Additional notes on Quadratic forms

Manuela Girotti

MATH 369-05 Linear Algebra I

1 Conics in R2
Definition 1. A conic is a curve in R2 described by a polynomial of degree 2 in two variables:

ax2 + bxy + cy 2 + dx + ey + f = 0

Theorem 2 (Reduction Theorem). Using a rotation and then a translation we can always
reduce any quadratic equation to standard form.
The standard forms for any possible type of conics are listed below, after the proof.

Sketch of the proof. Given the polynomial

b
    
 a x  x
0 = ax2 + bxy + cy 2 + dx + ey + f = x y b
 
2 + d e +f
2 c y y

consider the quadratic part

b
    
  a x   x
x y 2 = x y Q
b
2 c y y
and apply the Principal Axes Theorem: there exists an (orthogonal) change of variables
(
x = At + Bs
y = Ct + Ds
 
A B
where the matrix of coefficients is M = ∈ O(2) (a rotation) such that
C D
    
x  λ1 0 t
= λ1 t2 + λ2 s2
  
x y Q = t s
y 0 λ2 s
where λ1 , λ2 are the eigenvalues of Q. Substituting also in the linear part, we will obtain the
equation λ1 t2 + λ2 s2 + d (At + Bs) + e (Ct + Ds) + f = 0, meaning

λ1 t2 + λ2 s2 + (Ad + Ce) t + (Bd + De) st + f = 0.

At this point we can complete the squares and apply a new change of variables of the form
     
t ξ x
= + 0
s η y0

1
  
x0
where a fixed vector to be determined by completing the square (this change of variables
y0
represents a translation); in the end we reduce the equation to a purely quadratic one (plus a
constant):
λ1 ξ 2 + λ2 η 2 = k.

Important consequence:
The type of conic that we can obtain depends from the eigenvalues of Q!

1.1 Classification of Conics. Standard forms

Figure 1: Ellipses, Hyperbola and Parabola

We know that det(Q) = λ1 λ2 and thanks to the Reduction Theorem every conic can be reduced
into standard form
λ1 x2 + λ2 y 2 = k.
Without loss of generality, the constant k can be either zero k = 0 or it can be non-zero and
therefore normalize it to k = 1.

• if det(Q) > 0, then λ1 and λ2 have the same sign. Therefore,

x2 y2
- if k = 1 and λ1 , λ2 > 0, the conic we have is an ellipse: α2
+ β2
= 1.
2 y2
- if k = 1 and λ1 , λ2 < 0, we have an imaginary ellipse: − αx 2 − β2
=1
x2 y2
- if k = 0, we have two complex conjugate lines meeting in one real point: α2
+ β2
=
  
x y x y
α + i β α − i β = 0 (the set of real solutions is just one point (0, 0))

• if det(Q) < 0, then λ1 and λ2 have opposite sign. Therefore,

x2 y2 2 y2
- if k 6= 0, we have a hyperbola: α2
− β2
= 1 or − αx 2 + β2
=1

2
 2
  
x2
- if k = 0 we have a pair of real lines meeting in one point: −y
α2 β 2
= x
α + y
β
x
α − y
β =0

• if det(Q) = 0, then one of the eigenvalue is zero, say λ2 = 0 and λ1 6= 0 (they cannot be both
zero, otherwise we don’t have a conic to begin with). In this case we can complete the square
only for the first variable and we are left with something like

λ1 x2 + hs = k

for some constants h, k. Now,

- if h 6= 0, we can further reduce it to a parabola: λ1 x2 + y = 0.

- if h = 0, then we have the equation
k
x2 =
λ1
k
and, according to the sign of λ1 , we have:
q
k
(a) two parallel real lines for λ1 > 0: x = ± λk1
k
(b) one double line for λ1 = 0: x2 = 0
q
k k
(c) two parallel complex conjugate lines for λ1 < 0: x = ±i λ1

Conclusion: these are all the possible conics (up to congruence and similarity). There are no
others types of conics in the plane.

Why “conics”? They are called conics because they are all slices of the 3D cone

x2 + y 2 = z 2 .

3
2 Quadric surfaces in R3
Definition 3. A quadric surface is a surface in R3 described by a polynomial of degree 2 in three
variables:
ax2 + by 2 + cz 2 + dxy + exz + f yz + gx + hy + jz + k = 0
or in matrix form
a d e 
      
 d 2 2 x x x  x
f 
     
x y z 2 b 2 y + f g j y  + k = x y z Q y  + f g j y  + k = 0
e f z z z z
2 2 c

The Reduction Theorem is still valid with the rotation matrix P ∈ O(3) (plus, eventually, a
translation). Therefore, every quadric surface in R3 can be reduced to a standard form (as listed
below). If all the eigenvalues are different from zero, the standard form looks like

λ 1 x2 + λ 2 y 2 + λ 3 z 2 = κ

where λ1 , λ2 , λ3 are the eigenvalues of Q, for some constant κ (that can be taken equal to either 0
or 1 without loss of generality). If one or more eigenvalues are zero, we need to be careful when
applying the Reduction Theorem (as we did in R2 when det(Q) = 0).

2.1 Classification of Quadric Surfaces. Standard forms

Important consequence: The type of quadric surface is again determined by the eigenvalues
of the matrix Q, more specifically by the sign of the eigenvalues or the vanishing of one or two
eigenvalues.

Definition 4. The signature of a real symmetric matrix Q is the triplet (p, q, r) that indicates
the number (counted with multiplicity) of positive, negative and zero eigenvalues of Q.

Depending on the signature we have:

• (3, 0, 0) (3 positive eigenvalues, 0 negative eigenvalues and 0 0-eigenvalue)

x2 y2 z2
- ellipsoid: α2
+ β2
+ γ2
=1
x2 y2 z2
- imaginary cone with vertex at the origin: α2
+ β2
+ γ2
=0

• (2, 1, 0) (2 positive eigenvalues, 1 negative eigenvalue and 0 0-eigenvalue)

x2 y2 z2
- hyperboloid of one sheet: α2
+ β2
− γ2
=1
x2 y2 z2
- cone with vertex at the origin: α2
+ β2
− γ2
=0

• (1, 2, 0) (1 positive eigenvalue, 2 negative eigenvalues and 0 0-eigenvalue)

x2 y2 z2
- hyperboloid of two sheets: α2
− β2
− γ2
=1

• (0, 3, 0) (0 positive eigenvalues, 3 negative eigenvalues and 0 0-eigenvalue)

4
 Figure 2: Ellipsoid (left) and Hyperboloid of 1 sheet (right)

Figure 3: Cone (left) and Hyperboloid of 2 sheets (right)

2 y2 z2
- imaginary ellipsoid: − αx 2 − β2
− γ2
=1

5
• (2, 0, 1) (2 positive eigenvalues, 0 negative eigenvalues and 1 0-eigenvalue)
x2 y2
- elliptic paraboloid: z = α2
+ β2
x2 y2
- elliptic cilinder: α2
+ β2
=1

Figure 4: Elliptic paraboloid (left) and Elliptic cylinder (right)

  
x2 y2
- two complex conjugate planes containing the z-axis: α2
+ β2
= x
α + i βy x
α − i βy =0
(the set of real solutions is the line (0, 0, t))

• (1, 1, 1) (1 positive eigenvalue, 1 negative eigenvalue and 1 0-eigenvalue)

x2 y2
- hyperbolic paraboloid: z = α2
− β2
x2 y2
- hyperbolic cilinder: α2
− β2
=1
  
x2 y2 x y x y
- two planes containing the z-axis: α2
− β2
= α + β α − β =0

• (0, 2, 1) (0 positive eigenvalues, 2 negative eigenvalues and 1 0-eigenvalue)

2 y2
- imaginary cilinder: − αx 2 − β2
=1

• (1, 0, 2) and (0, 1, 2) (1 positive eigenvalue, 0 negative eigenvalue or 0 positive eigenvalues, 1

negative eigenvalue, and 2 0-eigenvalues)

- parabolic cilinder: y = αx2 + β

x2
- two parallel planes: α2
= 1 ⇒ x = ±α

6
 Figure 5: Hyperbolic paraboloid (left) and Hyperbolic cylinder (right)

Figure 6: Parabolic cylinder

- two coincident planes: x2 = 0

2
- two parallel complex conjugate planes: − αx 2 = 1 ⇒ x = ±αi

Conclusion: these are all the possible quadric surfaces (up to congruence and similarity).
There are no others types of quadrics in the 3D-space.

Note 5. The classification of quadric hypersurfaces in Rn for general n follows the same criteria
(study of eigenvalues of the quadric and their mutual sign and -possible- vanishing).

7
3 How to identify a conic in R2 ? Example.
Step 0. You have the conic
√
3x2 − 2xy + 3y 2 − 8 2x + 10 = 0
and you rewrite it in matrix form
 √
    
  3 −1 x  x
x y + −8 2 0 + 10 = 0
−1 3 y y

Step 1 - Eigenvalues. Consider only the quadratic part, in particular the matrix
 
3 −1
Q=
−1 3
and study its eigenvalues:
det (λI − Q) = (λ − 3)2 − 1 = λ2 − 6λ + 8 = 0
The eigenvalues are: λ1 = 2 > 0 and λ2 = 4 > 0 (guess: this could be an ellipse or 2 complex lines
intersecting in one point).

Step 2 - Eigenvectors. Orthogonally diagonalizes Q. First of all we look for a basis for each
eigenspace Wλ .
λ = 2:
      
x −1 1 x 0
(2I − Q) = =
y 1 −1 y 0
meaning solving the system
(
−x + y = 0
−x + y = 0
The solutions are
    
x 1
W2 = , x∈R =
x 1
λ = 4:
      
x 1 1 x 0
(4I − Q) = =
y 1 1 y 0
meaning solving the system
(
x+y =0
x+y =0
The solutions are
    
−y −1
W4 = , y∈R =
y 1

8
Step 3 - Gram-Schmidt. 2
    Apply Gram-Schmidt procedure to find an orthonormal basis for R .
1 −1
The vectors and are already mutually orthogonal, so we just need to normalize them.
1 1
    
~ 1 1 ~ 1 −1
B = b1 = √ , b2 = √
2 1 2 1

Step 4 - Orthogonal matrix P and orthogonal change of variables. Construct the matrix
" 1 #
h i √ − √1
P = ~b1 ~b2 = 12 1
2
√ √
2 2

P is a 2 × 2 orthogonal matrix such that

 
T 2 0
P QP = .
0 4

Apply the change of variables

(
x= √1 t − √1 s
   
x t 2 2
=P meaning
y s y= √1 t + √1 s
2 2

to the conic. This way, we get rid of the cross product in the quadratic part:
    T        
  x t t   T t   2 0 t
x y Q = P Q P = t s P QP = t s
y s s s 0 4 s

Applying this change of variables to the whole conic will give us:
√
0 = 3x2 − 2xy + 3y 2 − 8 2x + 10
2 2
√
      
1 1 1 1 1 1 1 1 1 1
=3 √ t− √ s −2 √ t− √ s √ t + √ s + 3 √ t + √ s − 8 2 √ t − √ s + 10
2 2 2 2 2 2 2 2 2 2

= . . . expand and simplify . . .

= 2t2 + 4s2 − 8t + 8s + 10

Step 5 - Complete the square. Complete the square for the variables t and s:

0 = 2t2 + 4s2 − 8t + 8s + 10 = 2 t2 − 4t + 4 + 4 s2 + 2s + 1 − 8 − 4 + 10

= 2 (t − 2)2 + 4 (s + 1)2 − 2

and obtain

2 (t − 2)2 + 4 (s + 1)2 = 2.

This is an ellipse centered at the point (2, −1) with semi-axes α = √1 and β = 12 .
2

9
Note 6. The final change of variables (the translation) that will transform the conic into standard
form will be
(
t=ξ+2
s=η−1

Substituting this expression into the conic above, we get (dividing by 2)

ξ 2 + 2η 2 = 1.

10