This scene shows the band that traps the Taylor error. Raise the degree and the band shrinks to zero.
The remainder has a ceiling. So you can promise in advance how large the error can get. The true curve always stays inside the band.
For example, measuring with a π ruler promises an error of at most a millimeter. The Taylor remainder makes the same kind of promise. Higher degree narrows the promised range.
n=3: over the T3β of sin, the same remainder argument gives ex the bound 24eββ0.1133.Show proof
Every derivative of ex is ex, so f(k)(0)=1. Hence T3β(x)=1+x+2x2β+6x3β and T3β(1)=1+1+21β+61β=38ββ2.6667.
By Taylor's theorem (Lagrange remainder), R3β(1)=4!ecββ 14 for some cβ(0,1). Since ecβ€e, β£R3β(1)β£β€24eββ0.1133.
The true error is eβ38ββ0.0516<0.1133, inside the bound β a remainder bound is a safe overestimate.
Use the remainder bound (n+1)!β£xβ£n+1ββ0 to prove the Maclaurin series of sinx converges to sinx for every x.
β£Rnβ(x)β£β€(n+1)!β£xβ£n+1βnβββ0forΒ everyΒ fixedΒ x
n=9: on [βΟ,Ο] the envelope's half-width narrows to Ο10/10!β0.0258.Show proof
Every derivative of sin is Β±sin or Β±cos, so β£f(n+1)(c)β£β€1. The Lagrange remainder is Rnβ(x)=(n+1)!f(n+1)(c)βxn+1, hence β£Rnβ(x)β£β€(n+1)!β£xβ£n+1β.
The case x=0 is trivial and separate: the series is identically 0=sin0 (here anβ=0 identically, so the ratio below would be 0/0).
Assume xξ =0 and put anβ=(n+1)!β£xβ£n+1β; then anβan+1ββ=n+2β£xβ£β.
Once n+2>2β£xβ£ we get an+1ββ€21βanβ, so for nβ₯N:=β2β£xβ£β, anββ€aNβ(21β)nβNβ0 β the factorial beats the power geometrically.
Hence β£Tnβ(x)βsinxβ£β0: the Maclaurin series converges to sinx for every x.
If f(x)=βk=0ββckβ(xβa)k can be differentiated termwise, prove ckβ=k!f(k)(a)β (uniqueness of Taylor coefficients).
n=5: the coefficients of T5β(x)=xβx3/6+x5/120 match k!f(k)(0)β.Show proof
Assume f(x)=βk=0ββckβ(xβa)k converges near a. By the power series theorem, a power series is termwise differentiable to all orders inside its radius of convergence, so the repeated termwise differentiation used below (differentiating j times for arbitrary j) is justified β this theorem is cited here to align the hypotheses.
Differentiating j times gives f(j)(x)=βk=jββk(kβ1)β―(kβj+1)ckβ(xβa)kβj (terms with k<j vanish in the differentiation).
Setting x=a kills every term with k>j via (xβa)kβj; only k=j survives, giving f(j)(a)=j!cjβ, i.e. cjβ=j!f(j)(a)β β the Taylor coefficients are unique.