Documentation

TauCeti.RepresentationTheory.Quiver.Reflection.Indecomposable

Reflecting an indecomposable representation at a sink #

The Bernstein-Gelfand-Ponomarev reflection at a sink i replaces the vertex space Mᵢ by the kernel of the sum TauCeti.incomingSum of the arrows into i, and it acts on dimension vectors by the simple reflection sᵢ whenever that sum is onto (TauCeti.dimVector_reflectRep). This file discharges that hypothesis: the sum of the arrows into a sink is onto for every indecomposable representation except the vertex simple Sᵢ, which is the representation the reflection annihilates.

One construction does the work. A linear endomorphism π of Mᵢ that the paths through i do not see extends by the identity at the other vertices to an endomorphism TauCeti.vertexEnd of the whole representation, idempotent as soon as π is. The category of representations is abelian, hence idempotent complete, so an indecomposable object carries no idempotent endomorphism besides 0 and the identity (TauCeti.idempotent_eq_zero_or_id_of_indecomposable), and therefore π itself is 0 or the identity. Applied to the projection onto the range of the incoming sum along a complement, this says the range is all of Mᵢ unless M vanishes away from i; applied to the projection onto a line, it says a representation concentrated at i is a line there, hence is Sᵢ, provided i carries no nontrivial closed path.

TauCeti.vertexEnd is stated at a vertex, not at a sink: naturality needs only that no path from elsewhere into i sees π before it, that no path from i to elsewhere sees π after it, and that π commutes with the maps of closed paths at i. At a sink the second condition is vacuous and all closed paths are trivial; the dual construction at a source, in TauCeti.RepresentationTheory.Quiver.Reflection.Source.Indecomposable, is the same theorem with the first condition vacuous instead.

Main definitions #

Main results #

Implementation notes #

TauCeti.sinkIdempotent_eq_id_or_eq_zero_of_indecomposable takes its hypothesis on π in the form "π fixes the image of each arrow into i" rather than "π fixes the range of TauCeti.incomingSum". The two agree on a finite quiver, by TauCeti.map_toPath_mem_range_incomingSum, but the arrow form needs no finiteness, and neither then do the construction, its recognition lemmas, or the statement that an indecomposable representation concentrated at a sink is a line there. Finiteness enters only where TauCeti.incomingSum does.

The results comparing M with TauCeti.simpleRep k Q i are stated in the last section, where the quiver, its arrows and the field all live in one universe. This is forced, not chosen: the vertex spaces of a reflected representation are cut out of a product indexed by the arrows, so TauCeti.reflectRep needs them in max v w x, while Sᵢ puts the field itself at i and so needs them in the universe of the field. A statement mentioning both therefore has to identify the two, and the general statements above -- which name no vertex simple -- are proved in the wider setting and specialize to it.

As in the neighbouring files, a vertex i : Q is used as an object of the free category CategoryTheory.Paths Q, which is Q only by unfolding a semireducible definition. Rewriting cannot see through that identification, so every step that branches on equality of vertices is factored through a lemma quantified over Q, and the components of TauCeti.vertexEnd are supplied by the private vertexEndApp; TauCeti.vertexEnd_app_self and TauCeti.vertexEnd_app_of_ne are the interface.

References #

This is the identity dim (C⁺ᵢ M) = sᵢ · dim M with its hypothesis discharged. See Bernstein--Gelfand--Ponomarev, Coxeter functors and Gabriel's theorem, and Derksen--Weyman, An Introduction to Quiver Representations, Ch. 2.

The endomorphism attached to an idempotent at a vertex #

noncomputable def TauCeti.vertexEnd {k : Type u} {Q : Type v} [Field k] [Quiver Q] (M : QuiverRep k Q) {i : Q} (π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)) (hin : ∀ (a : Q) (p : Quiver.Path a i), a ≠ i → ∀ (z : ↑(M.obj a)), π ((ModuleCat.Hom.hom (M.map p)) z) = (ModuleCat.Hom.hom (M.map p)) z) (hout : ∀ (b : Q) (p : Quiver.Path i b), b ≠ i → ∀ (y : ↑(M.obj i)), (ModuleCat.Hom.hom (M.map p)) (π y) = (ModuleCat.Hom.hom (M.map p)) y) (hcomm : ∀ (p : Quiver.Path i i), CategoryTheory.CategoryStruct.comp (M.map p) (ModuleCat.ofHom π) = CategoryTheory.CategoryStruct.comp (ModuleCat.ofHom π) (M.map p)) :
M ⟶ M

The endomorphism of M given by an endomorphism at one vertex. For a linear endomorphism π of Mᵢ that no path into i from elsewhere sees before it (hin), that no path out of i to elsewhere sees after it (hout), and that commutes with every closed-path map at i (hcomm), this is the endomorphism of M acting by π at i and by the identity at every other vertex.

The three hypotheses hold at a sink and, dually, at a source; they are what naturality asks for, one case of the square for each of the four positions of i relative to the ends of a path.

Instances For
    @[simp]
    theorem TauCeti.vertexEnd_app_self {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} {π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)} {hin : ∀ (a : Q) (p : Quiver.Path a i), a ≠ i → ∀ (z : ↑(M.obj a)), π ((ModuleCat.Hom.hom (M.map p)) z) = (ModuleCat.Hom.hom (M.map p)) z} {hout : ∀ (b : Q) (p : Quiver.Path i b), b ≠ i → ∀ (y : ↑(M.obj i)), (ModuleCat.Hom.hom (M.map p)) (π y) = (ModuleCat.Hom.hom (M.map p)) y} {hcomm : ∀ (p : Quiver.Path i i), CategoryTheory.CategoryStruct.comp (M.map p) (ModuleCat.ofHom π) = CategoryTheory.CategoryStruct.comp (ModuleCat.ofHom π) (M.map p)} :
    (vertexEnd M π hin hout hcomm).app i = ModuleCat.ofHom π

    At its vertex, TauCeti.vertexEnd acts by the given endomorphism.

    @[simp]
    theorem TauCeti.vertexEnd_app_of_ne {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} {π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)} {hin : ∀ (a : Q) (p : Quiver.Path a i), a ≠ i → ∀ (z : ↑(M.obj a)), π ((ModuleCat.Hom.hom (M.map p)) z) = (ModuleCat.Hom.hom (M.map p)) z} {hout : ∀ (b : Q) (p : Quiver.Path i b), b ≠ i → ∀ (y : ↑(M.obj i)), (ModuleCat.Hom.hom (M.map p)) (π y) = (ModuleCat.Hom.hom (M.map p)) y} {hcomm : ∀ (p : Quiver.Path i i), CategoryTheory.CategoryStruct.comp (M.map p) (ModuleCat.ofHom π) = CategoryTheory.CategoryStruct.comp (ModuleCat.ofHom π) (M.map p)} {a : Q} (ha : a ≠ i) :
    (vertexEnd M π hin hout hcomm).app a = CategoryTheory.CategoryStruct.id (M.obj a)

    Away from its vertex, TauCeti.vertexEnd acts by the identity.

    theorem TauCeti.vertexEnd_comp_self {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} {π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)} {hin : ∀ (a : Q) (p : Quiver.Path a i), a ≠ i → ∀ (z : ↑(M.obj a)), π ((ModuleCat.Hom.hom (M.map p)) z) = (ModuleCat.Hom.hom (M.map p)) z} {hout : ∀ (b : Q) (p : Quiver.Path i b), b ≠ i → ∀ (y : ↑(M.obj i)), (ModuleCat.Hom.hom (M.map p)) (π y) = (ModuleCat.Hom.hom (M.map p)) y} {hcomm : ∀ (p : Quiver.Path i i), CategoryTheory.CategoryStruct.comp (M.map p) (ModuleCat.ofHom π) = CategoryTheory.CategoryStruct.comp (ModuleCat.ofHom π) (M.map p)} (hidem : IsIdempotentElem π) :
    CategoryTheory.CategoryStruct.comp (vertexEnd M π hin hout hcomm) (vertexEnd M π hin hout hcomm) = vertexEnd M π hin hout hcomm

    TauCeti.vertexEnd is idempotent as soon as the endomorphism it is built from is.

    @[simp]
    theorem TauCeti.vertexEnd_eq_id_iff {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} {π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)} {hin : ∀ (a : Q) (p : Quiver.Path a i), a ≠ i → ∀ (z : ↑(M.obj a)), π ((ModuleCat.Hom.hom (M.map p)) z) = (ModuleCat.Hom.hom (M.map p)) z} {hout : ∀ (b : Q) (p : Quiver.Path i b), b ≠ i → ∀ (y : ↑(M.obj i)), (ModuleCat.Hom.hom (M.map p)) (π y) = (ModuleCat.Hom.hom (M.map p)) y} {hcomm : ∀ (p : Quiver.Path i i), CategoryTheory.CategoryStruct.comp (M.map p) (ModuleCat.ofHom π) = CategoryTheory.CategoryStruct.comp (ModuleCat.ofHom π) (M.map p)} :

    TauCeti.vertexEnd is the identity exactly when the endomorphism it is built from is.

    @[simp]
    theorem TauCeti.vertexEnd_eq_zero_iff {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} {π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)} {hin : ∀ (a : Q) (p : Quiver.Path a i), a ≠ i → ∀ (z : ↑(M.obj a)), π ((ModuleCat.Hom.hom (M.map p)) z) = (ModuleCat.Hom.hom (M.map p)) z} {hout : ∀ (b : Q) (p : Quiver.Path i b), b ≠ i → ∀ (y : ↑(M.obj i)), (ModuleCat.Hom.hom (M.map p)) (π y) = (ModuleCat.Hom.hom (M.map p)) y} {hcomm : ∀ (p : Quiver.Path i i), CategoryTheory.CategoryStruct.comp (M.map p) (ModuleCat.ofHom π) = CategoryTheory.CategoryStruct.comp (ModuleCat.ofHom π) (M.map p)} :
    vertexEnd M π hin hout hcomm = 0 ↔ π = 0 ∧ ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)

    TauCeti.vertexEnd vanishes exactly when the endomorphism it is built from vanishes and the representation is concentrated at its vertex: elsewhere it acts by the identity, which is zero only on a vanishing vertex space.

    Idempotents at a vertex of an indecomposable representation #

    theorem TauCeti.vertexIdempotent_eq_id_or_eq_zero_of_indecomposable {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} (hM : CategoryTheory.Indecomposable M) (π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)) (hin : ∀ (a : Q) (p : Quiver.Path a i), a ≠ i → ∀ (z : ↑(M.obj a)), π ((ModuleCat.Hom.hom (M.map p)) z) = (ModuleCat.Hom.hom (M.map p)) z) (hout : ∀ (b : Q) (p : Quiver.Path i b), b ≠ i → ∀ (y : ↑(M.obj i)), (ModuleCat.Hom.hom (M.map p)) (π y) = (ModuleCat.Hom.hom (M.map p)) y) (hcomm : ∀ (p : Quiver.Path i i), CategoryTheory.CategoryStruct.comp (M.map p) (ModuleCat.ofHom π) = CategoryTheory.CategoryStruct.comp (ModuleCat.ofHom π) (M.map p)) (hidem : IsIdempotentElem π) :
    π = LinearMap.id ∨ π = 0 ∧ ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)

    An idempotent at a vertex of an indecomposable representation is trivial. Let M be an indecomposable representation. An idempotent endomorphism of Mᵢ invisible to the paths through i, in the sense of TauCeti.vertexEnd, is either the identity, or zero — and in the second case M vanishes away from i.

    theorem TauCeti.sinkIdempotent_eq_id_or_eq_zero_of_indecomposable {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} (hi : Quiver.IsSink i) (hM : CategoryTheory.Indecomposable M) (π : ↑(M.obj i) →ₗ[k] ↑(M.obj i)) (hπ : ∀ (b : Q) (e : b ⟶ i) (z : ↑(M.obj b)), π ((ModuleCat.Hom.hom (M.map e.toPath)) z) = (ModuleCat.Hom.hom (M.map e.toPath)) z) (hidem : IsIdempotentElem π) :
    π = LinearMap.id ∨ π = 0 ∧ ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)

    An idempotent at a sink of an indecomposable representation is trivial. Let i be a sink of Q and M an indecomposable representation. An idempotent endomorphism of Mᵢ fixing the image of every arrow into i is either the identity, or zero — and in the second case M vanishes away from i.

    theorem TauCeti.exists_ne_zero_span_eq_top_of_forall_subsingleton {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} (hloop : ∀ (p : Quiver.Path i i), p = Quiver.Path.nil) (hM : CategoryTheory.Indecomposable M) (h : ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)) :
    ∃ (y : ↑(M.obj i)), y ≠ 0 ∧ k ∙ y = ⊤

    An indecomposable representation concentrated at one vertex is a line there: if M vanishes away from a vertex i carrying no closed path but the trivial one, its vertex space at i is spanned by a single nonzero vector.

    theorem TauCeti.finrank_end_eq_one_of_forall_subsingleton {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} (hloop : ∀ (p : Quiver.Path i i), p = Quiver.Path.nil) (hM : CategoryTheory.Indecomposable M) (h : ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)) :

    An indecomposable representation concentrated at a vertex with no nontrivial closed path has a one-dimensional endomorphism space.

    theorem TauCeti.dimVector_eq_single_of_forall_subsingleton {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} [DecidableEq Q] (hloop : ∀ (p : Quiver.Path i i), p = Quiver.Path.nil) (hM : CategoryTheory.Indecomposable M) (h : ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)) :
    (fun (j : Q) => ↑(dimVector M j)) = Pi.single i 1

    An indecomposable representation concentrated at a vertex carrying no nontrivial closed path has the corresponding simple dimension vector.

    theorem TauCeti.nonempty_iso_of_forall_subsingleton {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} (hloop : ∀ (p : Quiver.Path i i), p = Quiver.Path.nil) {N : QuiverRep k Q} (hM : CategoryTheory.Indecomposable M) (hN : CategoryTheory.Indecomposable N) (hMi : ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)) (hNi : ∀ (a : Q), a ≠ i → Subsingleton ↑(N.obj a)) :

    Two indecomposable representations concentrated at the same vertex are isomorphic, provided that vertex carries no closed path but the trivial one. Each is a line at that vertex and zero elsewhere, and no arrow joins the vertex to itself, so any linear isomorphism of the two lines is an isomorphism of representations.

    The surjectivity of the incoming sum #

    theorem TauCeti.incomingSum_surjective_or_forall_subsingleton {k : Type u} {Q : Type v} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} [Fintype Q] [(a b : Q) → Fintype (a ⟶ b)] (hi : Quiver.IsSink i) (hM : CategoryTheory.Indecomposable M) :
    Function.Surjective ⇑(incomingSum M i) ∨ ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)

    The sum of the arrows into a sink of an indecomposable representation is onto, unless the representation is concentrated at that sink. This is the surjectivity hypothesis of TauCeti.dimVector_reflectRep, discharged: the range of the incoming sum is a direct summand of Mᵢ, so a proper range would split M, and the only splitting an indecomposable representation admits leaves nothing outside the sink.

    The vertex simple as the exceptional case #

    theorem TauCeti.nonempty_iso_simpleRep_of_forall_subsingleton {k Q : Type u} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} (hloop : ∀ (p : Quiver.Path i i), p = Quiver.Path.nil) (hM : CategoryTheory.Indecomposable M) (h : ∀ (a : Q), a ≠ i → Subsingleton ↑(M.obj a)) :

    An indecomposable representation concentrated at one vertex is the vertex simple there, provided that vertex carries no closed path but the trivial one.

    theorem TauCeti.incomingSum_surjective_of_indecomposable {k Q : Type u} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} [Fintype Q] [(a b : Q) → Fintype (a ⟶ b)] (hi : Quiver.IsSink i) (hM : CategoryTheory.Indecomposable M) (hne : ¬Nonempty (M ≅ simpleRep k Q i)) :

    The sum of the arrows into a sink is onto for every indecomposable representation other than the vertex simple there. The vertex simple Sᵢ is genuinely excluded, by TauCeti.incomingSum_not_surjective.

    theorem TauCeti.dimVector_reflectRep_of_indecomposable {k Q : Type u} [Field k] [Quiver Q] {M : QuiverRep k Q} {i : Q} [Fintype Q] [(a b : Q) → Fintype (a ⟶ b)] [DecidableEq Q] (hi : Quiver.IsSink i) (hM : CategoryTheory.Indecomposable M) (hne : ¬Nonempty (M ≅ simpleRep k Q i)) (hfin : ∀ (e : (b : Q) × (b ⟶ i)), FiniteDimensional k ↑(M.obj e.fst)) :
    (fun (j : Q) => ↑(dimVector (reflectRep M hi) j)) = (vertexPreReflection Q i) fun (j : Q) => ↑(dimVector M j)

    The reflection at a sink acts on the dimension vector of an indecomposable representation by the simple reflection there, unless that representation is the vertex simple at the sink, which the reflection annihilates instead. This is the identity dim (C⁺ᵢ M) = sᵢ · dim M with its hypothesis discharged.