Kl.Graded Algebra

69 nodes 21 formalized 48 proved

Coefficient algebras and graded modules

This topic develops homogeneous components, finite module constructions, and the invariants used to compare sheaf kernels and images. The general constructions start with a commutative coefficient algebra and its grading. The geometric application chooses a basis of the extended root space \(V\), giving \[ A=\operatorname{Sym}_{\mathbb R}(V)\simeq\mathbb R[X_1,\ldots,X_N], \qquad N=\dim_{\mathbb R}V. \] Every variable and root label has degree one. The extra directions in the extended realization contribute variables as well.

The main route is: construct the graded module and its maps; use homogeneous generators to control its pieces; compute or bound its Hilbert series; then use exactness and positivity to turn numerical comparisons into statements about submodules.

Components and scalar degrees

A grading means that each vector has a unique finite decomposition \(m=\sum_nm_n\), with \(m_n\in M_n\). The map \(\pi_n(m)=m_n\) selects one component. A nonnegative grading has no components below zero; a bounded-below grading may begin in another integer degree.

Compatibility with the coefficient algebra is \[ A_iM_j\subseteq M_{i+j}. \] For a homogeneous scalar \(a\in A_i\), this gives \[ \pi_n(am)=a\,\pi_{n-i}(m). \] For a homogeneous vector \(m\in M_j\) and an arbitrary coefficient \(a\), it instead gives \[ \pi_n(am)=\pi_{n-j}(a)m. \] These two component calculations control homogeneous generators, maps and scalar quotients.

For a nonnegative coefficient grading, the source's augmentation is the ring map \[ \epsilon:A\to A,\qquad a\mapsto\pi_0(a). \] Only two degree-zero factors can contribute to the degree-zero part of a product, proving multiplicativity. Its kernel is the positive ideal \(A_+\). For polynomial coefficients, this is the ideal of zero-constant-term polynomials. When \(A_0=k\), the quotient \(A/A_+\) identifies with \(k\).

The finite module data used by sheaves

The ModuleData structure packages an \(A\)-module, its compatible \(k\)-scalar action, its component subspaces and decomposition, graded multiplication, nonnegative support and finite generation over \(A\).

Finite generation describes polynomial coefficients: \(A=k[X]\) is generated by \(1\) over \(A\), and has the infinitely many \(k\)-independent vectors \(1,X,X^2,\ldots\).

Our running example is \[ M=Ae_0\oplus Ae_2,\qquad \deg(e_0)=0,\quad\deg(e_2)=2. \] Its pieces are \[ M_n=A_ne_0\oplus A_{n-2}e_2. \] For \(A=k[X]\), their dimensions are \(1,1,2,2,\ldots\).

A degree-zero map sends each \(M_n\) into the same-degree target piece. A degree-\(r\) map satisfies \[ \pi_n(f(m))=f(\pi_{n-r}(m)). \] For a surjective degree-zero map, this supplies homogeneous preimages by projecting any preimage. For an injective degree-zero map, it detects the degree of a vector from its image.

Submodules, quotients and finite products

A homogeneous submodule \(S\) is closed under every \(\pi_n\). It has pieces \(S\cap M_n\), and the quotient \(M/S\) has pieces given by the images of \(M_n\).

The quotient component of \([m]\) is \([\pi_n(m)]\). Changing \(m\) by a vector of \(S\) changes its component by another vector of \(S\), so the class is well defined. If a sum of homogeneous quotient classes is zero, their representative sum lies in \(S\); projecting shows that every class is individually zero. This proves the quotient decomposition.

The component formula makes kernels and images homogeneous. Finite products use tuples of same-degree vectors and decompose coordinatewise. The shift convention is \[ M(d)_n=M_{n-d}, \] so a generator of degree \(r\) acquires degree \(r+d\). Nonnegative shifts preserve the support required by ModuleData.

The finite shifted free module with generator degrees \(d_i\) has \[ F_n=\prod_i A_{n-d_i}. \] Its general construction allows integer degrees. Nonnegative \(d_i\) give the bundled nonnegative free module.

Homogeneous covers and finite-dimensional pieces

Choose finite generators of a graded module \(M\). Split each into its finitely many homogeneous components. Those component vectors \(v_i\), with degrees \(d_i\), still generate \(M\), giving a surjection \[ F=\bigoplus_i A(d_i)\longrightarrow M,\qquad (a_i)\longmapsto\sum_i a_iv_i. \] The coefficient condition \(a_i\in A_{n-d_i}\) makes every summand have degree \(n\), so the map is degree zero.

For a polynomial ring in finitely many variables, each \(A_n\) has a finite monomial basis. The degree-\(n\) map \(F_n\to M_n\) is surjective by homogeneous preimages, hence \[ \dim_kM_n\le\sum_i\dim_kA_{n-d_i}<\infty. \] All sufficiently negative pieces vanish because \(n-d_i<0\) for every generator. This supplies the finite-dimensional pieces and lower support bound used in Hilbert series.

Polynomial presentations of root-label quotients

Let \(P\subseteq V\) be a linear subspace, and let \(I_P\) be the ideal generated by its degree-one vectors. The linear-quotient theorem gives \[ A/I_P\simeq\operatorname{Sym}(V/P). \] On generators it sends \([\iota(v)]\) to the generator associated with \([v]\in V/P\). Its reverse sends that generator to \([\iota(v)]\), independent of the representative.

A basis of \(V/P\) identifies the quotient with a polynomial ring, so it is a domain and \(I_P\) is prime. For a root line, this gives the edge coefficient ring \(A/(\alpha_e)\).

The symmetric grading is transported from polynomial coordinates through the chosen algebra isomorphism. Linear coefficient actions preserve degree one and therefore every total degree.

Prime labels and cancellation in an edge ring

For a vector subspace \(P\subseteq V\), the linear ideal \(I_P\) kills precisely the vector generators from \(P\), and \[ \operatorname{Sym}(V)/I_P\simeq\operatorname{Sym}(V/P). \] The forward map sends \([\iota(v)]\) to \(\iota([v])\); the inverse reverses this rule on generators. A polynomial presentation of the right side proves that the quotient is a domain.

For a nonzero vector \(a\), the generator \(\iota(a)\) is prime. Divisibility of \(\iota(v)\) by \(\iota(a)\) is equivalent to \(v\in ka\). A vector outside that line survives as a nonzero prime linear element in the quotient.

Apply this to embedded positive roots. Distinct roots remain independent, so their labels do not divide one another. In particular, another incident root label survives nontrivially in an edge ring \(A/(\alpha)\). Since that ring is a polynomial domain, this reduced label acts injectively on every free edge module.

Coordinate presentations of the edge ring

For an actual positive root \(a\), let \(\Phi:A_{\mathrm{sym}}\to A\) be the coordinate isomorphism and \(r_a=\Phi(\iota(i_0a))\). The two quotient presentations are \[ \overline A_a=A_{\mathrm{sym}}/(\iota(i_0a)), \qquad E_a=A/(r_a). \] The induced inverse-coordinate map \[ \psi_a:E_a\to\overline A_a,\qquad [f]\mapsto[\Phi^{-1}(f)] \] sends each reduced polynomial root to the same reduced symmetric generator. The edge-quotient coordinate theorem then identifies both with the polynomial ring on \(V/\mathbb R(i_0a)\). Distinct positive roots give prime reduced labels there.

Bases after a scalar quotient

Let \(M\) have a finite \(R\)-basis \(b_i\), and let \(q:M\to N\) be surjective with kernel \(aM\), where \(N\) has a compatible \(R/(a)\)-action. Reducing the coordinates gives a map \[ \rho:M\to(R/(a))^I \] with the same kernel. The rule \(j(q(m))=\rho(m)\) is therefore a linear isomorphism. Any quotient scalar has an \(R\)-representative, so \(R\)-linearity makes \(j\) linear over \(R/(a)\). It sends \(q(b_i)\) to the standard basis.

The scalar-kernel basis theorem gives this proof in detail. In a boundary sheaf it applies to the upper restriction, whose scalar kernel is the root label times the upper stalk. The actual edge-module theorem obtains a free module over the edge coefficient ring.

Linear reduction of a plane

If a linear map \(l:V\to S\) kills \(a\), and \(c=sa+rb\) lies outside the line \(ka\), then \(r\ne0\) and \[ l(c)=r\,l(b). \] The image of the nonzero field scalar \(r\) is a unit in the target algebra, with inverse the image of \(r^{-1}\). Thus these two reduced labels are associated scalars. The plane-reduction theorem supplies the coefficient and its unit explicitly; it later applies to root labels in a localized edge quotient.

Changing coefficients and transporting duals

For a \(k\)-algebra isomorphism \(e:A\to B\), the transported module has action \[ b\star m=e^{-1}(b)m. \] The underlying vectors and \(k\)-linear structure stay the same. A homogeneous basis keeps its degrees, with old coefficient \(a_i\) replaced by the new coefficient \(e(a_i)\).

A functional transports by \[ \lambda^e(m)=e(\lambda(m)). \] It is \(B\)-linear because \(e(\lambda(e^{-1}(b)m))=b\,e(\lambda(m))\). Its inverse sends a new functional \(\mu\) to \(m\mapsto e^{-1}(\mu(m))\). These maps identify the transported old dual with the new dual and turn a perfect pairing into the pairing with values \(e(\langle m,n\rangle)\).

The scalar twist by a ring automorphism \(\tau\) instead uses \(r\star y=\tau(r)y\). When it fixes the base field, it is coefficient transport with \(e=\tau^{-1}\). The two-stalk lattice uses an ordinary first coordinate and this twisted second coordinate. The divisibility of \(\tau(r)-r\) by the root label supplies the correcting vector that makes its scalar action preserve the pair relation.

Tensor extension of principal quotients

For any commutative ring homomorphism \(\varphi:R\to S\), the tensor-quotient theorem gives \[ S\otimes_R R/(a)\simeq S/(\varphi(a)), \qquad s\otimes[r]\longmapsto[s\varphi(r)]. \] The inverse is \([s]\mapsto s\otimes[1]\). A multiple of \(\varphi(a)\) maps to \(s\otimes[a]=0\), which makes the inverse well defined; tensor balancing verifies the other composite.

For \(S=A_{\mathfrak p_P}\), this map identifies scalar extension of the original edge coefficient ring with the quotient by the localized root label. The formula fixes the actual image of every scalar.

Homogeneous functionals and inverse degrees

If a homogeneous basis vector has degree \(d_i\), its coordinate functional on \(M_n\) reads a coefficient in \(A_{n-d_i}\). The functional degree is therefore \(-d_i\).

For a perfect pairing with degree shift \(L\), a degree-\(r\) functional is represented in degree \(r+L\). Decomposing the representative and testing against homogeneous vectors isolates that component; injectivity of the pairing map identifies it with the full representative.

If a homogeneous scalar \(a\) of degree \(d\) acts injectively and \(am\) has degree \(n\), component transport gives \[ a\,\pi_{n-d}(m)=\pi_n(am)=am. \] Injectivity yields \(m=\pi_{n-d}(m)\). The later costalk calculation uses these two operations: the pairing adds its shift, and removing an incoming-label product subtracts its degree.

Hilbert functions, formal series and numerical values

The Hilbert function records the dimensions of the pieces: \[ h_M(n)=\dim_kM_n. \] For a nonnegative grading, the formal Hilbert series is \[ \operatorname{Hilb}(M;z)=\sum_{n\ge0}h_M(n)z^n. \] Its coefficients specify the dimensions. At a positive real \(q\) where the series is summable, its numerical Hilbert value is \[ H_M(q)=\sum_{n\in\mathbb Z}h_M(n)q^n. \] The source implements these numerical terms and values, including intersections \(S\cap M_n\) with a fixed ambient grading.

For \(A=k[X]\), there is one monomial in each nonnegative degree, so \[ H_A(q)=1+q+q^2+\cdots=\frac1{1-q}\qquad(0<q<1). \] For \(N\) variables, exponent tuples factor into \(N\) independent geometric series, giving \(H_A(q)=(1-q)^{-N}\). Grouping those tuples by total degree gives the dimension series.

Convergence from the finite free cover

The shifted free cover has the convergent series \[ H_F(q)=\frac{\sum_iq^{d_i}}{(1-q)^N}\qquad(0<q<1). \] Its degreewise surjections give \[ 0\le h_M(n)q^n\le h_F(n)q^n. \] Comparison of nonnegative series therefore proves convergence for every finite graded polynomial module. It also proves convergence for the intersections with any subspace \(S\subseteq M\), since their dimensions are bounded by \(h_M(n)\).

This supplies the numerical values of partial sheaf kernels before the reverse-kernel argument establishes their freeness.

Characters count homogeneous basis generators

When \(M\) is finite graded free with nonnegative homogeneous basis degrees \(d_i\), its graded-rank polynomial, called the character in the source, is \[ \operatorname{grk}_A(M;z)=\operatorname{ch}(M;z)=\sum_i z^{d_i}. \] A generator of degree \(d_i\) contributes a copy of the coefficient series shifted by that degree. Therefore \[ \operatorname{Hilb}(M;z)=\frac{\operatorname{ch}(M;z)}{(1-z)^N}, \qquad H_M(q)=\frac{\operatorname{ch}(M;q)}{(1-q)^N}. \]

Two homogeneous bases give the same Hilbert value for every \(q\in(0,1)\). Cancelling the nonzero denominator gives equal numerator evaluations on an infinite set, hence equal integer polynomials. This proves basis independence.

For the running example over \(k[X]\):

Invariant Value for \(A\oplus A(2)\)
Basis degrees \(0,2\)
Graded-rank polynomial \(1+z^2\)
Component dimensions \(1,1,2,2,\ldots\)
Numerical Hilbert value \((1+q^2)/(1-q)\)

Finite sums concatenate basis lists, shifts add to their degrees, and degree-preserving equivalences transport the lists. These operations give the corresponding character identities.

Minimal covers and the role of projectivity

Let \(J=(X_1,\ldots,X_N)\). Passing to \(M/JM\) removes all positive-degree scalar multiples and leaves a finite-dimensional real vector space. Choose a homogeneous basis of this quotient and lift its vectors to \(M\). These lifts define a map from a finite direct sum of shifted copies of \(A\) onto \(M\).

Graded Nakayama proves that this map is surjective. To see the mechanism, subtract a combination of the selected lifts from a homogeneous vector. The remaining vector lies in \(JM\), so it is expressed using positive-degree coefficients multiplying vectors of smaller degree. Repeat in those smaller degrees. The grading is bounded below, so the descent ends. The resulting map is a minimal graded free cover.

Projectivity is the splitting property: a surjection onto a projective module has a linear right inverse. If the target \(M\) above is projective, the free cover \(F\to M\) splits, with kernel \(K\). Minimality gives \(K\subseteq JF\); applying the projection onto \(K\) gives \(K=JK\).

The source's projective-to-free proof then uses a determinant argument. Finite generation writes generators of \(K\) as a matrix of \(J\)-coefficients times themselves. The determinant of the identity minus this matrix annihilates \(K\), but has constant term one and is therefore nonzero. Since \(K\) lies inside a free module over a domain, multiplication by that determinant is injective. Thus \(K=0\). The chosen lifts are a homogeneous free basis.

The reverse kernel argument first establishes projectivity, then applies this graded construction to obtain freeness.

Lifting maps and comparing minimal covers

A homogeneous basis turns a module map into a finite collection of prescribed vectors. To lift a map through a graded surjection, lift each basis image and retain the component of its required degree. The coefficient-degree formula then proves that the resulting linear map preserves degree.

For two minimal covers \(\pi:F\to T\) and \(\rho:G\to T\), a map \(f:F\to G\) covering the identity has a reverse homogeneous lift \(g\). The differences \(1-gf\) and \(1-fg\) land in the respective cover kernels, hence in \(JF\) and \(JG\).

A degree-zero self-map equal to the identity modulo the positive ideal is surjective by graded Nakayama. It is injective because the kernels of its powers stabilize in a Noetherian module. Thus \(gf\) and \(fg\) are invertible, which makes the specified map \(f\) bijective.

This comparison controls boundary-sheaf morphisms. Isomorphisms above a vertex identify its two boundary images; the remaining stalk maps become maps of minimal covers over a common target. Their bijectivity propagates downward from the normalized stalk.

A free cover with extra generators

Let \(\pi:P\to T\) be a minimal cover and \(f:F\to T\) a free graded cover. Homogeneous lifting gives \(g:F\to P\) and \(h_0:P\to F\) with \(\pi g=f\) and \(fh_0=\pi\). The difference \(1-gh_0\) maps into \(\ker\pi\subseteq JP\), so \(u=gh_0\) is invertible. Replacing \(h_0\) by \(h=h_0u^{-1}\) gives \(gh=1\).

Thus \[ F\simeq P\oplus\ker g,\qquad v\longmapsto(gv,v-hgv). \] The kernel is a graded projective summand and therefore graded free under the connected Noetherian-domain hypotheses. Its homogeneous generators record the extra shifts that must be added when a model decomposition extends over a new vertex.

For \(T=A/(\alpha)\), \(P=A\), and \(F=A\oplus A(2)\), with both covers given by the first coordinate modulo \(\alpha\), the remainder is \(A(2)\). This is the algebraic step behind the sheaf example \(O_y\oplus O_x(2)\).

Positive-ideal matrices and homogeneous degree

Let \(M\) have a homogeneous basis \(m_i\) of degrees \(d_i\), with coordinate functionals \(\lambda_i\). Suppose \(Q:M^*\to M\) maps \(\lambda_j\) into degree \(D-d_j\) and has image in \(A_+M\). Then \[ a_{ij}=\lambda_i(Q\lambda_j)\in A_{D-d_i-d_j}\cap A_+. \] The degree comes from applying a functional of degree \(-d_i\). Membership in \(A_+\) follows because an \(A\)-linear functional sends a sum of positive-ideal multiples to an element of that ideal.

If \(a_{ii}\ne0\), its degree is positive. Indeed negative pieces vanish, and a degree-zero element in \(A_+=\ker\epsilon\) equals its own zero augmentation. Therefore \(2d_i<D\). This is the positive-ideal diagonal-degree theorem. The local intersection map of a boundary stalk supplies its geometric application: boundary minimality gives the image containment and inverse duality gives \(D=\ell(b)-\ell(x)\).

Degreewise exactness and scalar quotients

Let \(f:M\to N\) be a surjective degree-zero map, with kernel \(K\). Homogeneous preimages make \[ 0\longrightarrow K_n\longrightarrow M_n\longrightarrow N_n\longrightarrow0 \] exact in every degree. Rank–nullity gives \(h_M(n)=h_K(n)+h_N(n)\). At a positive parameter with convergent source series, both right-hand series are dominated by the source, so summing gives \(H_M=H_K+H_N\).

For a homogeneous scalar \(a\) of degree \(d\) acting injectively, multiplication identifies \(M_{n-d}\) with \((aM)_n\). Thus \[ H_{aM}(q)=q^dH_M(q),\qquad H_{M/aM}(q)=(1-q^d)H_M(q). \] For a nonzero degree-one label, the second factor is \(1-q\).

The two normalizations used later are therefore \[ \operatorname{grk}_A(M;q)=(1-q)^N H_M(q) \] for an \(A\)-free module, and \[ H_{M/aM}(q)=(1-q)H_M(q) \] for a regular degree-one scalar. The first counts \(N\) coefficient variables; the second records one scalar quotient.

For \(A=k[X]\), \(M=A\oplus A(2)\), and \(a=X\), the quotient has one basis class in degree zero and one in degree two. Hence \[ H_{M/XM}(q)=1+q^2 =(1-q)\frac{1+q^2}{1-q}. \]

Actual images and changes of kernel conditions

For a homogeneous submodule \(S\subseteq M\) and a degree-zero map \(f:M\to N\), the restricted map onto \(f(S)\) gives \[ H_S(q)=H_{S\cap\ker f}(q)+H_{f(S)}(q). \] A homogeneous image vector has a homogeneous preimage inside \(S\), by component projection. This supplies exactness in each degree even when \(f\) is not onto the whole target.

In a sheaf, \(S\) is a partial outgoing kernel and \(f\) is one edge restriction. Adding that condition replaces \(S\) by \(S\cap\ker f\); its image term is exactly the Hilbert change.

Saturation and the reduced upper image

Suppose \(f:M\to N\) is linear and multiplication by \(\alpha\) on \(N\) is injective. For \(K=\ker f\), \[ \alpha m\in K\Longrightarrow \alpha f(m)=0 \Longrightarrow f(m)=0. \] Thus \(K\) is saturated with respect to \(\alpha\).

If a class from \(K/\alpha K\) maps to zero in \(M/\alpha M\), its representative is \(\alpha m\in K\). Saturation puts \(m\) in \(K\), proving injectivity of the reduced inclusion. If \(K\) is free over \(A\), reducing its basis gives a basis of \(K/\alpha K\) over \(A/(\alpha)\). The injected image is therefore an actual free quotient-ring module.

This is the algebra that identifies a free upper edge image from a free upper kernel and regularity on the remaining edge targets.

Projectivity from two presentations

A module is projective when its maps lift through linear surjections. Free modules have this property by lifting basis vectors. Direct summands of projective modules inherit it by extending a map along the retraction, lifting, and restricting along the inclusion.

Suppose \(f:F\to I\), \(g:E\to I\) are surjective, with \(F,E,\ker g\) projective. Choose lifts \[ a:F\to E,\quad b:E\to F,\qquad ga=f,\quad fb=g. \] The new kernel is a retract of \(F\oplus\ker g\): \[ i(x)=(x,a(x)),\qquad r(v,k)=v-ba(v)+b(k). \] The equation \(fr(v,k)=f(v)-ga(v)+g(k)=0\) puts the value in \(\ker f\), and \(ri(x)=x\). Therefore \(\ker f\) is projective.

For a finite free module over \(A/(\alpha)\), coordinate reduction from \(A^r\) gives a presentation whose kernel is \(\alpha A^r\). Over a domain with \(\alpha\ne0\), multiplication by \(\alpha\) identifies this kernel with \(A^r\). This supplies a concrete projective presentation for comparison.

Embedded kernels and images after scalar extension

For \(R\to S\) and an actual submodule \(P\subseteq M\), define \[ P_S=\operatorname{im}(S\otimes_R P\longrightarrow S\otimes_R M). \] This places the extended module inside a fixed ambient tensor module. If \(f:M\to N\), then \[ (f(P))_S=f_S(P_S) \] by expanding both sides into finite sums \(s\otimes f(p)\). This equality of images holds for any scalar extension.

If \(S\) is flat over \(R\), tensoring the exact sequence \(0\to\ker f\to M\to N\) gives \[ (\ker f)_S=\ker f_S. \] The embedded base-change theorem proves both assertions. Applying them to selected outgoing constraints and the two edge restrictions preserves the lower and upper images in one edge module. Localization is the case \(S=R_{\mathfrak p}\).

Prime denominators and membership in a target module

Let \(R\) be a UFD, let \(E\) be torsion-free, and let a vector \(x\in E\) have a nonzero multiplier \(c\) with \(cx\in M\). Prime-local membership supplies, for every prime element \(p\), a multiplier \(a\) with \(p\nmid a\) and \(ax\in M\).

To remove a factor \(p\) from \(c=pc_1\), set \(m=pc_1x\), \(n=ax\). They satisfy \(am=pc_1n\). In free coordinates of \(M\), primality makes every coordinate of \(m\) divisible by \(p\); division gives \(m=pm'\) inside \(M\). Torsion-free cancellation in \(E\) gives \(c_1x=m'\in M\). Repeating removes all prime factors.

For a projective \(M\), a section of its canonical free-module surjection supplies free coordinates and a retraction back to \(M\). The same argument divides there and then retracts, proving the projective-target version.

At a height-one prime \(\mathfrak p\), membership \(x/1\in M_{\mathfrak p}\) is precisely the existence of a denominator \(a\notin\mathfrak p\) with \(ax\in M\). This connects the module theorem to the local edge-image comparisons.

Defects and equality detection

For \(S\subseteq T\), put \(S_n=S\cap M_n\), \(T_n=T\cap M_n\). The quotient defects \(D_n=T_n/S_n\) have \[ \dim_kD_n=\dim_kT_n-\dim_kS_n. \] At a positive convergent parameter, \[ H_T(q)-H_S(q)=\sum_n\dim_kD_n\,q^n. \] Every term is nonnegative. Equality of the values makes every quotient dimension zero, so \(S_n=T_n\) in every degree.

If \(T\) is homogeneous, each vector of \(T\) is a finite sum of these common components and belongs to \(S\). Thus one such equality of values, together with the actual inclusion, gives \(S=T\).

For \(S=(X,Y)\subset T=k[X,Y]\), the only missing component is the constant one. The defect value is one. This is the same equality-detection mechanism applied to the two embedded edge images in the invariance argument.

For two surjective projections, a common endomorphism may act through different scalars on the two targets. Their difference annihilates the pair-projection quotient: subtracting one scalar multiple of a lift produces a relation vector with zero second projection. The word-module construction uses this calculation to obtain its reflection-root edge labels.

Reflected characters from dual bases

Suppose a linear equivalence from a module dual to a nonnegative graded module sends degree-\(r\) functionals to degree \(r+D\). A homogeneous basis of the original module with degrees \(d_i\) has dual basis degrees \(-d_i\). Its images therefore have degrees \(D-d_i\).

Each image basis vector is nonzero, so nonnegativity forces \(d_i\le D\). The character of the target is consequently \[ \sum_iq^{D-d_i} =\operatorname{reflect}_D\left(\sum_iq^{d_i}\right). \] This simultaneously proves the degree bound needed for reciprocal evaluation and supplies the outgoing-kernel character in the sheaf branch.

Hilbert numerators from degree shifts

If three dimension sequences satisfy \(f(n)=g(n)+h(n-1)\), their convergent sums satisfy \(H_f(t)=H_g(t)+tH_h(t)\). When the sums have polynomial numerators with a common nonzero factor, cancellation gives an equality of evaluations throughout \((0,1)\). The polynomial root bound then yields the formal identity \[ F(q)=G(q)+qH(q). \] This numerator theorem is the bridge from the real degreewise exact sequences of word stalks to their Hecke coefficient recursions.

Quotient scalar images and tensor denominators

An \(A\)-submodule of a module already acted on by \(C=A/(a)\) is stable under \(C\): choose an \(A\)-representative of each quotient scalar and use compatibility of actions. The scalar-image theorem identifies multiplication by \(z\in A\) with multiplication by its class in \(C\). If that action is injective, it transfers a free \(C\)-basis to the scalar image. This is the algebra used for the corrected upper edge image.

For a localization \(S=P^{-1}A\), the embedded tensor extension of a submodule consists of fractions with numerators in it. If \(1\otimes x\) belongs to that image, write it as \(m/s\). Equality of fractions supplies another \(t\in P\) with \(tm=(ts)x\), so \(ts\) sends \(x\) into the original submodule. The tensor-denominator theorem retains this second factor, which is needed for arbitrary modules.

Homogeneous image maps in the Hilbert calculation

A degree-\(d\) map sends the degree-\((n-d)\) part of a vector to the degree-\(n\) part of its image. For a homogeneous submodule \(S\), taking that component of any preimage makes the map \[ S_{n-d}\longrightarrow f(S)_n \] surjective. If \(f\) is injective, this is a piece isomorphism. Translation of the summable integer-indexed series gives \(H_{f(S)}(q)=q^dH_S(q)\), as proved in the homogeneous-image shift theorem.

For a degree-zero map with kernel \(aM\), saturation gives the restricted kernel \(aS\). Graded exactness then gives \(H_S=H_{aS}+H_{f(S)}\), and the scalar shift gives \(H_{f(S)}=(1-q^{\deg a})H_S\). The saturated-image Hilbert theorem states the exact action, regularity, finite-piece and convergence hypotheses. These are the two calculations used for corrected upper edge images.

Definitions and proofs by construction

The realization coefficient algebra

Grading and the positive ideal

Graded modules and maps

Changing coefficients and twisting

Bases, minimal covers and projectivity

Hilbert series, Hilbert values and characters

Algebra for support, quotients and comparison