Theorem One multiplicity family controls stalk and full-kernel characters
✓ theorem proved

One multiplicity family controls stalk and full-kernel characters

Let \((O_c)\) be a boundary-model family over \(A=k[X_1,\ldots,X_N]\), with \(k\) a field, on a graph with finitely many vertices. Let \(C\) be a model sum centered in \(U\).

Its construction gives polynomials \(\mu_c(q)\in\mathbb Z[q]\), with nonnegative coefficients and \(\mu_c=0\) for \(c\notin U\), such that \[ C_C(x;q)=\sum_c\mu_c(q)C_{O_c}(x;q). \] If every \(K_x(O_c)\) is free, then every \(K_x(C)\) is free and the same polynomials satisfy \[ K_C(x;q)=\sum_c\mu_c(q)K_{O_c}(x;q). \]

A graded sheaf isomorphism from another sheaf to \(C\) transports both formulas to that sheaf.

Proof

Step 1. Initialize the multiplicities

For the zero constructor, set \(\mu_c=0\) for every \(c\). All stalks and full outgoing kernels are zero, so both formulas hold. Their freeness and zero characters follow from the zero-module calculation.

Step 2. Update one center

Suppose the formulas hold for a preceding sum \(C\), and the next constructor adds \[ S=\bigoplus_{j=1}^mO_a(d_j) \] at \(a\in U\). Put \[ p(q)=\sum_{j=1}^m q^{d_j},\qquad \mu'_c(q)= \begin{cases} \mu_c(q)+p(q),&c=a,\\ \mu_c(q),&c\ne a. \end{cases} \] All exponents are nonnegative. The coefficients of \(p\) count the shifts of this block, so they are nonnegative; adding them preserves nonnegativity of \(\mu'_a\).

If \(c\notin U\), then \(c\ne a\) and its previous polynomial was zero, so \(\mu'_c=0\).

Step 3. Verify the stalk identity after the update

At a vertex \(x\), \[ (C\oplus S)^x=C^x\oplus\bigoplus_jO_a^x(d_j). \] Character additivity and the shift formula give \[ \begin{aligned} C_{C\oplus S}(x;q) &=C_C(x;q)+p(q)C_{O_a}(x;q)\\ &=\sum_c\mu'_c(q)C_{O_c}(x;q). \end{aligned} \] This proves the stalk identity by induction on the construction.

Step 4. Verify the kernel identity using the same update

Assume the model kernels are free. The kernel transport theorem gives \[ K_x(C\oplus S) \simeq K_x(C)\oplus\bigoplus_jK_x(O_a)(d_j). \] The induction hypothesis and the model assumptions make this a finite sum of free modules, so it is free.

Its character is \[ K_{C\oplus S}(x;q) =K_C(x;q)+p(q)K_{O_a}(x;q) =\sum_c\mu'_c(q)K_{O_c}(x;q). \] The update polynomial \(p\) is the same one used in Step 3, because the stalk and kernel summands come from the same added sheaves.

Step 5. Transport through a sheaf isomorphism

If \(B\simeq C\) as graded sheaves, its stalk maps preserve characters. The induced full-kernel isomorphisms do the same. Thus both identities transfer with the same multiplicities.

In particular, on a finite graph, the model-decomposition theorem applies to a free-stalk, flabby, globally generated sheaf with the upper-quotient rule. Together with free full kernels for the models and the given sheaf, it yields the source's simultaneous character-decomposition statement.

Step 6. Evaluate the running example

For \(C=O_y\oplus O_x(2)\), the multiplicities are \(\mu_y=1\), \(\mu_x=q^2\). At the lower vertex, \[ C_C(x;q)=1\cdot1+q^2\cdot1=1+q^2, \] \[ K_C(x;q)=1\cdot q+q^2\cdot1=q+q^2. \] Both formulas use the same recorded shifts.

References

The local source declarations below give the constructions and proofs.

Lean formalization

OAI.KLInvariance.MomentGraph.GradedSheaf.ModelSum.character: Decomposition.lean:23.

OAI.KLInvariance.MomentGraph.GradedSheaf.ModelSum.kernelFree: CharacterReciprocity.lean:694.

OAI.KLInvariance.MomentGraph.GradedSheaf.ModelSum.character_both: CharacterReciprocity.lean:710.

OAI.KLInvariance.MomentGraph.GradedSheaf.exists_character_decomposition: CharacterReciprocity.lean:1978.

Used by