1/399
Looks like no tags are added yet.
Three elementary row operations
Replacement (add a multiple of one row to another), interchange (swap two rows), scaling (multiply a row by a nonzero number). None of them change the solution set.
Echelon form (3 rules)
1) Zero rows at the bottom. 2) Each leading entry is to the right of the one above. 3) Zeros below each leading entry.
Reduced echelon form (2 extra rules)
4) Every leading entry is 1. 5) Each leading 1 is the only nonzero entry in its column.
Pivot column
A column of A that contains a pivot position (leading 1 in the RREF).
Basic vs. free variable
Basic: matches a pivot column. Free: matches a non-pivot column of the coefficient part; it can take any value.
When is a linear system consistent?
When the echelon form has no row [0 … 0 | b] with b ≠ 0 (the augmented column is not a pivot column).
Consistent system: one solution or infinitely many?
Exactly one if there are no free variables. Infinitely many if there is at least one free variable.
Span{v1, …, vp}
The set of all linear combinations c1v1 + … + cpvp.
b is in Span of the columns of A means …
Ax = b has a solution, and [A | b] has no pivot in the last column.
Columns of A (m×n) span ℝᵐ iff …
A has a pivot in every ROW; equivalently Ax = b is consistent for every b in ℝᵐ.
Columns of A are linearly independent iff …
Ax = 0 has only the trivial solution; A has a pivot in every COLUMN.
Solution set of Ax = b (when consistent)
x = p + v_h, where p is one particular solution and v_h is any solution of Ax = 0.
Homogeneous system Ax = 0 has a nontrivial solution iff …
There is at least one free variable.
Quick dependence tests
A set is dependent if it contains 0, if it has more than n vectors in ℝⁿ, or if one vector is a combination of the others. Two vectors: dependent iff one is a multiple of the other.
Linear transformation (2 rules)
T(u + v) = T(u) + T(v) and T(cu) = cT(u) for all u, v and scalars c. Consequence: T(0) = 0.
Standard matrix of T: ℝⁿ → ℝᵐ
A = [T(e1) T(e2) … T(en)]: column j is where T sends the j-th standard basis vector.
Standard matrix of R∘S∘T
[R][S][T]: the map applied first goes on the right.
T(x) = Ax is one-to-one iff …
Ax = 0 has only the trivial solution; pivot in every column; columns independent.
T(x) = Ax is onto ℝᵐ iff …
Columns of A span ℝᵐ; pivot in every row.
Find T(x) when T is given on non-standard vectors
Write x as a combination of the given inputs, then use linearity. Example: e1 = (1,2) − 2(0,1), so T(e1) = T(1,2) − 2T(0,1).
Inverse of a 2×2 matrix with rows (a, b), (c, d)
1/(ad − bc) times the matrix with rows (d, −b), (−c, a). Exists only when ad − bc ≠ 0.
How to find A⁻¹ for a larger matrix
Row reduce [A | I]. If the left becomes I, the right is A⁻¹. If a zero row appears on the left, A is not invertible.
(AB)ᵀ and (AB)⁻¹
(AB)ᵀ = BᵀAᵀ and (AB)⁻¹ = B⁻¹A⁻¹: the order reverses.
Invertible Matrix Theorem (core list)
For square A: invertible ⇔ n pivots ⇔ Ax = 0 only trivial ⇔ columns independent ⇔ columns span ℝⁿ ⇔ one-to-one ⇔ onto ⇔ det A ≠ 0 ⇔ 0 is not an eigenvalue ⇔ rank A = n ⇔ Nul A = {0}.
If AB + A = I, is A invertible?
Yes (A square): A(B + I) = I, so A⁻¹ = B + I.
If A² − 3A + 2I = 0, what is A⁻¹?
A(3I − A) = 2I, so A⁻¹ = (3I − A)/2.
det of 2×2 matrix with rows (a, b), (c, d)
ad − bc.
Cofactor expansion along row i
det A = a_i1 C_i1 + … + a_in C_in, with C_ij = (−1)^(i+j) det A_ij. Choose the row or column with the most zeros.
det of a triangular matrix
The product of the diagonal entries.
Effect of row operations on det
Replacement: no change. Interchange: sign flips. Scaling a row by k: det multiplied by k.
det(AB), det(Aᵀ), det(A⁻¹), det(cA)
det A · det B; det A; 1/det A; cⁿ det A (n×n).
det of B whose rows are combinations of rows of A
Write B = MA where row i of M holds the coefficients. Then det B = det M · det A.
Area or volume from a determinant
Area of parallelogram spanned by the columns of A = |det A|. A transformation x ↦ Ax multiplies areas by |det A|.
Subspace (3 conditions)
1) 0 is in H. 2) Closed under addition. 3) Closed under scalar multiplication. For ALL vectors and ALL scalars.
Fastest way to show H is a subspace
Write a general element as a span: s·v1 + t·v2 + …. Spans are always subspaces.
Fastest way to show H is NOT a subspace
Show the zero vector is not in H, or give one specific failure of closure.
Nul A (A is m×n)
{x : Ax = 0}, a subspace of ℝⁿ. Basis: the vectors multiplying the free variables in the parametric solution.
Col A (A is m×n)
Span of the columns = {Ax}, a subspace of ℝᵐ. Basis: pivot columns of the ORIGINAL A.
Row A
Span of the rows = Col Aᵀ, a subspace of ℝⁿ. Basis: nonzero rows of an echelon form.
Kernel and range of a linear T: V → W
Kernel: {v : T(v) = 0}, a subspace of V. Range: {T(v)}, a subspace of W.
Basis
A set that is linearly independent AND spans the space.
Dimension
The number of vectors in any basis. dim ℝⁿ = n, dim ℙₙ = n + 1, dim of m×n matrices = mn.
Rank–Nullity Theorem
rank A + dim Nul A = n (number of columns). Also dim Row A = dim Col A = rank A.
Basis Theorem
If dim V = p, any p independent vectors form a basis, and any p vectors that span V form a basis.
Spanning Set Theorem
A finite spanning set contains a basis: remove vectors that are combinations of the others.
Coordinate vector [x]_B
The weights c1, …, cn with x = c1b1 + … + cnbn. In ℝⁿ: x = P_B[x]_B, so [x]_B = P_B⁻¹x.
Coordinate mapping x ↦ [x]_B
A one-to-one linear map onto ℝⁿ (isomorphism). Lets you row reduce polynomials and matrices through their coordinate vectors.
P_{C←B}
Columns are [b1]C, …, [bn]_C. Then [x]_C = P{C←B}[x]B, and P{B←C} = (P_{C←B})⁻¹.
How to compute P_{C←B} in ℝⁿ
Row reduce [c1 … cn | b1 … bn] to [I | P_{C←B}]. The basis you convert TO goes on the left.
Eigenvector and eigenvalue
x ≠ 0 with Ax = λx. λ is an eigenvalue if (A − λI)x = 0 has a nontrivial solution.
Eigenspace for λ
Nul(A − λI): all eigenvectors for λ together with 0.
Characteristic equation
det(A − λI) = 0. For 2×2: λ² − (trace)λ + det A = 0.
Eigenvalues of a triangular matrix
The diagonal entries.
Algebraic vs. geometric multiplicity
Algebraic: times λ appears as a root of det(A − λI). Geometric: dim Nul(A − λI). Always 1 ≤ geometric ≤ algebraic.
Similar matrices
B = P⁻¹AP. Same characteristic polynomial and eigenvalues; usually different eigenvectors. Same eigenvalues does NOT imply similar.
Diagonalization Theorem
A (n×n) = PDP⁻¹ iff A has n linearly independent eigenvectors. Columns of P = eigenvectors; D = matching eigenvalues in the same order.
When is A diagonalizable (multiplicity test)?
For every eigenvalue, geometric multiplicity = algebraic multiplicity (eigenspace dimensions add to n).
n distinct eigenvalues ⇒ ?
Diagonalizable (sufficient, not necessary).
Steps to diagonalize A
1) Solve det(A − λI) = 0. 2) Basis for each Nul(A − λI). 3) Check dimensions add to n. 4) P = eigenvectors, D = eigenvalues. 5) A^k = PD^kP⁻¹.
Matrix of T relative to basis B, [T]_B
Columns are [T(b1)]_B, …, [T(bn)]_B. Apply T to each basis vector and rewrite the output in basis B.
If A = PDP⁻¹, what is D in terms of linear transformations?
D is the B-matrix of x ↦ Ax, where B is the basis of columns of P (eigenvectors).
Complex eigenvalue form for a real 2×2 matrix
If λ = a − bi with eigenvector v, then A = PCP⁻¹ with P = [Re v Im v] and C with rows (a, −b), (b, a).
What C with rows (a, −b), (b, a) does
Rotates by the angle φ of the point (a, b) and scales by r = √(a² + b²) = |λ|.
Complex eigenvalues of a real matrix come in …
Conjugate pairs: if λ has eigenvector v, then λ̄ has eigenvector v̄.
Dot product, length, distance
u·v = u1v1 + … + unvn = uᵀv; ‖v‖ = √(v·v); dist(u, v) = ‖u − v‖.
Orthogonal vectors
u·v = 0. Pythagorean Theorem: u ⊥ v iff ‖u + v‖² = ‖u‖² + ‖v‖².
W⊥
All vectors orthogonal to every vector in W. To find a basis: put a spanning set of W as ROWS of a matrix and compute its null space.
(Row A)⊥ and (Col A)⊥
(Row A)⊥ = Nul A; (Col A)⊥ = Nul Aᵀ.
Weights for an orthogonal basis
If y = c1u1 + … + cpup with {u_j} orthogonal: c_j = (y·u_j)/(u_j·u_j). No row reduction.
Orthonormal columns: U (m×n)
UᵀU = I_n, ‖Ux‖ = ‖x‖, (Ux)·(Uy) = x·y. UUᵀ is projection onto Col U.
Orthogonal matrix
SQUARE with orthonormal columns. U⁻¹ = Uᵀ and det U = ±1.
Orthogonal projection onto W (orthogonal basis u1, …, up)
ŷ = (y·u1/u1·u1)u1 + … + (y·up/up·up)up. With orthonormal U: ŷ = UUᵀy.
Best Approximation Theorem
proj_W y is the closest point in W to y. Distance from y to W = ‖y − proj_W y‖.
Orthogonal Decomposition Theorem
y = ŷ + z uniquely, ŷ in W and z = y − ŷ in W⊥. ŷ does not depend on which orthogonal basis you use.
Gram–Schmidt step
v_k = x_k minus the projections of x_k onto v1, …, v_(k−1). You may rescale each v_k to clear fractions.
QR factorization
A with independent columns: A = QR. Q = normalized Gram–Schmidt vectors (orthonormal columns); R = QᵀA, upper triangular with positive diagonal.
Normal equations
AᵀAx̂ = Aᵀb. Their solutions are exactly the least-squares solutions of Ax = b.
When is the least-squares solution unique?
When the columns of A are linearly independent (AᵀA invertible): x̂ = (AᵀA)⁻¹Aᵀb.
Least-squares error
‖b − Ax̂‖. The closest point in Col A is Ax̂ = b̂; x̂ itself is the list of weights in ℝⁿ.
Least squares with A = QR
Solve Rx̂ = Qᵀb by back-substitution.
Inner product axioms (5)
1) ⟨u,v⟩ = ⟨v,u⟩. 2) ⟨u+v,w⟩ = ⟨u,w⟩ + ⟨v,w⟩. 3) ⟨cu,v⟩ = c⟨u,v⟩. 4) ⟨u,u⟩ ≥ 0. 5) ⟨u,u⟩ = 0 only when u = 0.
Norm, distance, orthogonal in an inner product space
‖u‖ = √⟨u,u⟩; dist = ‖u − v‖; orthogonal means ⟨u,v⟩ = 0. Replace every u·v by ⟨u,v⟩ in all Chapter 6 formulas.
Evaluation inner product on ℙₙ
⟨p,q⟩ = p(t0)q(t0) + … + p(tn)q(tn) with n + 1 distinct points. Compute: table of values, then dot product of the value lists.
Why the evaluation inner product needs n + 1 distinct points
Axiom 5: a nonzero polynomial of degree ≤ n cannot vanish at n + 1 points.
Integral inner product on C[a,b]
⟨f,g⟩ = ∫ from a to b of f(t)g(t) dt.
Which ⟨u,v⟩ = uᵀAv are inner products on ℝⁿ?
A symmetric with all eigenvalues strictly positive.
Cauchy–Schwarz inequality
|⟨u,v⟩| ≤ ‖u‖‖v‖.
Triangle inequality
‖u + v‖ ≤ ‖u‖ + ‖v‖.
Best approximation to t² on [−1,1] by span{1, t} (integral inner product)
The constant 1/3 (from ⟨t²,1⟩/⟨1,1⟩ = (2/3)/2; the t-coefficient is 0).
Orthogonal basis of span{1, t, t²} in C[−1,1]
{1, t, t² − 1/3} (Legendre polynomials).
Show ⟨x,y⟩ = x²y² on ℝ is not an inner product
Axiom 3 fails: ⟨2·1, 1⟩ = 4 but 2⟨1,1⟩ = 2.
Symmetric matrix
Aᵀ = A (square, mirror image across the diagonal). Always symmetric: BᵀB, BBᵀ, uuᵀ, diagonal matrices.
Orthogonally diagonalizable
A = PDPᵀ with P orthogonal and D diagonal; equivalently ℝⁿ has an orthonormal basis of eigenvectors of A.
Symmetric ⇔ ?
Orthogonally diagonalizable (Theorem 2 of 7.1).
Eigenvectors of a symmetric matrix for different eigenvalues are …
Orthogonal. Proof: λ1(v1·v2) = (Av1)·v2 = v1·(Av2) = λ2(v1·v2), so (λ1 − λ2)(v1·v2) = 0.
Spectral Theorem (4 parts)
Symmetric A: (a) n real eigenvalues with multiplicity; (b) each eigenspace dimension = algebraic multiplicity; (c) eigenspaces mutually orthogonal; (d) orthogonally diagonalizable.
Steps to orthogonally diagonalize a symmetric A
1) Eigenvalues. 2) Basis of each eigenspace. 3) Gram–Schmidt INSIDE eigenspaces of dimension ≥ 2. 4) Normalize. 5) P = unit eigenvectors, D = eigenvalues, A = PDPᵀ.
Spectral decomposition
A = λ1u1u1ᵀ + … + λnununᵀ, where each ukukᵀ is the projection matrix onto Span{uk}.
Why orthogonally diagonalizable ⇒ symmetric
Aᵀ = (PDPᵀ)ᵀ = PDᵀPᵀ = PDPᵀ = A.
Singular values of A
σk = √λk where λk are the eigenvalues of AᵀA, in decreasing order. σk = ‖Avk‖.