GEG 311 Lecture _6: Multivariable functions analysis

Dr. Patrick S. Olayiwola

Dr. John Ogbemhe
Engr. Chidi Onyedikan
 Functions from ℝn to ℝm
Let 𝑈 be a subset of 𝑅𝑛, and let f: U → ℝ𝑚 be a function
from U to ℝ𝑚. Let set 𝑢1, … , 𝑢𝑛 be a basis of ℝ𝑛, and let
𝑣Ԧ 1, … , 𝑣Ԧ 𝑚 be a basis of ℝ𝑚. With the input variable
written as
𝑥Ԧ = 𝑥1, 𝑥2, … , 𝑥𝑛 = 𝑥1𝑢1 + 𝑥2𝑢2 + ⋯ + 𝑥𝑛𝑢𝑛
and the output as
f 𝑥Ԧ = f1 𝑥Ԧ , … , f𝑚 𝑥Ԧ
= f1 𝑥1, 𝑥2, … , 𝑥𝑛 𝑣Ԧ 1 + ⋯ + f𝑚 𝑥1, 𝑥2, … , 𝑥𝑛 𝑣Ԧ 𝑚
The functions f1 𝑥1, 𝑥2, … , 𝑥𝑛 𝑣Ԧ 1, … , f𝑚 𝑥1, 𝑥2, … , 𝑥𝑛 are
called the component functions of f w.r.t. the basis
𝑣Ԧ 1, … , 𝑣Ԧ 𝑚 .
Thus, f: ℝ𝑛 → ℝ𝑚 is called vector functions of a vector
variable.
 Functions from ℝn to ℝm
In any Euclidean spaceℝ𝑛, the standard basis is the set of
vectors written as 𝑒Ԧ1, … , 𝑒Ԧ𝑛 , where
0
⋮
𝑒Ԧ𝑖 = 1
⋮
0
So, if no basis is explicitly specified forℝ𝑛, then it is
assumed that we can use the standard basis.
Curves in ℝ𝒏. A parametrized curve into n-dimensional
space is a continuous function 𝑥: Ԧ 𝐼 → ℝ𝑛. Parametrized
curves are vector functions of a vector variable, but the
variable is taken from a subset of a one-dimensional
vector space.
 Functions from ℝn to ℝm
Nonlinear Coordinate Changes. A general change of
coordinate in ℝ2 is a function 𝑓: U → ℝ2, where U is the
subset of ℝ2 in which the coordinates are defined. E.g. the
change from polar coordinates to the Cartesian coordinates
is given by the function 𝐹: ℝ2 → ℝ2 defined by
𝐹 𝑟, 𝜃 = (𝑟𝑐𝑜𝑠𝜃, 𝑟𝑠𝑖𝑛𝜃).
In multivariable calculus, we study 𝐹: ℝ𝑛 → ℝ, which is
written as F 𝑥1, 𝑥2, … , 𝑥𝑛 . These functions are examples of
vector functions of a vector variable with codomain ℝ.
Example: Consider a function from ℝ2 to ℝ3 such that
2𝑥2(1−𝑥12) 4𝑥2𝑥1 1−𝑥22
𝐹 𝑥1, 𝑥2 = , ,
(1+𝑥12)(1+𝑥22) (1+𝑥12)(1+𝑥22) 1+𝑥22
Notice that the component functions satisfy
 Functions from ℝn to ℝm
𝐹12 + 𝐹22 + 𝐹32
4𝑥22 1 − 𝑥12 2 + 16𝑥12𝑥22 + 1 + 𝑥12 2 1 − 𝑥22 2
=
1 + 𝑥12 2 1 + 𝑥22 2
1+𝑥12 2
1+𝑥22 2
= =1
1+𝑥12 2
1+𝑥22 2

Thus, the image of 𝐹 lies on the unit sphere

𝕊2 = (𝑥, 𝑦, 𝑧) ∈ ℝ3|𝑥2 + 𝑦2 + 𝑧2 = 1
It is interesting to note that 𝐹 is not surjective onto 𝕊2.
Assuming 𝑥2 + 𝑦2 + 𝑧2 = 1, if 𝐹 𝑢, 𝑣 = (𝑥, 𝑦, 𝑧), then in
particular
1−𝑣2 1−𝑧
𝑧= 𝑣=
1+𝑣2 1+𝑧
which implies that −1 < 𝑧 ≤ 1, and hence, the point
(0,0, −1) is not an image of 𝐹. Furthermore, since
 Functions from ℝn to ℝm
2𝑣 2
𝑧2+ =1
1+𝑣2

For any fixed z, we have

1−𝑢2 2𝑢
𝑥= 1− 𝑧2 and y= 1 − 𝑧2
1+𝑢2 1+𝑢2
But then, if y=0, it is impossible to obtain 𝑥 = − 1 − 𝑧2.
Consequently, the image of F is all points on 𝕊2 except
points in the set 𝑥, 𝑦, 𝑧 𝜖 𝕊2|𝑥 = − 1 − 𝑧2, with z < 1 .

Fig. 1: Portion of the image for the example.

This shows the image of F over the rectangle
𝑥, 𝑦, 𝑧 𝜖 −2,5 × 0.5,5 .
 Functions from ℝn to ℝm
Note that when n or m >3, visualizing graphs of functions
becomes difficult. Recall also that graph of a function
𝑓: ℝn → ℝm is the subset of ℝn × ℝm = ℝn+m defined by
𝑥1, … , 𝑥𝑛, 𝑦1, … , 𝑦𝑚 ∈ ℝn+m| 𝑦1 𝑦𝑚 = 𝑓 𝑥1, … , 𝑥𝑛 .
,…,
Example:
𝐹: ℝ2 → ℝ ⇒ 𝐹 𝑥, 𝑦 = 𝑐𝑖 ⇒ 𝑥2+𝑦2+1 2y
with c = 0, ±0.2, ±0.4, ±0.6, ±0.8.

See Fig. 2 for the contour diagram of F.

𝐹: ℝ3 → ℝ ⇒ 𝐹 𝑥, 𝑦, 𝑧 = 𝑐𝑖
We may depict a function𝑓: Ԧ ℝ → ℝ2 as a plane curve and a
function𝑓:Ԧ ℝ → ℝ3 as a space curve, so that we plot the
image of 𝑓Ԧ in ℝ2 or ℝ3. In this way, visual information is lost
about the magnitude 𝑓′(t) Ԧ , intuitively speaking, how fast
one travels along the curve.
 Functions from ℝn to ℝm
𝑓Ԧ 𝑡 = 1 − 𝑐𝑜𝑠𝑡 𝑐𝑜𝑠𝑡, 1 − 𝑐𝑜𝑠𝑡 𝑠𝑖𝑛𝑡, 𝑠𝑖𝑛𝑡 .
Similarly, for the geometry of surfaces, we depict a function
Ԧ ℝ2 → ℝ3 by plotting its image in ℝ3. ( The graph of a
𝐹:
function of the form ℝ2 → ℝ3 is a subset of ℝ5, which is
quite difficult to visualize.

Fig. 2b: shows the image of the

Fig. 2a: contour so-called space cardioid, given
diagram of F by the function
 Functions from ℝn to ℝm
D𝐞𝐟𝐧 𝟏: Let 𝑓Ԧ and 𝑔Ԧ be two functions defined over a
subset 𝑈 of ℝn with codomain ℝm . Then we define the
following functions:
I. (𝑓Ԧ + 𝑔):
Ԧ 𝑈 → ℝm, where 𝑓Ԧ + 𝑔Ԧ 𝑥Ԧ = 𝑓Ԧ 𝑥Ԧ + 𝑔Ԧ 𝑥Ԧ
II. (𝑓Ԧ ∙ 𝑔):
Ԧ 𝑈 → ℝ, where(𝑓Ԧ ∙ 𝑔)
Ԧ 𝑥Ԧ = 𝑓Ԧ 𝑥Ԧ ∙ 𝑔Ԧ 𝑥Ԧ
III. If 𝑚 = 3, (𝑓Ԧ × 𝑔):
Ԧ 𝑈 → ℝ3, where(𝑓Ԧ × 𝑔)
Ԧ 𝑥Ԧ = 𝑓Ԧ 𝑥Ԧ × 𝑔Ԧ 𝑥Ԧ
D𝐞𝐟𝐧 𝟐: Let 𝑓Ԧ be a function from a subset 𝑈 ⊂ ℝn to ℝm,
and let 𝑔Ԧ be a function from 𝑉 ⊂ ℝm to ℝs. If the image of 𝑓Ԧ
is a subset of 𝑉, then the composition function 𝑓Ԧ ∘ 𝑔Ԧ is the
function 𝑈 → ℝs defined by (𝑓Ԧ ∘ 𝑔) Ԧ 𝑥Ԧ = 𝑔( Ԧ 𝑓Ԧ 𝑥Ԧ ).
D𝐞𝐟𝐧 𝟑: A function 𝐹: ℝn → ℝm is called a linear function if
𝐹 𝑥Ԧ + 𝑦Ԧ = 𝐹(𝑥)+Ԧ 𝐹(𝑦) Ԧ Ԧ 𝑦Ԧ ∈ ℝn
for all 𝑥,
𝐹(𝑘 𝑥)=k𝐹(
Ԧ 𝑥)
Ԧ for all k∈ ℝ 𝑎𝑛𝑑 𝑥Ԧ ∈ ℝn)
 Functions from ℝn to ℝm
If a function 𝐹: ℝn → ℝm is linear, then
𝐹 0 = 𝐹 0 − 0 = 𝐹(0)-𝐹(0)= 0,
And hence 𝐹 maps the origin of ℝn to the origin of ℝm.
If ℬ = 𝑢1, … , 𝑢𝑛 is a basis of ℝn, then any vector 𝑢 ∈ ℝn
can be written uniquely as a linear combination of vectors in
ℬ as
𝑢= 𝑐1𝑢1 + ⋯ + 𝑢 𝑛
That is,
𝑐1
[𝑢]ℬ= ⋮
𝑐𝑛
If, now, the basis ℬ is not specified we assume the standard
basis for the coefficients. If 𝐹 is a linear function, then
𝐹 𝑢 = 𝑐1𝐹 𝑢1 + ⋯ + 𝑐𝑛𝐹 𝑢𝑛 ,
 Functions from ℝn to ℝm
hence, to know all outputs of 𝐹 we need to know the
coefficients of [𝑢]ℬ and the outputs of the basis vectors of ℬ.
Suppose also that ℬ′ = 𝑣Ԧ 1, … , 𝑣Ԧ 𝑛 is a basis of ℝm. If the ℬ′ -
coordinates of the outputs of the vectors inℬ are
𝑎11 𝑎12 𝑎1𝑛
𝑎21 𝑎22 𝑎2𝑛
[𝐹(𝑢1)]ℬ′= ⋮ , [𝐹(𝑢2)]ℬ′= ⋮ ,…, [𝐹(𝑢 1)]ℬ′= ⋮
𝑎𝑚1 𝑎𝑚2 𝑎𝑚𝑛
Then the image of the vector 𝑢 ∈ ℝn is given by
𝑎11 𝑎12 𝑎1𝑛 𝑐1 𝑎11 𝑎12 𝑎1𝑛
𝑎21 𝑎22 𝑎2𝑛 𝑐2 𝑎21 𝑎22 𝑎2𝑛
[𝐹(𝑢)]ℬ′= ⋮ ⋮ ⋮ ⋮ . The matrix 𝐴 = ⋮ ⋮ ⋮
𝑎𝑚1 𝑎𝑚2𝑎𝑚𝑛 𝑐𝑛 𝑎𝑚1 𝑎𝑚2𝑎𝑚𝑛
Is called the ℬ, ℬ’-matrix of the linear function 𝐹. Thus, one
fundamental result is [𝐹(𝑢)]ℬ′=𝐴[𝑢]ℬ
For some 𝑚 × 𝑛 matrix 𝐴.
 Functions from ℝn to ℝm
Given a linear function 𝐹: ℝ2 → ℝ3, the image of is the set
Im 𝐹 = 𝐹(ℝn), and the kernel of 𝐹 is the zero set
ker𝐹 = ∈ ℝn|𝐹 𝑢 = 0
The image Im 𝐹 is a vector subspace of ℝm, and the kernel
is a subspace of ℝn. The rank of 𝐹is the dimension
dim(Im𝐹) . Hence,
rank(𝐹)≤ 𝑚𝑖𝑛 𝑚, 𝑛
So, a linear function 𝐹: ℝn → ℝm is surjective iff rank(𝐹)=m,
and 𝐹 is injective iff rank(𝐹)=n.
The rank is also useful for determining the linear
dependence between set of vectors. If 𝑢 1, … , 𝑢𝑛 is a set of
vectors in ℝm, then the matrix
𝐴 =(𝑢1, … , 𝑢𝑛)
where the 𝑢𝑖 are viewed as column vectors, represent a
 Functions from ℝn to ℝm

linear function 𝐹: ℝn → ℝm, with

Im𝐹 = 𝑠𝑝𝑎𝑛 𝑢1, … , 𝑢𝑛
Thus, the set of vectors 𝑢1, … , 𝑢 𝑛 is linearly
independent iff rank 𝐹 = 𝑛.
If 𝑛 = 𝑚, and
𝑑𝑒𝑡𝐴 ≠ 0, 𝑡ℎ𝑒𝑛 rank 𝐴 = 𝑛 = 𝑚,
and 𝑑𝑒𝑡𝐴 is the n-volume of the image under 𝐹 of
the unit n-cube. Consequently, if the columns of 𝐴 are
not linearly dependent, the n-volume of this
parallelepiped will be 0.

Hence, the following theorem,

 Functions from ℝn to ℝm
Theorem 4: For a linear system 𝐹: ℝn → ℝn with associated
square matrix 𝐴, the followings are true:
1. rank 𝐹 = 𝑛.
2. det 𝐴 ≠0.
3. Im 𝐹 = ℝn.
4. kernel 𝐹= 0 .
5. The column vectors of 𝐴 are linearly independent.
6. The column vectors of 𝐴 form a basis of ℝn.
7. The column vectors of 𝐴 span ℝn.
8. 𝐹 has an inverse function.
 Inverse and Implicit Functions
In a single-variable and multivariable calculus of a function 𝐹: ℝn → ℝ,
a critical is defined as a point 𝑎Ԧ = 𝑎1, … , 𝑎𝑛 such that the gradient of
𝐹 at 𝑎Ԧ is 0, i.e.,
𝜕𝐹 𝜕𝐹
∇𝐹(𝑎)=
Ԧ (𝑎),…,
Ԧ (𝑎)
Ԧ =0
𝜕𝑥1 𝜕𝑥𝑛
At such a point 𝐹 is said to have a flat tangent line or tangent plane,
and, 𝐹(𝑎)Ԧ is optimal, either a local minimum, local maximum, or a
“saddle point”.
Defn. 5: Let 𝑈 be an open subset of ℝn, 𝐹: 𝑈 → ℝ𝑚 a differentiable
function, and 𝑞 a point in 𝑈. 𝑞 is a critical point of 𝐹 if 𝐹 is not
differentiable at 𝑞 or if d𝐹𝑞 : ℝn → ℝm is not of maximum rank, i.e.,
d𝐹𝑞 < min(𝑚, 𝑛). If 𝑞 is a critical point of 𝐹 ,then 𝐹(𝑞) is a critical
value. If 𝑝 ∈ ℝm is not a critical value of 𝐹, even if 𝑝 is not in the image
of 𝐹), then 𝑝 is a regular value of 𝐹.
 Inverse and Implicit Functions
Proposition 6: Let 𝑈 be an open subset of ℝn, 𝐹: 𝑈 → ℝ𝑛
a differentiable function, and 𝑞 a point in 𝑈such that 𝐹 if 𝐹 is
differentiable at 𝑞 . The following are equivalent:
1. 𝑞 is a critical point of 𝑈.
2. 𝐽 𝐹 𝑞 = 0.
𝜕𝐹 𝜕𝐹
3. The set of partial derivatives (𝑞),…, (𝑞) is a linearly dependent set of
𝜕𝑥1 𝜕𝑥𝑛
vectors.
4. The differential d𝐹𝑞 is not invertible.
Proposition 7: Let 𝐹: 𝑈 → ℝ𝑚 be a function 𝑈 is an open subset of ℝn,
with 𝑛 ≠ 𝑚. Let 𝑞 ∈ 𝑈 such that 𝐹 is differentiable at 𝑞, and call 𝐴 the
Jacobian matrix 𝑑𝐹𝑞 𝑎𝑡 𝑞. Then the following are equivalent:
1. 𝑞 is a critical point of 𝑈.
2. The determinants of all the maximal square submatrices of 𝐴 𝑎𝑟𝑒 0.
3. The sum of the squares of the determinants of all the maximal squares of
𝐴 𝑖𝑠 0.
Furthermore, if 𝑛 > 𝑚, then 𝑞 is a critical point of 𝑈 iff det(𝐴𝐴𝑇) ≠0.
 Inverse and Implicit Functions
Example: For example, consider the function F : ℝ3 → ℝ2
defined by 𝐹 𝑥, 𝑦, 𝑧 = 𝑥 2 + 3𝑦 + 𝑧 3 , 𝑥𝑦 + 𝑧 2 + 1. The
2𝑥 3 3𝑧 3
Jacobian matrix for this function is 𝑑𝐹] = .
𝑦 𝑥 2𝑧
The maximal 2 × 2 submatrices are
2𝑥 3 2𝑥 3𝑧 3 3 3𝑧 3
, ,
𝑦 𝑥 𝑦 2𝑧 𝑥 2𝑧
So, since critical points occur where all these have
determinant 0, the critical points satisfy the system of
equations
2 2 2 2
𝑦= 𝑥 , 𝑦= 𝑥 ,
3 3
൞4𝑥𝑧 − 𝑥 2 𝑧 2 =
0, ⟺ 𝑥𝑧 2 − 𝑥𝑧 = 0,
𝑧 2 − 𝑥𝑧 = 0, 𝑧 2 − 𝑥𝑧 = 0,
 Inverse and Implicit Functions
Thus, the set of critical points of 𝐹 is
2 2 2 2 2
𝑥, 𝑥 , 𝜖ℝ3| 𝑥 𝜖ℝ− 0 ⋃ 𝑥, 𝑥 , 0 𝜖ℝ3| 𝑥 𝜖ℝ
3 𝑥 3
The set of critical values is then
8 2 2 3 4
3𝑥 2 + 3 𝑥 , 𝑥 + + 1 𝜖ℝ2| 𝑥 𝜖ℝ− 0 ⋃
𝑥 3 𝑥2
2 2 2
3𝑥 , 𝑥 + 1 𝜖ℝ2| 𝑥 𝜖ℝ
3
One important aspect of critical points already arises with real
functions. With a real differentiable function 𝑓 ∶ [𝑎, 𝑏]→ℝ, if 𝑓′ 𝑥0 =
0, one can show that f does not have an inverse function that is
differentiable over a neighborhood of x0. Conversely, if 𝑓′(𝑥0) ≠ 0, the
function 𝑓 has a differentiable inverse in a neighborhood of 𝑥0, with
−1 1
𝑓 ′ 𝑦0 = −1
𝑓′ 𝑓 𝑦0
 Inverse and Implicit Functions
Theorem: (Inverse Function Theorem). Let F be a function
from an open set U ⊂ ℝn to ℝn, and suppose that F is of
class Cr, with r ≥ 1. If q ∈ U is not a critical point of F, then
dFq is invertible and there exists a neighborhood V of q
such that F is one-to-one on V , F(V ) is open, and the
inverse function F −1 : F(V ) → V is of class Cr. Furthermore,
for all p ∈ F(V ), with p = F(q),
d 𝐹 −1 𝑝 = 𝑑𝐹𝑞 −1
Example: Consider the function 𝐹 𝑠, 𝑡 = (𝑠2 − 𝑡2, 2𝑠𝑡)
and 𝑞 = 2,3 . Note that 𝐹 is defined on all 𝑈 = ℝ2. The
Jacobian matrix is
2𝑠 −2𝑡
, i.e. 4 𝑠2 + 𝑡2 . 𝐹 satisfies the inverse
2𝑡 2𝑠
conditions at 0,0 .