Unique Factorization

6 Jan 2022


§ Motivation

Number theory is most easily described as the study of the natural numbers.

Some open problems in number theory include the twin prime conjecture, whether there exists an infinite number of prime pairs such that their difference is 2, or Goldbach's conjecture, whether every number can be written as the sum of two primes.

§ Unique Factorization in Z\mathbb{Z}

§ Integer Ring

Z\mathbb{Z}

Z={0,1,1,2,2,3,3,}\mathbb{Z} = \{0, -1, 1, -2, 2, -3, 3, \dots\} is the ring of integers with the usual definition of the sum and product operations.

  • If pp positive and prime, then p-p is also prime.
  • Every nonzero integer divides zero.
  • No number can be divided by zero.

§ Factorization

The division operator over Z\mathbb{Z} yields the following properties,

  1. a0,aa\forall a \neq 0, a | a (reflexive),
  2. If aba | b and ba,b | a, then a=±ba = \pm b (symmetric),
  3. If aba | b and bc,b | c, then aca | c (transitive),
  4. If aba | b and ac,a | c, then ab+c.a | b + c.

Let nZ{0}n \in \mathbb{Z}\setminus\{0\} and let pp be prime, then e,penpa+1n.\exists e, p^e | n \wedge p^{a+1} \nmid n.

Order

If p,e,nZ+,p, e, n \in \mathbb{Z}^{+}, then a,pe>n.\exists a, p^e \gt n.

We say that ee is the order of nn at pp and write ordpn=a.\text{ord}_p n = a.

  • Roughly, ee is the number of times that pp divides n.n.
  • For n=0,n = 0, we set ordp0=.\text{ord}_p 0 = \infty.
  • ordpn=0    pn.\text{ord}_p n = 0 \iff p \nmid n.

Every nonzero integer can be written as a product of primes.

Sps. there exists an integer NN such that it is the smallest positive integer that cannot be expressed as the product of primes. It follows that NN cannot be prime itself, so there must exist 1<m,n<N1 \lt m, n \lt N such that N=mn.N = mn. Notice that because m,nZ+m, n \in \mathbb{Z}^{+} and strictly smaller than N,N, m,nm, n must each be a product of primes. Contradiction. If m,nm, n are products of primes, then N=mnN = mn must be as well.

Factorization

We can collect prime terms and we can write n=p1e1p2e2prer,n = p_1^{e_1}p_2^{e_2}\dots p_r^{e_r}, or

n=(1)ε(n)ppa(p),n = (-1)^{\varepsilon(n)} \prod_p p^{a(p)},

where ε(n)\varepsilon(n) yields 0 or 1 depending on the sign of n,n, and the product is over all positive primes. The exponents a(p)a(p) are nonnegative integers, and a(p)=0a(p) = 0 for a countably infinite number of primes.

For every nonzero integer nn there exists a prime factorization uniquely determined by n,n,

n=(1)ε(n)ppa(p),n = (-1)^{\varepsilon(n)} \prod_p p^{a(p)},

where a(n)=ordpn.a(n) = \text{ord}_p n.

§ Greatest Common Divisor

If a,bZa, b \in \mathbb{Z} and b>0,b > 0, there exists q,rZq, r \in \mathbb{Z} such that a=qb+ra = qb + r with 0r<b.0 \leq r \lt b.

Consider all integers of the form axba - xb where xZ.x \in \mathbb{Z}. Let r=aqbr = a - qb be the least nonnegative element in this set. We claim that 0r<b.0 \leq r \lt b. If not, r=aqbbr = a - qb \geq b and so 0a(q+1)b<r.0 \leq a - (q + 1)b \lt r. Then rr is not the minimal nonnegative element, contradiction.

Ideals on Z\mathbb{Z}

If a1,a2,,anZ,a_1, a_2, \dots, a_n \in \mathbb{Z}, then define A=(a1,a2,,an)A = (a_1, a_2, \dots, a_n) to be the set of all integers of the form a1x1+a2x2++anxna_1x_1 + a_2x_2 + \dots + a_nx_n with x1,x2,,xnZ.x_1, x_2, \dots, x_n \in \mathbb{Z}.

Notice that the sum and difference of any two elements in AA is also in A.A. If aAa \in A and rZ,r \in \mathbb{Z}, then raA.ra \in A.

Then, AA is an ideal of ring Z.\mathbb{Z}.

If a,bZ,a, b \in \mathbb{Z}, then there exists dZd \in \mathbb{Z} such that (a, b) = (d).

Assume wlog. that aa and bb are not both zero, so that there are positive elements in (a,b).(a, b).

(\Rightarrow) Let dd be the smallest positive element in (a,b).(a, b). Clearly, (d)(a,b).(d) \subseteq (a, b).

(\Leftarrow) Sps. that c(a,b).c \in (a, b). By Lemma 2, there exists integers qq and rr such that c=qd+rc = qd + r with 0r<d.0 \leq r \lt d. Since both c,d(a,b)c, d \in (a, b) it follows that r=cqd(a,b).r = c - qd \in (a, b). And 0r<d0 \leq r \lt d implies r=0.r = 0. Therefore, c=qd(d).c = qd \in (d).

Greatest Common Divisor

Sps. a,b,dZ,a, b, d \in \mathbb{Z}, if dad | a and dbd | b and every other common divisor of aa and bb divides d,d, then we say that dd is the greatest common divisor (gcd) of aa and bb and write (a,b)=d.(a, b) = d.

If cc is another greatest common divisor, distinct from d,d, then we must have cdc | d and dc,d | c, so c=±d.c = \pm d. If there exists a gcd between two numbers, then it is determined up to sign.

Let a,bZ,a, b \in \mathbb{Z}, if (a,b)=(d),(a, b) = (d), then dd is the greatest common divisor of aa and b.b.

Since a(d)a \in (d) and b(d),b \in (d), we see that dad | a and db.d | b. Sps. cc is a common divisor. Then cc divides every number of the form ax+by,ax + by, particularly cd.c | d.

§ Coprimality

Coprime

Sps. a,bZa, b \in \mathbb{Z} such that (a,b)=±1.(a, b) = \pm 1. We say that aa and bb are coprime (or relatively prime)

Proposition 1

Sps. abca | bc and that (a,b)=1,(a, b) = 1, then ac.a | c.

Since a,a, bb coprime, there exists integers rr and ss such that ra+sb=1.ra + sb = 1. Therefore, rac+sbc=c.rac + sbc = c. Since aa divides the left hand side, we get that ac.a | c.

Corollary 1

If pp prime and pbc,p | bc, then either pbp | b or pc.p | c.

The divisors of pp are ±1\pm 1 and ±p.\pm p. Then (p,b){1,p}(p, b) \in \{1, p\} i.e. either p,bp, b coprime or pb,p | b, in which case we are done. Otherwise, if (p,b)=1,(p, b) = 1, then by proposition 1 we get that pc.p | c.

Then, if pp prime, pbp \nmid b and pc,p \nmid c, then pbc.p \nmid bc.

Corollary 2

Sps. pp prime and that a,bZ,a, b \in \mathbb{Z}, then ordp(ab)=ordpa+ordpb.\text{ord}_p (ab) = \text{ord}_p a + \text{ord}_p b.

Let α=ordpa\alpha = \text{ord}_p a and β=ordpb.\beta = \text{ord}_p b. Then a=pαca = p^{\alpha}c and b=pβd,b = p^{\beta}d, where pcp \nmid c and pd.p \nmid d. Then ab=pα+βcdab = p^{\alpha + \beta}cd and pcdp \nmid cd by corollary 1. Therefore, ordpab=α+β=ordpa+ordpb.\text{ord}_p ab = \alpha + \beta = \text{ord}_p a + \text{ord}_p b.

§ Unique Factorization Proof

Finally, the proof to Theorem 1. If we apply ordp\text{ord}_p to both sides of

n=(1)ε(n)ppa(n),n = (-1)^{\varepsilon(n)} \prod_p p^{a(n)},

and utilize the property of ordq\text{ord}_q given in corollary 2, we get

ordqn=ε(n)ordq(1)+pa(p)ordqp.\text{ord}_q n = \varepsilon(n) \text{ord}_q(-1) + \sum_p a(p)\,\text{ord}_q p .

Then, by definition of ordq,\text{ord}_q, we get ordq(1)=0\text{ord}_q (-1) = 0 and ordqp=0\text{ord}_q p = 0 if p,qp, q are distinct and 11 otherwise.

Then, the right hand side can be simplified into a(q),a(q), i.e. ordqn=a(q).\text{ord}_q n = a(q).

§ Unique Factorization in k[x]k[x]

§ Polynomial Ring

The unique factorization theorem can be generalized into other rings, not just the integers.

Polynomial Ring

The ring k[x]k[x] are polynomials with coefficients in a field k.k.

§ Divisibility in k[x]k[x]

  • If f,gk[x],f, g \in k[ x ], then we say fgf | g iff hk[x],g=fh.\exists h \in k[ x ], g = fh.
  • If degf\deg f denotes the degree of f,f, then degfg=degf+degg.\deg fg = \deg f + \deg g.
  • degf=0\deg f = 0 iff ff is a nonzero constant.
  • fgf | g and gfg | f iff f=cgf = cg where cc is a nonzero constant.

Units

The only polynomials that divide all others are the nonzero constants, i.e. the units of k[x]k[ x ]

Irreductible Polynomials

A nonconstant polynomial pp is said to be irreductible iff qpq | p implies qq is either a constant or a constant times p.p.

These are analogous to the prime numbers in Z.\mathbb{Z}.

§ Factorization

Every nonconstant polynomial is the product of irreductible polynomials.

It is clear that polynomials with degree 11 are irreductible. Assuming that all polynomials with degree-bound nn are the product of irreductible polynomials, and that degf=n.\deg f = n. If ff is irreductible, we are done. Otherwise, f=gh,f = gh, where qdegg,degh<n.q \leq \deg g, \deg h \lt n. By the induction hypothesis gg and hh are themselves products of irreductible polynomials. Therefore, ff is also the product of irreductible polynomials as f=gh.f = gh.

Monic Polynomial

A polynomial ff is monic if its leading coefficient is 1.

Every polynomial, save for zero, is a constant times a monic polynomial.

Order in k[x]k[ x ]

Let pp be a monic irreductible polynomial. We define ordpf\text{ord}_p f to be the integer aa defined by the property that pafp^a | f but that pa+1f.p^{a+1} \nmid f. Such an integer must exist since the degree of powers of pp is monotonically increasing. Notice that ordpf=0    pf.\text{ord}_p f = 0 \iff p \nmid f.

Sps. fk[x],f \in k[ x ], then

f=cppa(p),f = c \prod_p p^{a(p)},

where the product is over all the monic irreductible polynomials, and cc is a constant.

The constant cc and the exponents a(p)a(p) are uniquely determined by f,f, and a(p)=ordpf.a(p) = \text{ord}_p f.

§ Greatest Common Divisor

Let f,gk[x].f, g \in k[ x ]. If g0,g \neq 0, then there exists h,kk[x]h, k \in k[ x ] such that f=hg+r,f = hg + r, where r=0r = 0 or r0r \neq 0 and degr<degg.\deg r \lt \deg g.

If gf,g | f, then h=f/gh = f / g and r=0.r = 0. Otherwise, let r=fhgr = f - hg be the polynomial of least degree among all polynomials of the form flgf - lg with lk[x].l \in k[ x ]. If degrdegg,\deg r \geq \deg g, then let the leading term of rr be axdax^d and that of gg be bxm.bx^m. Then rablxdmg=f(h+ablddm)gr - ab^{-l}x^{d-m}g = f - (h + ab^{-l}d^{d - m})g has degree less than r.r. Contradiction.

Ideals on k[x]k[ x ]

If f1,f2,,fnk[x],f_1, f_2, \dots, f_n \in k[ x ], then (f1,f2,,fn)(f_1, f_2, \dots, f_n) is the set of all polynomials of the form f1h1+f2h2++fnhn,f_1h_1 + f_2h_2 + \dots + f_nh_n, where h1,h2,,hnk[x].h_1, h_2, \dots, h_n \in k[ x ].

Then, (f1,f2,,fn)(f_1, f_2, \dots, f_n) is the ideal generated by f1,f2,,fn.f_1, f_2, \dots, f_n.

Given f,gk[x],f, g \in k[ x ], there exists dk[x]d \in k[ x ] such that (f,g)=(d).(f, g) = (d).

Sps. d(f,g)d \in (f, g) and it is an element of least degree.

(\Rightarrow) Clearly, (d)(f,g).(d) \subseteq (f, g).

(\Leftarrow) Let c(f,g).c \in (f, g). If dc,d \nmid c, then there exists polynomials h,rh, r such that c=hd+rc = hd + r with degr<degd.\deg r \lt \deg d. Since c,d(f,g),c, d \in (f, g), we get r=chd(f,g).r = c - hd \subseteq (f, g). Contradiction, rr has degree less than d.d. Therefore, dcd | c and c(d).c \in (d).

Greatest Common Divisor

Sps. f,gk[x].f, g \in k[ x ]. Then dk[x]d \in k [ x ] is said to be the greatest common divisor of ff and gg iff dfd | f and dgd | g and every other common divisor of ff and gg divides d.d.

The greatest common divisor of two polynomials is determined up to multiplication by a constant. If we require the gcd to be monic, then it is uniquely determined which is what we mean when we say the greatest common divisor.

Let f,gk[x].f, g \in k[ x ]. By Lemma 3, there exists dk[x]d \in k[ x ] such that (f,g)=(d).(f, g) = (d). Then, dd is a gcd of ff and g.g.

Since f(d)f \in (d) and g(d),g \in (d), we have dfd | f and dg.d | g. Sps. that hfh | f and hg.h | g. Then hh divides every polynomial of the form fl+gmfl + gm with l,mk[x],l, m \in k[ x ], particularly hd.h | d.

§ Coprimality

Coprime

f,gk[x]f, g \in k[ x ] are said to be coprime (or relatively prime) iff the common divisors of ff and gg are constants, i.e. (f,g)=(1).(f, g) = (1).

Proposition 2

If ff and gg coprime, we have that (f,g)=(1)(f, g) = (1) so there are polynomials l,ml, m such that lf+mg=1.lf + mg = 1. Thus, lfh+mgh=h.lfh + mgh = h. Since ff divides the left hand side, fh.f | h.

Corollary 1

If pp is an irreductible polynomial and pfg,p | fg, then pfp | f or pg.p | g.

Since pp is irreductible, either (p,f)=(p)(p, f) = (p) or (p,f)=(1).(p, f) = (1). In the former, pf,p | f, q.e.d.

In the latter case, p,fp, f coprime, and pfp | f follows from proposition 2.

Corollary 2

If pp is a monic irreductible polynomial and f,gk[x],f, g \in k[ x ], we have ordpfg=ordpf+ordpg.\text{ord}_p fg = \text{ord}_p f + \text{ord}_p g.

Refer to proof to corollary 2 for prop. 1.

§ Unique Factorization Proof

Finally, we have the tools to prove theorem 2. If we apply ordq\text{ord}_q to both sides of

f=cppa(p).f = c\prod_p p^{a(p)}.

We get

ordqf=ordqc+pa(p)ordqp.\text{ord}_q f = \text{ord}_q c + \sum_p a(p)\,\text{ord}_q p.

Clearly qcq \nmid c as cc is a constant, so ordqc=0.\text{ord}_q c = 0. Then ordqp=0\text{ord}_q p = 0 if qpq \neq p and 11 otherwise. Finally, we get that ordqf=a(q).\text{ord}_q f = a(q). This shows that the exponents are uniquely determined by ff and so is c.c.

§ Unique Factorization in a Principal Ideal Domain

The proof methods for unique factorization in Z\mathbb{Z} and k[x]k[ x ] are very similar. We can further generalize this notion of unique factorization. In this section, we will denote an integral domain with R.R.

§ Euclidean Domain

Euclidean Domain

RR is an Euclidean domain iff there exists a function λ:R{0}N{0},\lambda: R \setminus \{0\} \to \mathbb{N} \cup \{0\}, such that if a,bR,b0,a, b \in R, b \neq 0, there exists c,dRc, d \in R with the property a=cb+da = cb + d and either d=0d = 0 or λ(d)<λ(b).\lambda(d) \lt \lambda(b).

The rings Z\mathbb{Z} and k[x]k[ x ] are examples of Euclidean domains. For Z\mathbb{Z} we can consider λ\lambda to be the absolute value function, in k[x]k[ x ] λ\lambda is the function that assigns to every polynomial its degree.

Proposition 3

If RR is an Euclidean domain, and IRI \subseteq R is an ideal, then there exists aRa \in R such that I=Ra={rarR}. I = Ra = \{ra\,|\,r \in R\}.

Consider the nonnegative integers {λ(b)bI,b0}.\{\lambda(b)\,|\,b\in I, b \neq 0\}.

Since every set of nonnegative integers has a least element there is an aI{0}a \in I \setminus \{0\} such that bI{0},λ(a)λ(b).\forall b \in I \setminus \{0\}, \lambda(a) \leq \lambda(b).

We've claimed that I=Ra,I = Ra, clearly RaI.Ra \subseteq I. Sps. that bI,b \in I, then there exists c,dRc, d \in R such that b=ca+d,b = ca + d, where d=0d = 0 or λ(d)<λ(a).\lambda(d) \lt \lambda(a). Since d=bcaI,d = b - ca \in I, the inequality λ(d)<λ(a)\lambda(d) \lt \lambda(a) cannot be satisfied; so d=0d = 0 and b=caRa.b = ca \in Ra. Therefore, IRa.I \subseteq Ra.

§ Principal Ideal Domain

For elements a1,a2,,anR,a_1, a_2, \dots, a_n \in R, we define (a1,a2,,an)=Ra1+Ra2++Ran={i=1nriairiR}.(a_1, a_2, \dots, a_n) = Ra_1 + Ra_2 + \dots + Ra_n = \left\{\sum_{i=1}^n r_ia_i\,|\,r_i \in R\right\}. Then, (a1,a2,,an)(a_1, a_2, \dots, a_n) is an ideal.

If an ideal II is equal to (a1,a2,,an)(a_1, a_2, \dots, a_n) for some elements aiI,a_i \in I, we say that II is finitely generated. If I=(a)I = (a) for some aI,a \in I, we say that II is a principal ideal.

Principal Ideal Domain

If every ideal of RR is principal, then we say that RR is a principal ideal domain (PID).

Every Euclidean domain is a PID, but not every PID is an Euclidean domain, i.e. PID \supset Euclidean domain. However finding counterexamples to PID == Euclidean domain is somewhat hard. In practice, it is useful to show that a given ring is an Euclidean domain, which gives us the added bonus of showing that it is a PID.

§ Terminology

Division

If a,bR,b0,a, b \in R, b \neq 0, then we say that bb divides aa if cR,a=bc\exists c \in R, a = bc and write ba.b | a.

In terms of ideals, ab    (b)(a).a | b \iff (b) \subseteq (a).

Associate

a,bRa, b \in R are associates if a=bua = bu for some unit u.u.

A unit is defined as uRu \in R such that (u)=R.(u) = R. And in terms of ideals, aa and bb are associates iff (a)=(b).(a) = (b).

Irreducibility

An element pRp \in R is irreductible if apa | p implies that aa is either a unit or an associate of p.p.

Prime

A nonunit pR{0}p \in R \setminus\{0\} is said to be prime if pabp | ab implies that pap | a or pb.p | b.

In terms of ideals, pp prime iff ab(p)    a(p)b(p).ab \in (p) \implies a \in (p) \lor b \in (p).

We now make distinct the notion of irreductible and prime elements. These notions do not generally coincide, unlike in Z\mathbb{Z} and k[x]k[ x ] and later we will show that they coincide in a PID.

§ Greatest Common Divisor

Greatest Common Divisor

dRd \in R is said to be a (gcd) of two elements a,bRa, b \in R if

  1. dad | a and db,d | b,
  2. dadb    dd.d' | a \wedge d' | b \implies d' | d.

If both dd and dd' are gcds of aa and b,b, then dd is associate to d.d'.

Proposition 4

Let RR be a PID and a,bR.a, b \in R. Then aa and bb have a gcd and (a,b)=(d).(a, b) = (d).

Form the ideal (a,b).(a, b).

Since RR is a PID there exists dd such that (a,b)=(d).(a, b) = (d). Since (a)(d)(a) \subseteq (d) and (b)(d)(b) \subseteq (d) we have that dad | a and db.d | b.

If dad' | a and db,d' | b, then (a)(d)(a) \subseteq (d') and (b)(d).(b) \subseteq (d'). Then, (d)=(a,b)(d)(d) = (a, b) \subseteq (d') and dd.d' | d. Therefore (a,b)=(d).(a, b) = (d).

§ Coprimality

Coprime

aa and bb are coprime iff their common divisors are units.

Corollary 1

If RR is a PID and a,bRa, b \in R are coprime, then (a,b)=R.(a, b) = R.

Corollary 2

If RR is a PID, and pRp \in R is irreductible, then pp prime.

Sps. that pabp | ab and that pa.p \nmid a. Since pap \nmid a it follows that they only share units as common divisors. By corollary 1 (a,p)=R,(a, p) = R, so (ab,pb)=(b).(ab, pb) = (b). Since ab(p)ab \in (p) and pb(p)pb \in (p) we have (b)(p).(b) \subseteq (p). Therefore pbp | b and in a PID, prime elements     \iff irreductible elements.

In a PID we will use prime and irreductible interchangeably.

§ Order

Let (a1)(a2)(a3)(a_1) \subseteq (a_2) \subseteq (a_3) \subseteq \dots be an ascending chain of ideals. Then there exists kNk \in \mathbb{N} such that (ak)=(ak+l)(a_k) = (a_{k+l}) for l=0,1,2,.l = 0, 1, 2, \dots. That is, the chain breaks in finitely many steps.

Let I=i=1(ai),I = \bigcup_{i=1}^{\infty} (a_i), obviously II is an ideal so I=(a)I = (a) for some aR.a \in R. But ai=1(ai)a \in \bigcup_{i=1}^{\infty} (a_i) implies that a(ak)a \in (a_k) for some k.k. This shows that I=(a)(ak).I = (a) \subseteq (a_k). It then follows that I=(ak)=(ak+l)=I = (a_k) = (a_{k + l}) = \dots

Proposition 5

Every nonzero nonunit of RR is a product of irreductibles.

Let aR{0},a \in R \setminus \{0\}, where aa is a nonunit, then we want to prove the following cases.

  1. aa is divisible by irreductibles

    In the case that aa itself is irreductible, we are done. Otherwise, a=a1b1a = a_1b_1 where a1,b1a_1, b_1 are nonunits. If a1a_1 irreductible, done. Otherwise a1=a2b2a_1 = a_2b_2 where a2,b2a_2, b_2 are nonunits. If a2a_2 irreductible, done. Otherwise, continue as before.

    Notice that (a)(a1)(a2)(a) \subset (a_1) \subset (a_2) \subset \dots and by lemma 5 we have that this is a finite chain, so there must exist some kk such that aka_k is irreductible.

  2. aa is the product of irreductibles

    If aa is irreductible, done. Otherwise let p1p_1 be some irreductible such that p1a.p_1 | a. Then a=p1c1.a = p_1c_1. If c1c_1 is a unit, done. Otherwise, continue as before. Again, notice that (a)(c1)(c2),(a) \subset (c_1) \subset (c_2) \subset \dots, which is a finite chain by lemma 5. There must exist some kk for which a=p1p2pkck,a = p_1p_2\dots p_k c_k, where ckc_k a unit and pkckp_kc_k is irreductible.

We use the following lemma to construct the ord\text{ord} function akin to those for Z\mathbb{Z} and k[x].k[ x ].

Let pp be prime and a0.a \neq 0. Then there exists nZn \in \mathbb{Z} such that pnap^{n} | a but pn+1a.p^{n + 1} \nmid a.

Sps. that there does not exist an nn satisfying the lemma, then for each integer m>0m > 0 there would be an element bmb_m such that a=pmbm.a = p^mb_m. Then pbm+1=bmpb_{m+1} = b_m so that (b1)(b2)(b3)(b_1) \subset (b_2) \subset (b_3) \subset which would be an infinite chain which contradicts lemma 5.

Then the integer nn defined in lemma 6 is uniquely determined by pp and a,a, and we set n=ordpa.n = \text{ord}_p a.

§ Unique Factorization Proof

If a,bR{0},a, b \in R \setminus \{0\}, then ordp(ab)=ordpa+ordpb.\text{ord}_p (ab) = \text{ord}_p a + \text{ord}_p b.

Let α=ordpa\alpha = \text{ord}_p a and β=ordpb.\beta = \text{ord}_p b. Then a=pαca = p^{\alpha}c and b=pβdb = p^{\beta}d with pcp \nmid c and pd.p \nmid d. Thus ab=pα+βcd.ab = p^{\alpha + \beta}cd. Since pp prime, pcd.p \nmid cd. Therefore ordpab=α+β=ordpa+ordpb\text{ord}_p ab = \alpha + \beta = \text{ord}_p a + \text{ord}_p b

Let SRS \subset R be a set of primes with the properties

  1. pRsR,\forall p \in R \exists s \in R, such that pp and ss are associates.
  2. a,bS\nexists a, b \in S such that aa and bb are associates.

To generate the set SS we can choose a representative prime from each class of associate primes. In Z\mathbb{Z} we can construct the set SS by choosing all positive primes. In k[x]k[ x ] we chose SS to be the set of monic irreductible polynomials.

There does not exist a general way to construct SS for an arbitrary PID.

Let RR be a PID and SS a set of primes with the aforementioned properties.

Then if aR{0},a \in R \setminus \{0\}, then we can write

a=uppe(p),(†)a = u \prod_p p^{e(p)}, \tag{\dag}

where uu is a unit and the product is over all pS.p \in S. The unit uu and the exponents e(p)e(p) are uniquely determined by a.a. In fact, e(p)=ordpq.e(p) = \text{ord}_p q.

The existence of such a decomposition follows from proposition 5.

To show the uniqueness, let qSq \in S and apply ordq\text{ord}_q to both sides of eq. (\dag). Using lemma 7, we get

ordqa=ordqu+pe(p)ordqp.\text{ord}_q a = \text{ord}_q u + \sum_p e(p)\,\text{ord}_q p.

By definition of ordq\text{ord}_q we get that ordqu=0\text{ord}_q u = 0 and that ordqp=0\text{ord}_q p = 0 if qpq \neq p and 11 otherwise. So ordqa=e(q).\text{ord}_q a = e(q). Since exponents e(q)e(q) are uniquely determined, so is the unit u.u.

§ Z[i]\mathbb{Z}[i] and Z[ω]\mathbb{Z}[\omega]

§ Gaussian Integers

Gaussian Integers

Let i=1i = \sqrt{-1} and consider Z[i]C\mathbb{Z}[i] \subset \mathbb{C} defined by {a+bia,bZ}.\{a + bi\,|\,a,b \in \mathbb{Z}\}.

Clearly, Z[i]\mathbb{Z}[i] is closed under addition and attraction.

Also, if a+bi,c+diZ[i],a + bi, c+ di \in \mathbb{Z}[i], then (a+bi)(c+di)=ac+adi+bci+bdi2=(acbd)+(ad+bc)iZ[i].(a + bi)(c + di) = ac + adi + bci + bdi^2 = (ac - bd) + (ad + bc)i \in \mathbb{Z}[i]. So Z[i]\mathbb{Z}[i] is closed under multiplication and is a ring.

Since Z[i]C,\mathbb{Z}[i] \subset \mathbb{C}, it is an integral domain as well.

Proposition 6

Z[i]\mathbb{Z}[i] is an Euclidean domain.

For a+biQ[i]a + bi \in \mathbb{Q}[i] define λ(a+bi)=a2+b2.\lambda(a + bi) = a^2 + b^2.

Let α=a+bi\alpha = a + bi and γ=c+di\gamma = c + di and sps. γ0.\gamma \neq 0. α/γ=r+si,\alpha/\gamma = r + si, where r,sQ.r, s \in \mathbb{Q}. Choose integers m,nZm,n \in \mathbb{Z} such that rm12|r - m| \leq \frac{1}{2} and sn12.|s - n| \leq \frac{1}{2}. Set δ=m+ni.\delta = m + ni. Then δZ[i]\delta \in \mathbb{Z}[i] and λ((α/γ)δ)12λ(y)<λ(γ).\lambda((\alpha/\gamma) - \delta) \leq \frac{1}{2}\lambda(y) \lt \lambda(\gamma).

It follows that λ\lambda makes Z[i]\mathbb{Z}[i] an Euclidean domain.

§ Eisenstein Integers

The numbers ±1\pm 1 and ±i\pm i are the roots of x4=1x^4 = 1 over the complex numbers. Consider x3=1,x^3 = 1, since x31=(x1)(x2+x+1),x^3 - 1 = (x - 1)(x^2 + x + 1), the roots are 1,(1±3)/2.1, \left(-1 \pm \sqrt{-3}\right) / 2. Let ω=(1+3)/2,\omega = \left(-1 + \sqrt{-3}\right)/2, then ω2=(13)/2\omega^2 = \left(-1 - \sqrt{-3}\right)/2 and that 1+ω+ω2=0.1 + \omega + \omega^2 = 0.

Eisenstein Integers

Consider Z[ω]={a+bωa,bZ}.\mathbb{Z}[\omega] = \{a + b\omega\,|\, a, b \in \mathbb{Z}\}.

Clearly, Z[ω]\mathbb{Z}[\omega] is closed under addition and subtraction.

Also, (a+bω)(c+dω)=ac+(ad+bc)ω+bdω2=(acbd)+(ad+bcbd)ω.(a + b\omega)(c + d\omega) = ac + (ad + bc) \omega + bd\omega^2 = (ac - bd) + (ad + bc- bd) \omega. Thusly, Z[ω]\mathbb{Z}[\omega] is a ring.

Again, Z[ω]C\mathbb{Z}[\omega] \subset \mathbb{C} so it is an integral domain.

Additionally, Z[ω]\mathbb{Z}[\omega] is closed under complex conjugation. Since 3=3i=3i=3\overline{\sqrt{-3}} = \overline{\sqrt{3}\cdot i} = - \sqrt{3}\cdot i = -\sqrt{-3} we get that ω=ω2.\overline{\omega} = \omega^2. So if α=a+bωZ[ω],\alpha = a + b\omega \in \mathbb{Z}[\omega], then α=a+bω=a+bω2=(ab)bωZ[w].\overline{\alpha} = a + b\overline{\omega} = a + b\omega^2 = (a - b) - b\omega \in \mathbb{Z}[w].

Proposition 7

Z[ω]\mathbb{Z}[\omega] is an Euclidean domain.

For α=a+bωZ[ω]\alpha = a + b\omega \in \mathbb{Z}[\omega] define λ(α)=a2ab+b2.\lambda(\alpha) = a^2 - ab + b^2. Then, we can see that λ(α)=aa.\lambda(\alpha) = a\overline{a}.

Now, let α,βZ[ω]\alpha, \beta \in \mathbb{Z}[\omega] and sps. that β0.\beta \neq 0. Then α/β=αβ/ββ=r+sω,\alpha/\beta = \alpha\overline{\beta}/\beta\overline{\beta} = r + s\omega, where r,sQ.r, s \in \mathbb{Q}. We used the fact that ββ=λ(β)\beta\overline{\beta} = \lambda(\beta) is a positive integer and that αβZ[ω]\alpha\overline{\beta} \in \mathbb{Z}[\omega] since α,βZ[ω].\alpha, \overline{\beta} \in \mathbb{Z}[\omega].

Next we find mm and nn such that rm12|r - m| \leq \frac{1}{2} and sn12.|s - n| \leq \frac{1}{2}. Then let γ=m+nω.\gamma = m + n\omega. λ((α/β)γ)=(rm)2(rm)(sn)+(sn)214+14+14<1.\lambda((\alpha/\beta) - \gamma) = (r - m)^2 (r - m)(s - n) + (s - n)^2 \leq \frac{1}{4} + \frac{1}{4} + \frac{1}{4} \lt 1.

Let ρ=αγβ.\rho = \alpha - \gamma\beta. Then either ρ=0\rho = 0 or λ(p)=λ(β((α/β)γ))=λ(β)λ((α/β)γ)<λ(β).\lambda(p) = \lambda(\beta((\alpha/\beta) - \gamma)) = \lambda(\beta)\lambda((\alpha / \beta) - \gamma) \lt \lambda(\beta).

Then, Z[i]\mathbb{Z}[i] and Z[ω]\mathbb{Z}[\omega] are PIDs as well, which means the theorem of unique factorization is true.

§ Exercises

Exercise 1: Let aa and bb be nonzero integers. We can find nonzero integers qq and rr such that a=qb+r,a = qb + r, where 0r<b.0 \leq r \lt b. Prove that (a,b)=(b,r).(a, b) = (b, r).

Solution: Clearly aqb+r.a | qb + r. Let dd be the gcd of aa and b.b. Because da,d | a, we get dqb+r,d | qb + r, so dqbd | qb and dr.d | r.

As a result, any common divisor of aa and bb is also a divisor of r.r. Because r<b,r \lt b, there can be no common divisor between bb and rr greater than those of aa and b.b. Therefore, (a,b)=(b,r).(a, b) = (b, r).

Exercise 2: (this is a continuation to ex. 1) If r0,r \neq 0, we can find q1q_1 and r1r_1 such that b=q1r+r1b = q_1r + r_1 with 0r1<r.0 \leq r_1 \lt r. Show that (a,b)=(r,r1).(a, b) = (r, r_1). This process can be repeated. Show that it must end in finitely many steps. Show that the last nonzero remainder must equal (a,b).(a, b).

Solution: We've proved this in my cryptography class, see theorem 3.1 under the Euclidean algorithm.

Exercise 3 skipped.

Exercise 4: Let d=(a,b)d = (a, b) show that the Euclidean algorithm can be used to find mm and nn such that am+bn=d.am + bn = d.

Solution: The process is outlined under the Bézout Coefficients section from my cryptography notes.

Exercise 5 skipped.

Exercise 6: Let a,b,cZ.a, b, c \in \mathbb{Z}. Show that the equation ax+by=cax + by = c has integer solutions iff (a,b)c.(a, b) | c.

Solution: Let d=(a,c).d = (a, c).

(\Rightarrow) We can see that dd divides the left hand side which means that dd must divide c.c.

(\Leftarrow) Sps. dcd | c so c=dcc = dc' for some cZ.c' \in \mathbb{Z}. Then we can write au+bv=dau + bv = d for some m,nZ.m, n \in \mathbb{Z}. Then by multiplying the expression by cc' we get auc+bvc=dc=c.auc' + bvc' = dc' = c. Finally let x=ucx = uc' and y=vcy = vc' be integer solutions to ax+by=c.ax + by = c.

Exercise 7: Let d=(a,b)d = (a, b) and a=daa = da' and b=db.b = db'. Show (a,b)=1.(a', b') = 1.

Solution: Let d=(a,b),d' = (a', b'), so dad' | a' and db.d' | b'. Then ddadd' | a and ddb.dd' | b. If d1d' \neq 1 then dddd' is a common divisor of aa and bb that is larger than d.d.

Exercise 8: Let x0x_0 and y0y_0 be a solution to ax+by=c.ax + by = c. Show that all solutions have the form x=x0+t(b/d),y=y0t(a/d)x = x_0 + t(b / d), y = y_0 - t(a / d) where d=(a,b)d = (a, b) and tZ.t \in \mathbb{Z}.