Vous êtes sur la page 1sur 11

Miscellaneous

Daniel Gruszczynski
December 23, 2015
This document contains interesting blurbs and problems Ive come across
but had no home for. Each section is standalone and can be read (or skipped)
in any order.

1
1.1

Abstract Algebra
Eulers totient function

Definition 1. We define Eulers totient function as follows: For every


n Z+ , (n) is the number of positive integers a n such that gcd (a, n) = 1.
Property 1. (p) = p 1 for any prime p
Proof. By definition, p has no factors other than 1 and itself. Hence gcd (a, n) =
1 for all 1 < a < p. Thus, (p) = p 1.
Property 2. For any prime p and all a 1,
(pa ) = pa1 (p)
Proof. By property 1, pa 1 (pa ). Note that equality does not follow
because multiples of p are not relatively prime to pa . Excluding 1, there are
(pa /p) 1 = pa1 1 many factors. Therefore,
(pa ) = (pa 1) (pa1 1) = pa pa1 = pa1 (p 1) = pa1 (p).

Property 3. If a and b are relatively prime, then (ab) = (a)(b).


Proof. Notice that since gcd (a, b) = 1, lcm (a, b) = ab (and so we dont run
the risk of double counting).

1.2

Group theory

Here are some interesting examples of group homomorphisms.


Example 1. Define : R {1} by (x) = x/|x|. Then is a surjective
homomorphism since
(ab) =

ab
ab
=
= (a)(b)
|ab|
|a||b|

and (1) = 1. Because |x| is necessarily positive, the fibers of are the
positive (1) and negative (-1) reals, respectively.
Example 2. Define : R2 R by (x, y) = x + y. Then is surjective since,
for any a R, we have a = a + 0 = (a, 0). We could have also picked (0, a),
either of which corresponds to the fact that the x and y axes are homomorphic
to R by projection maps. Next,
((x, y) + (x0 , y 0 )) = (x + x0 , y + y 0 )
= (x + x0 ) + (y + y 0 )
= (x + y) + (x0 + y 0 )
= (x, y) + (x0 , y 0 )
and so is a homomorphism. The fibers of for a fixed c R correspond to
y = x + c, translations of the kernel y = x (for when c = 0). Explicitly,
ker () = {(x, x) | x R}. Hence, the fibers are all parallel lines.
Example 3. Another geometric example is : C R given by (a + bi) =
a2 + b2 , or equivalently (z) = |z|2 (squared modulus of the complex number).
First, is homomorphic since
((a + bi)(c + di)) = ((ac bd) + (bc + ad)i)
= (ac bd)2 + (bc + ad)2
= (ac)2 2abcd + (bd)2 + (bc)2 + 2abcd + (ad)2
= a2 (c2 + d2 ) + b2 (c2 + d2 )
= (a2 + b2 )(c2 + d2 )
= (a + bi)(c + di).
Then, ker () = {a2 + b2 = 1 | a, b R}, and so we have ker () ' S 1 .
Similarly, the fibers of are circles around the origin of all radii > 0. Hence,
the entire plane of the graph is covered except for the origin.
Example 4. Let F be a field and G be a subgroup of the 2 x 2 general linear
group consisting of upper triangular matrices. More explicitly,


a b
G={
| a, b, c F, ac 6= 0}.
0 c
2


a
Define : G F by (
0


b
) = a. Its clear that is surjective since
c


a 0
7 a.
0 1

Next, is a homomorphism because



  0 0
 0
a b
a b
aa
(
) = (
0 c
0 c0
0
= aa0

a
= (
0


ab0 + bc0
)
cc0
  0
b
a
)(
c
0


b0
).
c0

Fix some x F . The fiber of x is




x b
1 (x) = {
| b, c F, c 6= 0}
0 c
with the kernel being when x = 1.

2
2.1

Geometry
Affine Spaces

In linear algebra, we deal primarily with vector spaces. The points of a


vector space are treated as vectors emanating from a special point known as
the origin. Each vector is then represented by a tuple called coordinates that
are with respect to this origin and a basis. When we change bases, we keep the
same origin as before, and so linear combinations are preserved. More formally,
we can call the tuple F = (0, B) of the origin and basis a frame for our vector
space.
There are two observations to make here. First, if we were to pick a different
origin, say , then the location of points do not coincide with their vector
coordinates. This is because we are relying on the fact that a = 0a, where 0a
represents the difference between a and the origin. This means that for any
vector v in our initial frame
0v = 0 + v
The second observation is that linear combinations are preserved only in a
special case. Say a and b form vectors wrt 0. We want to represent a + b wrt
. Given the above facts,
a0 = a = 0a 0 = a

and likewise for b. So substituting for a and b, we have


a0 + b0 = (a ) + (b ) = a + b ( + )
But wait, this linear combination is also a vector! So we should have
a0 + b0 = a + b .
Hence, we get our desired properties only when
+ = 1.
Pn
If we extended this to n vectors, then of course wed require that i=1 i = 1.
For simplicity, well assume that all vector spaces are over R, and that families of scalars (i )iI have finite support. This means that if J I is finite,
then i = 0 for i I J.
Definition 2. An affine space is either the degenerate space , or a triple
hE, E, +i consisting of a nonempty set E of points, a vector space E of translations (free vectors), and an action + : E E E satisfying the following
conditions:
(A1) a + 0 = a for every a E
(A2) (a + u) + v = a + (u + v) for every a E, u, v E
(A3) For any two points a, b E, there exists a unique vector u E such that
a + u = b. We denote this u = ab.
One interpretation of this definition is that E and E provide different views
of the same space, and so we ought to be able to move back and forth between
points and vectors. To see this, fix any a E so that +(a, ) : E E is given
by u 7 a + u. Then, using the same a we can use (A3) to define b 7 ab from
E to E. Compose these mappings:
u 7 a + u 7 a(a + u)
Now, notice that a+u = a+a(a + u). Cancelling out a, we have u = a(a+u),
and so the above composition is an identity mapping on E. Reversing the order
of the composition yields
b 7 ab 7 a + ab,
the identity mapping on E.
Theorem 1 (Chasles Identity). Given points a, b, c E, ab + bc = ac.
Proof. We apply (A3) thrice: c = a + ac, c = b + bc, and b = a + ab. Then
c = b + bc = (a + ab) + bc = a + (ab + bc).
Since c = a + ac = a + (ab + bc), the equality follows.
4

Corollary 1. The following are true:


Given any a E, aa = 0.
For any a, b E, ba = ab
For any a, b, c, d E, ab = dc bc = ad. This is the parallelogram law.
Proof. Since a + 0 = a and a = a + aa, aa = 0. Next, Chasles Identity gives
us ab + ba = aa = 0. Hence ba = ab. Last,
ab + bc = ac = ad + dc
gives us our result.
Definition 3. For
Pany family of weighted points (ai , i )iI in E,
we say the point iI i ai is the barycenter of this family.

iI

i = 1,

A difference between points and vectors is in how we interpret their affine


(linear) combinations. An affine combination of points is like a scale: each point
is given an associated mass (normalized to be a proportion of 1) and balancing
this out is akin to the center-of-mass (COM), hence the name barycenter. A
linear combination of vectors is the scaling of a parallelepiped.
Definition 4. Given an affine space, we define an affine frame with origin a0
is a family (a0 , ..., am ) of m + 1 points in E such that B = (a0 a1 , a0 a2 , ..., a0 am )
is a basis in E. The resulting pair (a0 , B) is also called the affine frame.
A family of two points is affine independent iff ab 6= 0. Equivalently, a 6= b.
Of course, the resulting subspace is the set of points
(1 )a + b,
the unique line segment connecting a and b. This is a convex set in one dimension. Now suppose we have a family of three points (a, b, c); this is affine independent iff ab and bc are linearly independent. Hence a, b, c are not collinear.
The result is
(1 )a + b + c.
If we hold = 0, then we get a line segment from a to b. If = 0, then we get
a line segment from a to c. Holding 1 = 0 means + = 1, and so
we form a line segement from b to c, thus forming a triangle (a convex shape in
two dimensions).
Definition 5. A subset V E is called convex if for any two points a, b V ,
then (1 )a + b V for [0, 1].

Definition 6. Given two affine spaces hE, E, +i and hF, F, +0 i, a function f :


E F is called an affine map if f preserves barycenters.
That is, for every
P
family (ai , i )iI of weighted points in E such that iI i = 1, then
f(

i ai ) =

iI

i f (ai ).

iI

Proposition 1. Given any point a E, b F , and any linear map h : E F,


we claim that f : E F defined by
f (a + v) = b + h(v)
is an affine map. Intuitively, this says that an affine mapping of a translated
vector is the same as translating a vector under a linear mapping.
Proof. Pick a family of vectors (vi )iI in E and a family of scalars (i )iI whose
sum is 1. Then
X
X
i (a + vi ) = a +
i vi
iI

iI

and
X

i (b + h(vi )) = b +

iI

i h(vi ).

iI

Thus,
X
X
f(
i (a + vi )) = f (a +
i vi )
iI

iI

X
= b + h(
i vi )
iI

=b+

i h(vi )

iI

i (b + h(vi ))

iI

i f (a + vi )

iI

Lemma 1. Given an affine map f : E F , there exists a unique linear map


f : E F such that
f (a + v) = f (a) + f(v)
y = Ax + b where b 6= 0 unless there exists a fixed point a0 E.

2.1.1

Affine Groups

Definition 7. Given an affine space E, the affine group of E is the set of all
affine bijections f : E E. We denote this group by GA(E).
Since we have the general linear group GL(E) defined as the set of all bijective linear maps on vector space E, we can form a group homomorphism
L : GA(E) GL(E) by f 7 f. Then ker (L) is the set of translations on E.
2.1.2

Affine Geometry

Theorem 2 (Thales). Given any affine space E, if H1 , H2 , H3 are any three


distinct, parallel hyperplanes, and A and B are any two lines not parallel to Hi
(i [3]), letting ai = Hi A and bi = Hi B, then the following ratios are
equal:
a1 a3
b1 b3
=
=
a1 a2
b1 b2

2.2
2.2.1

Projective Geometry
Synthetic Treatment of Affine Geometry

Definition 8. An affine plane is a set A of elements called points, and a set


of subsets of A called lines satisfying the following axioms:
(A1) Given two distinct points P and Q, there is one and only one line containing both points.
(A2) Given a line l and a point P not on l, there is one and only one line which
is parallel to l and which passes through P .
(A3) There exists three non-collinear points.
Definition 9. Two lines are parallel if they are equal, or if they have no points
in common.
Definition 10. A set of points is collinear if there exists a line l containing
them all.
Proposition 2. The set of all lines parallel to l, [l], form an equivalence relation.
Proof. Since a line is equal to itself, lkl. If l1 kl2 , then l2 kl1 : l1 = l2 l2 = l1
or l1 l2 = = l2 l1 . Now suppose l1 kl2 and l2 kl3 . If l1 = l3 , we are done.
Otherwise, suppose there exists a point P l1 l3 . Then, P 6 l2 : otherwise
l1 = l2 = l3 . It follows by (A2) that there is only one line parallel to l2 containing
P , but again this is a contradiction. Hence, l1 l3 = and l1 kl3 .
Proposition 3. Two distinct lines have at most one point in common.
Proof. If two lines pass through two distinct points, then by (A1), those lines
must be equal.
7

Proposition 4. There exists an affine plane with four points.


Proof. By (A3), there exists three non-collinear points P , Q, and R. By (A1),
there is a line through P and Q. Then, using (A2) there is a unique line r
through R parallel to P Q. Repeat this procedure to obtain line p through P
that is parallel to QR. Claim p 6 kr: otherwise P QkrkpkQR, which is impossible
since P Q 6= QR but Q P Q QR. Hence, p and r have a common point S.
Since S r, S 6= P and S 6= Q because P Q 6= r. And since S p, S 6= R.
Thus, P, Q, R, S are four distinct points.
Definition 11. A pencil of lines is a set of all lines through some point P .
Definition 12. Let A be an affine plane. Each distinct [l] is called an ideal
point, or point at infinity, in the direction of l. We write P = [l].

3
3.1

Optimization
Convexity

Definition 13. Let X be a convex set and f : X R. Then the function f is


convex if for all x, y X and t [0, 1],
f (tx + (1 t)y) tf (x) + (1 t)f (y)
The idea here is that the secant line segment from f (x) to f (y) lies above the
images of the points between x and y. If f is linear, then it is trivially convex.
Hence, convexity is a relaxation of linearity that requires our weights to sum to
one.

4
4.1

Graph Theory and Linear Programming


Matchings

For this section, we will consider finite, undirected, bipartite graphs. Formally,
this means that we have a graph G = (V, E) where |V | < , V = X Y for
disjoint sets X and Y , and if (x, y) E, then x X and y Y (or vice versa).
Now, it doesnt matter which set of vertices we call X or Y ... only that we are
consistent in our notation. Hence, well use X as our reference.
The goal is straightforward: find an injection f : X Y such that (x, f (x))
E. We call the graph of this function an X-matching. In general, a matching
is a set M of independent edges, and an X-matching is when every x X is an
endpoint for some edge in M . A perfect matching occurs when f is a bijection. (Recall that injections require that |X| |Y | and equality for bijections.)
As an aside, notice that we can reformulate this problem combinatorically:
If |X| = k, define sets A1 , ..., Ak where y Aj iff (xj , y) E for j [k]. Then,
letting S = {A1 , ..., Ak }, we seek a subset T Y so that f : T S forms
a bijection. We require that t f (t). In other words, we want to pick one
8

representative from each set so that no single element is representing multiple


sets.
A maximal matching is a matching M where, for any edge e E/M ,
M {e} is not a matching. Let M be the collection of all matchings of G. Then
a maximum matching is a matching M such that
M arg max {|M |}M M .
That is to say, M has the most number of edges possible for a matching in
G. We can then define (G) to be the matching number of G. Clearly,
maximum matchings are maximal, but maximal matchings can be suboptimal.
This heavily depends on which edges we pick.
We need one more concept. A vertex cover C is a set of vertices such that
every edge has a vertex in C as an endpoint. Subsequently, we define (G) to
be size of a minimum vertex cover. Thus, we arrive at our first theorem.
Theorem 3. In a bipartite graph G, (G) = (G).
Let A V . We call R(A) the range of A, the set of all vertices adjacent to
a vertex in A.
Theorem 4 (Halls Marriage Theorem). A bipartite graph G has an X-matching
iff for any subset A X, |R(A)| |A|.
Now, consider a bipartite graph with weighted edges. These weights could
represent compatability between people, similarity measures, or some sort of
likelihood (we exclude rankings from consideration). What we require is that
the sum of the weights of edges incident to a vertex equals 1. This has two
side effects. The first is obvious: this ensures a probability distribution for each
vertex. The second is less obvious: we must the same number of vertices in
X and Y . Why? Suppose |X| = n but |Y | = n + 1, and consider all edges
incident to X (the entire set). Then, by assumption, the sum of weights for
edges incident to X is n and for Y its n + 1. However, since edges are from X
to Y , the two sums should be equal. Hence, we have a contradiction.
This gives us a lead-in to a useful concept:
Definition 14. Let A Mnn (R). Then A is doubly stochastic if aij 0
for all i, j [n] and row and column sums equal one.
A bipartite graph meeting our criteria has a corresponding doubly stochastic
matrix since we define aij to be the weight between xi and yj . Implicitly, we
assume that any two edges not connected will have entry zero. Conversely, any
doubly stochastic matrix defines a bipartite graph.
Lemma 2. For any doubly stochastic matrix, its corresponding graph has a
perfect matching.
Proof. Suppose the contrary. By Halls Theorem, there exists A X such
that |R(A)| < |A|. Since G came from a doubly stochastic matrix, we have
9

P
w = iA,jR(A) wij = |A|. Now consider R(R(A)). Clearly A R(R(A)),
but equality may not follow (since a vertex in R(A) may have a neighbor not
in A). Since the sum of weights of edges incident to R(A) is |R(A)|, we have
|R(A)| w. Thus,
w |R(A)| < |A| = w,
a contradiction.
There is an interesting parallel here (huehuehue). In linear algebra, for a
subset S V of an inner product space, we have an orthogonal complement
S . This is the subspace of all vectors orthogonal to a vector in S. It follows
that the orthogonal complement of S must contain S, that is S (S ) , but
equality is not guaranteed. (It works when V is finite-dimensional)
Theorem 5 (Birkhoffs Theorem). Every doubly stochastic matrix is a convex
combination of permutation matrices.

Differential Geometry

This section takes us through the basics of differential geometry. We follow


Loring Tus book Intro to Manifolds.

5.1

Euclidean Spaces

In order to generalize calculus on Rn to manifolds, we must remove our dependency on ambient global coordinates. This requires more sophisicated machinery
and a whole lot of new terms.
First, some notation. Well use superscripts for coordinate indices. For
example, we write coordinates in Rn as x1 , ..., xn and points as p = (p1 , ..., pn ).
We assume p U , an open set in Rn .
Definition 15. Let k 0 be an integer. A real-valued function f : U R is
a member of C k at the point p U if all partial derivatives
j f
xi1 ...xij
exist for j [k] and are continuous at p. If f is a member of C k at p for all k,
then f is a member of C at p.
Of course, f C 0 is simply a function continuous at p.
Definition 16. Let k 0 be an integer. A vector-valued function f : U Rm
is a member of Ck if all of its component functions f i C k at p for i [m]. If
f is a member of Ck at p for all k, then f is a member of C at p. If f Ck
at p for every p U , then f Ck on U (we can extend this similarly to C on
U ).
The class of C and C functions at p or on U are called smooth.
10

Definition 17. We call a subset S Rn star-shaped wrt p S (the center)


if for every x S, the line segment joining p and x remains in S.

11

Vous aimerez peut-être aussi