Past exam of the commutative algebra course of the University of Oxford / 2023 (B2.2, A12191W1)

Source: released examination PDF, bundle pages 249–251; paper pages 1–3.

An ideal $\mathfrak{q}$ is $\mathfrak{p}$-primary if it is a proper primary ideal whose radical of an ideal is $\mathfrak{p}$. Explicitly,Equivalently, $ab\in \mathfrak{q}$ and $b\notin \mathfrak{p}$ imply $a\in \mathfrak{q}$: multiplication by every element outside $\mathfrak{p}$ is injective on the quotient ring $R/\mathfrak{q}$.

A minimal primary decomposition is a finite expressionin proper primary ideals such that the prime ideals ${\mathfrak{p}}_j=\sqrt{{\mathfrak{q}}_j}$ are pairwise distinct and no primary ideal can be omitted. More explicitly, for every $j$,Thus minimality means both irredundancy and distinct radicals of ideals; it does not assert that the primary ideals themselves are uniquely determined. For $I=R$, one may use the empty intersection convention.

In the terminology of a minimal primary decomposition, an associated prime of a primary decomposition of $I$ is a prime ideal occurring as $\sqrt{{\mathfrak{q}}_j}$ in such a decomposition. The first uniqueness theorem for primary decomposition says that this collection is independent of the chosen minimal primary decomposition. Among these prime ideals, an isolated prime of a primary decomposition is minimal under inclusion, equivalently a minimal prime over an ideal $I$; an embedded prime ideal strictly contains another member of the collection. ThusWhen $R$ is a Noetherian ring, these are precisely the associated primes of a module $R/I$. In that case the equivalent definition iswhere $(I:x)$ is a colon ideal. The decomposition terminology applies when a minimal primary decomposition exists; a general ring need not provide one for every ideal.

Suppose first that $\mathfrak{a}$ is $\mathfrak{m}$-primary. Since $R$ is a Noetherian ring, choose a finite generating set of an ideal $\mathfrak{m}=(x_1,\dots ,x_d)$. The equality of radicals of ideals $\sqrt{\mathfrak{a}}=\mathfrak{m}$ supplies positive integers $e_i$ with $x_i^{e_i}\in \mathfrak{a}$. SetEvery product of $n$ generators has some $x_i$ appearing at least $e_i$ times. Such products generate the product of ideals ${\mathfrak{m}}^n$, so ${\mathfrak{m}}^n\subseteq \mathfrak{a}$. Also $\mathfrak{a}\subseteq \sqrt{\mathfrak{a}}=\mathfrak{m}$. If $\mathfrak{m}=0$, take $n=1$ directly.

Conversely, suppose ${\mathfrak{m}}^n\subseteq \mathfrak{a}\subseteq \mathfrak{m}$. Taking radicals of ideals givesHence $\sqrt{\mathfrak{a}}=\mathfrak{m}$. The only maximal ideal containing $\mathfrak{a}$ is $\mathfrak{m}$: any such maximal ideal contains $\sqrt{\mathfrak{a}}$ and therefore equals $\mathfrak{m}$. By the correspondence theorem for ideals, the quotient ring $R/\mathfrak{a}$ is a local ring with maximal ideal $\mathfrak{m}/\mathfrak{a}$. Every element outside this maximal ideal is a unit.

If $ab\in \mathfrak{a}$ and $a\notin \mathfrak{a}$, then $b$ cannot be outside $\mathfrak{m}$, since multiplying by its inverse in the quotient ring would give $a\in \mathfrak{a}$. Thus $b\in \mathfrak{m}$ and $b^n\in \mathfrak{a}$, proving the required primary ideal condition. We have proved the maximal-primary ideal criterion

Put $A=\mathbb{C}[X,Y]$, $P=(X)$ and $Q=(X^2,Y)$. The quotient ring $A/P\cong \mathbb{C}[Y]$ is an integral domain, so $P$ is a prime ideal and hence a primary ideal. Put $\mathfrak{m}=(X,Y)$. ThenThe polynomial ring $A$ is a Noetherian ring by the Hilbert basis theorem, and $\mathfrak{m}$ is a maximal ideal since $A/\mathfrak{m}\cong \mathbb{C}$. Therefore the maximal-primary ideal criterion from c makes $Q$ $\mathfrak{m}$-primary. In particular its radical of an ideal is $\mathfrak{m}$, distinct from $P$.

Neither primary ideal can be omitted: $X\in P\setminus Q$, while $Y\in Q\setminus P$. Thus the given intersection is already a minimal primary decomposition. To express the ideal by generators, write an element of $P\cap Q$ as $Xh=X^{2}u+Yv$. Reducing modulo the prime ideal $(X)$ gives $Y\bar{v}=0$ in the integral domain $\mathbb{C}[Y]$, so $v=Xw$. Cancelling $X$ in the integral domain $A$ gives $h=Xu+Yw$. HenceBy the first uniqueness theorem for primary decomposition, the associated primes of a module $A/J$ are exactly the two radicals of ideals. Their strict inclusion $P⊊\mathfrak{m}$ givesOne can also see both associated primes of a module directly through annihilators:For the first colon ideal, reducing $hY\in J$ modulo $(X)$ forces $h\in (X)$, and the reverse inclusion follows from $XY\in J$. For the second colon ideal, cancel $X$ in $hX\in X(X,Y)$. Both represented elements of the quotient module $A/J$ are nonzero.

Because $(r)$ is a proper prime ideal, each $(r^k)$ is proper and its radical of an ideal isIndeed, $(r^k)\subseteq (r)$ gives one inclusion of radicals of ideals, while $(rc{)}^k\in (r^k)$ gives the other.

We prove the saturation condition for a primary ideal: if $ab\in (r^k)$ and $a\notin \mathfrak{p}$, then $b\in (r^k)$. For $k=1$ this is the defining property of a prime ideal. For $k>1$, write $ab=r^{k}c$. Since $a\notin (r)$, the prime ideal condition gives $b=r{b}_1$. The element $r$ is a non-zero-divisor, so cancellation givesInduction gives $b_1\in (r^{k-1})$, hence $b\in (r^k)$. Together with the computed radical of an ideal, this proves

Since the radical of an ideal $\sqrt{\mathfrak{q}}$ is $(r)$, some positive power of $r$ belongs to $\mathfrak{q}$. Choose the least such exponent $k$. Then $(r^k)\subseteq \mathfrak{q}$ and $r^j\notin \mathfrak{q}$ for $0\le j<k$, with $r^0=1$ excluded because the primary ideal $\mathfrak{q}$ is proper.

Take $x\in \mathfrak{q}$. If $x\notin (r^k)$, choose the largest $j\in \{0,\dots ,k-1\}$ for which $x\in (r^j)$, and write $x=r^{j}u$. Maximality of $j$ implies $u\notin (r)=\sqrt{\mathfrak{q}}$: otherwise $x\in (r^{j+1})$, a contradiction, including when $j=k-1$. The primary ideal condition applied to $r^{j}u\in \mathfrak{q}$ therefore forces $r^j\in \mathfrak{q}$, contrary to the choice of $k$.

Consequently every $x\in \mathfrak{q}$ lies in $(r^k)$, andTogether with i, this classifies the primary ideals for a regular principal prime. The exponent is unique: if $(r^k)=(r^{k+1})$, then $r^k=r^{k+1}c$, and cancellation of the non-zero-divisor $r^k$ would give $1=rc$, contradicting that $(r)$ is proper. No Noetherian hypothesis is needed.

The image $f/1$ is a unit in the localization at a prime ideal $R_{\mathfrak{p}}$, because $f\notin \mathfrak{p}$. The universal property of localization, applied to the ring homomorphism $R\to R_{\mathfrak{p}}$ and the multiplicative subset $\{1,f,f^2,\dots \}$, gives a unique ring homomorphism fixing the image of $R$:Fixing the image of $R$ is exactly the required compatibility with the algebra over a ring structures. Uniqueness also follows directly: such a ring homomorphism must send $1/f$ to the inverse of $f/1$, and these elements together with $R$ generate $R_f$.

Choose $f=1$, so that $R_f=R$ and $f\notin \mathfrak{p}$. For an integral domain, the canonical ring homomorphism $R\to R_{\mathfrak{p}}$ is injective: if $a/1=0$, the vanishing criterion in a module localization gives $sa=0$ for some $s\notin \mathfrak{p}$. Such an $s$ is nonzero, so the absence of zero divisors gives $a=0$. Thus

Consider the kernel of a ring homomorphismSince $R$ is a Noetherian ring, this ideal has a finite generating set of an ideal, say $K=(a_1,\dots ,a_t)$. For each $i$, choose $s_i\notin \mathfrak{p}$ with $s_i{a}_i=0$, and putThen $f\notin \mathfrak{p}$ by the prime ideal condition, and $fK=0$. If $K=0$, take the empty product $f=1$.

If $a/f^n$ maps to zero under the ring homomorphism ${\theta }_f$ from a, then $a/1=0$ in $R_{\mathfrak{p}}$ because $f/1$ is a unit there. Hence $a\in K$, so $fa=0$. The vanishing criterion in a module localization now gives $a/f^n=0$ already in $R_f$. ThereforeThis is the finite generation clears localization kernels argument: the finite generating set of an ideal permits a single element outside $\mathfrak{p}$ to annihilate the entire kernel of a ring homomorphism.

For $n\ge 1$, let $K_n=\ker ({\varphi }^n)$, where the power denotes repeated composition of the ring endomorphism. These ideals form an ascending chain:The ascending chain condition for a Noetherian ring gives $K_N=K_{N+1}$ for some $N\ge 1$. Take $a\in \ker \varphi$. Since $\varphi$ is surjective, its iterate ${\varphi }^N$ is surjective, so choose $b\in R$ with ${\varphi }^N(b)=a$. Thenso $b\in K_{N+1}=K_N$. Therefore $a={\varphi }^N(b)=0$.

Thus the kernel of a ring homomorphism $\varphi$ is zero. This proves that a surjective endomorphism of a Noetherian ring is an automorphism, and in particular

Use the direct product of ringswith coordinatewise addition and multiplication and unit $(1,1,\dots )$. Define the shiftCoordinatewise operations show that $g$ preserves addition, multiplication and the unit, so it is a ring endomorphism. It is surjective because $(0,b_0,b_1,\dots )$ maps to any prescribed $(b_0,b_1,\dots )$. Its kernel of a ring homomorphism isso it is not injective. This is the noninjective surjective shift on a product of fields.

This ring fails the Noetherian hypothesis in c. Indeed, let $e_n$ have a single $1$ in coordinate $n$. The ideals $I_n=(e_0,\dots ,e_n)$ form the strictly ascending chainsince $e_{n+1}\notin I_n$.

A ring is reduced if its only nilpotent element is zero. Equivalently,The second expression says that its nilradical vanishes. A reduced ring may have zero divisors; for example, the direct product of rings $\mathbb{R}\times \mathbb{R}$ is reduced but $(1,0)(0,1)=0$.

Write $N=\sqrt{(0)}$. Every element of the extension of an ideal $N{R}_{\mathfrak{p}}$ is a fraction $a/s$ with $a\in N$ and $s\notin \mathfrak{p}$: a finite sum of such fractions has a common denominator and numerator still in the ideal $N$. If $a^n=0$, then $(a/s{)}^n=0$, soConversely, let $a/s$ be a nilpotent element of the localization at a prime ideal $R_{\mathfrak{p}}$. For some $n\ge 1$, $a^n/s^n=0$. The vanishing criterion in a module localization supplies $t\notin \mathfrak{p}$ with $t{a}^n=0$. It follows thatso $ta\in N$. Since $ts\notin \mathfrak{p}$,This proves that the nilradical commutes with localization:The multiplication by $t$ is necessary: the numerator $a$ of a nilpotent element in a localization need not itself be nilpotent in $R$.

If $R$ is reduced, its nilradical $N$ is zero. By b, every localization at a prime ideal has zero nilradical, so in particular $R_{\mathfrak{m}}$ is reduced for every maximal ideal $\mathfrak{m}$.

Conversely, suppose every such localization is reduced. Take a nilpotent element $x\in R$. Then $x/1$ is nilpotent in every $R_{\mathfrak{m}}$, hence zero there. To see directly that $x=0$, suppose otherwise and consider the proper annihilatorChoose a maximal ideal $\mathfrak{m}$ containing this annihilator. The equality $x/1=0$ in $R_{\mathfrak{m}}$ would imply $sx=0$ for some $s\notin \mathfrak{m}$, by the vanishing criterion in a module localization. But then $s\in {\mathrm{Ann}}_R(x)\subseteq \mathfrak{m}$, a contradiction. This is the principle that localization detects zero elements. ThereforeThe argument also includes the zero ring, which is reduced and has no maximal ideals.

Let $N$ be the nilradical of $R$. By b, the reduced locus of a ring isThe Noetherian ring hypothesis makes the ideal $N$ finitely generated; write $N=(n_1,\dots ,n_t)$. Fix $\mathfrak{p}\in \mathcal{R}$. For every $i$, the vanishing criterion in a module localization gives $s_i\notin \mathfrak{p}$ with $s_i{n}_i=0$. The product $f=s_1\cdots s_t$ lies outside the prime ideal $\mathfrak{p}$ and satisfies $fN=0$. When $N=0$, take $f=1$.

For every prime ideal $\mathfrak{q}$ in the distinguished open setthe element $f/1$ is a unit in $R_{\mathfrak{q}}$. The equality $fN=0$ therefore implies $N{R}_{\mathfrak{q}}=0$, so $R_{\mathfrak{q}}$ is reduced. Thus $\mathfrak{p}\in D(f)\subseteq \mathcal{R}$, proving that $\mathcal{R}$ is a Zariski-open subset of the spectrum of a ring.

More precisely, let $A={\mathrm{Ann}}_R(N)$, the annihilator of a module $N$. The argument shows that $N{R}_{\mathfrak{p}}=0$ if and only if some $f\in A$ lies outside $\mathfrak{p}$. HenceEquivalently, the complement is the support of a module $N$. This is vanishing of a finitely generated module near a prime applied to $N$: its finite generation lets the separately chosen denominators be combined into one element annihilating all of $N$.

Set $L=M/\mathfrak{a}M$, a quotient module. The exactness of localization gives, for every maximal ideal $\mathfrak{m}$,If $\mathfrak{m}\supseteq \mathfrak{a}$, the hypothesis $M_{\mathfrak{m}}=0$ makes this quotient module zero. If $\mathfrak{m}⊉\mathfrak{a}$, choose $s\in \mathfrak{a}\setminus \mathfrak{m}$. Its image in $R_{\mathfrak{m}}$ is a unit, so $\mathfrak{a}{R}_{\mathfrak{m}}=R_{\mathfrak{m}}$ and the displayed quotient module is again zero. ThusFor completeness, if $0\ne \ell \in L$, its annihilator is a proper ideal, and hence lies in a maximal ideal $\mathfrak{m}$. The vanishing criterion in a module localization shows $\ell /1\ne 0$ in $L_{\mathfrak{m}}$: any element annihilating $\ell$ lies inside $\mathfrak{m}$. This contradicts the vanishing just proved. Hence $L=0$, andThe use of localization detects zero elements is valid for every module; neither finite generation nor a Noetherian hypothesis is needed.