Theorem Indecomposability of the normalized boundary sheaf
✓ theorem proved

Indecomposability of the normalized boundary sheaf

Over a nonnegatively graded domain, the normalized minimal-boundary sheaf on a finite ordered graph has no nontrivial graded idempotent endomorphism: \[ p^2=p\quad\Longrightarrow\quad p=0\ \text{or}\ p=1. \] Equivalently, its identity is a primitive projector, so the sheaf cannot split into two nonzero graded direct summands.

Proof

Step 1. Determine the idempotent on the top stalk

Let \(p:B\to B\) be a graded sheaf endomorphism with \(p^2=p\). At the top stalk \(B^b=A\), an \(A\)-linear endomorphism is multiplication by \(c=p_b(1)\). Idempotence gives \(c^2=c\), or \(c(c-1)=0\). Since \(A\) is a domain, \(c=0\) or \(c=1\).

We first prove that \(p_b=0\) forces the entire endomorphism to vanish.

Step 2. Propagate zero through the edge modules above a vertex

Assume by downward induction that \(p_y=0\) for every \(y>x\). For any outgoing edge \(e:x\to y\), the upper map \(B^y\twoheadrightarrow B^e\) is surjective. Given \(z\in B^e\), lift it to \(v\in B^y\). Commutation of the upper square gives \[ p_e(z)=p_e(\rho_{y,e}(v))=\rho_{y,e}(p_y(v))=0. \] Thus all those edge maps vanish.

Step 3. Use minimality to eliminate p at x

For \(v\in B^x\), commutation of the lower square gives \(\rho_{x,e}(p_xv)=p_e(\rho_{x,e}v)=0\). Hence \(\operatorname{im}p_x\) lies in the full outgoing kernel, and minimality at a non-top vertex puts it inside \(JB^x\).

Let \(N=\operatorname{im}p_x\). Since \(p_x\) is idempotent, it acts as the identity on \(N\). Applying \(p_x\) to \(N\subseteq JB^x\) gives \(N=p_x(N)\subseteq Jp_x(B^x)=JN\). Thus \(N=JN\). The map is graded, so its image is graded and bounded below. Graded Nakayama gives \(N=0\), hence \(p_x=0\).

Step 4. Finish both alternatives

Finite downward induction starts with the zero top map and proves every stalk map zero. Surjectivity of all upper restrictions then makes every edge map zero.

If instead \(p_b=1\), apply the zero-top argument to \(1-p\). Its square is itself because \(p^2=p\), and its top map is zero. Therefore \(1-p=0\), so \(p=1\).

A direct sum decomposition would produce the projection onto one summand, an idempotent. The conclusion says one summand must be zero, proving indecomposability in the stated graded sense.

References

Related paper context.

Lean formalization

OAI.KLInvariance.MomentGraph.BoundarySheaf.primitive_identity: BalancedPairings.lean:1974.

Used by