Class groups of cluster algebras Daniel Smertnig University of - - PowerPoint PPT Presentation

class groups of cluster algebras
SMART_READER_LITE
LIVE PREVIEW

Class groups of cluster algebras Daniel Smertnig University of - - PowerPoint PPT Presentation

Class groups of cluster algebras Daniel Smertnig University of Waterloo with Ana Garcia Elsener (Graz, Austria) and Philipp Lampe (Kent, UK) arXiv:1712.06512 Cluster algebras Cluster algebras were introduced by Fomin and Zelevinsky in 2002.


slide-1
SLIDE 1

Class groups of cluster algebras

Daniel Smertnig

University of Waterloo with Ana Garcia Elsener (Graz, Austria) and Philipp Lampe (Kent, UK) arXiv:1712.06512

slide-2
SLIDE 2

Cluster algebras

Cluster algebras were introduced by Fomin and Zelevinsky in 2002. More than 600 preprints on the arXiv. Connections to many difgerent area of mathematics: T

  • tal

positivity, combinatorics, T eichmüller theory, representation theory, knot theory, Lie algebras, … Defjned via combinatorial data: Quivers and mutations.

slide-3
SLIDE 3

Quivers

Quiver: fjnite directed graph (for us:) no loops or 2-cycles parallel arrows allowed. 1

1 2

1 2

1 3 4 2 acyclic 1 3 4 2 with cycle(s)

slide-4
SLIDE 4

Quiver mutations I

Mutation of a quiver 𝑅 at vertex 𝑗. 1 2 3

  • 1. For arrows 𝑘 → 𝑗 → 𝑙, add arrows

𝑘 → 𝑙. 1 2 3

  • 2. Flip all arrows incident

with 𝑗. 1 2 3 = 2 1 3

  • 3. Remove 2-cycles.

Parallel mutation of seed: In {𝑦1, … , 𝑦𝑜} replace 𝑦𝑗 by 𝑦′

𝑗:

𝑦𝑗𝑦′

𝑗 = ∏ 𝑘→𝑗

𝑦𝑘 + ∏

𝑗→𝑘

𝑦𝑘. {𝑦1, 𝑦2, 𝑦3} ⇝ {𝑦1, 𝑦1+𝑦3

𝑦2

, 𝑦3}

slide-5
SLIDE 5

Quiver mutations II

2 1 3 {𝑦1, 𝑦1+𝑦3

𝑦2

, 𝑦3}

  • 1. For arrows 𝑘 → 𝑗 → 𝑙, add arrows

𝑘 → 𝑙. 2 1 3

  • 2. Flip all arrows incident with 𝑗.

2 1 3

  • 3. Remove 2-cycles.

2 1 3 ⇝ 1 2 3 𝑦1𝑦′

1 = 𝑦′ 2 + 𝑦3, so new seed { 𝑦1+(1+𝑦2)𝑦3 𝑦1𝑦2

, 𝑦1+𝑦3

𝑦2

, 𝑦3}.

slide-6
SLIDE 6

Cluster algebras

Let 𝑅 be a quiver on vertices {1, … , 𝑜} and {𝑦1, … , 𝑦𝑜} an initial seed. Mutation yields a (possibly infjnite) collection of seeds. Each element of a seed is a cluster variable. Defjnition The cluster algebra 𝐵 = 𝐵(𝑅) is the subalgebra of ℤ(𝑦1, … , 𝑦𝑜) generated by all cluster variables. ℤ[𝑦1, … , 𝑦𝑜] ⊂ 𝐵 ⊂ ℤ(𝑦1, … , 𝑦𝑜).

slide-7
SLIDE 7

Cluster algebras

Example (𝐵3) 𝐵3: 1 2 3 𝐵(𝐵3) = ℤ[𝑦1, 𝑦2, 𝑦3, 1 + 𝑦2 𝑦1 , 𝑦1 + 𝑦3 𝑦2 , 1 + 𝑦2 𝑦3 , 𝑦1 + (1 + 𝑦2)𝑦3 𝑦1𝑦2 , (1 + 𝑦2)𝑦1 + 𝑦3 𝑦2𝑦3 , (1 + 𝑦2)(𝑦1 + 𝑦3) 𝑦1𝑦2𝑦3 ]

slide-8
SLIDE 8

Laurent phenomenon

Theorem (Fomin, Zelevinsky 2002) Denominators of cluster variables are monomials, hence 𝐵 ⊆ ℤ[𝑦±1

1 , … , 𝑦±1 𝑜 ].

slide-9
SLIDE 9

Finite type classifjcation

Theorem (Fomin, Zelevinsky 2003) Cluster algebras of fjnite type (=having fjnitely many cluster variables) are classifjed by Dynkin diagrams. 𝐵𝑜 𝐶𝑜 𝐷𝑜 𝐸𝑜 𝐹6 𝐹7 𝐹8 𝐺4 𝐻2

slide-10
SLIDE 10

Goal Understand factorizations of elements into atoms (irreducible elements) in cluster algebras.

1 When is 𝐵(𝑅) factorial (a UFD)? 2 What happens if it is not?

slide-11
SLIDE 11

Factorizations: earlier results

Theorem (Geiß, Leclerc, Schröer, 2012)

1 Cluster variables are (pairwise non-associated) atoms. 2 If 𝐵 is factorial, all exchange polynomials

𝑔𝑗 ∈ ℤ[𝑦1, … , 𝑦𝑜] with 𝑦𝑗𝑦′

𝑗 = 𝑔𝑗 are irreducible and

pairwise distinct. Example If 𝑔𝑗 = 𝑕1 ⋯ 𝑕𝑙, then 𝑦𝑗𝑦′

𝑗 = 𝑕1 ⋯ 𝑕𝑙.

If 𝑔𝑗 = 𝑔𝑘 for 𝑗 ≠ 𝑘, then 𝑦𝑗𝑦′

𝑗 = 𝑦𝑘𝑦′ 𝑘.

Theorem (Lampe, 2012, 2014) Classifjcation of factoriality for simply-laced Dynkin types (𝐵𝑜, 𝐸𝑜, 𝐹𝑜).

slide-12
SLIDE 12

Acyclic cluster algebras

𝑦′

𝑗... obtained from initial seed {𝑦1, … , 𝑦𝑜} by mutation at 𝑗:

𝑦𝑗𝑦′

𝑗 = 𝑔𝑗

with 𝑔𝑗 = ∏

𝑘→𝑗

𝑦𝑘 + ∏

𝑗→𝑘

𝑦𝑘 exchange polynomials. Theorem (Berenstein, Fomin, Zelevinsky 2006; Muller 2014) Let 𝑅 be acyclic. Then 𝐵 = ℤ[𝑦1, 𝑦′

1, … , 𝑦𝑜, 𝑦′ 𝑜] ≅ ℤ[𝑌1, 𝑌′ 1, … , 𝑌𝑜, 𝑌′ 𝑜]/(𝑌𝑗𝑌′ 𝑗 −𝑔𝑗).

𝐵 is fjnitely generated, noetherian, integrally closed. Corollary (Locally) acyclic cluster algebras are Krull domains.

slide-13
SLIDE 13

Krull domains

Theorem Let 𝐵 be a Krull domain with divisor class group 𝐻 = 𝒟(𝐵) and 𝐻0 = { [𝔮] ∶ 𝔮 divisorial [=height-1] prime } ⊆ 𝐻. Then there exists a transfer homomorphism 𝜒∶ (𝐵 ∖ {0}, ⋅) → ℬ(𝐻0), with ℬ(𝐻0) the monoid of zero-sum sequences over 𝐻0. Corollary

1 𝐵 is factorial (= a UFD) if and only if 𝐻 is trivial. 2 Factorization theory of 𝐵 determined by 𝐻 and 𝐻0.

slide-14
SLIDE 14

First Main Result

Theorem (Garcia Elsener, Lampe, S., 2017) Let 𝐵 = 𝐵(𝑅) be a Krull domain (e.g., 𝑅 acyclic), and {𝑦1, … , 𝑦𝑜} a seed. Then 𝐻 = 𝒟(𝐵) ≅ ℤ𝑠 for some 𝑠 ≥ 0, and every class contains infjnitely many prime divisors (𝐻0 = 𝐻). 𝐵 is factorial if and only if 𝑠 = 0. 𝑠 = 𝑢 − 𝑜 with 𝑢 the number of height-1 primes containing

  • ne of 𝑦1, … , 𝑦𝑜.
slide-15
SLIDE 15

Consequences

Corollary For 𝑅 acyclic1, the necessary conditions of Geiß–Leclerc–Schöer are suffjcient for 𝐵(𝑅) to be factorial. Corollary Acyclic cluster algebras with (invertible) principal coeffjcients are factorial. Corollary If 𝐵 = 𝐵(𝑅) is a Krull domain but not factorial, then Kainrath’s Theorem applies: for every 𝑀 ⊆ ℤ≥2 there exists 𝑏 ∈ 𝐵 with L(𝑏) = 𝑀.

1without isolated vertices

slide-16
SLIDE 16

Goal Get explicit description of the rank 𝑠 of 𝒟(𝐵) ≅ ℤ𝑠, directly in terms of 𝑅. Restrict to 𝑅 acyclic.

slide-17
SLIDE 17

Exchange matrix

For a quiver 𝑅, defjne Signed adjacency matrix: skew-symmetric 𝑜 × 𝑜-matrix 𝐶 = 𝐶(𝑅), with 𝑐𝑗𝑘 = #{arrows 𝑗 → 𝑘} − #{arrows 𝑘 → 𝑗}. a vector 𝑒 ∈ ℤ𝑜 with 𝑒𝑗 the gcd of the 𝑗-th column of 𝐶.

slide-18
SLIDE 18

Partners

Defjnition Vertices 𝑗, 𝑘 ∈ [1, 𝑜] are partners if the following equivalent conditions hold.

1 Exchange polynomials 𝑔𝑗, 𝑔𝑘 have a common factor. 2 there exist odd 𝑑𝑗, 𝑑𝑘 ∈ ℤ: 𝑑𝑘𝑐∗𝑗 = 𝑑𝑗𝑐∗𝑘. 3 v2(𝑒𝑗) = v2(𝑒𝑘) and 𝑐∗𝑗/𝑒𝑗 = ±𝑐∗𝑘/𝑒𝑘.

Partnership is an equivalence relation on [1, 𝑜]: Partner sets.

slide-19
SLIDE 19

Example

Example 1 2 3 𝐶 = ⎛ ⎜ ⎜ ⎜ ⎜ ⎝ 1 −1 1 −1 ⎞ ⎟ ⎟ ⎟ ⎟ ⎠ 𝑒 = (1, 1, 1) Partner sets: {1, 3}, {2}.

slide-20
SLIDE 20

Main result for acyclic quivers

For a partner set 𝑊 ⊆ [1, 𝑜] and 𝑒 ≥ 1, let c(𝑊, 𝑒) = #{𝑗 ∈ 𝑊 ∣ 𝑒 divides 𝑒𝑗}. (Recall: 𝑒𝑗 is gcd of the 𝑗-th column of adjacency matrix 𝐶) Theorem (Garcia Elsener, Lampe, S. 2017) Let 𝑅 be acyclic and 𝐵 = 𝐵(𝑅). Then 𝒟(𝐵) ≅ ℤ𝑠 with 𝑠 = ∑

𝑊 a partner set

𝑠𝑊, where 𝑠𝑊 = ∑

𝑒≥1 𝑒 odd

(2c(𝑊,𝑒) − 1) − #𝑊.

slide-21
SLIDE 21

Corollary: fjnite type

Corollary If 𝑅 is acyclic and without parallel arrows, then 𝐵(𝑅) is factorial if and only if there are no partners 𝑗 ≠ 𝑘. Corollary For the cluster algebras of Dynkin types: T ype 𝐵𝑜 is factorial if 𝑜 ≠ 3, and 𝒟(𝐵3) ≅ ℤ. T ype 𝐶𝑜 is factorial if 𝑜 ≠ 3, and 𝒟(𝐶3) ≅ ℤ. T ype 𝐷𝑜 is factorial. T ype 𝐸𝑜 has 𝒟(𝐸𝑜) ≅ ℤ for 𝑜 > 4, and 𝒟(𝐸4) ≅ ℤ4. T ypes 𝐹6, 𝐹7, and 𝐹8 are factorial. T ype 𝐺4 is factorial. T ype 𝐻2 has 𝒟(𝐻2) ≅ ℤ.

slide-22
SLIDE 22

Summary

For cluster algebras that are Krull domains, the class group is always of form ℤ𝑠. For acyclic cluster algebras, 𝑠 can be expressed directly in terms of the quiver and is trivial to compute. Similar results hold

  • ver fjelds of characteristic 0 as ground ring, and

for skew-symmetrizable cluster algebras with (invertible) frozen variables. Open questions How to determine 𝑠 in the locally acyclic case? When is 𝐵(𝑅) a Krull domain? [completely] integrally closed?