Cheat Sheet: Comp Sc BSc, LinAlg - Brian Funk, 21.04.

2001 - 22-918-18-957
ˆ A is regular and A−1 = AH (AT ) Def outer product: m-vector x and n-vector
ˆ AAH (AAT ) = I y: xy T
ˆ A−1 is unitary (orthogonal) S 2.14: A m × n-matrix has rank 1 if it is the
Complex Numbers ˆ AB is unitary (orthogonal) outer product of an m-vector ̸= 0 and n-vector
Def Zerodiviser: If AB = 0 ⇔ A,B ̸= 0
Zerodiviser, Nullteiler S 2.15: The orthogonal projection Py x of the
Def transposes: (AT )ij = Aji n-vector x onto y is defined as:
y
Py x = ∥y∥2 · yy H x = uuH = Pu where u = ∥y∥
1
Def conjugate transposed: AH = (A)T = AT
Basics Def symmetric: AT = A ⇔A symmetric Def projections matrix: Py = 1
∥y∥2
· yy H It has
Compability Conditions: br+1 = ... = bm = 0 Def skew-symmetric: AT = −A ⇔A the properties: PyH
= Py
S 1.1: Ax = b hat min eine Lösung ⇔ r = m skew-symmetric
Def hermitian: AH = A ⇔A hermitian (hermitian/symmetric) and Py2 = Py
Def : z = a + bi ⇔ ℜ(z) = a, Im(z) = b oder r < m + V B dann: r = n ⇔ 1 Lösung , (idempotent)
r < n ⇔ ∞ Lösungen S 2.6: Also accounts for AT instead of AH . α
Def : z = a + bi ⇔ z̄ = a − bi ⇔ r · e2π−ϕ simplifies to α
z = r · cos(ϕ) Cor 1.7: For a quadratic SLE with n
Def : p + i · sin(ϕ)
√ equations and n variables we have the ˆ (AH )H = A Inverse
Def : |z| =r = x2 + y 2 = z · z̄ ˆ (αA)H = αAH
y following set of equivalence, of wich ONLY one
arctan x

y
1. Q of them can be true; So, EITHER ˆ (A + B)H = AH + B H
Def : ϕ = arctan x +π 2./3. Q i Rank(A) = n (A is regular) ˆ (AB)H = B H AH Def invertible: ∃A−1 ⇔ A−1 · A = A · A−1 = I
arctan y + 2pi 4. Q

ii for every b there exist at least one S 2.17: A is invertible⇔ ∃X : AX = I ⇔ X is
x S 2.7: For symmetric matrices A and B it
solution holds that: AB = BA ⇔ AB is symmetric It unique ⇔ A is regular
iii for every b there exists exactly one holds for arbitrary matrix C that: C T C and S 2.18: If A, B are regular:
Operations solution CC T are symmetric. The same holds for the ˆ A−1 is regular and A−1−1 = A
iv the corresponding homogeneous system hermitian case ˆ AB is regular and (AB)−1 = B −1 A−1
has only the trivial solution ˆ AH is regular and (AH )−1 = (A−1 )H
OR the following equivalences hold
Def : z1 ± z2 : (x1 + x2 ) ± i(y1 + y2 ) v Rank(A) < n (A is singular)
Scalarproduct and Norm S 2.19: If A is regular das LGS Ax = b has the
unique solution x = A−1 b
Def : vi for some b there exits no solution row 
Finding an inverse [A|I] −−−→ I|A−1 if

z1 · z2 : (x1 + i · y1 ) + (x2 + i · y2 ) = r1 · r2 ei(ϕ1 +ϕ2 ) vii for no b a unique solution exists op
z1 r1 i(ϕ1 −ϕ2 ) z ·z¯ Def euclidian scalarproduct:
Def : z2 : r2 e = |z1 |22 viii for some b infinity many solution exists
 
in Pn a b
⟨x, y⟩ = xT y =
Pn
k=1 xk · yk −
−→ k=1 xk · yk A= and det(A) ̸= 0 ⇔ A is invertible
√ 2
ix the corresponding homogeneous system c d
Def : n a ⇔ a = z ⇔ |a| · eiϕ = r n · eiωn ⇔ r =
n R
p
n ϕ+2kπ has non-trivial solutions S 2.9: 
d −b

|a|, ω = n S1 ⟨x, y + z⟩ = ⟨x, y⟩ + ⟨x, z⟩(linear in 2nd ⇔ A−1 = ad−bc
1
−c a
Matrices and Vectors factor)
S1 ⟨x, αy⟩ = α ⟨x, y⟩ (linear in 2nd factor) A=

a11 a12

⇔ A−1
 −1
a11 a−1
12

Polynomials S2 for E = R: a21 a22 −1
a21 −1
a22
⟨x, y⟩ = ⟨y, x⟩ (symmetric)
Definitions S2’ for E = C:
The roots of a complex polynomial√
are ⟨x, y⟩ = ⟨y, x⟩(hermitian) Orthogonal and unitary matrices
b± b2 −4ac S3 ⟨x, x⟩ > 0, ⟨x, x⟩ = 0 ⇔ x = 0(positiv
pairwise conjugated. Def : z = 2a A m × n matrix hat m row (Zeilen)↓ definite)
Def : az n + c = 0 ⇔ z = n − a Def unitary/orthogonal: AAH = I, AAT = I ⇔
p c
and n columns (Spalten)→ Cor 2.10:
, in which the i,j element gets noted by ai,j or S4 for E = R: linear in 1st factor A is unitary/orthogonal ⇔ det(A) = ±1
(A)i,j ⟨w + x, y⟩ = ⟨w, y⟩ + ⟨x, y⟩ S 2.20: A,B are unitary/orthonormal:
Def nullmatrix: Has in every entry 0 ⟨αx, y⟩ = α ⟨x, y⟩ ˆ A is regular and A−1 = AH
Def diagonalmatrix: Has in every entry 0 S4’ for E = C: conjugate-linear in 1st factor ˆ AAH = In
except for the diagonal: (D)ij = 0 for i ̸= j one ⟨w + x, y⟩ = ⟨w, y⟩ + ⟨x, y⟩ ˆ A−1 is unitary/orthogonal
can write Diag(d11 , · · · , dnn ) ⟨αx, y⟩ = α ⟨x, y⟩ √ ˆ AB is unitary/orthogonal
Def identity: The identity is written as
p
Def norm: ∥x∥ = ⟨x, x⟩ = xT x = ˆ columns are orthonormal
In = Diag(1, · · · , 1) It holds that AI = IA = A pPn in
2 −
qP
n 2
S 2.21: Images from unitary/orthonormal
Def upper triangular matrix: We have k=1 (|xk |) − →
R k=1 xk matrices are conformal (längen-winkeltreu)
 
(R)ij = 0 for i > j (Rechtsdreiecksmatrix) S 2.11: | ⟨x, y⟩ | ≤ ∥x∥ · ∥y∥( Cauchy-Schwarz cosϕ sinϕ
Def lower triangular matrix: We have Def 2d rotation: R(ϕ) =
inequality, ”=” holds when y is a multiple of x −sinϕ cosϕ
(R)ij = 0 for i < j (Linksdreiecksmatrix) or vice verca) Def 3d rotation:
 
Def Matrix-set: The set of m × n-matrices is Def CBS: CBS is a property of the scalar 1 0 0
written as: Em×n For vectors we have: En , product: CBS squared yields: Rx (ϕ) = 0 cosϕ −sinϕ , Ry (ϕ) =
where E is R or C | ⟨x, y⟩ |2 ≤ ⟨x, x⟩ ⟨y, y⟩ 0 sinϕ cosϕ
Def matrix multiplication: If C = AB then S 2.12: For the euclidian norm holds:

cosϕ 0 sinϕ

one can write P N1 ∥x∥ > 0, ∥x∥ = 0 ⇔ x = 0(positiv definit)  0 1 0  , Rz (ϕ) =
n Pn
Cij = (AB)ij = k=1 (A)ik (B)kj = k=1 aik bkj N2 ∥αx∥ = α ∥x∥ (homogeneous) −sinϕ 0 cosϕ
S 2.1: N3 ∥x ± y∥ ≤ ∥x∥ + ∥y∥ (Triangle-inequality)  
ˆ (αβ)A = α(βA) cosϕ −sinϕ 0
Def : Angle ϕ between x, y: sinϕ cosϕ 0
ˆ (A + B) + C = A + (B + C) Re(⟨x,y⟩) in ⟨x,y⟩
, ϕ = arccos −
−→ arccos ∥x∥·∥y∥ 0 0 1
ˆ (αA)B = α(AB) ∥x∥·∥y∥
SLE ˆ (AB) · C = A · (BC)
R
Def : x, y are orthogonal: ⟨x, y⟩ = 0 ⇔ x ⊥ y
ˆ (α + β)A = αA + βA
ˆ (A + B) · C = AC + BC
S 2.13: ∥x ± y∥2 = ∥x∥2 + ∥y∥2 ⇔ x ⊥ y
(Pythagoras) Def p-norm:
LU-Decomposition
ˆ A · (B + C) = AB + AC 1 The LU-decomposition is useful when multiple
ˆ α(A + B) = αA + αB ∥x∥p = (|x1 |p · · · |xn |p ) p SLE have the same A
Gauss Algorithm ˆ A+B =B+A ˆ Find P A = LR
S 2.20: Let A and B be unitary(orthogonal). Outer Product and Projections ˆ solve Lc = P b
It holds: ˆ solve Rx = c
 1 bring into row echelon form
Lem 4.8: Every set {v1 , · · · , vm } ⊂ V with ˆ F : X 7→ Y bijective (isomorphism) ⇔ 2 create a vector for every row, which
|Bv | < m is linear dependant Rank F = dim X = dim Y does not have a pivot. The dimensions
Cor 4.10: in an finite vectorspace, a set with ˆ F : X 7→ Y bijective (automorphism) ⇔ of the vectors are E 1×n [3]Solve SLE
n independant vectors is basis of V if Rank F = dim X, ker F = {0} Ax = 0 with the yielded vectors.
dim(V ) = n Cor 5.10: 4 Write the solution as vector
Def : The coefficients ξk are coordinates of x ex
with respect to a basis B ξ = (ξ1 , · · · , ξn )T is a ˆ Rank(G ◦ F ) ≤ min(RankF, RankG) −1 −4 7 3 −1 −4 7 3
3 0 −6 0 0 −12 15 9
coordinate vector ˆ G is injective Rank(G ◦ F ) = RankF −3 4 1 −3 −−→ 0 0 0 0
−−→
Def : Two subspaces U, U ′ ⊂ V are ˆ G is surjective Rank(G ◦ F ) = RankG 1 −4 3 3
(1)
x4 =α
0 0 0 0
(2)

complementary if every v ∈ V has a unique −1 −4 7 3 x3 =β

representation in U andU ′ . Namely, 0 −12 15 9
0 0 ↑ ↑
−−→ x2 =
5β+3α
v = u ∈ U + u′ ∈ U ′ . → V = U
L ′ (3) 4
U 0 0 0 0 x1 =2α
−−→
Matrices as linear mapping (4)

Linear Maps span

( 2
5
4
0
3
4
)

1 0
0 1

Def columnspace: The columnspace Def Maps: Let X, Y be vector spaces with
Definitions
Vectorspaces (Spaltenraum) of A is the subspace
ℜ(A) = im(A) = span{a1 , · · · , an }
dimX = n, dimY = m
ˆ F : X 7→ Y a linear map
Def : A vectorspace V Def nullspace: The nullspace (Nullraum) of A ˆ A : En 7→ Em′ , ξ 7→ η *
Def linearity: F : V → W is linear: is a subspace N (A) = kerA = L0 (Ax = 0)
over K is a non-empty set, on which
ˆ F (v + w) = F (v) + F (w) ˆ B : En 7→ Em , ξ ′ 7→ η ′ *
vectoraddition and scalarmultiplication is Def : # free variables = dimN (A) ˆ T : En 7→ En ′ , ξ 7→ ξ′ **
ˆ αF (v) = F (αv) S 5.12: Rank A =r: and L0 Solution of
defined
Def injective: ∀x, x′ ⊂ X : f (x) = f (x′ ) ⇔ x = x′ ˆ S : Em 7→ Em ′ , η 7→ η ′ **
Def Axioms: Ax = 0 ⇒ dimL0 = dimN (A) = dim(KerA) = n − r *Abbildungsmatrix
Def surjective: ∀y ⊂ Y, ∃x ⊂ X, f (x) = y S 5.13: Rank A ∈ M m×n :
V1 : x + y = y + x **Transformationsmatrix
V2 : (x + y) + z = x + (y + z) Def bijective: surjective and injective⇔ f −1
V3 : ∃0 ∈ V : x + 0 = x exists ˆ pivots in Row-echelon-form
V4 : ∀x∃ − x : x + (−x) = 0 ˆ dim(im(A)) of A : En 7→ Em
V5 : α(x + y) = α · x + α · y Matrix representation ˆ dimension of the linear independent
V6 : (α + β)x = αx + βx columns/rows
V7 : (αβ)x = α(βx) Cor 5.14: RankAT = RankAH = RankA
V8 : 1 · x = x Let F be a linear map X → Y . One can write S 5.16: for A ∈ Em×n and BinEp×m :
S 4.1: F (bi ) ∈ Y as aPlinear combination of the basis
m
i : 0·x=0 of Y: F (bi ) = k=1 ak,l · ck Def : The matrix ˆ RankBA ≤ min(RankA, RankB)
ii : ·0 = 0 Am×n witht the elements ak,l is a matrix ˆ RankB = m ≤ p ⇒ RankBA = RankA
iii : α · x = 0 → x = 0 ∨ α = 0 (Abbildungsmatrix) with respect X, Y ˆ RankA = m ≤ n ⇒ RankBA = RankB S 5.20: RankF
iv : (−α)x = α(−x) = −(α · x) F (x) = y ⇔ Aξ = η [H] Cor 5.17: From S.5.16 it follows for quadratic   = r has the mappingmatrix*
Ir 0
Def polynomial space: Pn is defined as S all matrices. A ∈ Em×m and BinEm×m A=
0 0
polynomials of degree n. Further: P = ∞ n=0 Pn
S 4.1: I a vectorspace the following holds for ˆ RankBA ≤ min(RankA, RankB)
a scalar α and x ∈ V : ˆ RankB = m ⇒ RankBA = RankA Vector spaces with scalar
ˆ 0x = 0 ˆ RankA = m ⇒ RankBA = RankB
ˆ 0α = 0 S 5.18: For quadratic matrix En×n the products
ˆ αx = 0 ⇒ α = 0 or x = 0 following statements are equivalent:
ˆ (−α)x = α(−x) = −(αx)
S 4.12: {b1 , · · · , bn } ⊂ V is a basis of V ⇔ ˆA is regular Definitions
Pn x ∈ V can be uniquely represented
every vector ˆRankA = n
as: x = k=1 ξk bk ˆColumns are linearly independent
S 4.2: ∀x ∈ V, ∀y ∈ V ∃z ∈ V : x + z = y where z ˆRows are linearly independent Def Norm: A norm is a function
is unique and z = y + (−x) ˆkerA = N (A) = {0} | · | : V → R, x → ∥x∥ in a vector space which
Def isomorphism: F is bijective ⇔ F is an
ˆA is invertible satisfies:
Subspace isomorphism
ˆImA = ℜ(A) = En N1 ∥x∥ > 0, ∥x∥ = 0 ⇔ x = 0 (positiv definit)
S 5.19: For Ax = b, b ̸= 0 with the solution x0 N2 ∥αx∥ = α ∥x∥ (homogenous)
Def automorphism: F is isomorphism and N3 ∥x ± y∥ ≤ ∥x∥ + ∥y∥ (Triangle-inequality)
and L0 the solutionset is defined by
X = Y ⇔ F is an automorphism A normed vector space
Def : A subspace (unterraum) U is a Lb = x0 + L0 and is called affine subspace
S 5.1: F is isomorphism ⇔ F −1 exists and is has a norm Def scalar product: is a function
non-empty subset of V. It is closed under (not a real subspace since 0 ∈
/ Lb )
an isomorphism and linear ⟨·, ·⟩ : V × V → E, x, y 7→ ⟨x, y⟩, which satisfies:
vector addition and scalar multiplication. U dim(Im(A)) = n − dim(ker(A)) = n − (n − r) = r
contains the zero-vector S1 ⟨x, y + z⟩ = ⟨x, y⟩ + ⟨x, z⟩(linear in 2nd
S 4.3: Every subspace is a vectorspace Kernel, Image and Rank factor)
Def spanning set: The vectors v1 , · · · , vn are a RC Find Basis of Im A= R(A): S1 ⟨x, αy⟩ = α ⟨x, y⟩ (linear in 2nd factor)
spanning set (erzeugendes System) of V, if 1 bring into row echelon form S2 ⟨x, y⟩ = ⟨x, y⟩(symmetric, hermitian)
∀w ∈ span{v1 , · · · , vn } Def Kern: kerF = {x ∈ X|F (x) = 0} 2 mark rows with pivots S3 ⟨x, x⟩ > 0, ⟨x, x⟩ = 0 ⇔ x = 0(positiv
S 5.6: F injective ⇔ kerF = {0} 3 marked columns in the normal form are definite))
Def Image: ImF = {F (x)|x ∈ X} a Basis Def unitsphere: the set {x ∈ V | ∥x∥ = 1}
Linear dependency, basis, dimensions S 5.6: F surjective ⇔ imF = Y ex Def induced norm: The length
ker A
−1 −4 7 3 −1 −4 7 3 p of a vector is
3 0 −6 0 0 −12 15 9 defined as: ∥·∥ : V 7→ R, ∥x∥ 7→ ⟨x, x⟩
is the solution set of Ax = 0. Im(A) −3 4 1 −3 −−→ 0 0 0 0
−−→
(1) (2) Def angle ϕ: ϕ = ∢(x, y), 0 ≤ ϕ ≤ π is defined
1 −4 3 3 0) 0 0 0
set of all b, such that Ax = b is solvable S 5.7: −1 −4 7 3
(
−1 −4 ⟨x,x⟩ ℜ⟨x,y⟩
Def linear dependency: Vectors v1 , · · · , vn dimX − dim(kerF ) = dim(imF ) = Rank(F ) ↑ −12 15 9 3 0 by: ϕ = ∥x∥·∥y∥ = ∥x∥·∥y∥
0 ↑
−−→
0 0 (3)
span −3 4
arePlinearly dependent Def : The rank F is equal to dim(im(F )) 0 0 0 0 1 −4
Def orthogonal vectors: two vectors x, y are
n
⇔ k=1 αk · vk = 0 → α1 = · · · = αn = 0 Cor 5.8: orthogonal ⇔ ⟨x, y⟩ = 0
Def dimension: the dimension of V is ˆ F : X 7→ Y injective ⇔ Rank F = dim X Def orthogonal sets : two sets X, Y are
dimV = |spanV | (dim{0} = 0) ˆ F : X 7→ Y surjective ⇔ Rank F = dim Y RC Find Basis of ker A = N (A), A ∈ E m×n : orthogonal ⇔ ∀x ∈ X, ∀y ∈ Y ⟨x, y⟩ = 0
S 6.1: All matrices are unitary/orthogonal Cor 6.12: ˆ if n = dimX = dimY < ∞ RC Least Squares with SVD:
|⟨x, y⟩|2 ≤ ⟨x, x⟩ ⟨y, y⟩ = ∥x∥2 · ∥y∥2 (Cauchy ⟨x, y⟩v = ξ H η = ⟨ξ, η⟩v = ξ ′ , η ′ v = ξ ′H η ′ ⇒ T is 4 F is isomorphism 2

Schwarz Inequality) conformal (längen-winkeltreu) 5 {b1 , · · · , bn } is orthonormal basis of X H H

S 6.2: ⇔ {F (b1 ), · · · , F (bn )} is a orthonormal ∥Ax − b∥22 = Σ Vx −U b = ∥Σy − c∥22 ;x∗ =
Note Convention: A representation of a | {z }
∥x ± y∥2 = ∥x∥2 + ∥y∥2 ⇔ x ⊥ y(Pythagoras) basis of Y
|{z}
vector with respect ot the basis B1 is written y c
Def orthogonal basis: as [v]B1 6 F −1 is unitary/orthogonal 2

⇔ ∀i, ∀j, i ̸= j : ⟨bi , bj ⟩ = 0 7 The mapping matrix V Σ+ U H b ⇒ ∞ solutions, here: smallest 2-

Therefore: [v]B2 = M at(B1 )B2 [v]B1 and norm (y ∗ = Σ+ U H b) Where Σ+ is the pseu-
Def orthonormal basis: ⇔ orthogonal basis [v]B1 = M at(B2 )B1 [v]B2 where M at(B2 )B1 is the (Abbildungsmatrix) A is
with vectors of length 1 unitary/orthogonal doinverse of Σ, hence it holds
matrix of change ob basis from B1
to B1 Hence:
S 6.3: A set M of pairwise orthogonal vectors M at(B1 )B2 = [b1 ]B2 |· · · | [bn ]B2
are linearly independant if 0 ∈ /M
S 6.4: LetP{b1 , · · · , n} a orthonormal basis, RC Calculating the matrix of F with respect Determinants
x ∈ V: x = n ¯ to Basis B: We have a function F : X 7→ X and
k=1 ⟨bk , x⟩ bk → ξk = ⟨bl , x⟩ Def :
S 6.5: from ξk = ⟨bl , x⟩v , ηk = ⟨bl , x⟩v follows
Pn H
the basis of X := B:
1 calculate for all ∀a ∈ B: F (ba )
Least Squares det (a11 ) = a11 , det

a11 a12
a21 a22

= a11 a22 − a12 a21 ,
⟨x, y⟩v = k=1 ξn ηk = ξ η = ⟨ξ, η⟩En Which 2 solve for all F (ba ) = αa b1 + βa b2 · · · γa bn  
implies that if a basis in V is orthonormal the a11 a12 a13 a11 a22 a33 +a21 a32 a13
3 write coordinate vectors as: ξa = Let Ax = b a overdetermined SLE ( Equations ¿
scalar product is valid in V det a21 a22 a23  = +a31 a12 a23 −a13 a22 a31
(αa βa · · · γa )T Variables). No exact solution exists.
From that follows: a31 a32 a33 −a12 a21 a33 −a11 a23 a32
4 write matrix as F[B] = ξ1 |··· | ξn → x∗ = argminx∈En ∥Ax − b∥22 ⇒ (Ax − b) ⊥ ℜ(A)
∥x∥v = ∥ξ∥En , ∢(x, y)v = ∢(ξ, η)En , x ⊥ y ⇔ ξ ⊥ η Pn Pn
S 8.12: det(A) = i=1 aki Kki = i=1 ail Kil for
Def Pseudoinverse: If Rank A = n: a fixed k and l.
RC Gram-Schmidt: k+i
ˆ b1 = ∥aa1∥ RC Compute Basistransformationsmatrix from A+ = (AH A)−1 AH ⇒ A+ A = I Def cofactor Kki : Kki = (−1) det(A[k,i] )
1 v B to S : Def A[k,i] : Is defined as the matrix A without
ˆ bek = ak − k−1
P
Since S is a standardbasis we have: Def normalequations: (AT A)x = AT y
j=1 ⟨bj , ak ⟩v · bj the k − th row and i − th column
1 S → B is given by B = (s1 |s2 | · · · |sn ) RC Least Square Method for functions: Def : detA = 0 ⇔ A is singular
ˆ bk =
bf
k
∥bfk ∥v (Columns of B are the basis vectors of Def : detA ̸= 0 ⇔ A is regular
We assume that ker(A) = {0} and AH A is reg-
ex B) S 8.3:
2/3 ular
2 3 2 2 3 2 Compute inverse of B to get B → S i det(A) is linear in every row
A = 2 4 −
→ a1 = 2/ 2 = 2/3 , a
f 2 = 4 − 0 bring problem in a form where every-
1 1 1 1 1 1/3 1 ii swapping two rows changes the sign of
thing is numerically determined except
−1/3 −1/3 −1/3 −1/3
 2/3 
3
RC Basistransformationsmatrix from 2 × 2 the coefficients det(A)
2/3 , 4 = 2/3 , a2 = 2/3 / 2/3 = 2/3 7→
1/3 1 −2/3 −2/3 −2/3 −2/3 matrices with respect to the standard basis: 1 calculate AT y iii det(I) = 1
2/3 −1/3 2 calculate AT A S 8.4:
A = 2/3 2/3
The standard  of 2 ×
basis  2 matrices
 is given iv if A has a row with 0 ⇒ det(A) = 0
3 solve the equation (AT A)x = AT y
 
2/3 −2/3 1 0 0 1 0 0 0 0
by: S = , , , 4 calculate error r = y − Ax v det(γA) = γ n det(A)
0 0 0 0 1 0 0 1
ex vi if A has to equal rows ⇒ det(A) = 0
S 6.6: After k-steps the set {b1 , · · · , bk } is and the
  new
 Basis
  is  defined
  B =
by:
The equation is given y(t) = x1 t + x2 t2 We have vii adding a multiple of a row to another
pairwise orthonormal.{b1 , · · · , bk } is a basis ⇔ a b e f i j m n
, , , Then tn (1, 2, 3, 4) and y(t)n = (13.0, 35.5, 68.0, 110.5) −
→ row doesnt affect the det
{a1 , · · · , ak } is a basis Every vectorspace (̸= ∞) c e g h k l o p 0
Qn
viii is A a diagonalmatrix: det(A) = Q i=1 aii
has a orthonormal basis the matrix (Abbildungsmatrix) is defined by 1 1 13.0
2 4 35.5
→ AT y = 2535.0
730.0
→ AT A iv is A a triangularmatrix: det(A) = n i=1 aii
Cor 6.7: To a vector space with scalar a e i m A = 3 9 , y = 68.0 − − =
 
4 16 110.5
1 2 Cor 8.10: Every statement of Satz 8.3 and 8.4
product with finite or countably infinite many b f j n 30 100
F = 100 354 − 30 100
→ 100 730
→
7 x = 7.9355, x2 = also holds for columns instead of rows
dimensions a orthonormal basis exists. c g k o 354 2535.0 1
3 S 8.5: Using Q Gauss on A results in:
Def orthogonal complement: U ⊥ is the d h l p 4.919 det(A) = (−1)v n k=1 rkk
orthogonal complement of a subspace U .
L ⊥ where v is # swappings of rows and rkk are the
U U =V
RC Prove that B is Basis: RC Least Square Method for 2D-points: diagonal elements of the row echelon form S
S 6.9: For a complex matrix with rankA = r it
S denotes the standard basis. 0 Write X- / Y-coordinate alternately in 8.7: det(AB) = det(A) · det(B)
holds:
1 compute for all ∀a ∈ B: ba = αs1 + the form (x, 0), (0, y) in A for every Cor 8.8: if A is regular ⇒ det(A−1 ) = det(A) 1
ˆ N (A) = ℜ(AH )⊥ ⊂ En
βs2 · · · γsn point. Write X- / Y-coordinate alter-
ˆ N (AHL ) = ℜ(A)⊥ ⊂ Em S 8.9: det(AT ) = det(A) and det(AH ) = det(A)
2 Since span(B) = span(S), B has to be ba- nately in y.
ˆ N (A) L ℜ(AH ) = En sis. 1 Rest as usually
Def det of block
 matrices:
A C
ˆ N (AH ) ℜ(A) = Em ex det
0 B
= det(A) · det(B)
ˆ dimℜ(A) = r The points P = {(−1, 1), (1, 1), (1, −1), (−1, −1)}
RC Prove that F is a bijective mapping Def det of unitary matrices: Let A be
ˆ dimℜ(AH ) = r F : X 7→ Y with the basis X for X and Y for Y : should be transformed with respect to unitary/orthogonal |det(A)| = ±1 proof:
ˆ dimN (A) = n − r 1 calculate for all ∀a ∈ X : F (xa ) the squared distance to the points P ′ =
det(U T U ) = det(I) = (det(U ))2 = 1 ⇒ det(U ) = ±1
ˆ dimN (AH ) = m − r 2 solve for all F (xa ) = αa y1 + βa y2 · · · γa yn {(0, 2), (1, 3), (0, −2), (−1, −3)}. The transforma-
 
p
s ·px
Those are the fundamental subspaces 3 write coordinate vectors as: ξa = tion is defined as T (P ) = T px = sx
(αa βa · · · γa )T 
−1 0

y
 
0
y ·py
Eigenvalues and -vectors
Change of Basis 4 write matrix as F[X ] = 1 ξ |··· | ξn
 0 1   2  Def eigenvector: A number λ ∈ En is called
5 As F[X ] is quadratic and has full rank
 1 0    1  eigenwert of a linear mapping: F : X if
   
we have that dim(X ) = dim(Y) and thus
 0 1  sx  3  ∃v ∈ V, v ̸= 0 such that F (v) = λv. v is an
   
B, B′ Pare orthonormalbasis. Hence: by Cor.5.8 that F is bijective Ax = y ⇒  =
1 0 s  0  eigenvector. The set of all eingenvectors,
  
y
b′k = j=1
 
n Tjk bj Matrix for change of basis T :  0
 −1 
−2
  which corespond to λ form a subspace
T −1 = T H since both basis are orthonormal. −1 0  −1 Eλ = {v ∈ V |F (v) = λv}
Therfore it holds that: unitary/orthogonal mapping 0 −1 −3 Def spectrum: The set of all eigenvalues of F
ˆ ξ = T ξ′ is called spectrum
ˆ ξ′ = T −1 ξ Def : ξ ∈ En is a eigenvector of λ ⇔ Aξ = λξ
ˆ B =, B′ T Def unitary: A linear mappingF : X 7→ Y is RC Least Squares with QR-decomposition: Lem 9.1: A linear map F and its matrix
ˆ B′ =, BT H unitary/orthogonal if ⟨F (v), F (w)⟩y = ⟨v, w⟩x We have the normal equations AH Ax = representation have the same eigenvalues and
S 4.13: Let ξ= (ξ1 · · · ξn )T be a coordinate S 6.13: AT b ⇒ (QR)H (QR)x = (QR)H b ⇒ RH QH QRx = the eigenvectors are connected by the
vector of an arbitrary vector v ∈ V with 1 F is isometric (längentreu): RH QH ⇒ RH Rx = RT QT b ⇒ Rx = QH b There- coordinaterepresentation kv
respect to the old basis ∥F (v)∥y = ∥v∥X fore we have: Lem 9.2: λ is eigenvalue ⇔ ker(A − λI) is
. Let ξ’= (ξ1′ · · · ξn
′ T
) be the new representation 2 F is conformal (winkeltreu): 1 compute QR-decomposition of A singular (Eλ = ker(A − λI))
of a Pvector x with respect to the new basis. v ⊥ w ⇔ F (v) ⊥ F (w) 2 solve Rx = QT b Def multiplicity: The geometric multiplicity
n P n ′ ′ of λ = dim(Eλ )
x= i=1 ξi bi = k=1 ξk bk 3 kerF = {0}, F is injective
Def characteristic polynomial: It is defined RC Eigenvaluedecomposition with SVD: RC SVD of A ∈ Em×n with AAH : ˆ if positiv definit ⇒ regular (L.3.7)
by XA (λ) = det(A − λI) Pn=0 The SVD is given by A = U ΣV H 1 Calculate (AAH ) ∈ Em×m Def unitary/orthogonal: AH A = I ⇔A is
Def Trace: tr(A) = k=1 akk 1 Expand U ΣV H ⇒ U IΣV H ⇒ U I1 I2 ΣV H 2 find eigenvalue of √ AA
H
√
unitary/orthogonal
ˆ A is regular (S.2.20)
S 9.5: λ ∈ E is eigenvalue of A ⇔ XA (λ) = 0 where U I1 = V and I1 I2 = I 3 write Σr = diag( λ1 , · · · , λn ) ∈ Em×m ˆ A−1 = AH (S.2.20)
Lem 9.4: 2 Calculate (U I1 ) (I2 Σ) V H 4 rewrite: Σ ∈ Em×n : Σ = Σ0r 00 ˆ A−1 is unitary (S.2.20)
XA (λ) = (−1)n · λn (−1)n−1 · tr(A) · λn−1 + ˆ A and B is unitary ⇒ AB is unitary (S.2.20)
5 find eigenvectors of AAH ⇒ v1 , · · · , vr ˆ
· · · det(A)λ0 = an ·n +an−1 · λn−1 + · · · det(A) 6 norm eigenvectors and compute V =
det(A) = ±1
Lem 9.6: A (quadratic) matrix is singular if RC Composition of 1-rank- matrices: v vn
1 write A = V ΛV −1 ( ∥v1 ∥ | · · · | ∥v ) ∈ Rn×n
and only if it has 0 as an eigenvalue
Def algebraic multiplicity: is the multiplicity 2 rewrite A =
Pn T
k=1 Vk λk wk Vk = row (↓)
1 n
7 solve for U = AV Σ−1
∥
Multiple Choice
schmidt
of an eigenvalue in the char. polynomial. T
of V , wk = column (→) of V −1 ̸ U : Ur −−−−−−→ U ∈ Em×m
8 if Ur =
gram
S 9.13: geometric multiplicity ≤ algebraic
9 write A = U ΣV H General
multiplicity CIf the solutions of an SLE are
RC Powers of A: x1 = 0, x2 = 0, x3 = 1 the system has infinite
RC Find Eigenvalues and -vectors: 1 write A = V ΛV −1 many solutions
1 find char. polynomial XA (λ) = det(A − 2 calculate Am = V Λm V −1 RC SVD with Spectral decomposition C Let A be a real 2 × 4 matrix with rank 2. Then
λI) note: Λm = diag(am m A = V ΛV −1 : the SLE Ax = b has a non-trivial solution
11 , · · · , ann ) One has to sort the singular values of Λ ac- W If A is invertible it holds: ABA−1 = B
2 find roots of XA C En×n → E, A 7→ trace(A) is linear
cording to their value. Then
√ one can do:
3 for every λk find the solution for (A − W En×n → E, A 7→ det(A) is linear
1 We have U = V,√ Σ= Λ
λk I)x = 0 C Let D ∈ E2×2 , dim(KerD) = 2 only if D = 0 0
2 Rewrite: A = V ΛV −1 0 0
Singularvaluedecomposition C If A2 is invertible, so is A3
S 9.7: for similar matrices C = T −1 AT (C and Cor 11.4: If A ∈ Em×n and rank(A) = r then: det(A2 ) ̸== 0 ⇒ det(A) ̸= 0 ⇒ det(A3 ) ̸= 0
A are similar) holds that C If A is regular and A2 = A,then A = I
eiganvalues of AH A ∈ Em×m and AAH ∈ En×n W For linear dependant x, y, z it holds x = αy + βz
tr(A) = tr(C), det(A) = det(C), XA = XC and they Def SVD for AH A: Spectral-decomposition are the same but the mulitplicity of the Only x, y can be dependant
have the same eigenvalues exists for every matrix AH A. Since AH A is eigenvalue 0 is n − r or m − r C If A and A2 ∈ En×n and A2 is regular, A3 is
S 9.11: Eigenvectors for different Eigenvalues hermetian (hermetisch) and positive invertible
Def spectral norm: It is defined as: ∥A∥2 = σ1 C ∀x ∈ Rn , ∥Ax∥2 ≤ ∥A∥2 ∥x∥2
are linearly independant ⇒ max dimV different semidefinite.→ AH A has real, non-negative W Let f : Rn → R, f (x) := ∥Ax∥2 The function f is a
Eigenvalues eigenvalues λ ∈ R and λ ≥ 0 Therefore one can norm in Rn
Def : λ is eigenvalue for A ⇒ λq is eigenvalue rewrite:AH AV = V Λ −−−−→ AH AVr = Vr Σ2r ⇒
QR-Decomposition W ∥AB∥2 ≤ ∥A∥2
of Aq λ=σ 2 Given are the orthogonal matrices A and B with the
−1 H H −1 same dimension. Which of the following properties is
Note Trace: trace(A) is equal to the sum of VrH AH AVr = Σ2r ⇒ (Σr Vr A ) (AVr Σr ) =I
Def QR-Decomposition: A matrix A can be true?
all eigenvalues of A:trace(A) = λ1 + · · · + λn | {z }| {z } W The matrix product AB is orthogonal, but BA is
−1 =Ur composed as A = QR where Q is ortohogonal
Note Trace: det(A) is equal to the producct =Ur not orthogonal
and R is an upper triangular matrix. The W The matrix product BA is orthogonal, but AB is
of all eigenvalues of A:det(A) = λ1 · · · · · λn Def SVD: SVD exists for every matrix, such decomposition is unique if m ≤ n and not orthogonal.
that U, V are unitary and Σ is diagonal and Rank(A) = n C The matrix product AB and the matrix product
positive. A = U ΣV H it follows BA are orthogonal
Decompositions AAH = U Σ2m U H ,AH A = V Σ2n V H , AH = V ΣT U H W The matrix product AB and the matrix product
BA are not orthogonal It holds that AT = A−1
Def A invertible: If A is invertible RC QR-Decomposition:
BT = B −1 and further
A−1 = V Σ−1 U H 1 Gram Schmidt on the rows (↓) of A → Q (AB)T = B T AT = B −1 A−1 = (AB)−1 and also
S 11.11f: For Rank r it holds that: 2 solve R = QT A → R vice versa (BA)T = (BA)−1
Spectral-/Eigenvaluedecomposition ˆ {u1 , . . . , u1 }: Basis of Im(A) = R(A) ex We have Rn with the standard scalar product ⟨·, ·⟩
2 3 2/3 −1/3 and 2-norm. Let A be a real n × n matrix. Which
ˆ {ur+1 , . . . , um }: Basis of A = 2 4 −→ q1 = 2/3 , q2 = 2/3 7→ Q = statements are correct
1 1 1 D E
−1 Ker(AH ) = N (AH ) 1/3 −2/3 C ∀x, y ∈ Rn it holds that x, AT y = ⟨Ax, y⟩
Def : A = V ΛV (A and Λ are similar) 2/3 −1/3
Precondition for diagonalisation ˆ {v1 , . . . , v1 }: Basis of Im(AH ) = R(AH ) 2/3 2/3 − →R = QT A = 3 5
0 1
C AT = A−1 ⇒ ∀x, y ∈ Rn ⟨Ax, Ay⟩ = ⟨x, y⟩
: S 9.14: A ∈ C n×n
is diagonalisible ⇔ ∀ ˆ {vr+1 , . . . , vm }: Basis of Ker(A) = N (A) 2/3 −2/3 2 C AT = A−1 ⇒ ∀x ∈ Rn ∥Ax∥ = ∥x∥
C Let B be another real n × n matrix.
Eigenvalues (geom. mult. = alg. mult.) Def Automorphism: In a self-image
AT = A−1 and B T = B −1 ⇒
n×n
is unitary (AH = A)it (Selbstabbildung) it holds that the inverse of AB exists and it is: (AB)−1 =
S 9.15: If A ∈ C
holds that:
H
A = U ΣV H = V V U ΣV H = V R Σ V
H Definitions (AB)T
| {z } |{z} |{z} |{z} |{z} Def nullmatrix: Has in every entry 0
i all eigenvalues are real ˆ ∀A(A + 0 = 0 + A = A) (S.2.2)
R 1 2 3 4 Given are orthogonal matrices A ∈ Rn×n and B ∈ Rn×n .
ii the eigenvectors are pairwise orthogonal Def diagonalmatrix: Has in every entry 0 except for Which of the following statements are correct?
1,4 Change to orthonormal basis
iii an orthonormal basis U exists, which the diagonal: (D)ij = 0 for i ̸= j one can write
C The matrix AT is orthogonal
2 rotation, mirroring Diag(d11 , · · · , dnn )
consists of all the eigenvectors W The matrix A + B is orthogonal
3 scaling of unit-axes √ ˆ is symmetric W The matrix A + AT is orthogonal
iv for the unitary matrix U holds that ˆ det(A) = A11 · · · · Ann (S.8.4)
Def singular values: singular values: σi = λi C The matrix AB −1 is orthogonal
U H AU = Λ ˆ A−1 = diag( A1 · · · A 1 )
sorted in descending order σa ≤ σb · · · σr ≤ 0 · · · 11 nn
Cor 9.16: The previous statements is also
Def Eigenbasis: V is the orthonormal ˆ Am = diag(am m
11 , · · · ann ) Given is a lower triangular matrix A ∈ R3×3 whose
valid for real-symmetric matrices
eigenbasis of AH A such that: AAH = U Σ2m U H . Def identity: The identity is written as entries are non-negative integers and whose entries
In = Diag(1, · · · , 1) either occur only once or are equal to zero. Which of
RC Eigenvaluedecomposition: Similar for U as eigenbasis of AAH : ˆ AI = IA = A the following options are possible for the value of the
1 find the eigenvalues of A λk and write AH A = V Σ2n V H ˆ A−1 = I determinant det(A)?
Λ = diag(λ1 · · · λn ) Def upper triangular matrix: (R)ij = 0 for i > j W 5
2 find the according eigenvectors vk of λk RC SVD of A ∈ Em×n with AH A: ˆ is nilpotent W 7
Def lower triangular matrix: (R)ij = 0 for i < j W −2
write them as V = (v1 | · · · |vn ) (sorted 1 Calculate (AH A) ∈ En×n ˆ is nilpotent C 35 A is triangular⇒ det(A) = a1,1 a2,2 a3,3 ⇒ so
according to Λ) H
2 find eigenvalue of √ A A √ Def Zerodiviser: If AB = 0 ⇔ A,B Zerodiviser, det(A) = 0 or det(A) = a1,1 a2,2 a3,3
3 find inverse V −1 3 write Σr = diag( λ1 , · · · , λn ) ∈ En×n Nullteiler
Def symmetric: AT = A ⇔A symmetric
The dimensions of the subspace of all
skew-symmetric real 3 × 3 matrices is:
4 rewrite: Σ ∈ Em×n : Σ = Σ0r 00 ˆ A and B symmetric ⇒ AB = BA ⇔ AB is W 1
Cor 9.10: If A is diagonisable it can be symmetric (S.2.7) C 3
composed 5 find eigenvectors of AH A ⇒ v1 , · · · , vr ˆ if positiv definit ⇒ regular (L3.7)
Pn as a sum T
of 1-rank-matrices
6 norm eigenvectors and compute V =
W 6
: A= Vk λk wk with V = (V1 · · · Vn ) and v Def skew-symmetric: AT = −A ⇔A skew-symmetric W 9
k=1 vn
 T ( ∥v1 ∥ | · · · | ∥v ∥
) ∈ Rn×n ˆ tra(A) = 0 Let A ∈ R2×3 and b ∈ R2 . Assume a solution for
w1 1 n ˆ if A has odd order Ax = b exists
−1
 .  7 solve for U = AV Σ−1 – det(A)=0 C Ax = b has always ∞ solutions min 1 free
V =
 ..  from that follows: Avk = vk λk schmidt – do inverse exists variable
̸ U : Ur −−−−−−→ U ∈ Em×m

8 if Ur = – A is singular W The set of the solution (Lösungsmenge) of
T gram
wn Ax = b forms a line in 3D could also be a plance,
9 write A = U ΣV H if A has even order 1 or 2 free variables
and T
wk A T
= λk wk : wT is a left eigenvector ˆ inverse is skew-symmetric if it exists C Geometrically Ax = b coresponds to an
Def hermitian: AH = A ⇔A hermitian intersection of two planes in 3D
 inverse follows directly from the definition of Let A ∈ E2×2 with Rank(A) = 1 and T race(A) = 5. What Let A ∈ Rm×n be such that Ax = 0 has only the trivial
orthogonal mappings. are the eigenvalues solution. Then it holds that:
Rank W 1 is an eigenvalue C dimIm(A) = n
W Let B ∈ E3×1 and C ∈ E1×3 : BC can have rank 3. C 0 is an eigenvalue W dimIm(A) = 1
Det W 2 is an eigenvalue C dimKer(A) = 0
W Let Q be unitary and A ∈ En×n , W −5 is an eigenvalue W dimKer(A) = 1 The kernel of A is exactly the
det(QA) = det(A) |detQ| = ±1 C 5 is an eigenvalue Since det(A) = 0 and solution set of the system of equations Ax = 0.
Vectorspaces trace(A) = 5 ⇒ 0 = λ1 · λ2 and 5 = λ1 + λ2 Since Ax = 0 has only the trivial solution,
W Let V be a vector space over R with scalar C if A, B, P ∈ En×n and P is invertible with
dimKer(A) = 0. Furthermore, it holds that
product ⟨·, ·⟩ and let F : V 7→ V be a linear map. A = P BP −1 then: det(A) = det(B) dimKer(A) + dimIm(A) = n. Therefore
If it holds that ∀v ∈ V, ⟨v, F (v)⟩ = 0, then F is dimIm(A) = n.
necessarily the null map, i.e., F (v) = 0 for all Given is a matrix A ∈ Rn×n with entries aij = ij and
Decompositions
v ∈ V . C A matrix A ∈ Rn×n with n eigenvalues has 2n
n > 1. Which statement is correct? Which of the following statements with A ∈ Rn×n is
W Given a vector space V with a norm ∥·∥. For all W det(A) = 1 normed spectral decompositions since the sign
generally true
u, v ∈ V , we have ∥v∥ ≤ ∥v + u∥. C det(A) = 0 can be changed n-times
C im(A) = im(2A)
counterexample: u = −v = (1, 1)T W det(A) = (−1)n C ker(A))ker(2A)
CLet S ⊂ V and W be a subspace of V : W det(A) = (−2)n Let A ∈ Rm×n m ≥ n be a matrix with rank k. Denote
S ⊂ W ⇒ span(S) ⊂ W the QR-decomposition of A as A = QR, where W im(A) = im(A2 )
Which of the following statements are not correct for
C In a vector space of finite dimension with scalar Q ∈ Rm×k has orthonormal columns, and R ∈ Rk×n is W im(A) = im(A + I)
arbitrary n × n-matrices A and B?
product, one can complete any set of C det(A + B) = det(A) + det(B) an upper (right) triangular matrix. Which one of the W im(A) = im(AT )
orthonormal vectors to form an orthonormal W det(AB) = det(BA) following statements is always true? W ker(A))ker(A2 )
basis. W If A is singular then AB is also singular W Rank A < Rank R W ker(A))ker(A + I)
C A vector space of finite dimension with scalar W det(AAT A) = (det(A))3 W QQT = I W ker(A))ker(AT )
product has an orthonormal basis. Corollary 6.7 W If A has linearly independant columns, we have
W Consider the vector space Rn with the Euclidean Let A, B ∈ Rn×n with AB = −BA Rank R = m
scalar product. The scalar product of two unit
vectors can be arbitrarily large. Cauchy
Schwarz, S 6.1: ⟨v, w⟩2 ≤ ⟨v, v⟩ ⟨w, w⟩ = 1 · 1 = 1
C det(AB) = det(−BA)
W det(A)det(B)) = −det(A)det(B) n has to be even
C If A has linearly independant columns, we have
Rank R = n
Let A ∈ Rm×n with linearly independant columns
Proofs
⇒S8.4 v 1) Prove that AH A and AAH have the same eigenvalues
W We again consider Rn with the Euclidean scalar W Either A or B has a zero-determinant and A = Q1 R1 = Q2 R2 , two QR-Decompositions of A
product. Can we find any number of pairwise W A and B have to be singular C QT We have that AH A and AAH are similar with
orthogonal unit vectors in this vector space? 1 Q2 is orthogonal T = A and T −1 = AH
W ABx = 0 has more than one solution
C QT 1 Q2 is a upper and lower triangular matrix Therefore by Satz 9.7 they have the same
(Lösungsschar)
and therefore a diagonal matrix eigenvalues (and also the same trace and det)
Let V the standardvectorspace of all 2 × 2 matrices. C ABx = c can have no, one and ∞ many
W QT 1 Q2 = I 2) Let Q ∈ En×n be an orthogonal matrix. Prove that,
Which of the following are subspaces of V ?(B = 1 2 ) solutions, if c ∈ R2 , c ̸= 0
3 4 W Rank(R1 ) = m if n is odd, that at least one of the matrices (Q + I) and
W {A W It has to be A = 0 or B = 0
n ∈ V|A is invertible}
o C Rank(R1 ) = n (Q − I) singular.
W A ∈ V|A2 = 0 Which of the following statements are correct for an
W Rank(R2 ) = m Let KQ (x) be the characteristic polynomial of Q.
C Rank(R2 ) = n
arbitrary n × n-matrix A and for arbitrary n?
n o
C A ∈ V|AT = A λ is an eigenvalue of Q ⇔ λ is a root of KQ (x)
W det(2A) = 2det(A) C QT 1 Q2 is regular
n o Q is orthogonal ⇒ |λ| = 1
C A ∈ V|AT B = BA W det(−A) = det(A) Let A, B ∈ Rn×m , B is regular and B = QR By Lemma 9.2 we have: λ is eigenvalue
C det(A4 ) = det(A)4 W f : Rn 7→ R, x 7→ ∥Ax∥2 , is a norm in Rn ⇒ (A − λI) is singular
Consider the vector space F of functions of R 7→ R with W Let A be a triangular matrix with the property therefore with λ = ±1 at least one of them has
C If ̸ ∃k, such that Ak is invertible, so is A not
the operations addition(f + g)(x) = f (x) + g(x) and scalar ai,j = 0f ori + j > n + 1 (so there are zeros at the to be singular
invertible
multiplication (λf )(x) = λf (x). This includes the bottom right). The determinant can be C Vx ∈ Rn , ∥Ax∥2 ≤ ∥A∥2 · ∥x∥2 3) Prove that for an orthogonal matrix Q it holds that
subspace P 2 = {a0 + a1 x + a2 x2 |ai ∈ R} of polynomials calculated using the formula W det(B) = det(R) ∥Qx∥2 = ∥x∥2
of degree ≤ 2. Which of the following statements are det(A) = a1,n · a2,n1 · · · an,1 . C ∥B∥2 = ∥R∥2 1 p 2
q
S.2.6
correct? C g : Rn 7→ R, x 7→ ∥Qx∥2 , is a norm in Rn ∥Qx∥2 = ⟨Qx, Qx⟩ = (Qx)T Qx =
Let A, P, Q ∈ Rn×n where P is permutation matrix
C Span{x + 1, x1, x2 + 1, x2 1} is equal to P2 . W AB is regular, but BA is not necessarily
q
and Q is a unitary matrix. S.2.20 p 2 1
xT QT Qx xT Ix =
p
= ⟨x, x⟩ = ∥x∥2
W x + 1, x1, x2 + 1, x2 1 ∈ P2 are linearly independent W det(P A) = det(A) W ∥AB∥2 ≤ ∥A∥2
1 = def of norm, 2 def of scalar product
W x + 1, x1, x2 + 1, x2 1 ∈ P2 form a generating set C det(P AP ) = det(A)
4) Prove that, if λ is an eigenvalue of orthogonal Q then
(spanning set) of F . W det(QA) = det(A)
λ = ±1
C x + 1, x1, x2 + 1, x2 1 ∈ P2 form a generating set
Basis 2 N2
W The transformation matrix of a basis Qv = λv(1) ⇔ ∥Qv∥ = ∥λv∥ ⇔ ∥v∥ = ∥λv∥ ⇔
(spanning set) of P2 .
Eigenvalue/Eigenvectors transformation between orthonormal bases is ∥v∥ = |λ| ∥v∥ ⇔ |λ| = 1
W The polynomials x + 1, x1, x2 + 1, x2 1 ∈ P2 form a the identity matrix. The transformation matrix 1 = def eigenvalue, 2 = as proven before, N2 =
basis of P2 C with XA (λ) = (λ − 1)3 + 3 is A ∈ E3×3 invertible of base transformation between orthonormal norm is homogeneous
Let V, W be finite dimensional vector spaces over a 0 isn’t eigenvalue ⇒ A is regular bases is orthogonal, the identity matrix is only 5) Prove that for a arbitrary matrix A with its
space K. Let F : V 7→ W be a linear mapping and W Let v1 and v2 be eigenvectors of A, so is v1 + v2 one possibility. eigenvalue λ it holds that (A − λI) is singular
(v1 , · · · , vn ) a basis of V . Then it holds that: an eigenvector C The inverse of the transformation matrix of a Let v be the eigenvector to the coresponding λ
W F (v1 ), · · · , F (vn ) are linearly independent if F is C If A ∈ En×n and XA (λ) = (λ − 1)n + 2, A is base transformation between orthonormal bases in (A − λI)
surjective invertible is its Hermitian transpose.
C Similar matrices have the same eigenvalues 1
C F (v1 ), · · · , F (vn ) are linearly independent if F is C A ∈ Rn×n is an orthogonal matrix if and only if (A − λI)v = (Av − λIv) = Av − λv = λv − λv = 0v
injective W Similar matrices have the same eigenvectors its columns form an orthonormal basis of R with As v is eigenvector we have
C F (v1 ), · · · , F (vn ) form a generating end system W Every n × n matrix has linear independant respect to the Euclidean scalar product. L.9.6
if F is surjective eigenvectors v ̸= 0(2) ⇒ λ = 0 ⇔ (A − λI) is singular
C The change of basis matrix is unitary (if E = C)
W F (v1 ), · · · , F (vn ) form a generating end system W Eigenvectors, which correspond to the same 1 = as v is eigenvalue, 2 = def eigenvalue
or orthonormal (if E = R) if both bases are
if F is injective eigenvalue are always linear dependant orthonormal. 6) Let A ∈ Rn×n be a real matrix and x ∈ Rn be a
C F (v1 ), · · · , F (vn ) form a basis if and only if F is C If a real matrix has a eigenvector, it follows that vactor.Prove that: ∥Ax∥2 ≥ σmin ∥x∥2 . Where σmin is
an isomorphism the matrix has infinity many eigenvectors the smallest singular value of A
Let V, W be two real vector spaces with scalar C Every rotation in R3 has the eigenvalue λ = 1 Let A = U ΣV T be the SVD of A
products, let B be an orthonormal basis of V and let W If λ1 with v and λ2 with w, so is (λ1 + λ2 ) a
Procedures
W The Gram-Schmidt orthogonalization method 1 2
F : V 7→ W be an orthogonal mapping. Which of the eigenvalue with eigenvector v + w ∥Ax∥2 = U ΣV T x = ΣV T x ≥
can be used to compute an equally large set of 2 2
following statements is true? Let A ∈ R3×3 with eigenvalues λ1 , λ2 , λ3 linearly independent vectors from a set of N2
W F is an isomorphism. F is only an isomorphism C A is diagonisable if all eigenvalues are different linearly dependent vectors. Σmin V T x = σmin IV T x =
2 2
if dimV = dimW < ∞ W if A is diagonisable, all eigenvalues have to be W Let v1 , · · · , vn ⊂ Rn be a set of n vectors. Using 3
C ∥F (v)∥W = ∥v∥v for all v ∈ V . Follows directly different the Gram-Schmidt process, we can always |σmin | IV T x VTx
= = |σmin | ∥x∥2
from the definition of the scalar product induced W A is diagonisable if it has 3 eigenvectors They 2 2
produce n unit-length and pairwise orthogonal 1 = U is orthogonal and proof 3, 2 =
norm and the orthogonality of F. have to bo linearly independant vectors. Gram Schmidts needs linear
Σmin = diag(σmin · · · ), 3 = V T is orthogonal
W If dimV, dimW < ∞, it is possible that C If λ1 = 2, λ2 = −2, λ3 = 1 and B = A3 − 3A3 then independant vectors
dimV > dimW With dimV > dimW there is no and proof 3
is B diagonisable
orthogonal mapping F : V 7→ W . 7) Prove that for a regular matrix A the two SLE
W If AP = P D and D is a diagonalmatrix then the Let A ∈ Rn×n , m < n. Let Ax = b be a system of linear
W F is not injective. columns of P are eigenvectors of A Only if the equations and let x be a solution in the least squares Ax = b and AT Ax = AT b yield the same solution
C is angle-preserving, i.e., for all v, w ∈ V it holds eigenvalues of A are on the diagonal of D sense. Which statement is always correct? 1
that ∢(F (v), F (w)) = ∢(v, w). AT Ax = AT b ⇒ A−T AT Ax = A−T AT b ⇒ Ax = b
W The vector (bAx) is orthogonal to the row space 1 = Since A is regular
C F is an isomorphism to the image of F . F is Let A ∈ Rn×n be positiv definite and symmetric.
injective as shown before and obviously of A. S.8.9
surjective to its image. Since the inverses of
Further, let λ1 , · · · , λn be the eigenvalues to the
CThe vector (bAx) is orthogonal to the column det(A) ̸= 0 ⇒ det(AT ) ̸= 0, hence AT is regular
eigenvectors v1 , · · · , vn as well
linear mappings are also linear, F is therefore an space of A. ⇒ normal equations
isomorphism W A2 has at least one eigenvalue with a strictly W x is in the null space of A. 8) For A ∈ Rn×n holds that A = AT . Prove that all
W The set of images F (B) is an orthonormal basis positive imaginary part W The solution x does not always exist eigenvalues of A2k k ∈ N are not negative.
C it holds that λj > 0 for all j = 1, · · · , n
of W . Since F is not necessarily an isomorphism, Lets define the SVD as follows: A = V ΣV −1
the image space can be smaller than W . W A hast at least one eigenvalue which satisfies:
Further we know A is orthogonal
C The set of images F (B) is an orthonormal basis geom.mult. < alg.mult Kernel/Image S.2.20
of Im(F )This follows from the isomorphism W The eigenvalues are pairwise distinct: λj ̸= λi , if A = V ΣV −1 ⇒ A = V ΣV T
property of F from the question above j ̸= i W If the nullspace of an 8 × 7 matrix is One has to prove inductively that An = V Σn V T
C If it exists,F −1 : Im(F ) 7→ V is orthogonal. It C There exist positiv real numbers α > 0 such that 5-dimensional, the rowspace has dimension 3 B.C for n = 1 it holds that: A = V Σ1 V T
exists as seen above. The orthogonality of the v T Av ≥ v T v for all v ∈ Rn n − dim(Ker(A)) = 7 − 5 = 2 = dim(Im(A)) I.H Assume it holds for any k ∈ N
 I.H n=1
I.S k 7→ k + 1: Ak+1 = AAk = AV Σk V T = Therefore at least one solution exists. Since 2 = since C = T −1 AT ∃a, such that ϕ(a) = x Therefore,
S.2.20 rank(A) < n ⇔ dim(N (A)) > 0 and as every solution of Det: αx = αϕ(a) = ϕ(αa) ∈ Im(ϕ) since ϕ is linear
V ΣV T V Σk V T = V ΣΣk V T = V Σk+1 V T the system is in Sp + αSh for any α. Where Sp is a 1 S.8.7
The eigenvalue for An can be found in the det(C) = det(T −1 AT ) = 11) Let B = (1, x, x2 ) and B ′ = (x + 1, x − 1, x2 ).The
arbitrary particular solution and Sh is the homogeneous
diagonal of Σn . With σq as the original solution. det(T −1 ) · det(A) · det(T ) = columns (spalten ) of T  are the elements of B ′ in the
eigenvalues of A, one can write:

9) Prove Satz 9.7. Namely, proof for any two similar C.8.8 1 −1 0
Σ2 k = diag(σ12k · · · σ 2k ) Therefore no eigenvalue det(T −1 ) · det(T ) · det(A) = 1 · det(A) = det(A) basis B Then T ′ = 1 1 0 by inverting
n matrices that the characteristict polynomial and det is B →B
 
can be negative. equal 1 = def of C = T −1 AT 0 0 1
 
9) Prove that AH Ax = AH b has infinitely many Assume A and C are similar. Therefore we have 10) let V and W be two vectors spaces. Let ϕ : V 7→ W 0.5 0.5 0
that C = T −1 AT . T we get T ′ = −0.5 0.5 0 The
solutions. be a linear mapping. Show, that Im(ϕ) is a subspace of B→B ′ B →B
1 2 W 0 0 1
AH Ax = AH b ⇒ AH (Ax − b) = 0 ⇒ Characterstic Polynomial: XC (λ) = (C − λI) =
mapping matrix D ′ is then  given by
1 S.8.7 1 im(ϕ) isn’t empty: 0 = ϕ(0) since ϕ is linear
AH (Ax − (b⊥ + b∥ )) = AH ((Ax − b⊥ ) − b∥ ) = det(T −1 (AT − λT )) = det(T −1 )det(AT − λT ) =

0.5 0.5 1
2 For x, y ∈ Im(ϕ) holds that x + y ∈ im(ϕ): ∃a, ∃b
D′ = T DT ′ = −0.5 −0.5 1
2 S.8.7
det(AT − λT )det(T −1 ) = det((AT − λT )T −1 ) = such that ϕ(a) = x and ϕ(b) = y Therefore, B→B ′ B →B
AH (−b⊥ ) ⇒ −(AH b⊥ ) = 0 x + y = ϕ(a) + ϕ(b) = ϕ(a + b) ∈ Im(ϕ) since ϕ is
0 0 0
1 as b⊥ ∈ R(A) it follows: Ax − b⊥ = 0 1
det(AT T −1 − λT T −1 ) = det(A − λI) = XA (λ) linear
H
2 b ∈ N (A ) 1 = def of characteristic polynomial 3 For x ∈ Im() and α ∈ R holds that αx ∈ Im() :