A PPLICATION OF E LLIPTIC C URVES TO F ERMAT ’ S L AST T HEOREM

Elise Alvarez-Salazar, Caroline Baldan, and Kelly Stump

2022 Mathematics Directed Reading Program, University of California - Santa Barbara

Abstract Important Theorems Example of Finding Rational Solutions

For this year’s Directed Reading Program, we studied elliptic curves and Mordell-Weil draws further conclusions about the previously created Let us consider the elliptic curve E : y 2 = x3 − x. Applying our
methods for finding all their rational solutions. The three theorems about abelian group structure, stated below: code to E , we see that (0, 0), (1, 0), (−1, 0) and the point at infinity
to be mentioned all tell us that the abelian group over E(Q) has a rich make up E(Q)torsion, where each non-identity element has order
group structure. Using this knowledge, we tackle the specific case of E(Q) is a finitely generated abelian group. In other words, there are
2. We find that the discriminant ∆E = 64. Thus, the only prime
n = 4 of Fermat’s Last Theorem. points P1, ..., Pn such that any other point Q ∈ E(Q) can be expressed
of bad reduction to consider is p = 2. We determine that 2 is of
as a linear combination
multiplicative bad reduction. Thus, by Thm. 2.6.4 in [1], we see
Preliminary Information Q = a1P1 + ... + anPn that
for some ai ∈ Z RE ≤ m + 2a − 1 = 0
Definition: An elliptic curve over Q is a smooth cubic projective curve E Thus, E(Q) ∼
= Z2 × Z2.
From this theorem, and facts we know concerning finitely generated
defined over Q, with at least one rational point O ∈ E(Q) that we call the
abelian groups, we find that: Fermat’s Last Theorem (n = 4)
origin.
We will focus on elliptic curves of Weierstrass Form: E(Q) ∼
= E(Q) torsion ⊕ ZRE

y 2 = x3 + Ax + B where A, B ∈ Z Continuing on, we will refer to RE as the rank. We can reach further con- Problem Statement: Let n = 4. Are there any solutions to
clusions about the group structure created over E(Q)torsion with Mazur’s an + bn = cn where a, b, c ∈ Z with abc ̸= 0?
theorem stated in [1] as Thm 2.4.2.
Defining P + Q Solution: We claim that there are no non-trivial solutions. We
Finding Rational Solutions are given the equation a4 + b4 = c4, when
2(b2+c2) 4b(b2+c2)
The operator for E(Q) shall be defined as follows: x = a2 and y = a3 are substituted in, we get the elliptic
curve E : y 2 = x3 − 4x.
For P, Q ∈ E(Q), where P ̸= Q, we find the secant line which intersects The natural continuation of the process of finding rational solutions for
both P and Q, Y : y = ax + b. Solving for the third point of intersection E is to next explore methods to calculate E(Q)torsion and ZRE . Applying our given algorithm to this elliptic curve, we find that
of Y with our curve E , labelled R, we see that P + Q is the reflection of E(Q)torsion = {(0, 0), (2, 0), (−2, 0), O }. Note that these torsion
Specifically for calculating E(Q)torsion, we have a theorem from Nagell- points correspond to trivial solutions of a4 + b4 = c4.
R over the x-axis.
Lutz:
For the free part, an attempt to bound the rank proves insufficient
Let E/Q be an elliptic curve with Weierstrass equation y 2 = x3 + Ax + B as the prime of bad reduction is additive. Thus, we move onto
where A, B ∈ Z Then, every torsion point P ̸= O of E satisfies: use of the algorithm in 2.9 of [1] which tells us that the rank is 0.
(1) The coordinates of P are integers, i.e. x(P ), y(P ) ∈ Z. Thus, E(Q) ∼ = E(Q)torsion.
(2) If P is a point of order n ≥ 3 then 4A3 + 27B 2 is divisible by y(P )2.
(3) If P is of order 2 then y(P ) = 0 and x(P )3 + Ax(P ) + B = 0.
Acknowledgements
We have come up with two methods from our readings for trying to cal- We would like to thank the DRP team for organizing this year’s
culate the rank. The first uses Theorem 2.6.4 in [1]. And the other program. We would also like to thank our DRP Mentor, Marcos
possible solution is found in section 2.9 of [1]. Reyes, for guiding us through our project. He was a wonderful
resource while reading through our elliptic curve texts and taught
Scan the following QR code to be taken to our algorithm that will find the
us well.
torsion points of assorted elliptic curves:
For the case where P = Q, we consider the tangent line rather than the References
secant and a similar procedure follows to find 2P . Note that every point
has an inverse and our identity is the point at infinity, O . Thus, we see
that for this defined + operator, we generate an abelian group on E(Q). [1] Á. Lozano-Robledo, Elliptic Curves, Modular Forms and their
L-functions, American Mathematical Soc., 2011
[2] J. Silverman, J. Tate, Rational Points on Elliptic Curves,
Springer-Verlag, New York, 2015
 Alexander Polynomial the Great
Alycia Doucette and Elizabeth Benda - Mentored by Melody Molander
2022 Mathematics Directed Reading Program. University of California - Santa Barbara

Introduction The Alexander Polynomial Other Polynomial Representations

What is a knot? Simply speaking, a knot is a The Alexander polynomial was a method invented in 1928 as a way to represent knots and links as polynomial The other polynomial representations we looked at
closed curve in space that does not intersect it- equations. It is an invariant for all representations of knots and links up to the same orientation. The Alexander were the Jones polynomial and the HOMFLY poly-
self in any way. Knots have many applications to polynomial is dependent on the orientation of the knot or link being assessed. The formula to compute the nomial. The Jones polynomial, V (t), is derived us-
1
other fields of science and are fun for mathemati- Alexander polynomial was refined by John Conway in 1969, and is now based on the following two rules: ing three rules, and the base variable t 2 . All prime
cians to study. One of the main questions posed (1) knots with 9 or fewer crossings have a distinct Jones
∆(⃝) = 1
when studying knots is how to tell whether or 1
polynomial. The HOMFLY polynomial, unlike the
− 21 (2)
not two different projections are the same knot. L+ L− L0 ∆(L+) − ∆(L−) + (t − t )∆(L0) = 0
2
other two, is multivariable. However, it does main-
A tool that has developed as a way to distinguish tain a similar structure to that of the Alexander
two knots from each other is representing knots as The main tool used to compute the Alexander polynomial is called the resolving tree. The resolving tree polynomial, using L+, L−, and L0. Knots under
polynomials. In this poster we will focus on one is an easy way to break a knot down into a series of unknots and trivial links. In order to create the resolving both the HOMFLY and Jones polynomials are not
of the three major polynomial representations of tree, you choose one crossing of the knot, and determine whether it is an L+, L−, or L0 crossing. From there, affected by orientation, however, when computing
knots, the Alexander polynomial. the chosen crossing is broken down into two new knots. These new knots are dependent on what type of the HOMFLY of a link, orientation between the two
crossing the original one is. links does affect the result.

Definitions Resolving Tree of the Figure-Eight Knot

Consider the rules of the HOMFLY polynomial:
• Projection: A two-dimensional picture P (⃝) = 1 (1)
representation of a knot. −1
αP (L+) − α P (L−) = zP (L0) (2)
• Orientation: A direction in which you travel
around the knot. The Alexander and Jones polynomials can be de-
• Crossing number: The least number of rived from the HOMFLY rules as follows:
− 21 1
crossings that occur in any projection of a ∆(t) = P (α = 1, z = t − t ) 2
1 1
particular knot. −1 −
V (t) = P (α = t , z = t 2 − t 2 )
• Link: A set of knotted loops tangled up together.
• Unknot: The unknot is also known as the Conclusion
trivial knot, and it looks as follows:
Each polynomial representation of knots has its own
benefits and drawbacks. While the HOMFLY poly-
nomial comes the closest to distinguishing between
all knots and links, there is not currently any poly-
nomial representation of knots that can completely
distinguish all knots and links. Knots are the best!
(a) Hi, I’m an oriented (b) Hi, I’m an oriented
unknot! trefoil! Alexander Polynomial of the Figure-Eight Knot
References and Acknowledgements
1
− 21
∆(L+) − ∆(L−) + (t − t )∆(L0) = 0
2
It was fascinating to read and learn about how knots,
∆(L+) = ∆(⃝) = 1 simple strings in space, can be transformed into differ-
1
− 1 1
− 1 ent polynomials. We would like to thank our graduate
∆(L0) = ∆(⃝ ∪ ⃝) − (t 2 − t 2 )∆(⃝) = −(t 2 − t 2 )
1
mentor Melody Molander, and the DRP, for creating this
− 12 1
− 21 1
− 12
(c) Hi, I’m a link! (d) Hi, I’m a crossing! ⇒ ∆(L−) = ∆(L+) + (t − t )∆(L0) = 1 + (t − t )(−t + t )
2 2 2 space for us to explore and grow our interests in mathematics.
−1
= 3−t−t
Adams, Colin. The Knot Book. American Mathematical So-
Since 3 − t − t−1 ̸= 1 we know that the figure-eight knot is not a projection of the unknot. ciety, 2004.
 I NTRODUCTION TO A LGEBRAIC N UMBER T HEORY
Robin Lee, Mr. Mulun Yin
University of California, Santa Barbara

√
Introduction Example
√ of Minimal Polynomials: If α = 2 , then f (x) = x 2 − 2 is the minimal polynomial Because dα is a root of the new equation, xn + dan−1xn−1 + · · · + dn−1a1x +
of 2, because all the coefficients in f (x) are ∈ Q, it is monic as the leading coefficient is dna0 = 0. This means that dai ∈ Z, because we multiply ai by its common
1, of the smallest degree (2), and α is a root. Similarly, the minimal polynomial of i is x2 + 1. denominator multiple and ai ∈ Q. Thus, all the coefficients are integers and dα
In this poster, we will overview the fundamentals of Algebraic Number Theory,
is an algebraic integer.
focusing on the basic definitions of rings and fields, algebraic numbers, and alge-
braic integers. Therefore, ∀α ∈ A, ∃d ∈ Z st dα ∈ I,i.e., dα is an algebraic integer.
Field of Algebraic Numbers
This means that every algebraic number α is an algebraic integer divided by an
Rings and Fields Utilizing our newfound knowledge of Algebraic Numbers and Minimal Polynomials, we can integer, which is analogous to a rational. Namely,
finally discuss the Field of Algebraic Numbers.
algebraic numbers = algebraic integers
integers
As the most fundamental concept of Algebraic Number Theory, rings and fields Let us define the set A of algebraic numbers. We actually know that set A is a field, but this
are algebraic structures that contain two binary operations (addition and multipli- will be proven later using field extension. Because it is a field, we can infer that it has the We define Z[α] as the smallest ring containing Z and α, which is analogous to
cation) with properties similar to those for integers Z. In [1], we can define a ring same properties as the ones we have mentioned in the "Rings and Fields" section. As such, Q(α). Similar to the lemma in the field extension, α is an algebraic integer if and
as a non-empty set R with addition and multiplication. Assuming R is a ring, we if α and β are algebraic numbers, then so are the following: only if Z[α] is a finitely generated Z module. Now we are trying to prove that this
mean it has the following characteristics: α + β , α − β , αβ , α
β where β ̸= 0. a ring using this lemma.
• a set closed under addition a + b ∈ R and multiplication ab ∈ R
This
√ is√important, because for example, assume we want to find the minimal polynomial of Here R is a finitely generated Z module means every element in R can be writ-
• commutative under addition a + b = b + a 2 + 3 to check using definition whether it’s algebraic. This may be difficult to compute ten uniquely as a linear combination of fixed n elements. For example, every
at
√ first√glance, but using our knowledge of the field of algebraic numbers, we already know element in Z[i] is in the form a + bi where a, b ∈ Z. We will be able to show every
• associative under addition a + (b + c) = (a + b) + c and multiplication a(bc) = 2 + 3 is an algebraic number.
(ab)c √ Even√though we did not find the minimal polynomial, we element in Z[α] is the root of a monic polynomial with coefficients ∈ Z, i.e., they
know this is algebraic, as both 2 and 3 are algebraic. are all algebraic integers.
• contains the additive identity a + 0 = a, ∀a ∈ R,for some 0 ∈ R
Analogous to the lemma in Field Extension, because α and β are algebraic
• contains additive inverses: ∀a ∈ R, ∃s ∈ R such that a + s = 0
Field Extension integers, we know that Z[α] is finitely generated and Z[β] is finitely generated.
• contains the multiplicative identity 1 ∗ a = a ∗ 1 = a, ∀a ∈ R, for some 1 ∈ R Thus, Z[α, β] is finitely generated. Our previous proofs suggest that α + β ∈
Z[α, β], which means Z[α + β] ⊆ Z[α, β] and Z[α + β] is finitely generated.
√ In our case, a field extension of Q can be defined as Q(α), denoted by Q(α)/Q, where Q(α)
Example of Rings: Z, Q, R, C, Z[ 2], Z[i] Therefore, α + β must be algebraic.
is the smallest field containing Q and α (an algebraic number); there are a few examples in We can utilize this proof with α − β , αβ , because they are all in Z[α, β]
√ √ the Rings and Fields section. An element of Q(α) is a polynomial with "variable" α (though
An element of the ring Z[ 2] is a + b 2 where a, b ∈ Z.
An element of the ring Z[i] is a + bi where a, b ∈ Z. α is fixed), with coefficients in Q. Therefore, the set of all algebraic integers forms a ring.
An example
√ of the
√ ring’s addition
√ and multiplication properties is: √ √
(1 + √2) + (2 +√ 2) = 3 + 2√ 2, and Note that we are able to combine two elements in Q(α) as they are both polynomials and Again, is 2 + 3 an algebraic integer? We know the answer is yes, despite the
√ √ follow the usual rules for scalar multiplication and addition for polynomials. As such, Q(α) is
(1 + 2) ∗ (2 + 2) = 2 + 2 2 + 2+2=4+3 2 fact that we didn’t even compute it’s minimal polynomial!
a vector space over Q. Furthermore, the degree of the√field extension is defined to be the
Similar to rings, fields not only contain the same properties as a ring, but also dimension of the√ Q vector space Q(α). Referring to Q( 2), the dimension is 2, because we
have a basis {1, 2} that consists of 2 elements.
contain multiplicative inverses (in addition to additive inverses) and is commuta- Integers in Number Fields
tive under multiplication. In other words, a field F is a unique configuration of a
commutative ring that contains at least two elements such that every non-zero There exists a lemma that states α is algebraic if and only if the field extension Q(α)/Q has
element in F is both commutative under addition and multiplication. Furthermore, a finite degree. As a consequence, let us look at the integers in a number field K , which is by
a field contains a multiplicative inverse. definition, K ∩ I (namely, algebraic integers that are in K ):
Using the aforementioned lemma, because α and β are algebraic, we know that [Q(α) : Q]
√ and [Q(β) : Q] are both finite. Thus, [Q(α, β) : Q] must also be finite. As we can infer that
Example of Fields: Zn where n is a prime and positive integer, Q, Q( 2), Q(i) If we take α, β ∈ K ∩ I (integers in K , as we just defined), then we can prove
α + β ∈ Q(α, β), this implies Q(α + β) ⊆ Q(α, β) and [Q(α + β) : Q] is finite. We know from that their sum α + β ∈ K ∩ I .
√ √ the aforementioned lemma that α is algebraic if and only if the field extension Q(α)/Q has
An element of the field Q( 2) is a + b 2 where a, b ∈ Q.
An element of the field Q(i) is a + bi where a, b ∈ Q. a finite degree. Therefore, α + β must be algebraic. Note that in order to show algebraic Because they are in the intersections of K and I , α, β ∈ K and α, β ∈ I .
numbers make a field, we just need to show that they are closed under the operations, since Furthermore, since I is a ring as proven above, α + β ∈ I . Similarly, K is a field
those axioms(say, associativity) are all inherited from C. (closed under addition), so α + β ∈ K .
We can utilize this proof with α − β , αβ , α
β (where β ̸= 0), because they are all ∈ Q(α, β).
Algebraic Numbers and Minimal Polynomials Hence, algebraic numbers form a field. Therefore, α + β ∈ K ∩ I . This is similar to α − β and α ∗ β , as both are in K ∩ I .
√ √ In conclusion, K ∩ I forms a ring.
Diving deeper into our understanding of fields and rings, it is imperative we first Example of Field√Extension: The field extension Q( 2) = {a + b 2|a, b ∈ Q} and the √
overview an essential element to utilizing Algebraic Number Theory: Algebraic degree is 2, and 2 is algebraic. We can test that u + v and uv are still in the form a + b 2 Example: √ √
Numbers. According to [2], we can say that a complex number α is algebraic if where a, b ∈ Q when u, v are. The integers in Q( 2) is Z[ 2]
it is the root of a polynomial with specifically integer coefficients, and transcen- √ √ √ The integers in Q(i) is Z[i]
• (a1 +1 b 2) + (a2 + b2 2) = (a1 + a2) + (b1 + b2) 2
dental if it is not. Furthermore, in the following proof, we can conclusively prove √ √ √
that there are only a countably large amount of Algebraic Numbers. • (a1 + b1 2)(a2 + b2 2) = (a1a2 + 2b1b2) + (a1b2 + a2b1) 2
Given any polynomial with integer coefficients: Further Applications
p(X) = C0X d + C1X d−1 + · · · + Cd = 0. We define a number field K as an extension of Q of finite degree.
With all these definitions, we could study number theory, say the theory of prime
with Ci ∈ Z and C0 ̸= 0, we can define the "height" H(p) as:
numbers, in a much broader context. Some familiar results about Z are still true
H(p) = d + |C0| + · · · + |Cd| ∈ Z Integrality and the Ring of All Algebraic Integers in this new setting, but some are not (as an example, unique factorization of an
integer into primes fail in general). These discoveries lead us to the modern
Such that given any n ∈ Z, there are only finitely many such polynomials whose
algebraic number theory...
heights are ≤ n. So, every polynomial with integer coefficients (which corre- The ring of all algebraic integers I can be defined as an algebraic number α where the And yes, we are also able to show that the integers in a number field are always
sponds to an algebraic number) can thus be controlled by an integer, but Z is minimal polynomial of α over Q has coefficients in Z. Thus, it is a subset of algebraic finitely generated—just as all the existing examples suggest.
countably infinite—proving that Transcendental Numbers not only exist, but are numbers, and in the following sections, we will prove that it forms a ring.
also more prevalent than their Algebraic counterparts as C is uncountable.
Note: From the aforementioned properties, we can conclude that every rational First and foremost, suppose α is a root of xn + an−1xn−1 + · · · + a1x + a0 = 0 where ai ∈ Q.
m , where m, n ∈ Z, is algebraic, since it is always a root of nX − m = 0 References
n Then, we have d = common multiple of denominators of ai, then dn(αn + an−1αn−1 + · · · +
a1α + a0) = 0. Thus:
With our definition of Algebraic Numbers established, we are able to quickly per- [1] David R. Finston and Patrick J. Morandi. “Abstract Algebra: Structure and Application”. In:
ceive the definition of the Minimal Polynomial of an algebraic number α. The dn(αn + an−1αn−1 + · · · + a1α + a0) = 0 (2010).
minimal polynomial of α is a (unique) polynomial that consist of the following at- => (dα)n + dαn−1(dα)n−1 + · · · + dn−1a1(dα) + dna0 = 0 [2] Frazer Jarvis. “Algebraic Number Theory”. In: (2010).
tributes: (1) coefficients are in Q, (2) leading coefficient is 1 (monic), (3) smallest
possible degree, and (4) α is a root.
 E LLIPTIC C URVE C RYPTOGRAPHY
Rocky Beaty
Mentor: Mychelle Parker
University of California, Santa Barbara | 2022 Directed Reading Program

What is an Elliptic Curve? Group Law Example

Definition 1 An elliptic curve over a field K is defined by an equation There is a convenient way of defining an addition operation for two points in E(K) to give a Note: It is possible to turn the geometric interpretation of point addition and point
third point in E(K). With this operation, the set of points in E(K) forms an abelian group, doubling into algebraic formulas by solving the cubic equations.
E : y 2 = x3 + ax + b (1)
Let K = F97 and take
where ∞ serves as the identity. The addition operation has a clear geometric interpretation.
where a, b ∈ K and ∆ ̸= 0 where ∆ is the discriminant of E and is defined as First, notice that any line will intersect an elliptic curve E at most 3 times. Given any two E : y 2 = x3 + 2x + 3.
distinct points, P = (x1, y1) and Q = (x2, y2), on E , then P + Q = R = (x3, y3) is found by Consider P = (3, 6), one can calculate the multiples of P using the mentioned
∆ = −16(4a3 + 27b2). drawing a line through P and Q, find the third point this line intersects E . Then to obtain R algebraic formulas to obtain:
reflect this point about the x-axis. Doubling a point P is the same, though the tangent line
at the point P is used. Note: P − Q is performed by taking −Q = (x2, −y2) ∈ E(K). 0P = ∞ 1P = (3, 6) 2P = (80, 10) 3P = (80, 87) 4P = (3, 91)
Note: There is a more general form of the equation: 5P = ∞ 6P = (3, 6) 7P = (80, 10) 8P = (80, 87) 9P = (3, 91)
This pattern continues, so we see that 5P = ∞ =⇒ P is a generator of order
y 2 + a1xy + a3y = x3 + a2x2 + a4x + a6.
n = 5, and forms the cyclic subgroup
However, if the characteristic of K ̸= 2 or 3, then the equation can be expressed
⟨P ⟩ = {∞, P, 2P, 3P, 4P }.
as in (1). This assumption applies to all elliptic curves used in cryptography, and
thus equation (1) is sufficient for us. Now, consider the following problem. Let

Definition 2 Let K be a field over which an elliptic curve is defined. Then the P = (3, 6), d = 3, Q = dP = 3P = (80, 87), k = 9
K-rational points, denoted E(K), are all points on E with coordinates in K , along
and suppose the encoded message is M = (24, 2). Using the algebraic formu-
with the point at infinity denoted ∞. The order of the curve, #E(K), is the total
las, one can calculate C1 and C2,
number of points on the curve.
C1 = kP = 9(3, 6) = (3, 91)
Elliptic curves can be defined over infinite fields such as R or Q, or they can be
Fig. 3: Point Addition and Point Doubling. [1] C2 = M + kQ = (24, 2) + 9(80, 87) = (24, 2) + (80, 10) = (92, 16).
defined over finite fields such as Z/pZ or Fq . Consider the following graphs of
various elliptic curves: From this abelian group comes the basis for the scheme of elliptic curve cryptography. The recipient receives C1 and C2, and then computes
dC1 = d(kP ) = k(dP ) = kQ = (80, 10)
What is Elliptic Curve Cryptography?
so
M = C2 − kQ = (92, 16) − (80, 10).
Elliptic Curve Cryptography (ECC) is a modern public-key cryptography technique based on
Notice that −kQ = −(80, 10) = (80, −10) where −10 ≡ 87 (mod 97) hence
the mathematics of elliptic curves over finite fields. ECC relies on the difficulty of solving the
−kQ = (80, 87). So we get
Elliptic Curve Discrete Logarithm Problem.
Definition 3 The elliptic curve discrete logarithm problem (ECDLP) is: given an elliptic curve
M = (92, 16) + (80, 87) = (24, 2)
E defined over a finite field Fq , a point P ∈ E(Fq ) of order n, and a point Q ∈ ⟨P ⟩, find the as desired. An attacker wishing to recover M would likely know
integer l ∈ [0, n − 1] such that Q = lP . The integer l is called the discrete logarithm of Q to E(F97), P, n, Q, C1 and C2. However, even with this information it is computa-
Fig. 1: Elliptic Curves over R. [1]
the base P , denoted l = logP Q. tionally infeasible to compute kQ due to the cyclic nature of ⟨P ⟩ and assuming k
Simply put, the ECDLP is the problem of finding an integer n such that Q = nP . It exploits is sufficiently random.
the fact that, as shown above, it is rather easy to double a point P ∈ E(K) together, but
it is thought to be very difficult to figure out how many times the point was doubled. The Why ECC?
essential pieces of a secure ECC scheme are:
1. Elliptic Curve E(Fp) over finite field Fp, p prime ECC is often preferred over RSA schemes because of the security and perfor-
2. P : generator - P ∈ E(Fp) is a generator mance it offers using smaller key sizes. A common ECC key size of 256-bits is
3. d : private key - d ∈ Z is selected uniformly at random from the interval [1,n-1] equivalent to a 3072-bit RSA key.
4. Q : public key - a point Q = dP ∈ E(Fp)
5. k : random integer - used to increase security of encryption scheme References
In the ECC scheme, a sender’s message is represented as a point M , and encrypted by
adding it to kQ, where Q = dP is the intended recipient’s public key. The sender transmits [1] Elliptic Curves over Finite Fields. https://graui.de/code/elliptic2/. Accessed: 2022-
the points C1 = kP, C2 = M + kQ to the recipient who uses their private key d to compute 05-02.
[2] Darrel Hankerson, Alfred Menezes, and Scott Vanstone. Guide to Elliptic Curve Cryptogra-
Fig. 2: Elliptic Curve over finite field F29 . [2] dC1 = d(kP ) = k(dP ) = kQ phy. Spring, 2004.

and can then easily recover M = C2 − kQ. An attacker would have to find kQ, which is
computationally infeasible using the public information.
 A LGEBRAIC N UMBER T HEORY AND A PPLICATIONS
Yanbo Cheng, Mychelle Parker
UC Santa Barbara

Motivations Dedekind domain The Unit theorem

It is known that a prime p can be written in the form p = x2 +y 2 with x, y ∈ Z if and In this section, we denote OK to be the ring of integers of an algebraic number field K . From Minkowski Theory, we derive the Dirichlet’s Unit Theorem by studying the
only if p ≡ 1 mod 4. Since we can factorize such p in Z[i] as p = (x + iy)(x − iy), Such a ring has the following main properties: exact sequence:
∗ →Γ→0
1 → µ(K) → OK
it is natural to think of the prime elements in Z[i]. We then want to relate the field Theorem 7. OK is a neotherian ring. It is integrally closed and every nontrivial prime ideal
Q(i) to Z[i], and a proposition was found that illustrates such relationship. Where OK ∗ is the group of units and µ(K) is the roots of unity that lie in K , and
of OK is a maximal ideal.
Proposition 1. Γ is the image λ(OK ∗ ) defined by
Theorem 8. Every ideal of a of a Dedekind domain O that is nonzero and not the a ̸= O
Z[i] = {x ∈ Q(i) : x2 + ax + b = 0 for some a, b ∈ Z} R]+
Y
admits a factorization in to nonzero prime ideal of O: λ(a) = (log|τ (a)|)τ ∈ [
This proposition can be seen as a motivation to study the properties of algebraic τ
a = p1 · · · pn
integers of an algebraic number field. Where τ run over the complex embeddings τ : K → C.
This factorization is unique up to reordering. ∗ of O is the direct product of the finite
Theorem 12. The group of units OK K
Introduction We see that this is similar to a unique factorization domain in which every element admits a group µ(K), which is the group of roots of unity are in K , and a free abelian
factorization into a product of a unit and irreducible elements which is unique up to associ- group of rank r + s − 1. Where r is the number of real embeddings σ : K → R
We first establish some basic principles of algebraic number theory. ation and reordering. and s is the number of pairs of complex conjugate embeddings σ, σ : K → C.
Then we can thus look at the properties of the extensions of Dedekind domains.
Definition 2. An algebraic number field K is a finite extension of Q. A element Let o be a Dedekind domain with field of fraction K , let L|K be a field extension with integral This theorem give us a way to express any units u in OK uniquely in the form
α ∈ K is called an algebraic integer if f (α) = 0 for some monic polynomial closure O. Then we can decompose prime ideals of o in O i i i r+s−1
u = ξu11 u22 · · · ur+s−1
f (x) ∈ Z[x].
Theorem 9. Let p be a prime ideal of o, then where ξ is a root of unity and u1, u2 · · · are units of OK that can be seen as a
Definition 3. Let A ⊆ B be a ring extension. Then, b ∈ B is integral over A if e
f (b) = 0 for some monic polynomial f (x) ∈ A[x]. We then define the integral pO = P11 · · · Penn basis of the free abelian group mentioned above.
closure to be the set A = {b ∈ B : b integral over A}. A is then called integrally with fi = [O/Pi : o/p], we have the fundamental identity:
closed if A = A. Applications
n
X
As in linear algebra, traces and norms play an important role in algebraic number fiei = [L : K]
theory. We thus give their definition. i=1 One of the applications of the Dirichlet’s Unit Theorem is the solution of Pell’s
equations.
Definition 4. For a finite field extension L|K . The trace of an element α ∈ L
is the trace of the endomorphism ψ : L → L, ψ(x) = αx where L is seen as a P-adic numbers Corollary 13. There exists infinily many pairs of solutions x, y ∈ Z to the equa-
K -vector space. The norm pf α is then the determinant of ψ , that is: tion
x2 + ny 2 = 1
T rL|K (α) = T r(ψ), NL|K (α) = det(ψ) Now we introduce another topic, which are the p-adic numbers. We give two definitions of
with n < 0 not a perfect square and n ∈ Z.
the p-adic integers Zp.
There is an extra property in of traces and norms in a separable extension L|K
that uses field embeddings from L into an algebraic closure K of K . This is a direct application of√the Dirichlet’s Unit Theorem on the quadratic ex-
tension K|Q, where K = Q( −n), and we use the fact that r = 2, s = 0, thus
Definition 10. Zp can be defined as the projective limit of if the rings Z/pnZ, and thus
Proposition 5. Let L|K be a separable extension, and define the set Σ = {σ : r + s − 1 = 1.
L → K a field embedding }. Then we have: ∞ An application of the p-adic numbers is the following proposition:
Zp = lim Z/pnZ = {(xn)n ∈ Z/pnZ : xn+1 ≡ xn mod pn}
Y
n←
X
T rL|K (α) = σ(α) n=1 Proposition 14. Let f (x1, · · · , xn) be a polynomial with coefficients in integer.
σ∈Σ Then we have the equivalence:
We could also define Zp through Cauchy sequences.
Define the p-adic absolute value ||p as follows: f (x1, · · · , xn) ≡ 0 mod pn is solvable for all n ≥ 1
Y
NL|K (α) = σ(α) ′
σ∈Σ Let a = cb , b, c ∈ Z, we can find some integer n such that a = p c′ where (b′c′, p) = 1. Then
n b ⇐⇒ f (x1, · · · , xn) = 0 is solvable in p-adic integers
we have |a|p = 1 . We can thus define a metric using || just like what we did using the
We then give the definition of a Dedekind domain, which is the main object that pn p Thus, the application of p-adic number also gives a way to solve problems in
algebraic number theory studies. normal absolute value ||. Thus, we can define the p-adic numbers using Cauchy sequence elementary number theory.
with respect to the metric ||p. The induced metric on Zp is d(x, y) = |x − y|p for x, y ∈ Zp.
Definition 6. A Dedekind domain is a neotherian, integrally closed integral do-
main in which every nonzero prime ideal is maximal. Acknowledgements
Definition 11. Let R be the ring of Cauchy Sequence with respect to ||p, and m be the ideal
The product and sum of ideals defined such that of nullsequence, that is, the Cauchy sequences that converges to zero. Then we define the
p-adic numbers Qp as This is a poster of the Directed Reading Program in 2022. I would like to thank
a + b = {a + b : a ∈ a, b ∈ b}
X Qp = R/m Mychelle Parker for being my mentor in this program.
ab = { aibi : ai ∈ a, bi ∈ b, ∀i ∈ I}
Then, define the p-adic integers as
i∈I References
The importance of Dedekind domain is due to the fact that it gives unique prime Zp = {x ∈ Qp : |x|p ≤ 1}
factorization of prime ideals.
Jürgen Neukirch, Algebraic Number Theory
 Brouwer’s Fixed Point Theorem with
Application to Game Theory
Ruizhe Qian
Mentor: Pranav Arrepu

Introduction Lemma. ψ∗ is a homomorphism of groups. randomizing over the set of available actions according to
Proof. ψ∗([f ][g]) = ψ∗([f ∗ g] = [ψ(f ∗ g)] = [ψf ∗ ψg] = some probability distribution. Such strategy is called mixed
We will prove Brouwer’s Fixed Point Theorem by using strategy. We can define mixed strategy as follows. In pris-
[ψf ][ψg] = ψ∗[f ]ψ∗[g].
fundamental groups. Then, we will show the application of oner’s dilemma, mixed strategy can be like one person has
Brouwer’s Fixed Point Theorem to the game theory, namely By proving this lemma, we can give ψ∗ a name.
50% chance confessing 50% chance not confessing.
the Nash Equilibrium. Definition (Induced Homomorphism). The homomorphism
ψ∗ : π(X, x) → π(Y, ψ(x)) defined by ψ∗[f ] = [ψf ], where Definition (Mixed Strategy). Let (N, (A1, . . .Q
, An), O, µ, u) be
Brouwer’s Fixed Point Theorem in R ψ : X → Y is a continuous map, is called the induced ho- a normal form game, and for any set X let (X) be the set
momorphism. of all probability distributions over
Q X. Then, the set of mixed
Theorem (Brouwer’s Fixed Point Theorem). Given that set strategies for player i is Si = (Ai).
K ⊂ Rn is compact and convex, and that function f : K → What if we have ψ as a homeomorphism?
Definition (Mixed Strategy Profile). The set of mixed strat-
K is continuous, then there exists c ∈ K such that f (c) = c. Corollary. If ψ : X → Y is a homeomorphism then ψ∗ :
egy profiles is simply the Cartesian product of the individual
This is generalized statement of Brouwer’s Fixed Point π(X, x) → π(Y, ψ(x)) is an isomorphism.
mixed strategy sets, S1 × · · · × Sn.
Theorem in R. In this poster, we will explore the proof of a The last piece of the puzzle is the fundamental group of
simple case, D2 ⊂ R2. D2 is homeomorphic to any closed the circle S 1, which turns out to be Z. Let’s consider a map By si(ai) we denote the probability that an action ai will
and bounded compact subset of R2. But we will use the e be played under mixed strategy si. The subset of actions
generalized version of this theorem to prove the existence that are assigned positive probability by the mixed strategy
R → S1 si is called the support of si.
of Nash equilibrium.
t → e2πit Definition (Support). The support of a mixed strategy si for
Algebraic Topology Preliminaries a player i is the set of pure strategies {ai|si(ai) > 0}.
Note that e−1(1) = Z ⊂ R. If we are given f : I → S with
We establish our theory from homotopy, an important f (0) = f (1) = 1, there is a unique map f˜ : I → R with Now, we want to introduce the concept of expected utility,
equivalence relationship in topology, f˜(0) = 0 and ef˜ = f . f˜ is the lifting map of f . The integer a basic notion in decision theory,
Definition (Homotopy). [3] Two continuous maps f0, f1 : f˜(1) ∈ e−1(1) = Z is defined to be degree of f . If f0 and f1 Definition (Expected Utility of a Mixed Strategy). Given a
X → Y are said to be homotopic if there is a continu- are equivalent paths in S 1, then f˜0(1) = f˜1(1). As a result, normal form game (N, A, u), the expected utility ui for player
ous map F : X × I → Y such that F (x, 0) = f0(x) and the function π(S 1, 1) → Z, where [f ] 7→ degree(f ), is isomor- i of the mixed strategy profile s = (s1, . . . , sn) is defined as
F (x, 1) = f1(x). Then, we say f1(x) ≃ f0(x) phism, which means the fundamental group of the circle is
the set of integers. n
Definition (Homotopic Relative). Suppose that A is a sub- X Y
set of X and that f0 and f1 are two continuous functions ui(s) = ui(a) sj (aj )
from X to Y . We say f0 and f1 are homotopic relative to A Algebraic Proof for the Main Theorem a∈A j=1
if there is a homotopy F : X × I → Y between f0 and f1
such that F(a, t) does not depend on t for a ∈ A. Proof. Suppose to the contrary that x ̸= f (x) for all x ∈ D2. Then, we want to look at games from an individual
Then, we may define a function ψ : D2 → S 1 by setting ψ(x) agent’s point of view. This is going to lead us to the
Homotopy type, also known as homotopy equivalence,
to be the point on S 1 obtained from the intersection of the most influential concept in game theory, the Nash Equi-
following from homotopy, is an important tool to classify
line segment from f (x) to x extended to meet S 1. We want librium. Assume an agent knew how others were go-
topological space. ing to play, his strategy becomes simple. Define s−i =
to show ψ is continuous. Let’s write ψ explicitely in coordi-
Definition (Homotopy Equivalence). Two spaces X and Y nates, y = ψ(x). The condition the ray meets the boundary (s1, . . . , si−1, si+1, . . . , sn), a strategy profile s without i’s
are homotopic equivalent if there exists continuous maps is strategy. We can write s = (si, s−i). If the agents other
f : X → Y and g : Y → X such that |y + t(x − y)|2 = 1. than i were commit to play s−i, a utility-maximizing agent I
g ◦ f ≃ Id : X → X would face the problem of determining his best response.
It is a quadratic equation with the solution in
f ◦ g ≃ Id : Y → Y p Definition (Best Response). Player i’s best response to the
−2(x − y)y ± 4((x − y)y)2 − 4|x − y|2(|y|2 − 1) strategy profile s−i is a mixed strategy s∗i ∈ Si such that
The maps f and g are then called homotopy equivalences. t± =
2|x − y|2 ui(s∗i , s−1) ≥ ui(si, s−i) for all strategies si ∈ Si. Obviously,
We only interested in the solution y + t+(x − y). Therefore, in prison’s dilemma, confess is the best response.
ϕ is continuous. Define i : S 1 → D2, denote the inclu- Finally, we will go to the most important definition, the
sion, then ψi = Id and we have the commutative diagram. Nah equilibrium. Its existence is one of the most well known
S1
Id
S1 application of Brouwer’s Fixed Point Theorem.
i ψ
Definition (Nash Equilibrium). A strategy profile s =
(s1, . . . , sn) is a Nash equilibrium if for all agents i, si is a
D2 best response to s−i
This leads to another commutative diagram, When both prisoners confess, the game attain the Nash
Id
π(S 1, 1) π(S 1, 1) equilibrium.
i∗
ψ∗
Existence of Nash Equilibrium
π(D2, 1)
Spaces that are homotopy equivalent to a point are given where ψ∗ and i∗ denote induced homomorphism. But Theorem (Nash 1951). Every game with a finite number of
a special name. The identity function of this space is homo- π(D2, 1) = 0 since D2 is contractible, and so we get an- players and action profiles has at least one Nash equilib-
topic to the constant map. other commutative diagram due to isomorphism. rium
Definition (Contractible). A space X is said to be con- Z
Id
Z Proof. Given a strategy profile s ∈ S, for all i ∈ N and
tractible if it is homotopy equivalent to a point. i∗ ai ∈ Ai we define
ψ∗
By using constant map and inclusion map, the following
result can be easily derived. 0 ψi,ai (s) = max{0, ui(ai.s−i) − ui(s)}
Remark. Dn is contractible and any convex subset of Rn is This is impossible. Therefore, we prove the Brouwer’s Fixed
contractible. Point Theorem in two dimension. a function denoting the change of utility after each itera-
Consider that the cylinder, C and the circle S are a pair tion of strategy. We then define the function f : S → S by
of homotopy equivalent spaces. Define i : S → C as the Game Theory Preliminaries f (s) = s′, where
inclusion. This motivates the following definition. Now we move on to application to the game theory. ′ si(ai + ψi,ai (s))
Definition (Retraction). A subset A of a topological space si(ai) = P
We want to introduce the abstract notion of a normal form bi∈A si(bi) + ψi,bi (s)
X is called a retract of X if there is a continuous map game, the following definitions are from [2]. We will use si(ai + ψi,ai (s))
r : X → A such that r ◦ i = Id : A → A, where i : A → X prisoner’s dilemma to illustrate those definitions. In pris- =P
is the inclusion map. The map r is called a retraction. oner’s dilemma, two prisoners are interrogated separately. bi∈A ψi,bi (s) + 1
Before we step into the definition of the fundamental If both of them confess, they get sentence for 3 years. If one Intuitively, this function maps a strategy profile s to a new
group, we want to give the definition of some related con- confess, the other does not, the people who confess gets 1 strategy profile s′ in which each agent’s actions that are bet-
cepts. years of sentence, the other gets 10 year. If both of them do ter responses to s receive increased probability mass.
Definition (Path). A continuous mapping f : [0, 1] → X is not confess, they are innocent. The function f is continuous since each ψi,ai is continuous.
called a path in X. Definition (Normal-form game). A (finite, n-person) normal- Since S is convex and compact and f : S → S, by Brouwer’s
Definition (Path Equivalent). Two path f, g in X are said to form game is a tuple (N, A, O, µ, u), where Fixed Point Theorem, f must have at least one fixed point.
be equivalent if f and g are homotopic relative to {0, 1}. We • N is a finite set of n players, indexed by i. First if s is a Nash equilibrium then all ψ’s are 0, making s a
write f ∼ g fixed point of f .
• A = (A1, . . . , An), where Ai is a finite set of actions (or
Definition (Loop). A path is said to be closed if f (0) = f (1). pure strategies; we will use the terms interchangeably) Conversely, consider an arbitrary fixed point of f, s. If s is
If f (0) = f (1) = x then we say that f is based at x. available to player i. Each vector a = (a1, . . . , an) ∈ A is a fixed point, then s′(a′i) = s(a′i). It follows that ψi,bi (s) = 0,
Now, we have the definition of the fundamental group. called an action profile (or pure strategy profile); which only happens when no player can enhance their util-
Definition (Fundamental Group). [1] The fundamental ity. Therefore, s′ = f (s) is the Nash equilibrium
• O is a set of outcomes;
group of a space X will be defined so that its elements are
• µ : A → O determines the outcomes as a function of the
loops in X starting and ending at a fixed basepoint x ∈ X
but two such loops are regarded as determining the same
action profile; and References
element of the fundamental group if one loop is homotopy • u = (u1, . . . , un) where ui : O → R is a real valued utility
equivalent to the other in space X. We denote this group function for player i [1] Allen Hatcher. Algebraic Topology. 2001.
as π(X, x). In prisoner’s dilemma, N = 2. A is confess or not con- [2] Alber Xin Jiang and Kevin Leyton-Brown. “A Tutorial
We will explore the effect of continuous map between fess. O is what happens if both of prisoners confess, not on the Proof of the Existence of Nash Equilibrium”. In:
topological spaces ψ : X → Y has upon fundamental confess and etc. In this way, µ and u is easy to understand. (2007).
groups. Consider ψ∗ : π(X, x) → π(Y, ψ(x)) where ψ∗[f ] = While players can select a single action to play, which is the [3] Czes Kosniowski. A First Course in Algebraic Topology.
[ψf ], f is a path in X. pure strategy, they can also follow another type of strategy: Cambridge University Press, 1980.
 A Bite-sized Introduction to Fluid Mechanics
Carlos Ortiz, advised by Pranav Arrepu
2022 Math Directed Reading Program, UC Santa Barbara

Introduction The Navier-Stokes Equations of Motion for Viscous Fluids

Consider the deriva ve in the first term on (7):
The focus of fluid mechanics is the measurement of observables related to a fluid. Liquids and gases are D J1i D ∂x1
 
∂ D x1 ∂u1 ∂u1 ∂xl ∂u1 The Euler equa ons assume stresses incident only normal to the surface of a fluid. Real fluids, however,
examples of fluids, and their observables include temperature, pressure and density, to name a few. To = = = = = Jli. (8) are hardly as ideal. We can correct the equa ons of mo on by modifying the stress tensor σij to contain
Dt D t ∂ai ∂ai D t ∂ai ∂xl ∂ai ∂xl
approach this focus through first principles, the descrip on of fluids is idealised by the no on of a con nuum, addi onal stresses unrelated to pressure. By Cauchy’s Theorem (see [2]), one can prove that these non-
which neglects the microscopic structure of fluids as separate molecules. ”Infinitesimal” volume-elements of We interchange the deriva ves since the ini al posi on a is me-independent. The first term of (7) now pressure forces, represented by the deviatoric tensor, act linearly on the normal vector. Then we can split up
the fluid (called fluid parcels) are then understood to be large enough to contain many molecules, but small expands completely as the stress tensor into a sum of two terms: (i) the stresses normal to the surfaces of the fluid given by the
D J1i ∂u1
rela ve to the varia on in the length-scale of the fluid proper es. In this way, observables are understood to ijk J2j J3k = ijk JliJ2j J3k . (9) pressure; and (ii) the stresses ac ng at arbitrary direc ons along the surface of the fluid. Mathema cally,
Dt ∂xl
be averaged values over a fluid parcel. With this discussion, we are ready to form a mathema cal approach σij = −pδij + dij , where dij denotes the deviatoric tensor. The implementa on of this stress tensor to obtain
to fluid theory. For l 6= 1, ijk = 0 by defini on since then i = j or i = k. An analogous approach can be made for the the equa ons of mo on is analogous to the method in obtaining the Euler equa ons. But before that, we
other two terms of (7), admi ng one final term by virtue of index repe on. Thus, reducing (7) gives need to first obtain the form of the deviatoric tensor.
To ”Think” Eulerian or Lagrangian DJ ∂ul
 
= ijk J1iJ2j J3k = J(∇ · u). (10) By a physical argument, we find that the deviatoric,
Dt ∂xl and hence the stress tensor, is symmetric. From Fig-
There are two natural methods for studying fluid proper es. In the Lagrangian approach, we follow a fluid ure 1, one can see that the torque about the z-axis of
With this result, we can observe how a fluid property q(X(a, t), t) changes with me over a fluid parcel:
parcel as it moves and measure the observables along the mo on of the parcel. Suppose, at some ini al a cube centered at the origin is α3(σ21 − σ12), where
Dq DJ
Z  
D D
Z Z
me t0, a fluid occupies an open set S0 of Rn. We could then label a fluid par cle on the fluid (say with some q(X(a, t), t) d3x = qJ d3a = J +q d3a (11) α is the side-length of the cube. From elementary
dye) at the posi on a ∈ S0, and follow the par cle over me. At some later me t, the fluid occupies the D t St D t S0 S0 D t Dt
physics, we know that the moment of iner a of such
set St, and the par cle’s posi on is given by X(a, t) ∈ St. The Lagrangian coordinate X(a, t) depends on the Dq q
Z   Z  
D a cube is of order α4, so that the angular accelera on
= + q∇ · u J d3a = + q∇ · u d3x (12)
me as well as on the ini al posi on to dis nguish between fluid par cles. S0 D t St D t is propor onal to α−1(σ21 − σ12). In the limit of a fluid
In the Eulerian approach, we instead consider a fixed point x in space and measure the fluid proper es at D
Z Z 
Dq
 parcel α → 0, the angular accelera on remains finite
3
this point as func ons of me (being careful to ensure that x remains in St – otherwise no fluid is at the =⇒ q(X(a, t), t) d x = + q∇ · u d3x. (13) only if σ21 = σ12. A similar computa on of the torque
Dt St St Dt
point!). An observable, q, is then a func on of posi on and me: q = q(x, t). about the other axes allows one to conclude that the
The result of equa on (13) is known as the Reynolds Transport Theorem. When the observable q = tensor is symmetric. Figure 1.
Surely, there must be some rela on between the two methods! In fact, the most obvious one is the concept
ρ(X(a, t), t), we have For momentum to be conserved, the force on the
of velocity: we have D
Z Z 
D ρ

∂X ρ d3x = + ρ∇ · u d3x. (14) fluid must be propor onal to the gradient of the velocity. See the discussion on sec on 6.1 of [1]. As
= u(X(a, t), t), (1) D t St St D t
∂t a result, the deviatoric tensor is a linear func on of the deforma on tensor Dij = (1/2)(∂ui/∂xj + ∂uj /∂xi).
where u is the flow velocity in the Eulerian viewpoint at the Lagrangian coordinate X. This follows merely by If we assume that mass is conserved, then (14) must be zero. The equa on above must hold for all fluid This also means that dij and Dij are simultaneously diagonalisable. By permu ng the eigenvalues under
construc on. But what about other observables? In the Lagrangian perspec ve, any observable a ached parcels, which is only true if the integrand itself is zero: rota ons, the requirement that dij be isotropic forces the deviatoric tensor to take the form
∂ui ∂uj
 
to a fluid parcel just depends on the me explicitly. In the Eulerian viewpoint, however, the observables Dρ
+ ρ∇ · u = 0. (15) dij = λ(∇ · u)δij + µ + . (20)
depend on me explicitly and implicitly through the posi on. By the Chain rule, Dt ∂xj ∂xi
d q(X(a, t), t) ∂q ∂X With the full form of the deviatoric tensor, we can derive the equa ons of mo on. By Newton’s second
= + ∇q = ∂tq + u · ∇q. (2)
dt ∂t ∂t law, we obtain
R (17) except now with our corrected stress tensor. The only new addi on is the deviatoric
tensor term ∂St (d · n̂) d2x, which, by the Divergence theorem, becomes St (∇ · d) d3x. Consider just one
R
The special operator
D
≡
∂
+ u · ∇, (3) The Euler Equations of Motion component (using index nota on and Einstein summa on conven on):
Dt ∂t ∂dij ∂ ∂ ∂uj ∂ 2ui
The result (15) is known as the con nuity equa on, and, together with conserva on of momentum, we (∇ · d)i = = λδij (∇ · u) + µ +µ 2 (21)
is called the material deriva ve, and denotes the Lagrangian me deriva ve in Eulerian variables. ∂xj ∂xj ∂xj ∂xi ∂xj
can arrive at the so-called Euler equa ons. Newton’s second law relates the material deriva ve of the
∂ ∂ ∂uj
momentum of a fluid to the net external force on the fluid. The net force per unit volume can be expressed = λδij (∇ · u) + µ + µ∇2ui (22)
A Useful Identity: The Derivative of the Determinant of the Jacobian generally as ∂xj ∂xi ∂xj
∂ ∂ ∂ ∂
 
∂σij
One useful way to describe how fluid parcels transform is by the Jacobian J. (For simplicity, I will work in Fi = fi + , (16) = λ δi1 + δi2 + δi3 (∇ · u) + µ (∇ · u) + µ∇2ui (23)
∂xj ∂x1 ∂x2 ∂x3 ∂xi
three-dimensions, and I’ll make heavy use of indicial nota on and Einstein summa on conven on.) Recall
Upon specifying the index i, the first and second terms of (23) are really the same thing, up to the viscosity
in the Lagrangian viewpoint that a fluid moves from a set S0 to St, which can be understood to occur via where σij is the stress tensor and fi is some external body force. Assuming an ideal fluid, the stress tensor
coefficients λ and µ. From this, we obtain that
a map Mt : S0 → St. We now introduce the Jacobian of this map, whose elements are given by is completely diagonal with σij = −pδij , so that ∇ · σ = −∇p, where p is the pressure exerted normal to

∂xi  the surface of the fluid. Hence, by Newton’s second law, ∇ · d = (λ + µ)∇(∇ · u) + µ∇2u. (24)
Jij = . (4) Z
Du 3
Z Z
∂aj t ρ dx= f d3x + (σ · n̂) d2x. (17) Returning to Newton’s second law, we have
St D t Du 3
St ∂St
Z Z
The determinant of the Jacobian can be wri en succinctly using the completely an symmetric tensor, ijk , 2

3
R 3 ρ dx= f − ∇p + (λ + µ)∇(∇ · u) + µ∇ u d x. (25)
as By the Divergence theorem, the far-right integral becomes St (∇ · σ) d x, so that St D t St
J = ijk J1iJ2j J3k . (5) Z
Du 3
Z Z
Since (25) must hold for any volume, it follows that
ρ dx= (f + ∇ · σ) d3x = (f − ∇p) d3x. (18)
With this, we can describe the deforma on of a fluid parcel from S0 → St by St D t St St Du
ρ = f − ∇p + (λ + µ)∇(∇ · u) + µ∇2u. (26)
Z Z Since this must hold for all such fluid parcels, we arrive at the second Euler equa on: Dt
d3x = J d3a. (6) And so, we arrive at the Navier-Stokes equa ons for a viscous fluid!
St S0 Du
ρ = f − ∇p. (19)
We might be interested in observing how volume integrals like (6) change over me. This raises an interim Dt
problem: what is the material deriva ve of J? Well, by the product rule, References
DJ D J1i D J2j D J3k
 
= ijk J2j J3k + J1i J3k + J1iJ2j . (7)
Dt Dt Dt Dt [1] Stephen Childress. An introduc on to theore cal fluid mechanics, volume 19. American Mathema cal Soc., 2009.
[2] Alexandre Joel Chorin and Jerrold E Marsden. A mathema cal introduc on to fluid mechanics, volume 3. Springer, 1990.
[3] Jerrold E Marsden and Thomas JR Hughes. Mathema cal founda ons of elas city. Courier Corpora on, 1994.
 Lebesgue Measure and Integration
Vardan Martirosyan Jordan Russo
University of California, Santa Barbara

Abstract and Background Properties of the Lebesgue Measure Lebesgue Integration

Traditionally, the integral of a non-negative single valued function is defined to be the area under There are several key properties that the Lebesgue Measure contains, which we will now describe:
the smooth curve of the function, from a start point a to an end point b on the real number line. Characteristic Functions For any set A, we define the characteristic function of A on the real
In undergraduate courses, this concept is formalized as the Riemman Integral. After proving nu- The Measure of An Interval is it’s Length Let I be an arbitrary non-empty interval. Then, I is numbers, denoted by χA, as: (
merous results and theorems relating to Riemman Integration, as well as extending it to multiple Lebesgue Measurable and: 1, if x ∈ A
dimensions, it is shown that the Riemman Integral has some limitations: namely, there are severe m(I) = l(I) (3) χA(x) = (6)
0, if x ∈
/A
issues when dealing with point-wise limits, and integrating sequences. To resolve these issues,
where l is the ’set’ length function described earlier.
the concept of the Lebesgue Measure and Lebesgue Integration is introduced at the end of un-
Lebesgue Measure is Translation Invariant Let E be a Lebesgue measurable set, and y be any Simple Functions Let φ be a real valued function defined on a measurable set E. It is called a
dergraduate and the beginning of graduate courses. For our project, we studied the Lebesgue
number. Then, the translation of E by y, E + y = {x + y|x ∈ E}, is also Lebesgue measurable simple function if it is measurable and takes a finite number of values. Any simple function can
Measure and Lebesgue Integration from the textbook ”Real Analysis” by H. Royden and P. Fitz-
and: be represented as a linear combination of characteristic functions:
patrick. n
m(E + y) = m(E) (4) X
ψ= ck · χEk , Ek = {x ∈ E| ψ(x) = ck } (7)
Definitions We display a picture to illustrate this property: k=1

We first define several important terms: Integration of Simple Functions For a simple function ψ defined on a set E where m(E) < ∞,
we defined the integral of ψ over E by:
Open, Non-Empty, and Bounded Sets A open set is a set such that, for any point in the set, Z Xn
and any given distance, a point of the set can be found between the given point and distance. ψ= ai · m(Ei) (8)
A non-empty set is a set that has at least one element contained within it. A bounded set is a E i=1
set that is of a finite size.
Lebesgue Integration For a bounded real-valued function f defined on a set E where m(E) < ∞,
Complement of a Set Let E be a set of points. The complement of E, denoted by E c, is the set
we define the lower and upper Lebesgue Integral, respectively, of f over E to be:
of points that are not in E. We note that E ∩ E c, the intersection of E and it’s complement, is Z   Z  
the empty set ∅. Additionally, the union of E and E c is all of the points U that are being
 
Figure 1. A picture displaying what translation invariance looks like. The picture comes from Wikipedia. sup ψ  ψ simple, ψ ≤ f on E and inf φ  φ simple, f ≤ φ on E (9)
looked at. E E
Lebesgue Measurable is Countably Additive Over Countable Disjoint Unions of Sets Let We say that f is Lebesgue Integrable over E when its lower and upper Lebesgue integrals over
Length {Ek }∞
k=1 be a countable disjoint collection of Lebesgue measurable sets. Then, we have that: E are equal. We call the common value the Lebesgue Integral of f over E, and denote it by:
Z
∞
Consider the extended real number line, which spans R, the set of real numbers, combined with f (10)
m(∪∞
X
k=1Ek ) = m(Ek ) (5) E
−∞ and +∞. Let I be an interval on the extended real number line. We define the length of I
k=1
to be the difference of it’s endpoints if it is bounded, and to be ∞ if it is unbounded. We call the
We note that one of the key differences between the outer measure defined earlier and the
length function a set function, which is a function that assigns an extended real number to each
Lebesgue Measure is that in the equation above, the outer measure has sub-additive Advantages over the Riemman Integral
set in a collection of sets.
property, which is less powerful than the additive property stated above.
Monotone convergence Theorem Suppose we have a sequence of non-negative measurable
Outer Measure functions {fn} on a measurable set X such that fn converges pointwise to f almost everywhere
Before being able to define the Lebesgue Measure, we first have to define a separate measure, Lebesgue Measurable Functions and f1 ≤ f2 ≤ · · · ≤ fn. The Monotone Convergence Theorem gives us the following property
called the outer measure. Let A be a set of real numbers. Consider the countable collections for Lebesgue integration: Z Z Z
{Ik }∞ An extended real-valued function defined on a set E is said to be Lebesgue measurable, provided (11)
k=1 of nonempty, open, bounded intervals that cover A. For each collection, consider the lim
n→∞ X
fn = lim fn = f
sums of the lengths of the intervals within the collection. Since the lengths are forced to be it’s domain E is measurable, and it satisfies one of the following two conditions: X n→∞ X
positive numbers, each sum is uniquely defined independently of the order of the terms. We can Under the Riemman Integral, the ability to move the limit inside the integral requires uniform
For each real number c, the set {x ∈ E|f (x) > c} is measurable.
then define the outer measure of A, m∗(A), to be the infimum of all types of sums: convergence, while under the Lebesgue Intregral, we only require pointwise convergence. We
  For each real number c, the set {x ∈ E|f (x) ≥ c} is measurable. now give an example to illustrate the use of the Monotone Convergence Theorem. Let aij be
X ∞ 
m∗(A) = inf l(Ik )|A ⊆ ∪∞
k=1 Ik (1) an non-negative real valued sequence of numbers. Then, we have that:

k=1
 Characterizations and Properties of Measurable Functions X∞ X ∞ ∞ X
X ∞
aij = aij (12)
A function f is measurable if and only if for each open set O, the inverse image of O under f i=0 j=0 j=0 i=0
Measurable Functions and the Lebesgue Measure
is measurable.
Let E be a set. We define E to be measurable if for any set A, we have the following to be true: A real valued function that is continuous on it’s measurable domain is measurable.
m∗(A) = m∗(A ∩ E) + m∗(A ∩ E c) (2) A monotone function that is defined on an interval is measurable. Acknowledgements
All sets that satisfy the above equation make up a Borel sigma algebra. Then, the Lebesgue Linear combinations, products, and compositions of finite measurable functions on the same
Measure is the restriction of the set function outer measure to this class of measurable sets. We set E are also measurable on the set E. We would like to Pranav Arrepu for his guidance in helping us understand the material during this
denote the Lebesgue measure by m, and write that m(E) = m∗(E). We note that the Lebesgue A non-negative measurable function is the limit of a sequence of simple functions. program. Additionally, we would like to thank the Directed Reading Program at UCSB for giving
measure is not defined on all subsets of R: only those that satisfy the above equation. (A proof us the opportunity to participate in this program.
of why not all subsets of R are measurable comes from Vitali’s Theorem).
 Sticking a ‘PINN’ in It: A Physics-Informed Neural Network Approach to PDEs
Yan Lashchev 1 Bill Nguyen 1 Mentor: Rafael Lainez Reyes 1
1Department of Mathematics, University of California - Santa Barbara

Abstract PINN Approach PINN Approximation and Evolution of Loss

Recent successes in neural networks have greatly encouraged their use in solving classical prob- For the solu on of the PDE (1) now proceeds by minimiza on of the loss func onal
lems in applied mathema cs, as the networks allow for rapid prototyping with usable es ma ons.
   
φθ (X) := φrθ (X r ) + φ0θ X 0 + φbθ X b , (3)
This holds especially true in areas involving high dimensional par al differen al equa ons (PDEs),
such as quantum physics and fluid dynamics. Here, we present a neural network architecture, the where X denotes the collec on of training data and the loss func on φθ contains the following
physics-informed neural network (PINN), and implement a specific method, the con nuous me terms:
approach.
The Mean Squared Residual
Background Nr
(a) View 1 (b) View 2 (c) Loss
1
r r |rθ (tri , xri )|2 Test
X
φθ (X ) := Figure 1. The same cup of coffee. Mul ple mes.
We describe the PINN approach for approxima ng the solu on Nr
i=1
u : [0, T ] × D → R (?)  r r
Nr The chosen problem can be solved via separa on of variables. The idea is to assume the solu on
r
in a number of colloca on points X := ti , xi i=1 ⊂ (0, T ] × D, where rθ is the physics-
of an evolu on equa on u = u(t, x) can be wri en as
informed neural network (2), u(t, x) = F (t)G(x)
∂tu(t, x) + N [u](t, x) = 0, (t, x) ∈ (0, T ] × D, (1a)
u(0, x) = u0(x), x ∈ D, (1b) The Mean Squared Misfit w.r.t Initial and Boundary Conditions If we compute the corresponding par al deriva ves and replace in the PDE, we get
F 0(t) G00(x)
where N is a differen al operator ac ng on u, D ⊂ Rd a bounded domain, T denotes the final N0   Nb   =
  1    2  1    2 F (t) G(x)
me and u0 : D → R the prescribed ini al data. Based on the literature review conducted, we and φbθ X b := b, xb − u tb, xb 
X X
φ0θ X 0 := u t0 , x 0 − u x0  u t
 
 θ i i 0 i  θ i i b i i  The only way this equality is true for all t and x is if
N0 Nb

restrict our discussion to the Dirichlet case and define i=1 i=1
u(t, x) = ub(t, x), (t, x) ∈ (0, T ] × ∂D, (1c) n oNb F 0(t) = λF (t) and G00(x) = λG(x)
N
in a number of points X 0 := t0i , x0i i=10
⊂ {0} × D and X b := tbi, xbi


⊂ (0, T ] × ∂D,
where ∂D denotes the boundary of the domain D and ub : (0, T ] × ∂D → R the given boundary i=1 The boundary condi on becomes
where uθ is the neural network approxima on of the solu on u : [0, T ] × D → R. G(0) = G(π) = 0
data. The method constructs a neural network approxima on uθ (t, x) ≈ u(t, x) of the solu on of
(1), where uθ : [0, T ] × D → R denotes a func on realized by a neural network with parameters We can easily solve these ordinary differen al equa ons. By considering the cases λ > 0, λ = 0
θ. Example: Heat Equation
and λ < 0, we conclude λ = −n2, n ∈ N and (up to constants)
 
A classical problem in the domain of PDEs, the heat equa on governs the temperature distribu on F (t) = exp −n2t
Continuous Time Approach and G(x) = sin(nx)
of a rod of length l :
∞
(t, x) ∈ R+ × (0, l)
 
ut = kuxx
X
Since the equa on is linear, by the principle of superposi on u(t, x) = cn exp −n2t sin(nx)
u(t, 0) = u(t, l) = 0 t≥0 n=1
∞
u(0, x) = f (x) x ∈ (0, l). X
Finally, since u(0, x) = sin(3x) = cn sin(nx) with c3 = 1 and cn = 0 if n 6= 3. Hence,
If k, called the conduc vity is a constant the rod is isotropic; if k = k(x) it is anisotropic or n=1
heterogeneous medium.
u(t, x) = exp(−9t) sin(3x)
Application
True Solution
With respect to the fi ng, we choose k = 1, l = π, and f (x) = sin(3x) for the demo of the PINN.

We assume that the colloca on points Xr as well as the points for the ini al me and boundary
data X0 and Xb are generated by random sampling from a uniform distribu on.

Figure 1) Neural network architecture of the PINN approach

The (strong) residual of a given neural network approxima on of (?) with respect to the PINN
approach above is
rθ (t, x) := ∂tuθ (t, x) + N [uθ ] (t, x) (2)
These networks are composi ons of alterna ng affine linear W ` · +b` and nonlinear func ons (a) PINN (b) True Solu on
σ `(·) called ac va ons, i.e.,
      References
uθ (z) := W Lσ L W L−1σ L−1 · · · σ 1 W 0z + b0 · · · + bL−1 + bL,

where W ` and b` are weight matrices and bias vectors, and z = [t, x]T . [1] Jan Blechschmidt Oliver Ernst. Three ways to solve par al differen al equa ons with neural networks — a review.
GAMM-Mi eilungen, 44(2), 2021.
Figure 2) Plot of the colloca on points (N = 10, 000)
[2] Peter Olver. Introduc on to par al differen al equa ons. Springer, 2020.

Mathematics Directed Reading Program 2022

 P RIMITIVE PARKING F UNCTIONS AND N ON -C ROSSING PARTITIONS
Yanru Liu, Mentored by Sam Sehayek
2022 Mathematics Direct Reading Program. University of California, Santa Barbara

Parking Functions Non-Crossing Partitions A New Proposition

Imagine living on a one-way street that dead-ends with n parking spots available. A partition of a finite set S is a collection {B1, · · · , Bk } of nonempty subset Bi ⊆ S s.t. B1 ∪ For maximal chains corresponding to the primitive parking functions, do they
You and your neighbors have n cars in total, and everyone has their preferred · · · ∪ Bk = S and Bi ∩ Bj = ∅ if i ̸= j . And in our research of primitive parking functions, we form a certain pattern?
spot to park. Without reversing, does there exist a solution that everyone can especially care about a special one: the non-crossing partition. A non-crossing partition
Lemma 3. The chain starts by merging 1 with some other element.
park without collision? In mathematics, this real life dilemma is called the parking of set {1, · · · , n} is a partition {B1, · · · , Bk } of {1, · · · , n} s.t. for a < b < c < d, a, c ∈ Bi,
problem. Consider this set up: and b, d ∈ Bj ⇒ i = j . This lemma is trivial as the primitive parking function is always starting with 1,
• There are n cars and n parking spots on a straight street (n is a positive and only the merging between 1 and some other element gives the label 1.
integer, n ∈ Z+; and i denotes the i-th spot, i ∈ {1, · · · , n})
Lemma 4. The primitive parking functions are always adding one single block to
• Ci is the i-th car to park, having preferred spot αi ∈ {1, · · · , n}. More than the other block.
one car can have the same preference.
This proposition can be verified via figure 4. From these lemmas, we can prove:
• If the preferred spot had already been occupied, then the car will move for-
Proposition 5. The subdiagram consisting only of nodes and edges from the
ward and park in the next available spot. No backward movement allowed.
primitive parking functions inside the non-crossing partition lattice is a coarsen-
If all n cars can be parked under these conditions, then the preference list α =
ing of the Boolean lattice of size equal to the length of these parking functions.
(α1, α2, · · · , αn) is a parking function. Fig. 2: Non-Crossing Partitions (Blue) & Crossing Partitions (Yellow) of {1, · · · , 11}.

Equivalently, an n-tuple of integers α = (α1, · · · , αn) is a parking function if and

A maximal chain of non-crossing partitions of {1, · · · , n + 1} is a sequence π0, · · · , πn of
only if βi ≤ i, where β = {β1, · · · , βn} is a reordering of α into weakly increasing
noncrossing partitions s.t. πi is obtained from πi−1 by merging two blocks of πi−1 into a
order. i.e. β1 ≤ · · · ≤ βn.
single block.
For n cars, how many parking functions are there?
A maximal chain of [n + 1] has n merging steps. If we pick a label for each step, there are
exactly n labels. Thus, it’s possible for us to connect parking function with maximal chains.
Theorem 2. There is a bijection between parking functions of length n and maximal chains
of NCn+1.
Here is the algorithm: Let A and B be the two blocks we’re going to merge at stage i, and
A contains the smallest element in the disjoint union A ∪ B . The label for this stage is the
largest element in A which is smaller than all elements in B .

Fig. 3: One of the Maximal Chain of {1, · · · , 5} and Its Associated Parking Function (1, 1, 3, 3).
(The top is 1 and the label goes in clockwise direction)
Fig. 1: Visual Representation of H.Pollack’s Proof.

Regard the elements of the group G = Z/(n + 1)Z as being the integers Every maximal chain is associating with a parking function, and only some of them are
0, 1, · · · , n. Let H be the (cyclic) subgroup of order n + 1 of the group Gn gener- associating with the primitive ones.
ated by (1, 1, · · · , 1). Each coset of H contains exactly one parking function. Let
f (n) be the number of parking functions of length n, hence we have
Theorem 1 (Konheim and Weiss, 1966). The number of parking functions of Fig. 6: Maximal Chains for Primitive Parking Functions of Length 4 (Up)
length n is & Boolean Lattice of size 4 (Bottom).
f (n) = (n + 1)n−1.
In other words, maximal chains of primitive parking functions of length n are look
just like the Boolean lattice of the same size with some relations removed.
Primitive Parking Functions
Acknowledgements
A parking function is called a primitive parking function if it is already in a
weakly increasing order.
I would like to thank my mentor Sam Sehayek, his knowledge and enthusiasm in
There is a well-known bijection between parking functions and labeled Dyke
mathematics deeply impressed me; he is such a good mentor and friend on my
paths, wherein each distinct labeling of the same Dyke path corresponds to a
way of learning mathematics.
permutation of the parking function. Thus, the primitive parking functions with
length n − 1 are in bijection with Dyke paths and can be enumerated by the n-th
Catalan number:   Reference
1 2n
Cn = .
n+1 n
Fig. 4: Primitive Parking Functions with Length 4 (black) in the Maximal Chains of Noncrossing Partitions of {1, · · · , 5} (grey).
Richard P. Stanley. Enumerative Combinatorics. Cambridge Press, 1998.
 A Brief Introduction to Network Theory
Ponokela DeMarzo
University of California, Santa Barbara

What is a Network? A Social Network Example Why are Networks Useful?

A network is a collection of nodes where pairs of nodes may be connected by edges. Networks are a powerful analytical tool which are used across many different disciplines with a
multitude of applications. Networks are an elegant representation of almost any system which
Networks can be visualized by drawing their graph structure, but they are also commonly repre-
consists of objects and connections between those objects, and when modeled this way, we can
sented by their adjacency matrix. The adjacency matrix for a network consisting of n nodes is
perform well-defined and meaningful calculations to analyze its structure.
an n x n matrix A with entries Aij = the number of edges connecting node i and j.
There are four primary categories in which we can sort networks: technological, information, so-
cial, and biological. Technological networks are physical networks which are typically responsible
  for the transfer of data or materials, such as the Internet, waterlines, or commercial airline flights.
0 1 1 0 1
1 Information networks can model the interaction of ideas, and are used to represent structures
0 1 1 0
  such as the World Wide Web or citation networks for academic papers. Social networks are
A=
1 1 0 0 0
0
 used to model people and their interactions, such as friendships in a workplace or followers on
1 0 0 0
social media. Even systems such as metabolic processes and food chains can be modeled by
1 0 0 0 0
biological networks.

Network Structure
Figure 1: The graph and adjacency matrix representations of the same network.
Beyond simply ranking nodes in accordance to their centralities, it also often important to be able
Edges can be weighted, where each edge is assigned a value that represents the "strength" to describe the overall structure of the network. Below are some basic, yet useful, measures of
of the connection, as well as directed, where a connection from node i to j does not imply a network structure:
connection from j to i. Networks are also not limited to only one type of node; however, since
many of the forms of analysis change, we will not be discussing them. Density: In a simple network consisting of n nodes, we can calculate the maximum possible
number of edges by counting the number of pairs of nodes given by n2 . The density of a


Figure 2: A social network constructed from anonymized friendship data collected by surveying a high school
network is the proportion of existing edges to the maximum possible edges.
Node Centrality math classroom. The density of the network is 20.2% and the clustering coefficient is 37.2%. The largest core is a
4-core which includes all nodes except 14 and 25. k-cores: A k-core is a connected set of nodes where each node is connected to at least k
A natural question arising from the gathered network data is determining the importance, or other nodes in the set.
centrality, of each node. This can give us an idea of which nodes have more influence over a
network. The four main centrality measures are defined below: Network Analysis
Degree Centrality: This is perhaps the simplest measure of centrality, calculated by counting
the number of edges attached to the node in question. In terms of the adjacency matrix A, Node Degree Eigenvector Closeness Betweenness Local Clustering
the degree centrality of a node i can be defined as follows: 1 5 0.1049 0.4615 0.0209 0.2 Figure 3: The nodes 1-4 form a 3-core, and the nodes 1-5 form a 2-core.
P 2 6 0.2153 0.4347 0.0116 0.4666
ki = Aij . 3 7 0.2595 0.5454 0.0525 0.4285
j k-components: A k-component is a set of nodes where each node is reachable from each of
4 4 0.0911 0.4347 0.0018 0.8333
5 10 0.3201 0.5084 0.0657 0.3777
the others by k node-independent paths.
Eigenvector Centrality: Unlike degree centrality, eigenvector centrality is primarily
6 7 0.2605 0.4477 0.0107 0.5714
concerned with the quality of connections, not quantity. To measure centrality this way, the 7 11 0.3491 0.4838 0.0573 0.3818
centrality of a node i will be proportional to the centrality of its neighbors, and thus is 8 6 0.1972 0.4918 0.0285 0.4
defined recursively like so: 9 6 0.0967 0.4687 0.0671 0.2666
10 5 0.1390 0.4347 0.0131 0.5
xi = κ−1 Aij xj .
P
11 5 0.1054 0.4687 0.0324 0.2
j
12 7 0.1116 0.4838 0.0767 0.2857 Figure 4: This network forms a 2-component.
Rewritten in matrix notation, this equation becomes 13 5 0.0918 0.4477 0.0169 0.6
14 2 0.0207 0.3370 0.0021 0
x = κ−1Ax, or Ax = κx. Local Clustering Coefficient: The local clustering coefficient is the proportion of the number
15 4 0.1432 0.4 0.0019 0.6666
16 6 0.1111 0.4761 0.0297 0.5333 of neighbors of a node i that are neighbors themselves. Visually, this can be thought of as
In this form, it is clear that x is an eigenvector of A; however, since there may be multiple the fraction of closed triangles out of all possible triangles with i as a vertex.
17 13 0.2783 0.625 0.2732 0.1538
eigenvectors, we generally define x and κ to be the leading eigenvector and eigenvalue.
18 7 0.1537 0.5 0.0474 0.2380
Closeness: This is a measure of the average distance from a node to other nodes. Suppose 19 5 0.1485 0.4225 0.0156 0.4
that dij is the shortest distance from node i to node j. Then the average distance from i to 20 8 0.2040 0.5555 0.0926 0.1428
21 6 0.2093 0.4285 0.0125 0.5333
every other node is 22 7 0.1516 0.5084 0.0817 0.1904
1 P 23 6 0.2025 0.5084 0.0603 0.3333
`i = n−1 dij . Figure 5: The local clustering coefficient of node 1 is 1/3.
j 24 7 0.2372 0.5 0.0573 0.4285
25 4 0.0481 0.375 0.0135 0.5
Since we want to consider nodes that are on average closer to all other nodes as being more 26 4 0.0840 0.4615 0.0137 0.3333
central, we define the closeness centrality as the inverse of `i so Clustering Coefficient: This is a generalization of the local clustering coefficient to the whole
27 4 0.0815 0.4411 0.0102 0
1 n−1  28 5 0.1052 0.4687 0.0316 0.3 network. For the whole network, it is the proportion of connected triples that are also closed
Ci = `i = P . triangles.
j dij 29 7 0.2432 0.4687 0.0247 0.4761
30 5 0.1447 0.4687 0.0171 0.3
Betweenness: This measures how often a given node lies on a shortest path between other 31 4 0.0742 0.4285 0.0065 0.5
nodes. Let nist be the number of shortest paths from s to t that pass through i, and let gst be
the total number of shortest paths from s to t. We can then define the betweenness
Table 1: The centrality and clustering measures for each node in Figure 2. The largest and smallest values in each References Guided by
column are bolded and the corresponding nodes are highlighted. Node 17 has the highest degree, closeness, and
centrality of a node i as follows: betweenness centralities, however, node 7 has the highest eigenvector centrality due to the importance of its
P nist neighbors (e.g., nodes 5 and 6). Node 4 has the highest clustering coefficient, indicating a tight-knit friend group,
Sanjay Kumar
xi = n12 gst . but has a low betweenness centrality because it is somewhat redundant in the network. Node 14 is the most
[1] Mark Newman.
UCSB
st Networks, Second Edition.
isolated. Oxford University Press, 2018.
 Differential Forms and Maxwell’s Equations
Samuel Zhang - Mentored by Yusen Xia
Department of Mathematics - University of California, Santa Barbara

Introduction Exterior Derivative

Differential manifolds are topological spaces that are locally ▶ Given a smooth differential k-form w, its exterior derivative is
homeomorphic to a vector space so that one may perform calculus defined as
n
on it, and a differential form allows one to define integrals over X
i1 ik
such manifolds. This poster is meant to revisit the Maxwell’s dw = dwi1i2...ik ∧ du ∧ ... ∧ du
Equations using such languages i1,...,ik =1
n n
X X ∂wi1i2...ik j
= du ∧ dui1 ∧ ... ∧ duik
i1,...,ik =1 j=1
∂uj
Smooth Manifolds and Tangent Map
▶ Given smooth differential k-forms w, η on a smooth manifold M and
a smooth scalar function f ,
▶ A topological space M is called an n-dimensional manifold if
▷ d(w + η) = dw + dη
∀p ∈ M there is a homeomorphism F : U → O such that U ⊂ Rn is
▷ d(f w) = df ∧ w + dη
non-empty and open, and O ⊂ M is an open subset containing p.
▷ d2w = d(dw) = 0
Such an F is called a local parametrization around p
▶ An n-dimensional manifold is called an n-dimensional smooth ▶ A connection between a differential form and exterior derivative in
manifold if there is a collection of local parametrizations R3 and usual multivariable calculus is shown below:
Differential Form on R3 Multivariable Calculus
Fα : Uα → Oα such that
f (x, y, z) f (x, y, z)
▷ ∪Uα = M (such parametrizations cover all of M )
▷ Any transition map Fα−1 ◦ Fβ is smooth on their domain
w = P dx + Qdy + Rdz F = P î + Qĵ + Rk̂

β = Ady ∧ dz + Bdz ∧ dx + Cdx ∧ dy G = Aî + B ĵ + C k̂

df ∇f
dw ∇×F
dβ ∇·G
d2f = 0 ∇ × ∇f = 0
d2w = 0 ∇ · (∇ × F ) = 0

Revisit Maxwell’s Equations

▶ The Maxwell’s Equations can be written in differential equations as:

Figure 1:The stereographic projection of a sphere, which is a parametrization of ρ
∇·E=
a sphere except at the north pole. The sphere is an example of a 2-dimensional ϵ0
smooth manifold ∇·B=0
▶ Given a smooth n-dimensional manifold M and a local ∂B
∇×E=−
parametrization F : U → M , the tangent space at p is defined as ∂t
1 ∂E
TpM = span{ ∂u∂ 1 (p), ..., ∂u∂ n (p)}, with each ∂u
∂
being the differential ∇ × B = µ0 j + 2
operator with respect to F (u1, ..., un) at p
i
c ∂t
▶ The first equation is the Gauss’s law on electric field, the second
▶ For a vector space V , define its dual space V ∗ = {T |T : V → R}
equation is a statement that an magnetic monopole does not exist
▷ Moreover, V ∗ is a vector space itself. Given a basis of V as
(it has been predicted in several models but not yet verified), the
{e1, ..., en}, a basis of V ∗ is {e1, ..., en} such that ei(ej ) = δij
third equation is the law of electromagnetic induction, and the fourth
▷ One can define the cotangent space of M at p as Tp∗M = (TpM )∗
equation is the Ampere’s circuit law with Maxwell’s correction
and any v ∗ ∈ Tp∗M is called a cotangent vector of M at p. In the
▶ Denote (t, x, y, z) ∈ R4 as (x0, x1, x2, x3) and take w be a k-form on
same fashion, one can express Tp∗M = span{du1, ..., dun} such
R4 (here k = 0, 1, 2, 3 or 4). Define the Hodge-star map from a
that dui[ ∂u∂ j (p)] = δij
k-form to (4-k)-form such that w ∧ ∗w = dt ∧ dx ∧ dy ∧ dz, or
−dt ∧ dx ∧ dy ∧ dz if w contains a dt term (this is known as the
volume form of the Minkowski spacetime)
Tensor Product and Wedge Product ▶ Express E, B, J as
∗ ∗ E = Exdx + Ey dy + Ez dz
▶ For V, W being two vector spaces, T ∈ V and S ∈ W , the tensor
B = Bxdy ∧ dz + By dz ∧ dx + Bz dx ∧ dy
product between T and S is defined as T ⊗ S : V × W → R such
that (T ⊗ S)(X, Y ) = T (X)S(Y ) J = −(Jxdy + Jy dz ∧ dx + Jz dx ∧ dy) ∧ dt + ρdx ∧ dy ∧ dz
▶ In the same setup, the wedge product is defined as Define F ≡ B + E ∧ dt. Together with the Hodge-star map, one can
T ∧S =T ⊗S−S⊗T rewrite the Maxwell’s equations as:
▷ One can see that a wedge product is alternating; T ∧ S = −S ∧ T dF = 0
▷ Also noted that T ∧ T = 0 d(∗F) = J

Differential Form
References
▶ Let M be a smooth manifold. The smooth differential k-form w on
[1] T.H. Fong.
M is defined as w : TpM × TpM × ... × TpM (k times)→ R such that Differentiable Manifolds & Riemannian Geometry.
for any
Pn local parametrization F : U → M , Hong Kong University of Science and Technology, 7th edition, 2021.
i1 ik
w = i1,...,ik =1 wi1i2...ik du ∧ ... ∧ du . The wi1i2...ik ’s are scalar functions
locally defined in F (U ) and are called the local components of w
Rb
▷ For example, in a f (x)dx the f (x)dx is a differential 1-form
 U NIVERSAL A PPROXIMATION T HEOREM
Xin Wang, Ruifan Wang
University of California, Santa Barbara

Multiplayer Feedforward Neural Networks Neural networks arise from the collection, Convolution Applied to a Specific Example
Σn = span{σ(w · x + θ) : w ∈ Rn, θ ∈ R}
A neural network is the interconnection of unit models characterized by a thresh- and the main result of the paper is that Σn is dense in C(Rn) if and only if σ is not an To illustrate the utility of the convolution, let f (x) := 1 − |x| when −1 ≤ x ≤ 1/2
2)
old value θ, a univariate activation function σ : R → R, and a vector of weights algebraic polynomial. This is a novel conclusion since the condition is very simple. and 0 otherwise. Let g(x) be a bump function, where g(x) = e −1/(1−x for
w = w1, ..., wn. Here, the value of n is determined by the dimension of the input- |x| ≤ 1 and 0 otherwise. The convolution: (f ∗ g)(x), is defined as,
vector x = x1, ..., xn. When we feed x into a unit, it computes σ(w · x − θ) and Z ∞ Z ∞
shoots the result to the next unit. A single hidden layer feedward neural network Reduced Case (f ∗ g)(x) = f (x − y)g(y)dy = f (y)g(x − y)dy
represents a f : Rn → R function −∞ −∞

k The equality of the integrals above follows by a change of variables. Since f and
X There are two directions to prove, one of which is not difficult: Σn is dense in C(Rn), then g are supported on a compact set, we will write the convolution as,
f (x) = βj · σ(wj · x − θj )
σ is not a polynomial. The rest of the proof aims to show that if σ is not a polynomial, then
j=1 Z 1/2
Σn is dense in C(Rn). The proof relies on some analysis tricks, which we summarize here. −1/(1−(x−y) 2)
(f ∗ g)(x) = (1 − |y|)e dy
Some common techniques to prove results like this are, −1

• Reduce the dimension of the space(s) considered. Then, we may use a numerical integrator to evaluate the convolution.
Here is a visual comparison between the discontinuous function f (x) and the
• Prove the result for well-behaved functions first. convolution of f (x) with g(x).
• Use a “smoothing" technique to deal with functions lacking regularity.

The complexity of the problem is reduced by showing first that if Σ1 = C(R), then
Σn = C(Rn). Then, Leshno et al. prove Σ1 = C(R) in the case that σ ∈ C ∞.

To show Σ1 = C(R) when σ ∈ C ∞, Leshno et al. show that Σ1 contains all polynomials.
The result follows then as a consequence of Weierstrass’s Theorem:
Theorem (Weierstrass’s Theorem[3]). If f is a continuous function on a compact set K , there
The most important application of neural networks is in machine learning, where exists a sequence of polynomials Pn such that
neural networks are “trained" to approximate a function. Thus, a fundamental lim Pn(x) = f (x)
n→∞
question for neural networks is whether they can approximate reasonable func- Remarks
tions to an arbitrary degree of accuracy. This depends on the activation function uniformly on K .
σ and is the subject of many papers, including the paper studied for this project.
It follows that Σ1 contains C(K), where all K ⊂ R. Hence, Σ1 is dense in C(R). In 1991, Hornik showed that the multilayer feed-ward architecture gives neural
networks the potential of being universal approximators[1]. Leshno et al. led the
study to a new dimension and discovered that a neural network does not need
Nonpolynomial Activation Function Generalized Case a continuous activation function to approximate some real-world functions in an
arbitrary accuracy. This endows the neural network a biological interpretation
Leshno et al. proved in their paper “Multilayer Feedforward Networks With a Non- because a real neuron is unlikely to have a continuous activation function. Later
From above steps, the "dense" argument can be easily achieved when σ is smooth. In
polynomial Activation Function Can Approximate Any Function" [2] that, under in the history, mathematicians extends the Universal Approximation Theorem
this section, the author generalizes the problem to the entire class of admissible activation
modest assumptions, a broad class of activation functions are suitable for build- by studying discontinuous functions, noncompact domains, and so on.
function by considering σ that is not smooth. The purpose of convolution σ ∗ φ is to deal
ing neural networks to approximate continuous functions. We studied this paper with the discontinuities and points where σ is not differentiable. In a way, the convolution
to understand the mathematics underlying the result. can overcome the limited differentiability of σ . We will discuss the merit of convolution in Acknowledgement
Definition (Notion of approximation). We say that a set F of functions in L∞
loc (R n) the next section.
n n
is dense in C(R ) if for every function g ∈ C(R ) and for every compact set We especially appreciate our mentor Zach Wagner for enlightenment and two
K ⊂ Rn, there exists a sequence of functions fj ∈ F such that By convolving σ with functions φ ∈ C0∞, the general case follows as a consequence of the
quarters of patient teaching for this project. We also thank every member con-
work for the reduced case: Σ1 is dense in C(R) so long as σ ∗ φ is not a polynomial for
lim ||g − fj ||L∞(K) = 0 tributed to the Directed Reading Program.
j→∞ some test function φ. The authors deal with this caveat using advanced techniques.

References

Colloquially, fj approximates g “arbitrarily well." Basically, Leshno et al. must knows what is the condition that makes σ ∗ φ a polynomial for
Definition. The admissible class of activation functions which Leshno et al. every test function φ. It turns out that this only occurs if σ is a polynomial almost everywhere,
denote by M is the set of locally bounded functions with a “small" number of which rules out any strange conditions where σ ∗ φ is a polynomial for some φ ∈ C0∞, yet [1] Kurt Hornik. “Approximation capabilities of multilayer feedforward networks”. In: Neural Net-
discontinuities: if σ ∈ M and K is the collection of discontinuities of σ , then K σ is not a polynomial. Their key argument is to show that if σ ∗ φ is a polynomial for every works 4.2 (1991), pp. 251–257.

has zero Lebesgue measure. test function φ, then the degree of σ ∗ φ is bounded by some m for every φ. From here, [2] Moshe Leshno et al. “Multilayer feedforward networks with a nonpolynomial activation func-
they conclude that since σ ∗ φ is a polynomial of degree at most m for every test function φ, tion can approximate any function”. In: Neural Networks 6.6 (1993), pp. 861–867.
andσ itself must be (almost everywhere) a polynomial of degree at most m. [3] Walter Rudin. Principle of Mathematical Analysis. McGraw-Hill, Inc., 1964, p. 159.
 A N I NTRODUCTION TO C RYPTOGRAPHY
Lainey Watlington
University of California - Santa Barbara

The Beginnings of Cryptography An Example in Python The Foundations of Modern Cryptography:

Elliptic Curves
Cryptography is the study of methods of sending messages in a Let us extend the idea of a digraph to a cryptosystem which enciphers a
disguised form so that only the intended recipients can remove the message of length n from an alphabet of any size. Let N represent the size
Elliptic Curve Cryptography
disguise and read the message. of the alphabet. Then, the enciphering function will be represented by
At the most basic level, a cryptosystem is the process of converting n−1 n−2 • An approach to public key cryptography which utilizes elliptic
N x1 + N x2 + · · · + N xn−1 + xn−1 = C curves over finite fields to create keys.
plaintext to a cyphertext using encryption and subsequently convert-
The Python code for an enciphering transformation of this form is as follows: • It is essentially impossible to find the discrete logarithm of a ran-
ing that cyphertext back to plaintext using decryption.
One of the earliest cryptosystems was created using digraphs, dom element of an elliptic curve with respect to a publicly known
which map two characters in a message to a number. Let us con- base point.
sider the 27 letter alphabet which contains letters A-Z and a blank. • The larger the elliptic curve, the more secure the cryptosystem is
Then, given any message, the following digraph can be used as an since the discrete logarithm becomes more difficult to compute.
enciphering function where x and y are two characters which occur An Elliptic Curve Over the Real Numbers
The deciphering transformation will subtract C mod N from C n times and
in succession in the message:
update C after each iteration. The Python code for a deciphering transfor-
27x + y = C mation of this form is:
The deciphering function is given by:
(
C mod 27 = x
C −x=y
Most early cryptosystems were based on a similar idea of using a
rule, or a key, to shift the letters in a message to a different location.
The idea was that only the person with the key would be able to
decipher the message. Primality and Factorization • Elliptic curves over the reals form an abelian group. Thus, if we
Breaking a Cryptosystem perform operations on two elements of the curve, we will end up
Cryptosystems have evolved over time to prevent people from breaking with another element on the curve.
them.
An Elliptic Curve Over the Complex Numbers
Cryptosystems were developed in order to help protect sensitive
• The easier it is to guess the enciphering key of a cryptosystem, the easier
information. In modern times, cryptography is widely used in the
it is to break the cryptosystem.
field of cybersecurity to protect people’s
• So, methods of creating difficult to guess keys were developed
• passwords
Public Key Cryptography: the enciphering and deciphering algorithms are
• credit card information
publicly known, but the enciphering and deciphering keys are concealed.
• identity information
Gaining access to the keys allows you to break the system.
• other sensitive forms of data
In an increasingly digital world, cryptosystems have become ex- How do we create difficult to guess keys?
tremely important in protecting this information. • Factoring primes is really difficult once we start dealing with very large
numbers. So, if we multiply two large primes together, factoring them be- • Elliptic curves over the complex numbers form a torus.
Cryptanalysis is the science of "breaking" the code of cryptosys- comes almost impossible without having access to a key. • We can think of plotting elements of the curve over the integer
tems. People do this in order to gain access to data that is not • The discrete logarithm problem is an idea based on the fact that if we lattice and then connecting all of the edges together.
intended for them. This begs the question, "How does one break a know y = bx, it is extremely difficult to solve for
cryptosystem?". To do so, one needs two types of information
x = logb y Acknowledgements
1. The general nature, or the structure of the system
Fermat Factorization provides a way of "breaking" some public key cryp-
2. The specific choice of certain parameters connected with the tosystems. If two primes are close enough together, this algorithm allows Reference Material: "A Course in Number Theory and Cryptogra-
given cryptosystem, like the shift parameter, also known as the one to efficiently calculate the two primes that have been multiplied together. phy" by Neal Koblitz
enciphering key This form of factorization is used to break RSA cryptosystems. Thank you to the UCSB Directed Reading Program and to my men-
tor Katherine Merkl for making this project possible.
 Isoperimetric Inequalities
Tyler Guo Mentor: Malik Tuerkoen
2022 Mathematics Directed Reading Program. University of California-Santa Barbara

Introduction Non-optimal Isoperimetric Inequality Coarea Formula for BV functions

The classical isoperimetric problem is stated as follows: Among all closed curves in the Let N ≥ 2. Then we have √ Let f ∈ BV (Ω) be a nonnega ve func on, and put
N −1
plane of fixed perimeter, which curve (if any) maximizes the area of its enclosed region? P (E) ≥ 2 N |E| N
Et := {x ∈ Ω : f (x) > t} for t ≥ 0.
This is equivalent to the problem: Among all closed curves in the plane enclosing a
for all measurable subsets E ⊂ RN with |E| < ∞. Then
fixed area, which curve (if any) minimizes the perimeter? The problem can be extended Z ∞
Var(f, Ω) = P (Et, Ω) dt.
to regions and surfaces in Rn. In this poster, we show that a sphere has the smallest 0
surface area with given volume by developing certain isoperimetric inequali es rela ng
Proof of the theorem
to the Ln measure of a sets and its perimeter.
The inequality holds trivially if P (E) = ∞. Suppose P (E) < ∞. We claim that for Optimal Functional Isoperimetric Inequality
Definitions u ∈ BV (RN ), N ≥ 2, we have
For f ∈ BV (RN ) we have
1
1
kuk NN−1 (RN ) ≤ √ Var(u, RN ). N
Var(f, R ) ≥ N ωN kf k
1
N
.
(1) For a func on u ∈ L (Ω, R), we define L 2 N N
L N −1
We obtain the inequality by applying this result to the func on 1E .
Z
1 N
Var(u, Ω) := sup{ u div ϕ dx : ϕ ∈ Cc (Ω, R ), kϕk∞ ≤ 1}.
Ω Proof of Claim: By standard approxima on arguments, one can show that there exists
We say u has bounded varia on in Ω if Var(u, Ω) < ∞. a sequence (un) such that un ∈ BV (RN ) ∩ Cc1(RN ) sa sfying Proof of the theorem
Moreover, we let BV (Ω) denote the space of func ons u ∈ L1(Ω) which have ku − unk1 → 0, Var(un, R ) → Var(u, R ). N N
bounded varia on in Ω. Since
We also set Hence it suffices to consider u ∈ Cc1(RN ). Integra on parallel to the j-th coordinate axis
−
yields kf k N ≤ kf +k N + f N ,
L N −1 L N −1 L N −1
BVloc(Ω) := {u ∈ L1loc(Ω) : Var(u, Ω0) < ∞ for every Ω0 ⊂⊂ Ω}. 1 Z
|u(x)| ≤ |∂j u(x1, . . . , xj−1, t, xj+1, . . . , xN )| dt := vj (x̂j ) and one can show that for f ∈ BV (RN ),
(2) For a Lebesgue measurable subset E of RN . The perimeter of E in Ω is defined by 2 R
Var(f, RN ) = Var(f +, R) + Var(f −, R).
P (E, Ω) := Var(1E , Ω). for x ∈ RN , j = 1, . . . , N .
1 Hence it suffice to consider the case where f is nonnega ve. In this case, the Coarea
We say E has finite perimeter in Ω if 1E ∈ BV (Ω); E has locally finite perimeter in Ω We then apply the Gagliardo’s Lemma to vj N −1
∈ LN −1(RN −1) and obtain formula and the isoperimetric inequality yields
if 1E ∈ BVloc(Ω). Z ∞ 1 Z ∞
N −1
N
N 1 N 1
N Var(f, R ) = P (Et) dt ≥ N ωN N
|Et| N dt.
0 0
Z Z
N Y Y Y 1
N −1 v N −1
|u(x)| dx ≤ vj (x̂j ) dx ≤ kvj kL1(RN −1))

N −1 =( N −1
Examples We now define
j
RN RN j=1 j=1 LN −1(RN −1) j=1
χ : [0, ∞) → R, χ(t) = kmin{f, t}k N .
L N −1
1 N N 1 Z N N 1 Z
N
≤( √
X X
(1) The distribu on func on ≤( kvj kL1(RN −1)) N −1 =( [ |∂j u(x)|] dx) N −1 |∇u| dx) N −1 . Then χ is con nuous, nondecreasing and hence a.e. differen able. Moreover, for t, h >
N j=1 2N R N
j=1 2 N R N

Fµ : R → R, Fµ(t) = µ((−∞, t]) 0, we have

N −1
of a probability measure µ on B(R) is a func on of bounded varia on in R. 0 ≤ χ(t + h) − χ(t) ≤ kmin{f, t + h} − min{f, t}k N ≤ k1Et hk N = h|Et| N ,
L −1
N L −1
N

(2) Suppose Ω ⊂ RN is bounded and u ∈ C 1(Ω). Then u ∈ BV (Ω) and which implies that χ is locally Lipschitz con nuous on (0, ∞) with χ (t) ≤ |Et| for 0 N −1
N

Optimal Isoperimetric Inequality

Z
Var(u, Ω) = |∇u| dx. a.e. t > 0. Hence χ sa sfies the assump ons of the Fundamental theorem of calculus .
Ω
Since 0 = χ(0) = lim+ χ(t), it follows that
N N
(3) Let QN := [0, 1] ⊂ R . Then Q has finite perimeter in R given by N For any measurable subset E ⊂ RN with |E| < ∞ we have t→0
1 N −1 1 Z b
0
Z ∞
N −1
P (QN ) = 2N. P (E) ≥ N ωN |E| N N , kf k N = lim [χ(b) − χ( )] = lim 1 χ (t) dt ≤ |Et| N dt.
L −1
N b→∞ b b→∞ b 0
(4) Let E ⊂ RN be a bounded open set with C 1-boundary. Then E has locally finite where ωN denotes the volume of the unit ball in RN , and the equality occurs if and
perimeter in Ω given by only if E is a ball.
P (E, Ω) = volN −1(∂E ∩ Ω). Acknowledgements
Proof of the theorem I would like to thank my mentor Malik Tuerkoen for his guidance and insight. I would also
Gagliardo’s Lemma like to thank the organizers of the UCSB Directed Reading Program for this wonderful
Suppose P (E) < ∞, we have 1E ∈ BV (RN ). Let E ∗ = Br (0), where r is chosen
Let N ≥ 2. For x ∈ RN , j = 1, . . . , N let xˆj := (x1, . . . , xj−1, xj+1, . . . , xN ). Moreover, such that |E| = |E ∗|. Then one can show there exists a sequence of sets (En) with opportunity.
let f1, . . . , fN ∈ LN −1(RN −1) be given, and let f : RN → R be defined by f (x) = P (En) ≤ P (E) and k1En − 1E ∗ k1 → 0. By lower semicon nuity,
f1(xˆ1) · · · fN (xˆN ). Then ∗
P (E ) ≤ lim inf P (E ) = lim inf Var(1 , R ) ≤ Var(1 , R ) = P (E).N N References
n→∞ n n→∞ E n E
N
f ∈ L1(RN ) and kf kL1(RN ) ≤
Y
kfj kLN −1(RN −1) . Moreover,
j=1 ∗ ∗ N −1
1
∗ N −1 1 N −1 [1] Lawrence Craig Evans and Ronald F Gariepy. Measure Theory and Fine Proper es of Func ons, Revised Edi on. CRC Press
P (E ) = volN −1(∂E ) = N ωN r = N ωN |E | N N = N ωN |E|N N . LLC, Oakville, 2015.