Crystal Bases | Kashiwara Operators, Weight Graphs and q→0 Combinatorics

A crystal basis keeps the combinatorial skeleton of a quantum-group representation while discarding most of its q-dependent coefficients. Vectors become nodes, Kashiwara operators become coloured arrows, weights label the nodes, and tensor-product decomposition becomes a graph problem.

The phrase “q→0” must be interpreted carefully. One does not usually take an arbitrary matrix representation and substitute q=0 into every coefficient. Crystal theory first chooses a suitable lattice over functions regular at q=0. Only after that controlled integral structure is in place does one reduce modulo q and obtain the crystal basis.

This guide follows the BTT routes through Quantum Groups, Kac–Moody Algebras and Tensor Products and Duality. The final Batch 08 article uses the same highest-weight combinatorics to study Littlewood–Richardson multiplicities.

The q→0 idea · Kashiwara operators · sl2 crystals · Tensor products · Tableaux · Global bases · Affine crystals · Practice

Why a representation may have a simpler combinatorial shadow

A quantum-group module can contain coefficients such as q-integers, powers of q and rational functions of q. Those coefficients are essential to the full linear representation, but many structural questions—weights, highest-weight components and tensor-product multiplicities—can survive after the coefficients are compressed.

Crystal theory builds that compression in a controlled way. The resulting graph remembers which lowering operation can move one basis element to another and how the weight changes.

It is therefore a decategorification-like move inside representation theory: much linear data is discarded, but carefully chosen discrete information remains exact.

The q→0 lattice

Let A₀ be the subring of rational functions in q that are regular at q=0. For a suitable integrable U_q(g)-module M, a crystal lattice L is an A₀-submodule with enough vectors to recover M after extending scalars.

Reduce L modulo qL. The quotient L/qL is now a vector space over the ground field at q=0. A crystal basis is a pair (L,B), where B is a basis of L/qL satisfying compatibility conditions with weights and Kashiwara operators.

The order of operations matters:

quantum module → choose q-regular lattice → reduce modulo q → discrete crystal basis.

Naively substituting q=0 into a coefficient such as q⁻¹ would be meaningless. The lattice is what makes the limit structured.

The crystal graph data

A crystal B carries:

• a weight map wt:B→P;
• raising operators ẽ_i;
• lowering operators f̃_i;
• integer functions ε_i(b);
• integer functions φ_i(b).

The operators can send a node to another node or to a formal zero symbol 0 when no move exists.

Draw an i-coloured arrow

b —i→ b′

when f̃_ib=b′. Then ẽ_ib′=b.

Kashiwara operators come from simple-root strings

Fix one simple root i. Inside an integrable quantum-group module, vectors can be decomposed into U_q(sl₂)-type strings for the i-th simple-root direction.

Very schematically, a vector is written as a sum of divided powers f_i^(r)u with e_iu=0. Kashiwara’s operators shift the string index:

ẽ_i: r↦r−1,   f̃_i: r↦r+1.

The q-dependent coefficients are removed while the position along the string survives.

ε and φ measure room to move

For a normal highest-weight crystal, define

ε_i(b)=max{r≥0: ẽ_i^rb≠0},

φ_i(b)=max{r≥0: f̃_i^rb≠0}.

They satisfy

φ_i(b)−ε_i(b)=⟨h_i,wt(b)⟩.

Thus the weight coordinate equals “downward room minus upward room” in the i-string.

The sl2 crystal B(n)

The irreducible U_q(sl₂) module of highest weight n has n+1 crystal nodes. Label them b₀,b₁,…,b_n, with

f̃b_r=b_r+1,   ẽb_r=b_r−1

when the indicated node exists.

The graph is one chain:

b0  →  b1  →  b2  → ... →  bn

Its node data is

wt(b_r)=n−2r,   ε(b_r)=r,   φ(b_r)=n−r.

Check the identity: φ−ε=(n−r)−r=n−2r, which is the simple-coroot weight coordinate.

The two-node fundamental crystal

For n=1, write u=b₀ and v=b₁. Then

u  →  v
wt    +1    -1
ε      0     1
φ      1     0

This tiny graph will be enough to recover the decomposition 2⊗2=3⊕1 combinatorially.

Tensor-product crystals

For crystals B₁ and B₂, the underlying set of the tensor product is B₁×B₂, written b₁⊗b₂. We use the convention

ẽ_i(b₁⊗b₂)=ẽ_ib₁⊗b₂ if φ_i(b₁)≥ε_i(b₂), otherwise b₁⊗ẽ_ib₂;

f̃_i(b₁⊗b₂)=f̃_ib₁⊗b₂ if φ_i(b₁)>ε_i(b₂), otherwise b₁⊗f̃_ib₂.

The strict versus non-strict inequality is part of this convention. Some references reverse tensor order or swap the inequality convention. Mixing rules from two conventions can split the graph incorrectly.

Worked decomposition B(1)⊗B(1)

There are four nodes:

u⊗u, u⊗v, v⊗u, v⊗v.

Begin at u⊗u. Since φ(u)=1>ε(u)=0, f̃ acts on the first factor:

u⊗u → v⊗u.

At v⊗u, φ(v)=0 is not greater than ε(u)=0, so f̃ acts on the second factor:

v⊗u → v⊗v.

That gives a three-node component with weights 2,0,−2: the crystal B(2).

The remaining node u⊗v is isolated. For ẽ, φ(u)=1≥ε(v)=1, so the operator attempts to raise u and gives zero. For f̃, 1>1 is false, so it attempts to lower v and also gives zero.

Thus

B(1)⊗B(1)≈B(2) ⊔ B(0).

This is the crystal version of the sl₂ decomposition

V(1)⊗V(1)≈V(2)⊕V(0),

or 2⊗2=3⊕1 in dimension language.

Highest-weight nodes identify components

A node b in a highest-weight crystal is highest weight when every ẽ_ib=0.

Each connected component of a tensor product of integrable highest-weight crystals contains a highest-weight node. Its weight identifies the irreducible highest-weight crystal represented by that component.

Therefore tensor multiplicity becomes a counting problem:

count highest-weight nodes of weight ν in B(λ)⊗B(μ) → multiplicity of V(ν).

A second sl2 tensor product

Consider B(2)⊗B(1). It has 3·2=6 nodes. Two nodes are highest weight under the tensor rule:

• b₀⊗u has weight 3;
• b₀⊗v has weight 1.

The first generates a four-node B(3) component. The second generates a two-node B(1) component.

Hence

B(2)⊗B(1)≈B(3)⊔B(1),

matching 3⊗2=4⊕2 for sl₂.

Characters can be read from crystals

If B is the crystal of an integrable highest-weight representation with a compatible global basis, then the character is obtained by counting nodes by weight:

ch V=Σ_b∈Be^wt(b).

For B(2), the weights are 2,0,−2, so

ch V(2)=e²+1+e⁻²

in a one-variable shorthand.

A crystal can therefore recover weight multiplicities even though the actual basis vectors and q-dependent coefficients have been discarded.

Type A crystals and semistandard Young tableaux

For representations of sl_n or GL_n, many highest-weight crystals have a tableau model. Nodes are semistandard Young tableaux of a fixed shape λ with entries in {1,…,n}.

Rows are weakly increasing and columns are strictly increasing. The weight records how many 1s, 2s, …, n’s appear.

Kashiwara operators f̃_i and ẽ_i change selected entries i↔i+1 according to a signature rule applied to a chosen reading word.

The signature algorithm is convention-sensitive because different authors choose different reading words. A tableau calculation should state its reading convention before locating the surviving i or i+1 symbol.

The fundamental sl3 crystal

The defining three-dimensional representation has crystal nodes labelled 1,2,3 with coloured arrows

1  --1-->  2  --2-->  3

The first simple-root lowering operator changes 1 to 2; the second changes 2 to 3.

Tensor powers of this tiny crystal already generate semistandard-tableau combinatorics after the appropriate highest-weight components are selected.

Littlewood–Richardson multiplicities appear as crystal component counts

For type A, irreducible polynomial GL_n representations are indexed by partitions. Their crystals can be modelled by semistandard tableaux.

The tensor product B(λ)⊗B(μ) decomposes into connected highest-weight components. The number of components isomorphic to B(ν) is the Littlewood–Richardson coefficient c^ν_λμ.

This gives a representation-theoretic meaning to the tableau rule developed in the next article:

LR coefficient = number of appropriate highest-weight crystal components.

Crystals retain tensor structure but not every linear map

A crystal knows weights, simple-root moves and tensor-product combinatorics. It does not by itself record all matrix coefficients of the quantum-group action.

Two different q-dependent bases can have the same crystal after reduction. The crystal is therefore a structural invariant, not a complete replacement for the representation.

This compression is exactly why crystals are powerful computationally: one can solve many decomposition and multiplicity questions with discrete graph algorithms.

Global and canonical bases lift the crystal back to q-dependent vectors

Kashiwara’s global basis and Lusztig’s canonical basis provide q-dependent basis vectors whose reduction recovers the crystal basis under standard compatibility results.

The crystal graph can therefore be viewed as the skeleton of a much richer basis with positivity and integrality properties.

This relation is not a unique reconstruction from the bare graph alone. The global-basis construction also uses bar involution, integral lattices and the ambient quantum group.

Canonical basis positivity

Canonical/global bases often have remarkable positivity properties in suitable settings: structure constants or transition coefficients can lie in nonnegative Laurent-polynomial rings after the correct normalization.

Such positivity is one reason these bases connect to geometry, categorification and cluster algebra structures.

Do not infer positivity for every arbitrary coordinate expansion. Positivity statements depend on the chosen basis, algebra and operation.

Crystal operators are not ordinary Lie algebra matrices

The symbols ẽ_i and f̃_i act on crystal nodes or the zero symbol. They are not linear operators on the original complex vector space after q has been discarded.

For example, f̃(b)=b′ records that one global-basis vector lies next to another in the i-string. It does not specify the scalar coefficient multiplying the corresponding quantum-group basis vector under f_i.

This distinction prevents a common error: using a crystal arrow as though it were a matrix entry equal to 1 in the original representation.

Crystal reflection and Weyl symmetry

Because every i-string behaves like an sl₂ crystal, one can reflect a node across the midpoint of its i-string. These simple reflections combine into a Weyl-group action on normal crystals.

This makes the Weyl symmetry of integrable representations visible combinatorially. The weight of the reflected node is s_i(wt(b)).

The graph therefore contains a discrete model of the same Weyl reflections introduced in Kac–Moody representation theory.

Affine crystals

Integrable highest-weight representations of affine quantum groups have crystals with arrows coloured by the affine Dynkin nodes, including the extra 0-colour.

Highest-weight affine crystals are typically infinite because the corresponding representations are infinite-dimensional. Finite-dimensional representations of quantum affine algebras can nevertheless have finite crystals after the appropriate finite-dimensional theory is chosen.

Perfect crystals, Kirillov–Reshetikhin crystals and related finite combinatorial objects model portions of affine representation theory and solvable lattice systems.

These theories introduce additional hypotheses and are not implied by the finite-type tableau model alone.

Crystals and paths

Littelmann’s path model gives another combinatorial realization of highest-weight representation theory. Nodes are piecewise-linear paths in weight space, and root operators modify the paths.

Path crystals and Kashiwara crystals encode the same highest-weight representation data in compatible settings. The path model can make tensor products and branching rules geometrically intuitive.

The existence of multiple models is useful evidence that the crystal is an intrinsic combinatorial structure rather than one special tableau trick.

A crystal workflow

• 1. Fix the Cartan type and highest weight.
• 2. State the crystal convention. In particular, record tensor order and inequalities.
• 3. Label every node with its weight.
• 4. Compute ε_i and φ_i.
• 5. Draw f̃_i arrows and verify ẽ_i reverses them.
• 6. In tensor products, test the φ versus ε inequality before choosing a factor.
• 7. Locate highest-weight nodes.
• 8. Separate connected components.
• 9. Compare component sizes and weights with known irreducibles.
• 10. Remember that the crystal has forgotten scalar coefficients.

Common mistakes

• Substituting q=0 naively: crystal reduction requires a q-regular lattice.
• Treating crystal operators as linear matrices: they are combinatorial partial operators on nodes.
• Mixing tensor-product conventions: strict and non-strict inequalities matter.
• Forgetting the zero symbol: a missing arrow is part of the highest/lowest-weight data.
• Counting all tensor nodes as separate irreducibles: connected components, not individual nodes, represent summands.
• Confusing weight multiplicity with tensor multiplicity: one counts nodes of a weight; the other counts highest-weight components.
• Assuming every crystal is finite: affine highest-weight crystals are generally infinite.
• Assuming the graph uniquely reconstructs every q-dependent coefficient: it does not.

Practice questions

1. How many nodes are in B(5) for sl₂? 2. What is wt(b₃) in B(5)? 3. Find ε(b₃) and φ(b₃). 4. Why is b₀ highest weight?

5. In B(1)⊗B(1), what is f̃(u⊗u)? 6. What is f̃(v⊗u)? 7. Which node gives the B(0) component? 8. State the resulting decomposition.

9. What does a highest-weight node of weight ν in B(λ)⊗B(μ) signify? 10. What do tableaux entries count in the weight? 11. Why is q→0 not literal substitution into arbitrary formulas? 12. What information is lost when passing from a global basis to a crystal?

Worked answers

1. Six nodes: b₀ through b₅.

2. 5−2·3=−1.

3. ε=3 and φ=5−3=2.

4. There is no node above it, so ẽb₀=0.

5. Since 1>0, f̃ acts on the first factor: v⊗u.

6. Since 0>0 is false, f̃ acts on the second factor: v⊗v.

7. u⊗v is isolated and has weight zero.

8. B(1)⊗B(1)=B(2)⊔B(0).

9. It identifies one irreducible highest-weight component B(ν); counting such nodes gives the tensor multiplicity.

10. The number of 1s, 2s, … records the GL_n weight in the tableau model.

11. Coefficients can have poles at q=0. The crystal lattice selects combinations regular enough to reduce modulo q.

12. Scalar coefficients and most linear-combination data are forgotten; weights and simple-root connectivity remain.

Sources and further study

[1] Masaki Kashiwara, Crystal Bases and Categorifications, a survey of crystal bases, global bases and their modern extensions. [2] Kyeonghoon Jeong, Seok-Jin Kang and Masaki Kashiwara, Crystal Bases for Quantum Generalized Kac–Moody Algebras, for definitions, tensor-product rules and existence in a generalized Kac–Moody setting. [3] Arkady Berenstein and David Kazhdan, Lecture Notes on Geometric Crystals and Their Combinatorial Analogues. [4] Jin Hong and Seok-Jin Kang, Introduction to Quantum Groups and Crystal Bases, is a standard detailed reference for type-A tableaux and tensor products.

Representation Mathematics — Batch 08

Build the infinite-dimensional Lie algebra in Kac–Moody Algebras, specialise it to loops in Affine Lie Algebra Representations, and turn type-A highest-weight component counts into Littlewood–Richardson, Honeycomb and Saturation Theory. Return to the BTT Mathematics Learning Hub.