Past exam of the mathematics course of the University of Cambridge / 2004 / iii

The Hilbert basis theorem says that if $B$ is a commutative Noetherian ring, then $B[t]$ is Noetherian; iteration treats any finite number of variables. Here is a complete leading-coefficient proof. For an ideal $I\subseteq B[t]$, let $L_d$ be the set of degree-$d$ coefficients of its polynomials of degree at most $d$, including zero coefficients. These are ideals of $B$, and multiplication by $t$ gives $L_d\subseteq L_{d+1}$. The ascending chain condition makes the chain stationary from some $N$ onward. Choose finite generators for $L_0,\dots ,L_N$ and lift them to polynomials $f_{dj}\in I$ of degree at most $d$. If $f\in I$ has degree $m$, set $d=\min (m,N)$ and express its leading coefficient as ${\sum }_j{a}_{dj}{b}_j$, where $a_{dj}$ is the chosen degree-$d$ coefficient. Then $f-{\sum }_j{b}_j{t}^{m-d}{f}_{dj}$ has smaller degree. Induction proves that the finitely many lifts generate $I$. This proves the theorem, including coefficient rings with zero divisors.

Put $A=\mathbb{C}G$ and $L=\mathbb{C}[X,X^{-1}]$. The infinite dihedral group has normal forms $X^m$ and $X^{m}Y$. The relations move every $Y$ to the right and reduce its exponent modulo two; the action $X:n\mapsto n+1$, $Y:n\mapsto -n$ on $\mathbb{Z}$ shows that these forms are distinct. ThusThe Hilbert basis theorem makes $\mathbb{C}[X]$ Noetherian, and its localization $L$ is Noetherian: an ideal of a localization is generated by the images of the finite generators of its contraction. Since $A$ is free of rank two over $L$ on the left, it is a Noetherian module over $L$. Every left ideal of $A$ is an $L$-submodule, so all its ascending chains stabilize. Therefore $\mathbb{C}G$ is left Noetherian.

For the center of the infinite dihedral group algebra, write an element as $f(X)+g(X)Y$. Commuting it with $X$ gives $g(X)(X^{-1}-X)=0$ in the Laurent integral domain, forcing $g=0$. Commuting $f(X)$ with $Y$ then gives $f(X)=f(X^{-1})$. A symmetric Laurent polynomial is a linear combination of $1$ and $X^j+X^{-j}$, $j\ge 1$. With $z=X+X^{-1}$, the recursionexpresses $S_j=X^j+X^{-j}$ as a polynomial in $z$. Conversely every polynomial in $z$ is central. The highest Laurent exponent of a nonzero polynomial in $z$ shows that $z$ is transcendental. HenceThe identities $X^{-1}=z-X$ and $X^2-zX+1=0$ give $L=R+RX$. Independence follows by applying $X\mapsto X^{-1}$ to a relation $f(z)+Xg(z)=0$ and subtracting: $(X-X^{-1})g(z)=0$, so $g=f=0$. Consequentlyis a free $R$-module of rank four.

Let $M$ be a nonzero simple module on the left. For any $m\ne 0$, $M=Am$, so it is finite over $R$. Every central $r\notin P={\mathrm{Ann}}_R(M)$ acts bijectively: its kernel and image are $A$-submodules, and simplicity forces kernel zero and image $M$. To prove that $P$ is maximal, not merely prime, put $B=R/P$; then $M$ is a faithful finite $B$-module. For $0\ne b\in B$, $bM=M$. If $m_1,\dots ,m_d$ generate $M$, write $m_i=b{\sum }_j{c}_{ij}{m}_j$. The adjugate matrix identity for $I-bC$ shows that $\det (I-bC)$ annihilates every generator and hence $M$. Faithfulness makes this determinant zero in $B$. Its constant term is one, so it has form $1-bq$, proving $bq=1$. Thus $B$ is a field, and $P$ is a maximal ideal of $R$. This proves the general principle for simple modules over an algebra finite over its center in this instance.

Every maximal ideal of $\mathbb{C}[z]$ is $(z-\lambda )$ for some $\lambda \in \mathbb{C}$, by factorization of complex polynomials. The module $M=Am$ is therefore a quotient of the four-dimensional algebra $A/(z-\lambda )A$. In particular ${\dim }_{\mathbb{C}}M\le 4<\infty$, as required.

The actual simple dimensions are at most two. If $\lambda \ne \pm 2$, choose $u\ne \pm 1$ with $u+u^{-1}=\lambda$. The matrices $X=\mathrm{diag}(u,u^{-1})$ and $Y=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)$ give a representation of $A/(z-\lambda )A$ onto $M_2(\mathbb{C})$: polynomial functions of $X$ give the two diagonal matrix units, and multiplication by $Y$ gives the off-diagonal ones. Both algebras have dimension four, so this is an isomorphism. If $\lambda =\pm 2$, the element $X\mp 1$ has square zero and is conjugated to its negative by $Y$; its generated two-sided ideal is square zero. It annihilates every simple module, since its action on a simple module is either zero or the whole module, and the latter contradicts its square being zero. The remaining quotient is $\mathbb{C}[C_2]$, whose two simple modules have $Y=\pm 1$. Thus the generic simple modules have dimension two, and the four exceptional one-dimensional characters have $X=\pm 1$, $Y=\pm 1$.

If $M=0$, all its localizations are zero. Conversely, take $0\ne m\in M$. Its annihilator $I={\mathrm{Ann}}_R(m)$ is a proper ideal, so it is contained in a maximal ideal $P$, which is prime. The zero criterion for module localization says that $m/1=0$ in $M_P$ only if some $s\notin P$ satisfies $sm=0$. This would put $s\in I\subseteq P$, a contradiction. Therefore $M_P\ne 0$, provingIn fact maximal ideals suffice, and neither finite generation nor Noetherianity is needed for this implication.

Set $C=N/\phi (M)$. The natural map $N_P\to C_P$ is onto. Its kernel is the image of ${\phi }_P$: a fraction $n/s$ maps to zero precisely when $tn\in \phi (M)$ for some $t\notin P$. If $tn=\phi (m)$, then $n/s={\phi }_P(m/(ts))$. This proves the localized cokernel identity directly:Thus ${\phi }_P$ is a surjection for every $P$ exactly when $C_P=0$ for every $P$. Part (i) makes this equivalent to $C=0$, or to $\phi$ being a surjection. Hence surjectivity is detected at all prime localizations, and again maximal ideals alone would suffice.

Let $S=R\setminus P$. If $Q\subseteq P$ is a prime ideal, then $Q\cap S=\varnothing$. Its extension isIt is proper: $1=q/s$ would give $u(s-q)=0$ for some $u\in S$, and hence $us=uq\in Q$, impossible because $u,s\notin Q$. If a product $(a/s)(b/t)$ lies in $Q{R}_P$, clearing a denominator gives $uab\in Q$ for some $u\in S$. Primeness and $u\notin Q$ imply $a\in Q$ or $b\in Q$. Thus $Q{R}_P$ is prime. Its contraction is $Q$, since $r/1\in Q{R}_P$ implies $ur\in Q$ for some $u\in S$, forcing $r\in Q$.

Conversely, contract a prime ideal $\mathfrak{q}$ of $R_P$ to $Q=\{r:r/1\in \mathfrak{q}\}$. This is a prime ideal of $R$. It contains no element of $S$, since such an element becomes a unit and cannot belong to the proper ideal $\mathfrak{q}$. Hence $Q\subseteq P$. Also $r/s\in \mathfrak{q}$ exactly when $r/1\in \mathfrak{q}$, because $s/1$ is a unit, so $\mathfrak{q}=Q{R}_P$. These extension and contraction operations are inverse, giving the prime ideal correspondence for localization:

An associated prime of a module is an annihilator of a nonzero element. Suppose $Q={\mathrm{Ann}}_R(m)$ is prime and $Q\subseteq P$. Then $m/1\ne 0$ in $M_P$: a denominator $s\notin P$ killing $m$ would belong to $Q\subseteq P$. MoreoverIndeed, if $(a/s)(m/1)=0$, some $t\notin P$ has $tam=0$, so $ta\in Q$; primeness and $t\notin Q$ imply $a\in Q$. The reverse containment is immediate. This proves one inclusion in the localization of associated primes formula.

For the other, let $\mathfrak{q}={\mathrm{Ann}}_{R_P}(m/s)$ be an associated prime of $M_P$ and contract it to $Q$. By part (iii), $Q$ is prime, $Q\subseteq P$ and $Q{R}_P=\mathfrak{q}$. Since $s$ is invertible, $\mathfrak{q}$ is also the annihilator of $m/1$. The Noetherian ring hypothesis makes $Q=(a_1,\dots ,a_r)$ finitely generated. For each $a_i$, the equality $(a_{i}m)/1=0$ supplies $t_i\notin P$ with $t_i{a}_{i}m=0$. Put $t={\prod }_i{t}_i$, taking $t=1$ for the zero ideal. Then $Q$ annihilates $tm$, while $tm/1\ne 0$ because $t/1$ is a unit and $m/1\ne 0$.

If $b(tm)=0$, then $b/1$ annihilates $m/1$, so $b\in Q$. Therefore ${\mathrm{Ann}}_R(tm)=Q$, and $Q\in {\mathrm{Ass}}_R(M)$. We concludeThe single common denominator $t$ is the key point: one must turn a localized annihilator into the annihilator of one actual nonzero element of $M$, not merely assert that annihilators commute with localization.

Use generators $x,D$ with $Dx-xD=1$. The ordered monomials $x^i{D}^j$ form a basis: construct the Weyl algebra as the differential Ore extension of $\mathbb{C}[x]$ with $Df=fD+f^{\prime }$, which has unique ordered polynomial expressions. Giving both generators degree one produces the Bernstein filtration, whose associated graded ring is $\mathbb{C}[x,\xi ]$. If $a,b\ne 0$, their nonzero leading symbols have nonzero product in this polynomial integral domain. Thus $ab$ has nonzero leading symbol and cannot vanish. Hence $A_1(\mathbb{C})$ has no nontrivial zero divisors.

Here is the left-handed Ore theorem. Let $S$ be a multiplicative set of regular elements of a ring, containing $1$ and excluding $0$. There is an injective Ore localization whose elements are left fractions $s^{-1}a$ if and only ifThis is the left Ore condition. For a noncommutative domain, taking $S=A\setminus \{0\}$ gives a division ring exactly when this condition holds. With zero-divisor denominators a further left reversibility condition is needed; the regular-denominator statement is the one used here.

Necessity follows by writing $a{s}^{-1}=t^{-1}b$ in the quotient and multiplying by $t$ and $s$, giving $ta=bs$ in the embedded original ring. For a construction proving sufficiency, use the left Ore fraction construction: pairs $(s,a)$ represent $s^{-1}a$, and $(s,a)$ and $(t,b)$ are equivalent whenfor suitable $u,v\in A$. Reflexivity and symmetry are immediate. For transitivity, combine $us=vt=w$ and $u^{\prime }t=v^{\prime }q=w^{\prime }$ by a common left multiple $hw=k{w}^{\prime }\in S$. Then $(hv-k{u}^{\prime })t=0$, so regularity gives $hv=k{u}^{\prime }$, which combines the numerator equalities and proves transitivity.

Addition uses the common denominator $w$, with numerator $ua+vb$. To multiply $s^{-1}a$ by $t^{-1}b$, choose $v\in S,c\in A$ with $va=ct$; setThe Ore condition supplies a common refinement whenever a representative or an Ore relation is changed, and regularity cancels the shared denominator, proving independence of choices. The same refinements verify the ring axioms; the formulas are simply addition and multiplication after clearing left denominators. The map $a\mapsto (1,a)$ is injective, since equivalence to zero would give $ua=0$ with $u\in S$. Each $s\in S$ becomes invertible. A homomorphism inverting $S$ extends uniquely by $(s,a)\mapsto f(s{)}^{-1}f(a)$, giving the universal property. For all nonzero denominators in a noncommutative domain, a nonzero fraction has inverse $(s^{-1}a{)}^{-1}=a^{-1}s$, so the quotient is a division ring. This is the requested proof sketch, with the fraction equivalence and multiplication made explicit.

It remains to check the left Ore condition for $A_1(\mathbb{C})$. Its graded polynomial ring is Noetherian by the Hilbert basis theorem. Lifting homogeneous generators of a graded left ideal and successively cancelling leading symbols proves that the filtered algebra is a left Noetherian ring: each subtraction lowers the nonnegative degree, so it terminates.

In any left Noetherian noncommutative domain, nonzero principal left ideals intersect. Otherwise $Ab\cap Aa=0$ with $a,b\ne 0$ would makean infinite direct sum of nonzero left ideals. To verify directness, reduce a finite relation modulo $Aa$ to make its first summand zero, then cancel $a$ on the right and repeat. Its increasing finite partial sums contradict the ascending chain condition. A nonzero element of $Aa\cap As$ therefore has the form $ta=bs\ne 0$, with $t\ne 0$, proving the required condition. Consequently $A_1(\mathbb{C})$ has a classical ring of left quotients, and it is a division ring.

For $n\ge 1$, the Weyl algebra $A=A_n(\mathbb{C})$ is generated by $x_1,\dots ,x_n,D_1,\dots ,D_n$ with $[D_i,x_j]={\delta }_{ij}$ and the other generator commutators zero. Iterating the differential Ore extension construction gives the ordered basis $x^{\alpha }{D}^{\beta }$. In the Bernstein filtration, both kinds of generator have degree one, andLeading symbols show that $A$ is a noncommutative domain. The Hilbert basis theorem and homogeneous-generator lifting by decreasing degree make it both left and right Noetherian.

It is also a simple ring. In a nonzero two-sided ideal, choose a nonzero element and a monomial $x^{{\alpha }_0}{D}^{{\beta }_0}$ of maximal total degree with nonzero coefficient $c$. Applying $(\mathrm{ad}D{)}^{{\alpha }_0}(\mathrm{ad}x{)}^{{\beta }_0}$ eliminates every lower-degree term and every different term of that maximal degree. The result is the nonzero scalar $(-1{)}^{|{\beta }_0|}{\alpha }_0!{\beta }_0!c$. Thus the ideal contains $1$. Here the scalar extraction by iterated Weyl commutators uses characteristic zero. Despite simplicity, $A$ has no nonzero finite-dimensional modules: in such a representation $[D_1,x_1]=I$ would have trace $0=\dim M$, impossible over $\mathbb{C}$. Its interesting module theory is necessarily infinite-dimensional.

For a finitely generated left module $M$, choose a finite-dimensional generating subspace $M_0$ and set $M_j=F_{j}A{M}_0$. This is a good filtration of a module. Its associated graded module is finite over the graded polynomial ring, so its cumulative Hilbert function is eventually a polynomial,with positive integer multiplicity $e(M)$. These are the Bernstein growth dimension of a Weyl algebra module and its multiplicity of a filtered module. Two good filtrations bound each other after fixed shifts: lift finite homogeneous generators in one filtration, bound their degrees in the other, and use compatibility with $F_{j}A$; then reverse the roles. The growth degree and leading coefficient therefore do not depend on the good filtration of a module. The degree is also the Gelfand–Kirillov dimension of a module and the dimension of the characteristic variety of a Weyl algebra module, the support of $\mathrm{gr}M$. Since $M$ is a quotient of a finite sum of copies of $A$, $d(M)\le 2n$.

The essential lower bound is the Bernstein inequality for Weyl algebra modules:Here is a full proof. Multiplication gives a linear map $F_{j}A\to {\mathrm{Hom}}_{\mathbb{C}}(M_j,M_{2j})$. Prove it injective by induction on $j$. At $j=0$ a scalar killing $M_0\ne 0$ must vanish. If $a\in F_{j}A$ kills $M_j$, then for every generator $z$, its commutator $[a,z]\in F_{j-1}A$ kills $M_{j-1}$: both $azm$ and $zam$ are zero for $m\in M_{j-1}$. Induction gives $[a,z]=0$ for all generators. In the ordered basis, $[D_i,a]$ differentiates the $x_i$ exponents and $[x_i,a]$ differentiates the $D_i$ exponents with a minus sign. Vanishing of all these commutators in characteristic zero forces $a$ to be a scalar. It then vanishes because it kills $M_0$. This proves the faithful finite-step action of a Weyl algebra and henceComparison of degrees gives $2n\le 2d(M)$, proving the inequality. Equally, without invoking eventual polynomiality, this estimate directly bounds the Gelfand–Kirillov dimension of a module below by $n$. Right modules satisfy the same inequality through the filtration-preserving isomorphism $A\cong A^{\mathrm{op}}$, $x_i\mapsto x_i$, $D_i\mapsto -D_i$.

A nonzero finitely generated module with $d(M)=n$ is a holonomic Weyl algebra module; the zero module is conventionally included. This minimal-growth class is stable under submodules, quotient modules and extensions. For $0\to N\to M\to Q\to 0$, take the induced and quotient filtrations. Their graded sequence is exact; Noetherianity of $\mathrm{gr}A$ makes the induced filtrations good. Thus the Hilbert functions add andif both nonzero terms have the top dimension, their multiplicities add. For a holonomic Weyl algebra module, every nonzero subquotient has dimension $n$, by the upper bound inherited from $M$ and Bernstein's lower bound. Each nonzero factor in a strict chain therefore consumes positive integral multiplicity. This proves the multiplicity bounds the length of a holonomic Weyl module:In particular a holonomic module is both Noetherian and Artinian and has finite length, although its vector-space dimension is infinite. This distinction separates finite-dimensional representation theory from finite-length Weyl module theory.

The polynomial module $\mathbb{C}[x_1,\dots ,x_n]$, with $x_i$ acting by multiplication and $D_i$ by differentiation, is generated by $1$. Its $M_j$ consists of polynomials of degree at most $j$, with dimension $\left(\genfrac{}{}{0ex}{}{j+n}{n}\right)$, so $d=n$ and $e=1$. It is simple: differentiating a maximal-degree monomial of a nonzero polynomial isolates a nonzero constant, and multiplication by polynomials then generates the whole module. A second basic example is the delta moduleUsing the reverse ordered basis $D^{\beta }{x}^{\alpha }$, its vector space is $\mathbb{C}[D_1,\dots ,D_n]$; $D_i$ acts by multiplication and $x_i$ by negative differentiation. The same argument proves simplicity and holonomicity. These two examples are exchanged by the automorphism $x_i\mapsto D_i$, $D_i\mapsto -x_i$. Free modules, in contrast, have growth dimension $2n$ and are not holonomic. Thus the central organizing invariants are good filtration of a module, growth dimension, characteristic support and multiplicity, with Bernstein's estimate explaining why dimension $n$ is the natural smallest nonzero size.

Specify a commutative ground ring $k$ and a unital associative $k$-algebra $R$. For an $R$-bimodule $M$, the Hochschild cochain complex hasand coboundaryThe bimodule identities and associativity cancel the terms of ${\delta }^2$ in pairs. The Hochschild cohomology is $H{H}_k^q(R,M)=\ker {\delta }_q/\mathrm{im}{\delta }_{q-1}$; the paper's $H^q(R,R)$ uses the regular bimodule as coefficients.

Over a field, the bar resolution of an associative algebra explains its homological meaning. Put $R^e=R{\otimes }_k{R}^{\mathrm{op}}$. The degree-$q$ bar term is $R^{{\otimes }_k(q+2)}$, with the outer factors supplying the $R^e$ action, boundary obtained by multiplying adjacent factors with alternating signs, and augmentation multiplication $R\otimes R\to R$. Inserting $1$ at the left gives a $k$-linear contracting homotopy, proving exactness. The terms are free $R^e$-modules after choosing a $k$-basis for the inner tensor factors. Applying ${\mathrm{Hom}}_{R^e}(-,M)$ gives precisely the displayed cochain complex and thereforeThis ground-ring qualification in Hochschild cohomology matters for an arbitrary ring: one can take $k=\mathbb{Z}$, but the bar terms need not be projective over $R^e$. The ordinary cochain complex then computes relative Hochschild cohomology; the absolute Ext identification needs, for example, $R$ projective over $k$. Likewise the extension classification below applies to ground-ring-linearly split extensions. For algebras over a field these qualifications are automatic.

In degree zero, $(\delta m)(a)=am-ma$, so $H{H}^0(R,R)=Z(R)$. In degree one the cocycle equation isexactly the Leibniz rule for a derivation of an algebra. The coboundaries are the inner derivations $a\mapsto am-ma$. Thus the First Hochschild cohomology as outer derivations isIts significance is infinitesimal automorphisms modulo inner ones. Over the dual numbers $k[\epsilon ]/({\epsilon }^2)$, the map $a\mapsto a+\epsilon d(a)$ is multiplicative precisely when $d$ satisfies the Leibniz rule. Conjugation by $1+\epsilon m$ gives $a\mapsto a+\epsilon (ma-am)$, an inner derivation with the opposite sign convention. The quotient therefore measures infinitesimal symmetry not arising from conjugation.

Degree two controls extensions and changes of multiplication. A $k$-linear split square-zero extension of an algebra with prescribed kernel bimodule $M$ can be written as $E=R\oplus M$. Choose a unital section and let $f(a,b)$ be the difference between its product and its value on $ab$. ThenComputing both parenthesizations of a triple gives the associativity conditionnamely $\delta f=0$. A unital product has $f(1,a)=f(a,1)=0$, so one uses the normalized Hochschild cochain complex. Changing section by $g:R\to M$ gives $f\mapsto f+\delta g$. Conversely the displayed product for any normalized two-cocycle constructs such an extension, and a section change gives the corresponding extension isomorphism. This proves that second Hochschild cohomology classifies split square-zero extensions:The zero class is the split algebra extension with no extra product term. With $M=R$, this gives one meaning of the requested $H^2(R,R)$.

For a first-order associative deformation, retain the underlying $k$-module and put $a\ast b=ab+\epsilon f(a,b)$. Expanding $(a\ast b)\ast c=a\ast (b\ast c)$ modulo ${\epsilon }^2$ gives exactly $\delta f=0$. A coordinate change $T=1+\epsilon g$ transports the product to $T^{-1}(T(a)\ast T(b))$, whose first-order coefficient is $f+\delta g$. HenceThis classification is of infinitesimal deformations, not a guarantee that all classes extend to every order. For $a\ast b=ab+tf(a,b)+t^2{\mu }_2(a,b)$ modulo $t^3$, the next associativity equation isSubstitution of $\delta f=0$ shows the right-hand side is a three-cocycle. Its class in $H{H}^3(R,R)$ is the second-order obstruction to an associative deformation. It must vanish for a second-order extension to exist. At higher orders the same expansion gives further obstruction cocycles. If $H{H}^2(R,R)=0$, every first nonzero coefficient of a formal associative deformation is a coboundary and can be removed by a coordinate change. Repeating this order by order gives formal rigidity from vanishing second Hochschild cohomology, with the transformations converging in the $t$-adic topology. Vanishing of $H{H}^3$ instead removes these extension obstructions; these are distinct roles.

Examples make the two low degrees concrete. For $R=k[x]$ over a field, inner derivations vanish and every derivation is $h(x){\partial }_x$, determined by its value on $x$. The bimodule resolutionis exact: the middle tensor ring is the polynomial integral domain $k[u,v]$, multiplication identifies its quotient by $u-v$, and multiplication by $u-v$ is injective. Applying bimodule Hom gives zero differential. Consequently the Hochschild cohomology of a polynomial algebra in one variable isThis algebra has many infinitesimal automorphisms, while its ordinary associative multiplication is formally rigid.

For $R=k[x,y]$ in characteristic zero, the bilinear map $f(a,b)=({\partial }_{y}a)({\partial }_{x}b)$ is a two-cocycle: substitution in $\delta f$ and the two product rules cancel its four terms. It is not a coboundary, because every $\delta g$ on a commutative algebra is symmetric in its two inputs, whereas $f(y,x)-f(x,y)=1$. This class integrates to the Weyl deformation of a polynomial algebra in two variables, with $yx-xy=t$. The differential Ore extension construction supplies the ordered basis $x^i{y}^j$ over $k[t]$, so the $t$-adic completion has the fixed underlying module $k[x,y][[t]]$. Its first multiplication correction is exactly this $f$. This displays a genuinely noncommutative formal associative deformation detected by $H{H}^2$.

Finally, the Hochschild cup product $(f⌣g)(a_1,\dots ,a_{p+q})=f(a_1,\dots ,a_p)g(a_{p+1},\dots ,a_{p+q})$ descends to the graded cohomology algebra, and the Gerstenhaber bracket organizes the interaction of derivations and formal associative deformation cocycles. The calculations above explain why the first two positive degrees are especially useful: $H{H}^1$ measures outer infinitesimal symmetries, while $H{H}^2$ measures split extensions and infinitesimal associative structures, with degree three detecting the first obstruction to continuing them.

The Frattini subgroup is the intersection of the maximal subgroups of $G$, with $\Phi (1)=1$. A useful equivalent property isIndeed, a proper subgroup $H$ lies in a maximal subgroup, which also contains $\Phi (G)$. The Frattini subgroup is a characteristic subgroup, since automorphisms permute the maximal subgroups.

The Fitting subgroup is the largest nilpotent normal subgroup. To prove its existence, first form the p-core $O_p(G)$: the product of all normal p-subgroups. For two such subgroups, their product is normal and has order $|AB|=|A||B|/|A\cap B|$, a power of $p$. There are only finitely many subgroups, so $O_p(G)$ exists. For distinct primes $p,r$,Thus the product of the p-cores is a nilpotent normal subgroup, a direct product of groups of its Sylow subgroups. Conversely, a finite nilpotent normal subgroup $K$ has unique Sylow subgroups, each a characteristic subgroup of $K$. Each is normal in $G$ and lies in the corresponding p-core. Consequently the required largest subgroup exists, and

Put $\Phi =\Phi (G)$ and choose a Sylow subgroup $P$ of $\Phi$. Since $\Phi$ is normal, the Frattini argument gives $G=\Phi N_G(P)$: for $g\in G$, the Sylow subgroups $P^g$ and $P$ of $\Phi$ are conjugate by an element of $\Phi$, leaving an element of $N_G(P)$. The non-generator property of the Frattini subgroup proved above forces $N_G(P)=G$. Thus every Sylow subgroup of $\Phi$ is normal. Distinct such Sylow subgroups commute because their group commutators lie in their trivial intersection. Hence $\Phi$ is a nilpotent normal subgroup, and

Write $F=F(G)$, $C=C_G(F)$ and $D=C\cap F=Z(F)$. Both $C$ and $D$ are normal in $G$. Suppose $C>D$. Choose a minimal normal subgroup $K/D$ of $G/D$ contained in $C/D$. In a finite soluble group, such a minimal normal subgroup is elementary abelian: the last nontrivial term of its derived series is an abelian characteristic subgroup, so minimality makes the whole subgroup abelian; its primary components and subgroup of elements killed by $p$ then force one prime and exponent $p$.

It follows that $K^{\prime }\le D$. Moreover $K\le C$ centralizes $D\le F$, so $K^{\prime }\le Z(K)$. Therefore $K$ is a nilpotent group of class at most two. Since $K$ is normal in $G$, the definition of the Fitting subgroup gives $K\le F$. But then $K\le C\cap F=D$, a contradiction. This proves the Fitting subgroup is self-centralizing in soluble groups:

It suffices to centralize each minimal normal subgroup $M$. The subgroup $M\cap F(G)$ is normal in $G$, so it is either $1$ or $M$. In the first caseIn the second case $M$ is a nilpotent group. Its nontrivial group center is a characteristic subgroup of $M$, hence normal in $G$, so minimality makes $M$ abelian. The primary components, which are characteristic subgroups, and the subgroup killed by a prime show that $M$ is elementary abelian of exponent $p$. Thus $M\le O_p(G)$.

The finite p-group $O_p(G)$ acts by conjugation on $M$. Its orbits have prime-power sizes; hence the number of fixed elements is congruent to $|M|=0$ modulo $p$. The identity is fixed, so there is a nonidentity fixed element. Therefore $M\cap Z(O_p(G))\ne 1$. This intersection is normal in $G$, and minimality makes it $M$. The other p-cores commute with $O_p(G)$, and hence with $M$. Thus every factor of $F(G)$ centralizes $M$. Since the group socle is generated by these minimal normal subgroups,

Put $\Phi =\Phi (G)$ and let $H$ be the full preimage of $F(G/\Phi )$ in $G$. The preceding part gives $\Phi \le F(G)$, so the normal nilpotent image $F(G)/\Phi$ lies in $F(G/\Phi )$.

For the reverse inclusion, choose a Sylow subgroup $P$ of $H$. Its image $P\Phi /\Phi$ is the unique Sylow subgroup of the nilpotent group $H/\Phi$, hence a characteristic subgroup there and normal in $G/\Phi$. Thus $P\Phi ⊴G$. The subgroup $P$ is also a Sylow subgroup of $P\Phi$: the p-part of $\Phi$ lies in $P$, because $P\cap \Phi$ is a Sylow subgroup of the normal subgroup $\Phi$ of $H$. Apply the Frattini argument to $P\Phi$:The Frattini subgroup property forces $N_G(P)=G$. Hence every Sylow subgroup of $H$ is normal, so $H$ is a nilpotent group and $H\le F(G)$. Together the inclusions proveThis is Frattini lifting of nilpotence; simply calling an extension of nilpotent groups nilpotent would not establish it.

A soluble group is a group whose derived series $G^{(0)}=G$, $G^{(i+1)}=[G^{(i)},G^{(i)}]$ terminates at $1$. Equivalently, it has a finite subnormal series with abelian factors. For the reverse implication, if $G/N$ has derived length $r$, then $G^{(r)}\le N$; if $N$ has derived length $s$, then $G^{(r+s)}=1$. Iterating this observation through the series proves the equivalence. For a finite group, refining the abelian factors to composition factors says that solubility is equivalent to all composition factors being cyclic of prime order.

Subgroups and quotient groups of soluble groups are soluble, since their derived series lie in, or are images of, the original series. The same calculation proves closure under group extensions. Finite p-groups are nilpotent groups, hence soluble: their nontrivial centers allow induction to construct a central series. Every finite nilpotent group is a direct product of groups of its Sylow subgroups. The converse fails: $S_3$ has the abelian-factor series $1<C_3<S_3$, but its order-two Sylow subgroups are not normal. The symmetric group $S_4$ is also soluble, with series $1<V_4<A_4<S_4$ and abelian factors. A finite soluble group has elementary abelian minimal normal subgroups, by the argument in 1(c). In contrast, a nonabelian simple group is a perfect group and is not soluble.

The Fitting subgroup collects the normal nilpotent structure. For a nontrivial finite soluble group it is nontrivial, since it contains an elementary abelian minimal normal subgroup. Part 1(c) proves $C_G(F)\le F$, so conjugation embeds $G/C_G(F)$ in $\mathrm{Aut}(F)$. This shows why controlling a soluble group through normal abelian or nilpotent layers is effective, even when the whole group is not a nilpotent group.

For a set of primes $\pi$, a Hall subgroup of type $\pi$ has order divisible only by primes in $\pi$ and index divisible only by primes outside $\pi$. The Hall theorem for soluble groups says that a finite soluble group has such subgroups, any two are conjugate, and every subgroup of $\pi$-order is contained in one. These assertions generalize the Sylow theorems, which treat a single prime. Here is a proof of all three assertions.

First we need coprime splitting over an elementary abelian normal subgroup. Suppose $V⊴E$ is an elementary abelian p-group and $Q=E/V$ has order $h$ prime to $p$. A section $s:Q\to E$ defines an action on $V$ and an additive factor set $f(x,y)$ by $s(x)s(y)=f(x,y)s(xy)$. Associativity saysSumming over $z\in Q$ and dividing by $h$ in $V$ gives $f(x,y)=b(x)+xb(y)-b(xy)$, where $b(x)=h^{-1}{\sum }_{z}f(x,z)$. Replacing $s(x)$ by $-b(x)s(x)$ removes the factor set, giving a complement. Two complements differ by a map $d$ satisfying $d(xy)=d(x)+xd(y)$. Averaging over $y$ gives $d(x)=b-xb$, so the complements are conjugate by an element of $V$. This proves both splitting and conjugacy, without assuming that an arbitrary extension splits.

Induct on $|G|$, taking an elementary abelian minimal normal subgroup $V$ of exponent $p$. Choose a quotient Hall subgroup $\bar{H}\le G/V$ and let $K$ be its preimage. If $p\in \pi$, then $K$ itself is a Hall subgroup. If $p\notin \pi$, the preceding splitting argument gives a complement $H$ to $V$ in $K$, and that complement is a Hall subgroup of $G$.