Unit 9: Hyperplanes

Juan Luis Melero ans Eduardo Eyras

October 2018

1
Contents
1 Hyperplanes 3
1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Hyperplane equations . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Vector product . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Tangent hyperplane 11
2.1 Directional derivative . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Theorems related to the tangent hyperplane . . . . . . . . . . 12
2.2.1 Maximum value of a directional derivative . . . . . . . 12
2.2.2 The gradient is orthogonal to the level curve . . . . . . 12
2.3 Definition of tangent hyperplane . . . . . . . . . . . . . . . . . 14
2.4 Example of tangent hyperplanes and applications . . . . . . . 15
2.4.1 Tangent hyperplanes for contour lines . . . . . . . . . . 15
2.4.2 Lagrange multipliers . . . . . . . . . . . . . . . . . . . 16

3 Exercises 18

2
1 Hyperplanes
1.1 Definition
A hyperplane in an n−dimensional vector space Rn is defined to be the set
of vectors:  
x1
 .. 
u= . 
xn
satisfying the equation:

a1 x1 + · · · + an xn = b

where a1 , . . . , an and b are real numbers with at least a1 , . . . , an non-zero.

For example, in R2 a hyperplane is a line:

Figure 1: Graphical representation of the hyperplane equation x + y = 4

And for example, in R3 , a hyperplane is a 2D plane:

3
Figure 2: Graphical representation of the hyperplane equation 4x + y + z = 2

1.2 Hyperplane equations

To define the hyperplane equation we need either a point in the plane and a
unit vector orthogonal to the plane, two vectors lying on the plane or three
coplanar points (they are contained in the hyperplane). Is is usually built
from the combination of a point and a vector and it corresponds to a n − 1
dimensional vector subspace.

For the first case, we will get the equation from a point and an orthogonal
unit vector.
The vector p − p0 lies on the plane, thus it is orthogonal to n:

n ⊥ (p − p0 ) =⇒ hn, p − p0 i = 0

This provides the vector equation of the plane: hn, p − p0 i = 0. Given the
vector:  
a
n = b

c

4
Figure 3: Representation of the points and orthogonal vectors of a R3 hyper-
plane. p0 represents a vector indicating an specific known point in the plane.
p represents a vector indicating any point in the plane. n represents a unit
vector orthogonal to the plane.

Performing the scalar product and taking common factors, we end up with
the equation:
a(x − x0 ) + b(y − y0 ) + c(z − z0 ) = 0
This is the scalar equation of the plane. It is also possible to write it in terms
of a constant value D:

D = ax0 + by0 + z0 =⇒ ax + by + cz = D

This is the general equation of the plane.

As n defines the normal line to the hyperplane, we can define a normal line
with the normal vector:

 x(t) = x0 + at
r(t) = p0 + tn =⇒ y(t) = y0 + bt
z(t) = z0 + ct


 
a
for n = b .

c

5
Figure 4: Graphical representation of a line, which is a R2 hyperplane. v
represents the vector in the direction of the hyperplane. p represents the
point which the hyperplane goes through.

Example: line in R2 in the direction of ~v = (1, 1) and going through the

point p = (1, 0).
Any point in the line is given by the equation:
 
x = p1 + αv1 = 1 + α
r = p + αv =
y = p2 + αv2 = 0 + α

This is the parametric equation. We can isolate the parameter α and have
the equation:
 
x = p1 + αv1 = 1 + α
r = p + αv = →y =x−1
y = p2 + αv2 = 0 + α

This is the intrinsic equation.

We can build it as a hyperplane. Calculate a vector orthogonal to the line:

   √ 
u1 1/ √2
hn, vi = u1 + u2 = 0 → n = →n= , knk = 1
−u1 −1/ 2

So, with the orthogonal vector and the specific known point p, we can build
the equation for any point r of the hyperplane:

hn, ri = hn, pi

6
 √ √
x √ √
1
   
1/ 2 −1/ 2 = 1/ 2 −1/ 2 →
y 0
1 1
→ √ (x − y = √ = y = x − 1 the same as before
2 2
If we want to build the normal line through the point, we use the normal
vector and the parametric equation:
 √  (
1/ √2 x(t) = x0 + t √12
n= → r(t) = p0 + tn →
−1/ 2 y(t) = y0 − t √12

Using the same point p0 = (1, 0) as before, we get:

( )
x(t) = 1 + t √12
→y =1−x
y(t) = −t √12

Figure 5: Graphical representation of the normal line of an R2 hyperplane.

v represents the vector in the direction of the hyperplane. p represents the
point which the hyperplane and the normal line go through. y = 1 − x is the
equation of the normal line.

Example in R3 : consider the normal vector ~n and the vector p0 representing

a given point in the plane:
   
4 0
~n = −1 , p0 =  1 
6 −7

7
We can write the equation of the hyperplane defined by these two vectors:
hn, p − p0 i = 0 →
→ a(x − x0 ) + b(y − y0 ) + c(z − z0 ) = 0 →
→ 4(x − 0) − (y − 1) + 6(z + 7) = 0 →
→ 4x − y + 6z = −43
where p represents a vector for a generic point in the plane:
 
x
p = y 
z
As we said before:
a(x−x0 )+b(y−y0 )+c(z−z0 ) = 0 ↔ ax+by+cz = D, with D = ax0 +by0 +cz0
There is no point (x, y, z) can fulfill two conditions:
ax + by + cz = D
ax + by + cz = D0
D 6= D0
So there are no common points, i.e. no intersection → they are parallel.

A hyperplane in Rn can be represented by a single equation:

a1 x 1 + a2 x 2 + · · · + an x n = D
 
a1
 .. 
where n =  .  is a vector orthogonal to the hyperplane as we chose a
an
point p0 in the plane such that hn, p0 i = D.

All planes of that similar form, for any value of D are parallel to each other.
In particular, for D = 0 we have a1 x1 + a2 x2 + · · · + an xn = 0, which can be
re-written in temps of a 1 × n matrix AT = n:
 
x

 .1 
a1 . . . an  ..  = 0 → Au = 0
xn

8
Hence, the hyperplane is the kernel of the matrix A: Ker(A). The hyper-
plane is a vector subspace. In addition, considering that rank(A) = 1 and
dim(Rn ) = n then we see that the hyperplane has dimension n − 1:

rank(A) + dim(Ker(A)) = n → dim(Ker(A)) = n − 1

1.3 Vector product

Vector product or the cross product is an operation that allows to find a
vector perpendicular to two known vectors. This is useful here because to
build the hyperplane equation we may need it.

Given two vectors in R3 , u, v ∈ R3 and the canonical basis E3 = {e1 , e2 , e3 },

we define the vector product of these two vectors as:
 
e1 e2 e3
u × v = det u1 u2 u3  = (we solve it with the adjoint method)
v1 v2 v3
     
u2 u3 u1 u3 u1 u2
= det e − det e + det e
v2 v3 1 v1 v3 2 v1 v2 3
Properties:

1. u × v is orthogonal to both vectors

2. If u and v are linearly dependent, then u × v = 0

3. u × v = −v × u

Proof of property 1:

hu, u × vi = hu1 e1 , (u2 v3 − u3 v2 )e1 i+hu2 e2 , (u3 v1 − u1 v3 )e2 i+hu3 e3 , (u1 v2 − u2 v1 )e3 i =

u1 (u2 v3 − u3 v2 ) he1 , e1 i + u2 (u3 v1 − u1 v3 ) he2 , e2 i + u3 (u1 v2 − u2 v1 ) he3 , e3 i =

= u1 (u2 v3 − u3 v2 ) + u2 (u3 v1 − u1 v3 ) + u3 (u1 v2 − u2 v1 ) =
= u1 u2 v3 − u1 u3 v2 + u2 u3 v1 − u2 u1 v3 + u3 u1 v2 − u3 u2 v1 = 0
Similar for hv, u × vi.

9
Proof of property 2: (recall the properties of determinants)
   
e1 e2 e3 e1 e2 e3
u × v = det u1 u2
 u3  = det  u1 u2 u3  = 0
v1 v2 v3 αu1 αu2 αu3
Proof of property 3: (recall the properties of determinants)
   
e1 e2 e3 e1 e2 e3
u × v = det u1 u2 u3  = −det  v1 v2 v3  = −v × u
v1 v2 v3 u1 u2 u3
Example: determine the equation of the plane in R3 that contains the
points:
P = (1, −2, 0) = OP
Q = (3, 1, 4) = OQ
R = (0, −1, 2) = OR
First, we calculate two vectors in the plane:
     
3 1 2
P Q = OQ − OP = 1 − −2 = 3
    
4 0 4
     
0 1 −1
P R = OR − OP = −1 − −2 =  1 
2 0 2
Now we use the vector product of these two vectors to calculate an orthogonal
vector:  
e1 e2 e3
n = P Q × P R = det  2 3 4 
−1 1 2
     
3 4 2 4 2 3
= det − det + det = 2e1 − 8e2 + 5e3
1 2 −1 2 −1 1
 
2
So the vector n = −8 is orthogonal to the plane. It can be verified with

5
the scalar product:
 

2
hn, P Qi = 2 −8 5 3 = 0
4

10
The same for P R.

Now we can write the equation of the plane:

hn, ri = hn, pi
hn, p − p0 i = 0 → a(x − x0 ) + b(y − y0 ) + c(z − z0 ) = 0
Here we choose the normal vector n calculated before and any of the three
initial points,
 since
 they all are contained in the plane. We will choose the
1
point P = −2, but any of the points would have been correct. So we get
0
the equation:
2(x − 1) − 8(y + 2) + 5(z − 0) = 0 → 2x − 8y + 5z = 18

2 Tangent hyperplane
2.1 Directional derivative
Partial derivatives, and overall the gradient vector, can be interpreted as the
derivatives in the direction to the axis (the canonical basis vectors). For
example, ∂f∂x
(x0 , y0 ) is the slope of the tangent line to f , in the point (x0 , y0 ),
which is parallel to the OX axis. In case of two variable functions, it is not
a unique tangent line but, if it exists, is a whole tangent plane. If we want
to define the derivative not in the direction of the axis but a given direction,
then we calculate the directional derivative.

Directional derivative is defined as the derivative (the gradient) in the direc-

tion of a vector:
∂f ∂f
Du f = h∇f, ui = u1 + · · · + un
∂x1 ∂xn
u is generally taken to be a unit vector.
 
1
Example: consider the function f (x, y) = xy and the vector u = .
1
Calculate the directional derivative along u.
 
y
∇f =
x

11
 ∂f ∂f
Du f = h∇f, ui = u1 + · · · + u2 =y+x
∂x1 ∂x2

2.2 Theorems related to the tangent hyperplane

2.2.1 Maximum value of a directional derivative
Theorem: the maximum value, i.e. the maximal rate of change, of a di-
rectional derivative Du f, ∀u ∈ Rn is given by k∇f k and will occur in the
direction of ∇f .

Proof: we use the definition of the scalar product in terms of the angle
between the two vectors:

Du f = h∇f, ui = k∇f kkuk cos θ = k∇f k cos θ

The maximum value of the cosine occurs at zero angle:

max{cos θ} = 1 → θ = 0
θ

This implies that u is in the same direction as ∇f . If we maximize the

directional derivative in terms of the vector u:

max{Du f } = max{k∇f k cos theta} = k∇f k max{cos θ} = k∇f k

u u θ

So we get that the maximum value of the directional derivative is the norm
of the gradient, and it is parallel to the gradient:

max{Du f } = k∇f k at u k ∇f
u

Q.E.D

2.2.2 The gradient is orthogonal to the level curve

Theorem: consider the level curve of a function f , which is defined as
f (x, y) = k. The gradient ∇f at a point in the level curve p = (x0 , y0 ) is or-
thogonal to the level curve at that point f (x0 , y0 ) = 0, k ∈ R. This is similar
for level surfaces if we consider the function f (x, y, z) and the level surfaces
f (x, y, z) = k, k ∈ R. The gradient vector of the function ∇f at a point on
the level surface p = (x0 , y0 , z0 ) such that f (x0 , y0 , z0 ) = k is perpendicular

12
to the level surface at that point.

Proof: consider the level curve of a two variables function f (x, y) = k. Points
in the curve can be represented by r(t) = (x(t), y(t)) (recall the parameteri-
zation of a curve). To be in the level curve, the points of the function must
fulfill that f (x(t), y(t)) = k. The idea is to show that the gradient at a point
is orthogonal to the tangent line to the curve, which is defined as ∂r ∂t
. First,
we differentiate respect to the parameter t:

f (x(t), y(t)) = k
∂f ∂k
= = 0 k is a constant and does not depend on t
∂t ∂t
∂f ∂f ∂x ∂f ∂y
= + (recall the chain rule)
∂t ∂x ∂t ∂y ∂t
∂x
!  
∂f ∂x ∂f ∂y  ∂f ∂f  ∂t ∂r
+ = ∂x ∂y ∂y
= ∇f,
∂x ∂t ∂y ∂t ∂t
∂t
In summary:
   
∂f ∂f ∂r ∂r
=0 = ∇f, → ∇f, =0
∂t ∂t ∂t ∂t
We get that the gradient vector at a point p is orthogonal to the tangent
vector to any curve that passes through p on a surface, thence the gradient
is orthogonal to the surface.

We can generalize it to three variables function or even to n variable func-

tions. We define a level surface f (x1 , x2 , . . . , xn . Parameterized points in the
surface can be represented by r(t) = (x1 (t), x2 (t), . . . , xn (t)) and also must
fulfill that f (x1 (t), x2 (t), . . . , xn (t)) = k. Differentiating:
∂f
=0
∂t
And using the chain rule:
 ∂x1 
∂f ∂f ∂x1 ∂f ∂x2 ∂f ∂xn  ∂f   ∂t   
= + +· · ·+ = ∂x1 . . . ∂f  ...  = ∇f, ∂r
∂xn
∂t ∂x1 ∂t ∂x2 ∂t ∂xn ∂t   ∂t
∂xn
∂t

13
We get again that the gradient is orthogonal to the slope of the curve.

2.3 Definition of tangent hyperplane

With the definitions and theorems seen before we can now define the tangent
hyperplane in a point and get the equation. In the previous section we saw
the theorem that the gradient is perpendicular to the level curve or the level
surface of a function. If you recall the definition of hyperplanes and how to
get one, we need a point and an orthogonal vector. Given any point, now we
have a perpendicular vector to the function, which is the gradient.

Consider a function of two variables:

f : R2 → R
(x, y)→z = f (x, y)
We can transform this function into a ”level surface”, like it was the level
surface of a higher dimensional function:

z = f (x, y) → F (x, y, z) = f (x, y) − z = 0

Now we treat it as a level surface, so the gradient of the function F will be

orthogonal to any point of the function. Let F (x, y, z) define a surface. Then
the tangent plane to F (x, y, z) at a point (x0 , y0 , z0 ) is the plane with normal
vector ∇F (x0 , y0 , z0 ) that passes through the point (x0 , y0 , z0 ). F must be
differentiable at a point (x0 , y0 , z0 ). In general, the equation of the tangent
plane is:
∂F ∂F ∂F
h∇f, r − r0 i = (x0 , y0 , z0 )(x−x0 )+ (x0 , y0 , z0 )(y−y0 )+ (x0 , y0 , z0 )(z−z0 ) = 0
∂x ∂y ∂x
In particular, considering the previous function:
 ∂F   ∂f 
∂x ∂x
 ∂F   ∂f 
F (x, y, x) = f (x, y) − z → ∇F =  
 ∂y  = 
 ∂y 
∂F
∂z −1
Then we have:
∂f ∂f
z − z0 = (x0 , y0 )(x − x0 ) + (x0 , y0 )(y − y0 ) with z0 = f (x0 , y0 )
∂x ∂y

14
This is the equation of the tangent plane to surface z = f (x, y) at point
(x0 , y0 ).

Example: consider the surface z = f (x, y) = 3x2 − xy. We want to obtaing

the tangent plane at the point (1, 2, 1).

First, we define the surface as a level surface:

z = 3x2 − xy → F (x, y, z) = 3x2 − xy − z
We calculate the gradient at the given point:



∇F = 6x − y −x −1 → ∇F (1, 2, 1) = 4 −1 −1
Using the scalar product, we write down the equation of the tangent plane:
 

x−1
h∇F, r − r0 i = 4 −1 −1 y − 2 = 0 → 4x − y − z = 1
z−1

2.4 Example of tangent hyperplanes and applications

2.4.1 Tangent hyperplanes for contour lines
We can also calculate tangent hyperplanes of contour lines. Remember that
the gradient is perpendicular to the contour lines, hence the gradient will be
also the orthogonal vector of the tangent hyperplane.

Example: consider the function f (x, y) = x2 + y 2 , which represents a

paraboloid and its level curve f (x, y) = x2 + y 2 = k. Find the tangent
plane (line in this case) of this level curve: f (x, y) = x2 + y 2 = 8 at the point
p0 = (2, 2).

Calculate the gradient of the function at (2, 2):

   
2x 4
∇f = → ∇f (2, 2) =
2y 4
Write the equationof the tangent plane:
∂f ∂f
z − z0 = (x0 , y0 )(x − x0 ) + (x0 , y0 )(y − y0 )
∂x ∂y

15
Now z = 8 and z0 = f (2, 2) = 22 + 22 = 8:
 

x−2
8−8= 4 4 = 4x + 4y − 16 → x + y = 4
y−2

This can be graphically seen also. √The level curve f (x, y) = x2 + y 2 = 8

represents a circumference of radius 8. Drawing it and taking into account
that the gradient is perpendicular to the tangent vector we have what is
shown in Figure 6.

Figure 6: Tangent hyperplane (tangent line) of the circumference x2 +y 2 = 8.

Black line represents the tangent line with equation x + y = 4. Blue arrow
represents the gradient vector.

2.4.2 Lagrange multipliers

We can recover the definition of the Lagrange multipliers with the level curve
and tangent hyperplanes. The idea is that there are points in a function

16
f (x, y) at which the level curve f (x, y) = k coincide, and is tangent, with the
constraint. As the points are the same, the tangent lines are also the same
for f (x, y) and for the constraints G(x, y). Hence, to fins an optimal point
of f subject to the constraint G(x, y) we impose that the gradients must be
proportional to each other:

∇f = λ∇G and G(x, y) = 0

17
3 Exercises
Ex. 1 — Given a plane in R3 ,

−x + 2z = 10

and the line    

5 0
r = 2 + α −1
  
10 4
Determine whether they are orthogonal, parallel or neither.

Ex. 2 — Consider the surface x2 yz − y + z = 0. Obtain the tangent plane

at the point p0 = (1, 2, 3) and the normal line through that point.

Ex. 3 — Find the greatest and smallest values of f (x, y) = xy on the ellipse

x2 y 2
+ =1
8 2
Verify that
∇f = λ∇G
and the optimal point

18
References
[1] J; Gelonch J Bailo, E; Cecilia. Mathematics for biotechnology degree.

[2] Marc Peter Deisenroth; A Aldo Faisal and Cheng Soon Ong. Mathematics
for Machine Learning. 2018.

[3] James K. Peterson. Calculus For Biologists: A Beginning – Getting Ready

For Models and Analyzing Models. 2008.

[4] Kirsten ten Tusscher and Alexander Panfilov. Mathematics for Biologists.
2011.

19