order of the Taylor polynomial already in hand.
- Read the Lagrange error bound, the largest possible gap between that polynomial and the function's real value at x.
Worked example — bounding a cubic approximation at x=0.1
Take M=1, a=0, n=3, and evaluate at x=0.1. First |x−a|ⁿ⁺¹ = 0.1⁴ = 0.0001. Next (n+1)! = 4! = 24. Multiplying by M=1 and dividing gives bound = 1×0.0001 ⁄ 24 ≈ 0.0000041667 — a guarantee that a cubic Taylor polynomial built at a=0 cannot miss the true value at x=0.1 by more than about four millionths, so long as no derivative past third order ever exceeds 1 in size on that stretch.
Change the inputs to M=2, a=0, n=2, x=1, and the shape of the calculation stays identical even though every number is different: |x−a|ⁿ⁺¹ = 1³ = 1, (n+1)! = 3! = 6, so bound = 2×1 ⁄ 6 = 0.3333333333333333. A far looser guarantee than the first case, because x sat a full unit from the center instead of a tenth of one, and the polynomial itself stopped one term earlier.
Questions
What does the Lagrange error bound actually tell me?
It tells you the largest possible gap between a Taylor polynomial's estimate and the function's true value at a chosen point — a guaranteed ceiling, not the exact size of the error. The real error could sit anywhere from zero up to that ceiling; the bound only promises it won't go higher.
Where does the value of M come from?
M is something you supply, found by studying the (n+1)th derivative of the function you're approximating and determining the largest absolute value it reaches anywhere between the center a and the evaluation point x. Sometimes that maximum is obvious from the derivative's shape; other times a rougher, safely generous estimate is used instead.
Why is the bound exactly zero when x equals a?
Because |x−a| becomes zero, and raising zero to any positive power still gives zero, so the whole expression collapses to nothing regardless of M or n. That matches the underlying idea directly: at the center itself, the Taylor polynomial reproduces the function exactly, so there is no distance left to approximate across.
Does increasing n always shrink the bound?
Usually, since (n+1)! grows far faster than |x−a|ⁿ⁺¹ for a fixed x once n climbs past a few terms, but it is not automatic. If M grows sharply with each extra derivative taken, that growth can offset or even outweigh the shrinking effect of the factorial, so each case still needs checking on its own terms.
How does this differ from the exact Taylor remainder formula?
The exact remainder equals f⁽ⁿ⁺¹⁾(c)⁄(n+1)! times (x−a)ⁿ⁺¹ for some specific but unknown point c between a and x — a precise value nobody can generally locate without more information. The Lagrange bound sidesteps that unknown point entirely by replacing the derivative at c with M, its worst possible size over the whole interval, trading exactness for a number you can compute right away.
Can the Lagrange error bound come out negative?
No — every piece of the formula stays non-negative whenever M is non-negative, since |x−a|ⁿ⁺¹ is an absolute value raised to a power and (n+1)! is always a positive integer. A negative result would signal an invalid input rather than a real bound.