1 Linear Algebra

• Linear independence. v1 , . . . , vk are linearly independent iff the equation α1 v1 + · · · + αk vk = 0 has only
trivial solution α1 = · · · = αk = 0.
• Subspace (U, +, ·) of a vector space (V, +, ·) is a triple such that U ⊆ V , ∀u, v ∈ U , u + v ∈ U and ∀λ ∈ R
λ · u ∈ U . Affine subspace is a set W = U + u, where U is a subspace and u ∈ V . Subspace can be written
as a Span.
• Span({v1 , . . . , vk }) = {α1 v1 + · · · + αk vk : αi ∈ R}. Basic of a subspace U is a set of vectors B that are
linearly independent and such that Span(B) = U . The number of elements in a basis is called dimension
of U and denoted dim(U ). Equivalent definition of a basis: maximal subset of U of linearly independent
vectors, minimal subset spanning U .
To find a basis: write the subspace a span. Let S = Span({u1 , u2 , . . . , uk }). Arrange ui s as columns of
a matrix, call it A. Find rref (A). Identify pivot columns of rref (A). Select corresponding columns of A
as a basis for S.
• Equations of lines, planes, subspaces in normal form, in parametric form, norms, scalar product, vector
product.

• Linear transformations. A map L : U → V is a linear transform if L(u + v) = L(u) + L(v) and L(λu) =
λL(u) for all u, v ∈ U and λ ∈ R. Any linear transform is a matrix transform.
• Distances, projections, reflexions.
– If W is a subspace, w ∈ W is closest to v then w − v is orthogonal to W .
– If w1 , . . . , wm is an orthonormal basis of W , then projW v = (v · w1 )w1 + (v · w2 )w2 + · · · (v · wm )wm .
– If W = {w : w · a = ρ}, ||a|| = 1, then dist(v, W ) = |v · a − ρ|.
– Projection of a vector y onto a vector q is
q
projq (y) = (y · q) .
||q||2

– Distance from a point A to a plane P with normal vector n and containing a point Q is

|AQ · n|
dist(A, P) = .
n
– Distance from a point A to a line L in space that has a direction vector u and passes through a point
Q is
|AQ × u|
dist(A, L) = .
|u|
– Distance between lines L1 , L2 in space that have a direction vectors u1 , u2 and pass through points
Q1 , Q2 is
|Q1 Q2 · (u1 × u2 )|
dist(L1 , L2 ) = .
|u1 × u2 |
– Distance between two planes P1 , P2 with normal equations n · x = d and n · x = e is

|e − d|
dist(P1 , P2 ) = .
|n|

– Distance between a point A and a line Lplane in the plane with a normal vector n and passing through
a point Q is
|AQ · n|
dist(A, Lplane ) = .
|n|

1
 – Projection of a point A onto a line L in space with direction vector u and through the point Q.
u
projL (A) = proju (A − Q) + Q = ((A − Q) · u) + Q.
||u||2

– Projection of a point A onto a plane P with a normal vector n passing through a point Q. Let
d = dist(A, P ). Then
n
projP (A) = A ± d( ).
|n|
Check which belongs to the plane.
– Projection of a point A onto a line L in the plane with normal vector n. Let d = dist(A, L). Then
n
projL (A) = A ± d( ).
|n|

Check which belongs to the line.

• Matrix A.
– Rank of A is the number of pivots in the reduced row echelon form, or maximum number of linearly
independent rows, or maximal number of linearly independent columns.
– Determinant of A for square A: definition via row reduction, many properties.
– Eigenvalues, eigenvectors of A for square A.
– Determinant det(A) = 0 off columns are linearly dependent iff rows are linearly dependent iff Ax = 0
has infinitely many solutions iff rank(A) < n where A is n × n.
– Ax is a linear combination of columns with coefficients x1 , . . . , xn .
– Matrix product, properties, inverse, calculations. Determinant of a product.
– Diagonalization of a matrix.
– Matrices of rotation, dilation, projection, reflexion.
• Systems: procedure, number of solutions, understanding what is the set of solutions. Ker, Im of a linear
transform.

2 Differential equations
• Separable g(y)dy = f (x)dx. Integrate.
• Linear y 0 + p(x)y = q(x). Find integrating factor µ(x) = exp( p(x)dx). Multiply the equation by µ.
R

After simplifications it becomes:

(yµ)0 = qµ.
Integrate.
• Bernoulli y 0 + p(x)y = q(x)y n . Substitution
1
y = v 1−n .

This reduced the equation to linear.

• Linear higher order constant coefficients homogeneous. ak y (k) + ak−1 y (k−1) + · · · + a0 y = 0. Plug y = eλx .
The equation reduces to ak λk + ak−1 λk−1 + · · · + a0 λ = 0. Find all λ’s. For each λi find a solution yi .
General solution y = c1 y1 + · · · + ck yk .
Cases: non-repeated real, repeated real, complex, repeated complex. Example: λ1 = 5, λ2 = 5, λ3 =
5, λ4 = 1 − 6i, λ5 = 1 + 6i, λ6 = 1 − 6i, λ7 = 1 + 6i, λ8 = 0. Then y = c1 e5 x + c2 xe5x + c3 x2 e5x +
c4 ex cos(6x) + c5 ex sin(6x) + c6 xex cos(6x) + c7 xex sin(6x) + c8 .

2
• Linear higher order non-homogeneous. ak y (k) + ak−1 y (k−1) + · · · + a0 y = f (x).

ygen.nonhom = ygen.hom + yparticular.nonhomog. .

To find particular non-homogeneous, use method of undetermined coefficients. For example, when f (x) =
e3x , look for a solution candidate in the form y = Ae3x ; when f (x) = sin(3x), look for a solution
candidate in the form y = A sin(3x) + B cos(3x); when f (x) = x2 , look for a solution candidate in the
form y = Ax2 + Bx + C.
For the second order, one could use the method of variation of parameters. If y1 , y2 is a fundamental
y2 f
solution of homogeneous DE, then the general solution is y = u1 y1 + u2 y2 , where u01 = − W (y 1 ,y2 )
,
y1 f
u02 = W (y1 ,y2 ) .

• Reduction of order linear homogeneous. Given a solution y, look for another solution in the form y1 = vy,
plug y1 in the DE, obtain a lower order DE in terms of v, solve it, back substitute.
• Euler equations: special linear non-constant coefficients. ak y (k) + ak−1 y (k−1) + · · · + a0 y = f (x).

ygen.nonhom = ygen.hom + yparticular.nonhomog. .

To find ygen.hom use a candidate y = xr . Plug, obtain an equation on r, find all r, for repeated ones,
multiply by ln x.
Example: r1 = 2, r2 = 2, r3 = 2, r4 = 0, r5 = 4 − 7i, r6 = 4 + 7i. Then y = ygen.homogeneous =
c1 x2 + c2 x2 ln x + c3 x2 ln2 x + c4 + c5 x(4−7i) + c6 x(4+7i)x . Rearranging complex solutions, we get another
way to write the solution. y = c1 x2 + c2 x2 ln x + c3 x2 ln2 x + c4 + c5 x4 cos(7 ln x) + c6 x4 sin(7 ln x).

• Systems of linear homogeneous DEs constant coefficients. y 0 = Ay, where A is a square matrix and y is a
vector function.
Look for solutions in the form y = eλx v, where v is a vector. Plug in, get a new equation λv = Av, i.e., λ
is an eigenvalue, v is a corresponding eigenvector. For each λi find a solution yi . Then a general solution
is y = c1 y1 + c2 y2 + · · · + cn yn .
Eigenvalues λ = α ± iβ with eigenvectors v̄ = ā ± ib̄.
Then real solutions are given by

u(x) = eαx (cos(βx)ā − sin(βx)b̄)

v(x) = eαx (sin(βx)ā + cos(βx)b̄)

• Laplace transform method for solving linear DEs and systems of linear DEs with initial value conditions.
Apply Laplace transform to the DE with unknown y = y(x). The DE will be transferred into an algebraic
equation with unknown U = L(y) and coefficients in terms of s. Find U in terms of s. Find y = L−1 (U )
using the Table or convolution rules.

• The power series method for solving DEs. Given an initial value problem at x0 , represent a solution as a
power series expanded at x0 . I.e., y = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · . Plug in the DE. Starting with
a0 , a1 , find all other coefficients recursively.