SC505 - Linear Algebra & Optimization
Autumn 2019/20
In-Sem 1 4th September 2019
Duration: 90 minutes Total Marks 20
Note: You may use any result/example derived/stated/discussed in class without proof.
1. Let (V, +, ·) be a finite dimensional real vector space. Which of the following sets S associated
with V, with corresponding addition and scalar multiplication operations (+S , ·S ) defined below,
form a vector space? Verify only (i) closure with respect to addition and scalar multiplication, (ii)
additive identity and additive inverse, and their uniqueness (iii) distributivity of scalar multipli-
cation and addition. In case, any one of these is violated, do not check for the rest. [3+3]
(a) S = {W ⊂ V | W is a subspace of V } is the set of all subspaces of V, +S is the usual sum of
subspaces, and ·S is the scalar multiplication defined as:
∀ a ∈ R, ∀W ∈ S, a ·S W := { a · w | ∀w ∈ W }.
(b) Let the addition of any vector v ∈ V and any subspace W of V be defined as v + W := {v +
w | ∀w ∈ W }, and denoted as vW . Let U be a fixed subspace of V, and let S = {vU | ∀v ∈ V }
on which addition and scalar multiplication is defined as follows:
∀ pU , qU ∈ S, pU +S qU := ( p + q)U
∀ a ∈ R, ∀ pU ∈ S, a ·S pU := ( a · p)U .
2. If possible, find a linear transformation T : R3 → R2 such that T ((−2, 1, 0)) = (2, 4), T ((1, −3, 2)) =
(1, −2) and T ((−3, −1, 2)) = (−4, 6). Justify. [3]
3. Let A be a square matrix such that all its eigenvalues are zero. Show that A is diagonalizable if
and only if A = 0, the zero matrix. [4]
4. Let V be a finite dimensional vector space, and T ∈ L(V ) such that T 2 = T. Show that [3+2+2]
(a) V = R( T ) ⊕ N ( T ).
(b) T is a projection on R( T ) along N ( T ), i.e., ∀v ∈ R( T ), Tv = v, and ∀v ∈ N ( T ), Tv = θ.
(c) T is diagonalizable. Compute its diagonal representation.
PG Orientation!
1
Solutions
1. Sets S of a FDVS (V, +, ·).
(a) S = {W ⊂ V | W is a subspace of V } is the set of all subspaces of V.
Note that the subspace {θ } is the additive identity. In checking whether (S, +S ) is an abelian
group, specifically existence of additive inverse, notice that for any subspaces U, W ∈ S,
U ⊂ U + W and W ⊂ U + W. Thus the only subspace for which additive inverse exists is
{θ }. Hence, S is not a vector space.
(b) S = {vU | ∀v ∈ V }.
i. Additive closure: ∀vU , wU ∈ S, vU +S wU = (v + w)U ∈ S, since v + w ∈ V.
ii. Scalar multiplication closure: ∀ a ∈ R, ∀vU ∈ S, a ·S vU = ( av)U ∈ S, since av ∈ V.
iii. Additive Commutativity: ∀vU , wU ∈ S, vU +S wU = (v + w)U = (w + v)U = vU +S wU .
iv. Additive associativity: ∀vU , wU , qU ∈ S, vU +S (wU +S qU ) = vU +S (w + q)U = (v +
w + q)U = ((v + w) + q)U = (v + w)U +S qU = (vU +S wU ) +S qU .
v. Additive identity: Let Θ := θU = U. Note that ∀vU ∈ S, vU +S U = (v + θ )U = vU . In
fact, Θ = uU , for any u ∈ U.
vi. Additive inverse: Let vU ∈ S. Then vU +S (−v)U = (v + (−v))U = Θ.
vii. Multiplicative identity: ∀vU ∈ S, 1 ·S vU = (1 · v)U = vU .
viii. Distributive: ∀vU , wU ∈ S, ∀ a ∈ R, a ·S (vU +S wU ) = a ·S (v + w)U = ( a · (v + w))U =
( a · v + a · w ) U = ( a · v ) U + S ( a · w ) U = a · S v U + a · S wU .
ix. Distributive: ∀ a, b ∈ R, ∀vU ∈ S, ( a + b) ·S vU = (( a + b) · v)U = ( a · v + b · v)U =
( a · v )U + S ( b · v )U = a · S vU + b · S vU .
Thus S is a vector space, and is also called the quotient of V over U (denoted as V/U).
2. Let c1 , c2 , c3 be the three input vectors on which T is defined. Note that if the three vectors are LI,
then they can be mapped to arbitrary element by a linear transformation, and thus such a linear
transformation would exist. In this case, 2c1 + c2 − c3 = θ, implying that the set {c1 , c2 , c3 } is not
LI. Let the corresponding outputs be denoted by a1 , a2 , a3 . It is easy to see that 2a1 + a2 − a3 6= θ,
and therefore such a linear transformation is not possible.
3. ⇒ Since A is diagonalizable with all eigenvalues zero, the diagonal matrix Λ in A = MΛM−1 is
the zero matrix, and thus A = MΛM−1 = 0.
⇐ Given that A ∈ Rn×n is the zero matrix, Rn = N ( A) = N ( A − 0.In ), where In denotes the
n × n identity matrix. Thus any non-zero vector in Rn is an eigenvector of A with eigenvalue 0.
Let β be an arbitrary choice of basis of Rn , and let M be the matrix with vectors from β represented
in the standard basis in the columns. Then AM = MΛ, with Λ = 0, or A = M0M−1 . Thus A is
diagonalizable with all eigenvalues being zero.
4. Projection operator:
(a) V = R( T ) ⊕ N ( T ).
Let u ∈ R( T ) ∩ N ( T ). ∃v ∈ V, Tv = u, and Tu = θ. Now T 2 v = Tu = θ = Tv = u.
Therefore R( T ) ∩ N ( T ) = {θ }.
Let v ∈ V be an arbitrary vector in V. Then, Tv = w ∈ R( T ). Apply T on both sides
to get T 2 v = Tw = Tv (using T 2 = T). Thus T (v − w) = θ and v − w ∈ N ( T ). Let
v − w = n ∈ N ( T ). Thus, v = w + n with w = Tv ∈ R( T ) and n ∈ N ( T ), proving
V = R( T ) ⊕ N ( T ).
2
(b) T is a projection on R( T ) along N ( T ).
Let v ∈ R( T ). Thus, ∃u ∈ V, Tu = v. Then, T 2 u = Tv = Tu = v. Thus Tv = v. Also by
definition, ∀n ∈ N ( T ), Tn = θ. Thus T is a projection on R( T ) along N ( T ).
(c) T is diagonalizable. Let β = { β 1 , . . . , β p } be a basis of R( T ) and let γ = {γ1 , . . . , γq } be a
basis of N ( T ). Since V = R( T ) ⊕ N ( T ), α = β ∪ γ is a basis of V. Since ∀v ∈ R( T ), Tv = v,
R( T ) is the eigenspace of eigenvalue 1, and N ( T ) is the eigenspace of eigenvalue 0. Thus α
is an eigenbasis of T, and the diagonal representation of T, [ T ]αα contains p ones followed by
q zeros on the diagonal.
