3.

1 Taylor’s Theorem
3.1 Taylor’s Theorem
One-variable Taylor’s theorem:
let f ∶ X → R be differentiable up to order at least k, where X ⊂ R
let a ∈ R be a reference point
Taylor polynomial of degree k:

f ′′ (a) f (k) (a)

Pk (x) = f (a) + f ′ (a)(x − a) + (x − a)2 + ⋯ + (x − a)k
2 k!
Rk (x, a)
Taylor’s theorem: f (x) = Pk (x) + Rk (x, a) where lim =0
x→a (x − a)k

remainder term: if f is differentiable up to order at least k + 1, then

there exists ξ ∈ (a, x) or (x, a) such that

f (k+1) (ξ)
Rk (x, a) = (x − a)k+1
(k + 1)!
3.1 Taylor’s Theorem

Theorem 3.1
(Taylor’s theorem) Let f ∶ X → R be a differentiable function where
X ⊂ Rn is open. For any a ∈ X, define P1 (x) = f (a) + Df (a)(x − a).
Then
f (x) = P1 (x) + R1 (x, a)
where
R1 (x, a)
differentiable iff lim = 0.
x→a ∥x − a∥

so f (x) ≈ P1 (x) = f (a) + Df (a)(x − a) when x is close to a

*this is essentially the equation of tangent plane

x a vector
3.1 Taylor’s Theorem

Theorem 3.1
(Taylor’s theorem) Let f ∶ X → R be a differentiable function where
X ⊂ Rn is open. For any a ∈ X, define P1 (x) = f (a) + Df (a)(x − a).
Then
f (x) = P1 (x) + R1 (x, a)
where
R1 (x, a)
lim = 0.
x→a ∥x − a∥

Proof.

using the definitions, the result means

f (x) − [f (a) + Df (a)(x − a)]

lim =0
x→a ∥x − a∥

this is exactly the differentiability at a

3.1 Taylor’s Theorem

Definition 3.1
Let f ∶ X → R be a differentiable function where X ⊂ Rn is open. Let
a ∈ X and let h ∈ Rn . The incremental change of f from a to a + h is
defined by usually h is very small.
∆f = f (a + h) − f (a).
The total differential of f at a with a change h is defined by
*
df (a, h) = Df (a)h.
= a number, essentially a dot product

e.g. df ((a, b), (h, k)) = fx (a, b)h + fy (a, b)k

3.1 Taylor’s Theorem

by the first-order Taylor’s theorem, we have

lim(x->a) Remainder term /|x-a| = 0 P1(x)
f (a + h) − f (a) − Df (a)h
lim =0
h→0 ∥h∥

∆f − df (a, h)
this means lim =0
h→0 ∥h∥
this shows ∆f ≈ df (a, h) when ∥h∥ is small
3.1 Taylor’s Theorem

Example 3.1
A cuboid of size 1 × 2 × 3 m3 is to be made. Suppose there is an error of
at most 0.02 m for each side in the final product. Then the maximum
error of the volume V of the cuboid is approximately dV = 0.22 m3 .

Proof.

letting x, y, z be the side lengths, then V (x, y, z) = xyz

the goal is to maximize ∣∆V ∣ = ∣V (1 + h1 , 2 + h2 , 3 + h3 ) − V (1, 2, 3)∣
where ∣hj ∣ ⩽ 0.02 *Using total differential:
6 3 2 always a linear polynomial
we have DV (x, y, z) = (yz zx xy), so
with degree 1, hence compu-
tationally easy!
∆V ≈ dV = DV (1, 2, 3)h = (6 3 2) h = 6h1 + 3h2 + 2h3

the maximum error is ≈ 6(0.02) + 3(0.02) + 2(0.02) = 0.22

approximation
3.1 Taylor’s Theorem

Example 3.1
A cuboid of size 1 × 2 × 3 m3 is to be made. Suppose there is an error of
at most 0.02 m for each side in the final product. Then the maximum
error of the volume V of the cuboid is approximately dV = 0.22 m3 .

in fact, the maximum V is (1 + 0.02)(2 + 0.02)(3 + 0.02) = 6.222408

the minimum V is (1 − 0.02)(2 − 0.02)(3 − 0.02) = 5.782392
the original (ideal) value of V is (1)(2)(3) = 6
so the actual maximum error is

max {∣6.222408 − 6∣ , ∣5.782392 − 6∣} = 0.222408

3.1 Taylor’s Theorem

Theorem 3.2 use a degree k polynomial to approximate f

(Taylor’s theorem) Let f ∶ X → R be a function of class C k where
X ⊂ Rn is open. For any a ∈ X, define
k n
1
Pk (x) = f (a) + ∑ ( ∑ fxi1 xi2 ⋯xij (a)(xi1 − ai1 )⋯(xij − aij )).
j=1 j! i1 ,i2 ,...,ij =1
ipad
Then
f (x) = Pk (x) + Rk (x, a)
where
Rk (x, a)
lim k
= 0.
x→a ∥x − a∥
2
e.g. ∑ g(i1 , i2 ) = g(1, 1) + g(1, 2) + g(2, 1) + g(2, 2)
i1 ,i2 =1
=double summation
3.1 Taylor’s Theorem

e.g. for n = 2 and k = 2, we have

vector a = (a, b) vector x = (x, y)
P2 (x, y) = f (a, b) + fx (a, b)(x − a) + fy (a, b)(y − b)
1
+ [fxx (a, b)(x − a)2 + fxy (a, b)(x − a)(y − b)
2 *same if f is of class C^2
+ fyx (a, b)(y − b)(x − a) + fyy (a, b)(y − b)2 ]

if we let a = (a, b) and x = a + h = (a + h, b + k), then

P2 (x) = f (a) + fx (a)h + fy (a)k

1
+ (fxx (a)h2 + fxy (a)hk + fyx (a)kh + fyy (a)k 2 )
2
3.1 Taylor’s Theorem

e.g. for n = 2 and k = 3, we have

P3 (x, y) = f (a, b) + fx (a, b)(x − a) + fy (a, b)(y − b)

1
+ [fxx (a, b)(x − a)2 + fxy (a, b)(x − a)(y − b)
2!
+ fyx (a, b)(y − b)(x − a) + fyy (a, b)(y − b)2 ]
1
+ [fxxx (a, b)(x − a)3 + fxxy (a, b)(x − a)2 (y − b)
3!
+ fxyx (a, b)(x − a)2 (y − b) + fyxx (a, b)(x − a)2 (y − b)
+ fxyy (a, b)(x − a)(y − b)2 + fyxy (a, b)(x − a)(y − b)2
+ fyyx (a, b)(x − a)(y − b)2 + fyyy (a, b)(y − b)3 ]
3.1 Taylor’s Theorem

Proof.
(The case n = 2 and k = 2)
let a = (a, b) and x = (a + h, b + k)
define F (t) = f (a + th, b + tk), so F (0) = f (a) and F (1) = f (x)
F ′′ (0)
by Taylor’s theorem in one variable, F (1) ≈ F (0) + F ′ (0) +
2
by the chain rule, we have
d d
F ′ (t) = fx (a + th, b + tk) (a + th) + fy (a + th, b + tk) (b + tk)
dt dt
= fx (a + th, b + tk)h + fy (a + th, b + tk)k,
F ′′ (t) = fxx (a + th, b + tk)h2 + fxy (a + th, b + tk)hk
+ fyx (a + th, b + tk)kh + fyy (a + th, b + tk)k 2
3.1 Taylor’s Theorem

Proof.
(The case n = 2 and k = 2)
therefore, we have
F ′′ (0)
F (1) ≈ F (0) + F ′ (0) +
2
= f (a) + fx (a)h + fy (a)k
1
+ (fxx (a)h2 + fxy (a)hk + fyx (a)kh + fyy (a)k 2 )
2
3.1 Taylor’s Theorem

Proposition 3.1
Let f ∶ X → R be a function of class C k+1 where X ⊂ Rn is open. For
any a ∈ X, define
k n
1
Pk (x) = f (a) + ∑ ( ∑ fxi1 xi2 ⋯xij (a)(xi1 − ai1 )⋯(xij − aij )).
j=1 j! i1 ,i2 ,...,ij =1

Then there exists ξ lying on the segment between a and x such that
Error term:
n
1
f (x) = Pk (x)+ ∑ fxi1 xi2 ⋯xik+1 (ξ)(xi1 − ai1 )⋯(xik+1 − aik+1 ).
(k + 1)! i1 =1
i2 =1 unknown vector!
...
ik+1 =1 vector x

vector a
3.1 Taylor’s Theorem

e.g. for n = 2 and k = 2, the remainder term is

1
[fxxx (ξ)(x − a)3 + fxxy (ξ)(x − a)2 (y − b)
3!
+ fxyx (ξ)(x − a)2 (y − b) + fyxx (ξ)(x − a)2 (y − b)
+ fxyy (ξ)(x − a)(y − b)2 + fyxy (ξ)(x − a)(y − b)2
+ fyyx (ξ)(x − a)(y − b)2 + fyyy (ξ)(y − b)3 ]
3.1 Taylor’s Theorem

Definition 3.2
Let f ∶ X → R be a function of class C 2 where X ⊂ Rn is open. The
Hessian of f is defined by

⎛ fx1 x1 fx1 x2 ⋯ fx1 xn ⎞

=
⎜f fx2 x2 ⋯ fx2 xn ⎟
Hf = ⎜ x2 x1 ⎟.
⎜ ⋮ ⋮ ⋱ ⋮ ⎟
⎝fx x fxn x2 ⋯ fx x ⎠
n 1 n n
3.1 Taylor’s Theorem

for f ∶ R2 → R, x = (x, y), a = (a, b), we have

P2 (x) = f (a, b) + fx (a, b)(x − a) + fy (a, b)(y − b)

1
+ [fxx (a, b)(x − a)2 + fxy (a, b)(x − a)(y − b)
2
+ fyx (a, b)(y − b)(x − a) + fyy (a, b)(y − b)2 ]
x−a
= f (a, b) + (fx (a, b) fy (a, b)) ( )
y−b
1 f (a, b) fxy (a, b) x − a
+ (x − ay − b) ( xx )( )
2 fyx (a, b) fyy (a, b) y − b
1
= f (a) + Df (a)(x − a) + (x − a)T Hf (a)(x − a)
2
*the answer should be a polynomial in x, y if a is given
in fact, the same formula holds for any f ∶ Rn → R
3.1 Taylor’s Theorem

Example 3.2
Define f ∶ R2 → R by f (x, y) = ex sin (x + y) and consider a = 0. Then

2ex cos (x + y) ex (− sin (x + y) + cos (x + y))

Hf = ( x )
e (− sin (x + y) + cos (x + y)) −ex sin (x + y)

and
P2 (x, y) = x + y + x2 + xy.
3.1 Taylor’s Theorem

Proof.

given f (x, y) = ex sin (x + y)

we have Df = (ex (sin (x + y) + cos (x + y)) ex cos (x + y))
the Hessian Hf is

2ex cos (x + y) ex (− sin (x + y) + cos (x + y))

( x )
e (− sin (x + y) + cos (x + y)) −ex sin (x + y)

2 1
since Df (0, 0) = (1 1) and Hf (0, 0) = ( ), we obtain
1 0

x 1 2 1 x
P2 (x, y) = f (0, 0) + (1 1) ( ) + (x y) ( )( )
y 2 1 0 y
= x + y + x2 + xy *Polynomial in x, y
3.1 Taylor’s Theorem

Example 3.3
Define f ∶ R2 → R by f (x, y) = ln (x2 + y 2 + 1) and consider a = 0. For
any (x, y) with ∣x∣ , ∣y∣ ⩽ 0.1, we have

f (x, y) = x2 + y 2 + E(x, y),

where ∣E(x, y)∣ ⩽ 0.00083.

error term
=absolute value of the remainder of Taylor's Theorem.
3.1 Taylor’s Theorem

Proof.

given f (x, y) = ln (x2 + y 2 + 1)

we have
2x 2y
Df (x, y) = (fx fy ) = ( )
x2 + y 2 + 1 x2 + y 2 + 1
and

⎛ 2(−x + y + 1) −4xy
2 2
⎞
f fxy ⎜ (x2 + y 2 + 1)2 (x2 + y 2 + 1)2 ⎟
Hf (x, y) = ( xx )=⎜
⎜
⎟
⎟
fyx fyy ⎜ −4xy 2(x2 − y 2 + 1) ⎟
⎝ (x2 + y 2 + 1)2 (x2 + y 2 + 1)2 ⎠
3.1 Taylor’s Theorem

Proof.

using the second-order Taylor’s theorem, the main term is

x 1 x
P2 (x, y) = f (0, 0) + Df (0, 0) ( ) + (x y) Hf (0, 0) ( )
y 2 y
x 1 2 0 x
= 0 + (0 0) ( ) + (x y) ( )( )
y 2 0 2 y
= x2 + y 2

by proposition 3.1, the remainder term is

1
E(x, y) = (fxxx (ξ)x3 + 3fxxy (ξ)x2 y + 3fxyy (ξ)xy 2 + fyyy (ξ)y 3 )
3!
for some ξ = (ξ1 , ξ2 ) lying between (0, 0) and (x, y)
3.1 Taylor’s Theorem

Proof.

we compute

−4x(−x2 + 3y 2 + 3)
fxxx = ,
(x2 + y 2 + 1)3
−4y(−3x2 + y 2 + 1)
fxxy = fxyx = fyxx = ,
(x2 + y 2 + 1)3
−4x(x2 − 3y 2 + 1)
fxyy = fyxy = fyyx = ,
(x2 + y 2 + 1)3
−4y(3x2 − y 2 + 3)
fyyy =
(x2 + y 2 + 1)3
3.1 Taylor’s Theorem

Proof.

since ξ lies between (0, 0) and (x, y), we have ∣ξ1 ∣ , ∣ξ2 ∣ ⩽ 0.1
by the triangle inequality, we obtain
*absolute every term
−4ξ1 (−ξ12 + 3ξ22 + 3)
∣fxxx (ξ)∣ = ∣ ∣
(ξ12 + ξ22 + 1)3
∣4ξ1 ∣ (∣ξ12 ∣ + ∣3ξ22 ∣ + ∣3∣) ≤ (0.1)^2....
⩽
(ξ12 + ξ22 + 1)3 ≥ (0+0+1)^3

4(0.1)[(0.1) + 3(0.1)2 + 3]
2
⩽
(02 + 02 + 1)3
= 1.216

similarly, we find that ∣fxxy ∣ ⩽ 0.416, ∣fxyy ∣ ⩽ 0.416 and ∣fyyy ∣ ⩽ 1.216
3.1 Taylor’s Theorem

Proof.

by the triangle inequality, we finally obtain

∣E(x, y)∣
1
⩽ [∣fxxx (ξ)x3 ∣ + ∣3fxxy (ξ)x2 y∣ + ∣3fxyy (ξ)xy 2 ∣ + ∣fyyy (ξ)y 3 ∣]
6
1
⩽ [(1.216)(0.1)3 + 3(0.416)(0.1)3 + 3(0.416)(0.1)3
6
+ (1.216)(0.1)3 ]
⩽ 0.00083