A lot of content summarized from Mark Schmidt’s notes on Linear Algebra
Notation
Generally column major
 Scalar (1,1): $α$
 Column Vector (m, 1): $[a_{1}a_{2} ]$
 Row Vector (1, n): $[a_{1} a_{2} ]$
 Matrix (m, n): $[a_{1,1}a_{1,2} a_{2,1}a_{2,2} ]$
Operations
Transpose
$(A_{T})_{ij}=(A)_{ji}$
A matrix is symmetric if $A=A_{T}$
Vector Addition
Associative (brackets don’t matter) and commutative (order independent)
$a+b=[a_{1}a_{2} ]+[b_{1}b_{2} ]=[a_{1}+b_{1}a_{2}+b_{2} ]$
Scalar Multiplication
Associative (brackets don’t matter) and commutative (order independent)
$αb=α[b_{1}b_{2} ]=[αb_{1}αb_{2} ]$
Inner Product
Between two vectors of the same length, multiply each element together to get a scalar result
$a_{T}b=∑_{i=1}a_{i}b_{i}=γ$
A specific version of this is the dot product which can be expressed as the inner product between two vectors, $a⋅b=a_{T}b$
 Commutative: $a_{T}b=b_{T}a$
 Distributive across addition: $a_{T}(b+c)=a_{T}b+a_{T}c$
Outer Product
Between two vectors of the same length, create a matrix multiplying each combination of elements in each vector.
Given two vectors $u= u_{1}u_{2}⋮u_{m} $ and $v= v_{1}v_{2}⋮v_{n} $,
$u⊗v=A= u_{1}v_{1}u_{2}v_{1}⋮u_{m}v_{1} u_{1}v_{2}u_{2}v_{2}⋮u_{m}v_{2} ……⋱… u_{1}v_{n}u_{2}v_{n}⋮u_{m}v_{n} $
The resulting matrix $A$ is always rank1.
Multiplication
In general, we can multiply matrices A and B when the number of columns in A matches the number of rows in B
If A is (m, n) and B is (n, p), then AB is (m, p)
 Associative: $A(BC)=(AB)C$
 Distributive across addition: $A(B+C)=AB+AC$
 Generally not commutative: $AB=BA$
 Transposing reverses order: $(AB)_{T}=B_{T}A_{T}$
 Matrix powers don’t change order: $(AB)_{2}=ABAB$
 Matrixvector multiplication always yields a vector: $x_{T}Ay=x_{T}(Ay)=(Ay)_{T}x=y_{T}A_{T}x$
Properties
Vector Norm
A scalar measure of a vector’s length

$∥x∥≥0$

$∥x∥_{2}=x_{T}x$

Euclidean Norm (L2Norm): $∥x∥_{2}=∑_{i=1}x_{i} $
 Also note that $∥x∥_{2}=∥x∥_{2}=r_{T}r=⟨r,r⟩$

Manhattan Distance (L1Norm): $∥x∥_{1}=∣r_{1}∣+∣r_{2}∣$
 How many ‘blocks’ you need to traverse

L$∞$Norm: $∥x∥_{∞}=max(∣r_{1}∣,∣r_{2}∣)$
 How many blocks you have to walk in any one dimensions
Rank
 The dimension of the vector space generated (or spanned) by its columns.
 This corresponds to the number of linearly independent columns of A.
 This minimal set of vectors that span a space is called a basis
Orthogonal
If for some set of vectors $q$:
 $q_{i}q_{j}=0$, we call $q_{i}$ and $q_{j}$ orthogonal
 $q_{i}q_{j}=1$, we call $q_{i}$ and $q_{j}$ orthonormal
Inner product of square orthogonal matrices is the identity matrix: $Q_{T}Q=I=QQ_{T}$
Linear Dependence
A vector is linearly dependent on a set of vectors if it can be written as a linear combination of them
$c=α_{1}b_{1}+α_{2}b_{2}+⋯+α_{n}b_{n}$
A set of vectors is linearly dependent if and only if the zero vector can be written as a nontrivial combination of any of the vectors.
A matrix with fully independent columns has full column rank. If this is the case, $Ax=0$ implies that $x=0$
Special Matrices
Identity Matrix
1’s on the diagonal and 0’s otherwise. $I_{n}$ denotes an (n,n) identity matrix.
Multiplication by the identity matrix yields the original matrix. Columns of the identity matrix are called elementary vectors.
Diagonal Matrix
$D= d_{1}00 0d_{2}0 00d_{3} $
Spaces
Range (Columnspace)
Subspace spanned by the columns of a matrix.
A system $Ax=b$ is solvable if and only if b is in $A$‘s columnspace
Subspace
A nonempty subset of vector space that is closed under addition and scalar multiplication
Possible spaces of $R_{3}$
 0 Vector
 Any line or plane through the origin
 All of $R_{3}$
We say that the vectors generate or span the subspace when you can reach any point in the subspace through a linear combination of the vectors.
Matrices as transformation
Viewing $Ax=T(x)$
A linear transformation can’t move the origin. But, if there are linearly dependent columns, there are nonzero vectors that can be transformed to zero. The set of vectors that can be transformed to 0 is called the nullspace.
Null space: $N(A)$ is all $x$ such that $Ax=0$
Fundamental Theorem of Linear Algebra
 $r$ is the dimension of the columnspace which is the same as the dimension of the rowspace
 The nullspace is orthogonal to the rowspace
Inverses
We can find the inverses if and only if A is square and doesn’t have nullspace outside of the zero vector (otherwise we either lose information to the nullspace or can’t get to all vectors)
If the inverse exists, it is a unique matrix such that $A_{−1}A=I=AA_{−1}$
Some identities
 $(A_{−1})_{T}=(A_{T})_{−1}$
 $(γA)_{−1}=γ_{−1}A_{−1}$
 Assuming both $A_{−1}$ and $B_{−1}$ exist, $(AB)_{−1}=B_{−1}A_{−1}$
Special inverses of diagonal matrices
$D= d_{1}00 0d_{2}0 00d_{3} $
$D_{_{−}1}= 1/d_{1}00 01/d_{2}0 001/d_{3} $
Solving Linear Equations
Given A and b, we want to solve for x in $Ax=b$
Say, $[21 −11 ][xy ]=[15 ]$.
We can interpret this multiple ways:
 By rows: $x$ is the intersection of the hyperplanes $2x−y=1$ and $x+y=5$
 By columns: $x$ is the linear combination that yields the RHS in $x[25 ]+y[−11 ]=[15 ]$
 Transformation
$Ax=b$ generally has a solution when $b$ is in the columnspace of A. It has a single unique solution if the columns of A are linearly independent.
If $Ax=b$ has as solution we say it is consistent.
Basically, $x=A_{−1}b$
We can solve using Gaussian Elimination