Kac–Moody Algebras | Generalized Cartan Matrices, Roots and Highest-Weight Modules

Kac–Moody algebras extend the structure of finite-dimensional semisimple Lie algebras beyond finite type. Their input is a generalized Cartan matrix. Their generators satisfy Chevalley–Serre relations. Their root systems can contain infinitely many roots, including imaginary roots, while highest-weight representation theory still retains Weyl-group symmetry, integrability and a character formula.

The subject is valuable because it shows which parts of ordinary Lie theory survive when finite-dimensionality is removed. Triangular decomposition survives. Simple roots and coroots survive. Weyl reflections survive. Highest-weight modules survive. But the root system can become infinite, root multiplicities can exceed one, the Weyl group can be infinite, and new phenomena appear that have no finite-type analogue.

This guide develops the algebra from the matrix upward. It follows the earlier BTT route through Lie Group and Lie Algebra Representations, Category O and Verma Modules and Quantum Groups. The next article in this batch specialises the theory to affine Lie algebras.

Generalized Cartan matrices · Serre relations · Roots · Weyl group · Highest weights · Integrability · Weyl–Kac character formula · Practice

The first change: the Cartan matrix is allowed to leave finite type

Finite-dimensional complex semisimple Lie algebras are classified by Cartan matrices of finite Dynkin type. Kac–Moody theory begins by keeping the same matrix grammar while allowing matrices that are affine or indefinite.

A generalized Cartan matrix A=(a_ij) is an integer matrix satisfying:

• a_ii=2;
• a_ij≤0 when i≠j;
• a_ij=0 if and only if a_ji=0.

The asymmetry allowed by unequal negative off-diagonal entries is important. A matrix can still be symmetrizable even when it is not symmetric.

Symmetrizability

A generalized Cartan matrix is symmetrizable if there exists a diagonal matrix D with positive diagonal entries such that DA is symmetric.

For example,

A = [ 2  -1]
    [-2   2]

is not symmetric. But taking D=diag(2,1) gives

DA = [ 4  -2]
     [-2   2],

which is symmetric. Symmetrizability supplies an invariant bilinear form and supports the cleanest form of the Weyl–Kac character formula.

Three rank-two matrices show finite, affine and indefinite behaviour

Compare

finite A2:        [ 2 -1]     det = 3
                  [-1  2]

affine A1^(1):   [ 2 -2]     det = 0
                  [-2  2]

indefinite:       [ 2 -3]     det = -5
                  [-3  2]

For symmetric rank-two examples, positive definiteness signals finite type, positive semidefiniteness of corank one signals affine type, and an indefinite form signals indefinite Kac–Moody type.

The determinant alone is not a universal classification test for arbitrary nonsymmetric matrices, but in these examples it makes the transition visible.

From the matrix to generators and Serre relations

Choose a realization with a Cartan subalgebra h, simple roots α_i∈h* and simple coroots h_i∈h satisfying α_j(h_i)=a_ij.

The Kac–Moody algebra g(A) is generated by h together with Chevalley generators e_i,f_i. The defining relations include

• [h,h′]=0;
• [h,e_i]=α_i(h)e_i;
• [h,f_i]=−α_i(h)f_i;
• [e_i,f_j]=δ_ijh_i;
• (ad e_i)^1−aᵢⱼe_j=0 for i≠j;
• (ad f_i)^1−aᵢⱼf_j=0 for i≠j.

The last two are the Serre relations. The exponent is determined directly by the off-diagonal Cartan entry.

Worked Serre check for A2

For A₂, a₁₂=a₂₁=−1. Therefore 1−a₁₂=2 and the Serre relation is

(ad e₁)²e₂=0.

Expanding the iterated bracket gives

[e₁,[e₁,e₂]]=0.

The first bracket [e₁,e₂] is a root vector for α₁+α₂. Applying e₁ once more would attempt to create 2α₁+α₂, which is not a root in A₂.

The Serre relation therefore enforces the finite root-string length encoded by the Cartan matrix.

Worked Serre check for an affine edge

For A₁⁽¹⁾, the off-diagonal entry is −2. Hence

(ad e₀)³e₁=0,   (ad e₁)³e₀=0.

The longer Serre string reflects the double edge in the affine Dynkin diagram. The diagram is not merely decorative: it records exactly how many times one simple root can be added along another simple-root direction before the Serre relation forces zero.

Triangular decomposition survives

Let n₊ be generated by the e_i and n₋ by the f_i. Then

g(A)=n₋⊕h⊕n₊

as a vector space.

This is one of the most important structural survivals from finite-dimensional Lie theory. It permits highest-weight modules, Verma modules and PBW-type constructions even when g(A) is infinite-dimensional.

Root decomposition

The root lattice is

Q=⊕_iZα_i.

The adjoint action of h decomposes the algebra into weight spaces

g=h⊕⊕_α∈Δg_α,

where x∈g_α means [h,x]=α(h)x for every h∈h.

The nonzero α for which g_α≠0 are roots. Positive roots are nonnegative integer combinations of simple roots. Negative roots are their negatives.

Unlike finite type, root multiplicities dim g_α need not all equal one.

Real and imaginary roots

A real root is a root in the Weyl-group orbit of a simple root. Real roots retain many finite-type properties and have multiplicity one in the standard symmetrizable theory.

Roots that are not real are called imaginary. They occur only beyond finite type. Imaginary roots can have multiplicity greater than one, and their behaviour is one of the major new layers introduced by Kac–Moody algebras.

In untwisted affine algebras, a distinguished imaginary root δ appears. Multiples nδ are again roots, and real roots occur in infinite families shifted by multiples of δ.

This is the algebraic reason an affine root system repeats indefinitely rather than ending at a highest root.

The Weyl group still acts by reflections

For each simple root define a reflection on h* by

s_i(λ)=λ−λ(h_i)α_i.

The Weyl group W is generated by these reflections. Its Coxeter relations are determined by the products a_ija_ji.

In finite type W is finite. In affine and indefinite Kac–Moody theory it is usually infinite.

Worked reflection in A2

For A₂, α₁(h₁)=2 and α₂(h₁)=−1. Therefore

s₁(α₁)=α₁−2α₁=−α₁,

while

s₁(α₂)=α₂−(−1)α₁=α₁+α₂.

The simple reflection reverses its own root and sends the neighbouring simple root to the next positive root in the root string.

The Dynkin diagram controls the Coxeter relation

If a_ija_ji=0,1,2,3, then the order m_ij of s_is_j is respectively 2,3,4,6 in the familiar crystallographic cases.

For A₂, the product is 1, so (s₁s₂)³=1. For A₁⁽¹⁾, the product is 4, and the corresponding rank-two Weyl group is infinite dihedral rather than finite.

The jump from product 3 to product 4 is therefore a boundary between finite and infinite rank-two reflection behaviour.

Highest-weight modules

A highest-weight module of highest weight λ contains a nonzero vector v_λ such that

• e_iv_λ=0 for all i;
• hv_λ=λ(h)v_λ for h∈h;
• the entire module is generated from v_λ by U(g).

The Verma module M(λ) is the universal highest-weight module. Its simple quotient is denoted L(λ).

The construction is formally parallel to category O for finite-dimensional semisimple Lie algebras, but the set of roots below λ can now extend through infinitely many directions and multiplicities.

Dominant integral weights

A weight λ is dominant integral when

λ(h_i)∈Z_≥0

for every simple coroot.

Write fundamental weights Λ_i satisfying Λ_i(h_j)=δ_ij, after choosing the appropriate realization. Then every dominant integral weight has nonnegative integer simple-coroot coordinates.

These weights index the irreducible integrable highest-weight modules in the standard symmetrizable theory.

Integrability

A representation is integrable when each Chevalley generator e_i and f_i acts locally nilpotently and the Cartan acts with the expected weight decomposition.

Locally nilpotent means that for every vector v and every i, some sufficiently high power e_i^Nv is zero, and similarly for f_i.

This does not require the entire module to be finite-dimensional. Integrable highest-weight modules for affine Kac–Moody algebras are typically infinite-dimensional.

For a dominant integral λ, the simple highest-weight module L(λ) is integrable. Conversely, an integrable highest-weight module has dominant integral highest weight under the standard hypotheses.

Each simple root still generates an sl2-string

The generators e_i,f_i,h_i form an sl₂-subalgebra. Therefore an integrable highest-weight module restricts along each simple-root direction to finite sl₂ strings.

If λ(h_i)=m, then from the highest vector the i-string has length m downward before f_i kills the endpoint.

This local finite behaviour coexists with a globally infinite module because different simple-root directions can be combined indefinitely in affine or indefinite type.

Worked local string

Suppose λ(h₁)=2. Along the sl₂ generated by e₁,f₁,h₁, the highest vector produces

v, f₁v, f₁²v

and f₁³v=0 in the integrable simple module.

The three weights differ by α₁ steps. Even if the full module contains infinitely many weights obtained by combining other lowering directions, this one simple-root string remains finite.

Weight sets are Weyl-symmetric in integrable highest-weight modules

Integrability allows the simple sl₂ actions to exponentiate algebraically enough to implement Weyl reflections on weights. Therefore the weight set and character of an integrable highest-weight module are W-invariant in the appropriate sense.

This is a major structural payoff. Even when W is infinite, it still organises infinitely many weights into reflection orbits.

The Weyl–Kac character formula

For a symmetrizable Kac–Moody algebra and a dominant integral highest weight λ, the irreducible integrable character satisfies

ch L(λ)= [Σ_w∈Wε(w)e^{w(λ+ρ)−ρ}] / [∏_α∈Δ₊(1−e^−α)^{mult α}].

Here ε(w)=(−1)^ℓ(w), ρ is a weight satisfying ρ(h_i)=1, and imaginary-root multiplicities appear explicitly in the denominator.

In finite type this reduces to the ordinary Weyl character formula. In affine type both the Weyl group and positive-root set are infinite, so the character becomes a formal infinite expression whose cancellations encode deep combinatorics.

The denominator identity

Setting λ=0 gives the Weyl–Kac denominator identity

Σ_w∈Wε(w)e^wρ−ρ=∏_α∈Δ₊(1−e^−α)^{mult α}.

In affine examples, specialisations of such identities lead to classical q-series identities. This is one route from infinite-dimensional Lie theory into partition combinatorics and modular forms.

The exact specialisation depends on the affine algebra and chosen variables, so one should not identify every partition identity with one universal denominator specialisation.

Finite type versus affine type

Finite type has finitely many roots and a finite Weyl group. Affine type has infinitely many roots and an infinite affine Weyl group, but retains a controlled null direction generated by the imaginary root δ.

Affine algebras are therefore the first infinite-dimensional Kac–Moody algebras where representation theory remains highly structured and closely tied to loop algebras, modular forms, vertex operator algebras and conformal field theory.

The next Batch 08 guide develops this affine case directly from loops and central extensions.

Indefinite and hyperbolic behaviour

Indefinite Kac–Moody algebras move beyond the controlled affine null direction. Their imaginary-root systems and multiplicities become substantially more complicated.

Some indefinite matrices are hyperbolic: deleting any node of the Dynkin diagram leaves a finite or affine diagram. Hyperbolic Kac–Moody algebras appear in parts of mathematical physics, automorphic theory and conjectural symmetry structures.

The term hyperbolic should not be applied merely because a Cartan matrix has a negative determinant. The precise Dynkin-subdiagram criterion matters.

Relation to quantum groups and crystals

For a symmetrizable generalized Cartan matrix one can form the quantized enveloping algebra U_q(g). The highest-weight modules deform, and their crystal bases retain a combinatorial skeleton as q approaches zero in the precise Kashiwara sense.

This means the same Cartan matrix governs several layers:

• Kac–Moody Lie algebra g(A);
• quantum group U_q(g);
• highest-weight modules;
• crystal graphs and canonical/global bases.

The third article in this batch studies the crystal layer explicitly.

A compact calculation workflow

• 1. Verify the generalized Cartan conditions.
• 2. Test symmetrizability. Find D with DA symmetric when possible.
• 3. Write the Chevalley–Serre relations. The powers 1−a_ij are not optional.
• 4. Build simple reflections. Use s_i(λ)=λ−λ(h_i)α_i.
• 5. Separate real from imaginary roots. Real means Weyl-conjugate to simple.
• 6. For a highest weight, compute all simple-coroot coordinates.
• 7. Test integrability through local sl₂ strings.
• 8. State whether the algebra is finite, affine or indefinite before importing finite-type claims.
• 9. Include root multiplicities in character denominators.
• 10. Treat infinite Weyl sums and products formally unless convergence has separately been established.

Common mistakes

• Assuming every generalized Cartan matrix is symmetric: symmetrizable is the relevant broader condition.
• Forgetting the Serre exponent: it is 1−a_ij.
• Calling every root real: imaginary roots are a defining new feature beyond finite type.
• Assuming root multiplicity one: that is safe for real roots in standard symmetrizable theory, not for general imaginary roots.
• Equating integrability with finite dimension: affine integrable highest-weight modules are infinite-dimensional.
• Using a finite Weyl-group argument unchanged: affine and indefinite Weyl groups are infinite.
• Dropping multiplicities from the Weyl–Kac denominator: imaginary roots can occur with multiplicity greater than one.
• Calling a matrix hyperbolic from one numerical test: the Dynkin deletion criterion is part of the definition.

Practice questions

1. Verify that [[2,−1],[−2,2]] is a generalized Cartan matrix. 2. Find a positive diagonal D that symmetrizes it. 3. Write the Serre exponent for a₁₂=−3. 4. Compute s₁(α₂) in A₂.

5. Why is A₁⁽¹⁾ not finite type? 6. Define a real root. 7. What is new about imaginary roots? 8. If λ(hᵢ)=4, how long is the simple i-string from the highest vector in an integrable module?

9. State the dominant-integrality condition. 10. What role do root multiplicities play in the Weyl–Kac formula? 11. Why can an integrable module still be infinite-dimensional? 12. What does setting λ=0 in the Weyl–Kac formula produce?

Worked answers

1. The diagonal entries are 2, both off-diagonal entries are nonpositive integers, and neither off-diagonal entry is zero, so the zero-symmetry condition is satisfied.

2. D=diag(2,1) gives DA=[[4,−2],[−2,2]], which is symmetric.

3. 1−(−3)=4, so (ad e₁)⁴e₂=0.

4. α₂(h₁)=−1, so s₁(α₂)=α₂+α₁.

5. Its Cartan matrix is singular and its Weyl group is infinite; its root system contains infinitely many roots shifted by the null root δ.

6. A real root is in the Weyl-group orbit of a simple root.

7. Imaginary roots are not Weyl-conjugate to simple roots and can have multiplicity greater than one.

8. The sl₂ string has five weight vectors, from fᵢ⁰v through fᵢ⁴v, with fᵢ⁵v=0.

9. λ(hᵢ) must be a nonnegative integer for every simple coroot hᵢ.

10. They appear as exponents (1−e^−α)^{mult α} in the denominator.

11. Each simple-root direction may be locally finite while infinitely many combinations of lowering directions create infinitely many weight spaces.

12. The Weyl–Kac denominator identity.

Sources and further study

[1] Victor Kac, Infinite-Dimensional Lie Algebras, the standard reference for generalized Cartan matrices, Kac–Moody algebras, affine algebras and integrable highest-weight modules. [2] Nitu Kitchloo, MIT thesis notes, Kac–Moody groups and highest-weight modules, for the Chevalley–Serre presentation and triangular decomposition. [3] Gurbir Dhillon and Apoorva Khare, The Weights of Simple Modules in Category O for Kac–Moody Algebras, for modern highest-weight and integrability structure. [4] Victor Kac’s MIT Lie Algebras course archive provides the finite-type highest-weight prerequisites.

Representation Mathematics — Batch 08

Specialise infinite-dimensional Kac–Moody symmetry to loop algebras in Affine Lie Algebra Representations. Pass to the q→0 combinatorial skeleton in Crystal Bases. Compute tensor-product multiplicities through Littlewood–Richardson Coefficients, Honeycombs and Saturation. Return to the BTT Mathematics Learning Hub.