Categorification | Grothendieck Groups, 2-Representations and Higher Representation Theory

Categorification replaces a numerical or algebraic object with a category carrying more structure, together with a specified way to recover the original object. Numbers can become vector spaces, linear operators can become functors, and algebraic identities can become isomorphisms whose compatibility must itself be checked.

The word does not mean making an explanation more abstract without a target. A useful categorification names both directions: what richer structure is being built, and what operation returns the original invariant. The return operation is called decategorification. It deliberately forgets information, so different categories can have the same decategorified answer.

This guide develops several complete small examples. We compute Grothendieck groups, show how a non-split extension disappears from one such group, calculate a chain complex and its Euler characteristic, lift a two-by-two matrix to a functor, and derive a commutation relation from induction and restriction between symmetric groups. The final sections explain why higher representation theory also needs natural transformations and coherence.

Scope: complex linear categories and finite-dimensional examples unless stated otherwise. No claim of a physical higher-dimensional system or an automatically faster algorithm is intended. Helpful prerequisites are Tensor Products and Duality, Induction and Restriction, and Young Tableaux and Symmetric Group Representations.

Grothendieck groups · What exact relations forget · Chain complexes · Lift a matrix · Induction and restriction · 2-representations · Practice · Worked answers

Begin with an equation and ask what it has forgotten

The equation 2+3=5 can be lifted to an isomorphism between the direct sum of a two-dimensional vector space and a three-dimensional vector space and a five-dimensional vector space. The dimensions match, but the vector spaces also have vectors, linear maps, subspaces and possible additional structures.

Taking dimension forgets that extra information and returns the equation. This elementary lift illustrates the idea, but on its own it may add little mathematical value. A richer construction should reveal maps, extensions, gradings or other distinctions that the original number could not see.

For representation theory, the target is often an algebra acting on an abelian group or vector space. A categorification replaces that group with the Grothendieck group of a category and replaces operators by appropriate functors. Source [1] gives a precise framework for this kind of abelian categorification.

The Grothendieck group records objects subject to exact relations

For an essentially small abelian category A, form symbols [M] for isomorphism classes of objects. Impose [B]=[A]+[C] whenever 0→A→B→C→0 is a short exact sequence. The resulting abelian group is K₀(A).

The construction remembers additive information about objects. It does not preserve every morphism or every extension. If an exact sequence is non-split, its middle object still has the same K₀ class as the direct sum of the two end objects.

For a finite-length category, the classes of simple objects form a basis for K₀, and an object is represented by the multiplicities of its simple composition factors. This follows from the exact relations and the Jordan–Hölder theorem; it is part of the framework in source [1].

Compute K₀ of finite-dimensional vector spaces

Let Vec be the category of finite-dimensional complex vector spaces. Every vector space V is isomorphic to a direct sum of dim V copies of C. Therefore [V]=(dim V)[C].

Dimension is additive in exact sequences, so it defines a homomorphism K₀(Vec)→Z. Sending 1 to [C] gives an inverse. Thus K₀(Vec)≅Z. Tensor product supplies multiplication and satisfies [V⊗W]=[V][W], recovering ordinary integer multiplication.

Actual objects give nonnegative classes under this identification. The element −[C] exists in the Grothendieck group as a formal additive inverse, not as a vector space with negative dimension. Group completion creates formal differences; it does not create impossible objects.

Split Grothendieck groups ask a different question

For an additive category, the split Grothendieck group uses only relations [A⊕B]=[A]+[B]. It does not impose a relation for every non-split exact sequence.

In a semisimple category every exact sequence splits, so exact and split Grothendieck groups agree. In a non-semisimple category they can differ substantially. A statement involving K₀ must therefore specify which construction is intended.

This distinction is central to categorification. Some constructions deliberately count simple composition factors; others retain indecomposable direct-sum types. The right choice depends on which algebraic object one wants to recover.

A two-dimensional module reveals the difference

Let A=C[ε]/(ε²). A finite-dimensional A-module is a vector space with an operator N representing ε and satisfying N²=0. The regular module A has basis 1,ε, with N(1)=ε and N(ε)=0.

Let k=A/(ε), the one-dimensional module on which ε acts by zero. The span of ε inside A is another copy of k, giving the exact sequence 0→k→A→k→0.

The sequence does not split. On k⊕k the operator ε is zero, while on A it has rank one. Rank is invariant under change of basis, so A is not isomorphic to k⊕k.

Nevertheless exact K₀ imposes [A]=2[k]. The two modules have the same simple composition factors and therefore the same exact Grothendieck class. Their different attachment of those factors has been discarded.

Calculate both Grothendieck groups in this example

Because N²=0, every Jordan block has size one or two. Thus every finite-dimensional A-module decomposes into copies of k and A. These are the two indecomposable types.

The split Grothendieck group is Z[k]⊕Z[A], a free abelian group of rank two. The exact Grothendieck group is Z[k], because the non-split sequence supplies [A]=2[k]. The natural map from the first group to the second sends (a,b) to a+2b.

A module of dimension 5 with rank N=2 has two size-two blocks and one size-one block. Its split class is [k]+2[A], while its exact class is 5[k]. Knowing only the latter cannot recover the rank-two operator. The example identifies exactly what the chosen decategorification loses.

Exact functors descend to exact K₀

An exact functor F between abelian categories preserves short exact sequences. Therefore [M]↦[F(M)] respects the defining relations and gives a homomorphism on K₀. Composition of functors becomes composition of the induced homomorphisms. This is one of the basic mechanisms in source [1].

For a merely additive functor, the same conclusion is automatic for split K₀ but not for exact K₀. A functor can preserve direct sums while failing to preserve a non-split exact sequence.

The adjective exact is therefore not an optional technicality in an abelian categorification. It is what permits the proposed operator on classes to be well-defined.

A failed functor shows why exactness is needed

Apply Hom_A(k,−) to the dual-number example. Hom(k,k) has dimension one. Hom(k,A) also has dimension one, because the image of 1 must be annihilated by ε and therefore lie in the line Cε.

The quotient A→k sends ε to zero. Consequently the induced map Hom(k,A)→Hom(k,k) is zero, not surjective. The Hom functor is left exact here but not exact.

If one tried to define an exact-K₀ map using just the dimension of Hom(k,M), the relation [A]=2[k] would be sent to the false equation 1=2. Derived functors such as Ext measure the missing exactness. A lift of an operator must respect the relations of the decategorification actually being used.

Grading replaces integers by Laurent polynomials

For a finite-dimensional Z-graded vector space V=⊕ⱼVⱼ with degree-preserving maps, define its graded dimension as Σⱼ(dim Vⱼ)qʲ. Only finitely many coefficients are nonzero.

The Grothendieck ring of these graded vector spaces is Z[q,q⁻¹]. Define the grading shift V⟨1⟩ by (V⟨1⟩)ⱼ=Vⱼ₋₁. It moves a vector of degree j into degree j+1, so its class is q[V]. Tensor products add degrees, matching multiplication of Laurent polynomials.

For example, C in degree −1 together with two copies in degree 2 has graded dimension q⁻¹+2q². Forgetting the grading evaluates q at 1 and returns dimension 3. Keeping the grading distinguishes many spaces that ordinary dimension merges.

A grading shift is not a homological shift

An internal grading records weights such as the power of q. A homological or cohomological degree records position in a complex. Shifting the internal degree multiplies graded dimension by q; shifting a complex by one changes the sign of its Euler class.

In a bigraded homology theory, both operations may occur. A notation using the same bracket for both shifts without declaring the convention can change signs and powers throughout a calculation.

The distinction is especially important in quantum-group categorification and link homology. A coefficient q² does not mean a homological degree of two unless that additional identification has explicitly been made.

A complete chain-complex calculation

Consider a chain complex C₂→C₁→C₀ with C₂=C, C₁=C³ and C₀=C². Define d₂(t)=(t,0,0) and d₁(x,y,z)=(y,0). The composite d₁d₂ is zero, so the chain condition holds.

The kernel of d₂ is zero, hence H₂=0. The kernel of d₁ is the plane {(x,0,z)}, while the image of d₂ is its x-axis. Quotienting leaves the z-direction, so dim H₁=1.

The image of d₁ is the first coordinate line in C². Therefore H₀=C²/im d₁ is one-dimensional. We have computed homology dimensions (dim H₀,dim H₁,dim H₂)=(1,1,0).

The chain-space Euler characteristic is 2−3+1=0. The homology Euler characteristic is 1−1+0=0. The agreement is an exact calculation, but the single number zero is much less informative than the pair of surviving homology classes.

Why Euler characteristics agree

For a bounded finite-dimensional complex, write Zᵢ=ker dᵢ and Bᵢ=im dᵢ₊₁. Rank-nullity gives dim Cᵢ=dim Zᵢ+dim Bᵢ₋₁, and the definition of homology gives dim Zᵢ=dim Hᵢ+dim Bᵢ.

Substitute into the alternating sum of dim Cᵢ. Every dim Bᵢ appears once with sign (−1)ⁱ and once with the opposite sign from the adjacent degree. The boundary contributions cancel, leaving Σᵢ(−1)ⁱdim Hᵢ.

The cancellation explains why the Euler characteristic is unchanged by many alterations of a complex. It also explains why it cannot remember the separate homology groups: those groups have already been combined through an alternating sum.

Equal Euler characteristic does not mean equal homology

Compare C→C with identity differential and C→C with zero differential, using degrees 1 and 0. Both have chain Euler characteristic 1−1=0.

The identity complex is acyclic: H₁=0 and H₀=0. The zero-differential complex has H₁=C and H₀=C. They are not quasi-isomorphic, because their homology differs.

A categorification that retains the complex or homology distinguishes these two cases. Decategorification by Euler characteristic merges them. This elementary example is a precise answer to the question of what useful extra information a richer object can contain.

Graded Euler characteristic produces a polynomial invariant

For finite bigraded homology Hⁱ,ʲ, define the graded Euler characteristic Σᵢ,ⱼ(−1)ⁱqʲdim Hⁱ,ʲ. The homological degree i supplies the sign; the internal degree j supplies the power of q.

If the only nonzero groups are H⁰,⁰=C and H¹,²=C, the polynomial is 1−q². Adding one copy of C in each of H⁰,³ and H¹,³ changes the homology but not the polynomial, because the added q³ contributions cancel.

Thus even a full Laurent polynomial can forget information. A categorification is not automatically unique, and equal decategorifications do not imply equivalent richer objects.

Khovanov homology is a substantial example

Khovanov constructed a bigraded link homology whose graded Euler characteristic recovers a version of the Jones polynomial. The original construction is source [3], and source [4] surveys the relationship between the polynomial and the richer invariant.

In a standard unreduced normalization, the unknot has two generators in homological degree zero and internal degrees 1 and −1. Its graded Euler characteristic is q+q⁻¹. A convention normalizing the Jones polynomial of the unknot to 1 requires the corresponding adjustment; the two values should not be compared without recording normalization.

The theorem includes invariance under changes of link diagram, not just a way to assign a complex to one drawing. Building a diagram-dependent complex is only part of the task. One must prove that the permitted diagram moves preserve the appropriate homotopy or homology information.

Lift a two-by-two matrix to a functor

Let C=Vec⊕Vec. An object is a pair (V₀,V₁), and a morphism is a pair of linear maps. Its Grothendieck group is Z², with basis classes of (C,0) and (0,C).

Define an exact functor F(V₀,V₁)=(V₁,V₀⊕V₁). On a pair of maps (a,b), it acts by (b,a⊕b). The induced map on dimension columns is

[F] = [0  1]
      [1  1].

We have not simply named a matrix after a category. We have specified the functor on both objects and maps. Exactness follows because direct sums and the two projections preserve exact sequences of vector spaces.

Lift the matrix identity itself

Applying F twice gives F²(V₀,V₁)=(V₀⊕V₁,V₁⊕V₀⊕V₁). On the other hand, (Id⊕F)(V₀,V₁)=(V₀⊕V₁,V₁⊕V₀⊕V₁), up to the canonical associativity identifications of direct sum.

These identifications are natural in the pair (V₀,V₁), so F² ≅ Id⊕F as functors. Taking K₀ gives [F]²=I+[F]. The matrix equation is now the image of an actual functor isomorphism.

For example, an object with dimensions (2,3) is sent first to dimensions (3,5), then to (5,8). Id⊕F gives (2,3)+(3,5)=(5,8), an immediate numerical return check.

The same Fibonacci matrix does not automatically give the Fibonacci category

The matrix just used is the Fibonacci fusion matrix, but the construction has categorified a matrix action and one algebraic relation. It has not supplied the complete associator, rigidity, braiding and simple-object structure of the Fibonacci fusion category.

Here F is an endofunctor of a two-component vector-space category. In a fusion category, τ is an object in a rigid tensor category. One may relate such viewpoints through additional constructions, but they are not identical merely because their decategorified matrices agree.

This example separates a useful elementary lift from a stronger reconstruction claim. The extra categorical coherence is a mathematical requirement, not something that follows automatically from a familiar numerical pattern.

Induction and restriction categorify a commutation relation

Consider the direct sum of categories Rep(Sₙ) over all n≥0, using complex representations and objects supported in only finitely many n. Define F to induce from Sₙ to Sₙ₊₁ and E to restrict from Sₙ to Sₙ₋₁. On the n=0 component, E is zero.

Composition is read right to left: EF means first induce, then restrict. Mackey decomposition gives a natural isomorphism

EF ≅ FE ⊕ Id.

The symmetric-group form of this identity, together with diagrammatic natural transformations, is developed in Khovanov’s paper, source [2]. At n=0 it can also be checked directly: inducing from the trivial S₀ to the trivial S₁ and restricting back is the identity, while FE is zero.

Passing to K₀ yields [E][F]−[F][E]=I, the rank-one Weyl or Heisenberg commutation relation. It is not the sl₂ relation [e,f]=h. The right-hand side and the weight structure of the target algebra must be named correctly.

Why Mackey gives the extra identity summand

Regard Sₙ as the stabiliser of n+1 in Sₙ₊₁. Its double cosets in Sₙ₊₁ divide into two types: permutations that keep n+1 fixed, and those that move it into the first n positions.

The first double coset contributes the original representation unchanged. For the second, the relevant intersection subgroup is Sₙ₋₁, producing restriction followed by induction. Thus the two pieces are Id and FE.

The functor isomorphism comes from a decomposition of bimodules, so it is compatible with representation maps. Counting cosets alone suggests the dimension formula, but the bimodule construction supplies the naturality required for a categorification.

Work the relation on the trivial S2 representation

Start with the one-dimensional trivial representation of S₂. Inducing to S₃ gives the permutation representation on three cosets, which decomposes as the trivial plus the standard representation.

Restrict back to S₂. The S₃ trivial remains trivial, while the standard representation restricts to trivial⊕sign. Hence EF gives 2·trivial⊕sign.

In the other order, restriction to S₁ gives a one-dimensional trivial space, and induction to S₂ gives trivial⊕sign. Adding the original trivial representation produces 2·trivial⊕sign, exactly the first result.

The dimensions are 3=2+1. More importantly, the irreducible types and their multiplicities match. A dimension-only check would not establish the representation identity.

A second example using Young diagrams

Let V be the standard S₃ representation labelled by partition (2,1). Adding one box gives the S₄ shapes (3,1), (2,2) and (2,1,1), so FV is their direct sum.

Remove one box from each to restrict back. Shape (3,1) gives (3) and (2,1); shape (2,2) gives (2,1); shape (2,1,1) gives (2,1) and (1,1,1). Thus EFV has one trivial, three standard and one sign S₃ summand.

Restricting V first gives the two S₂ types (2) and (1,1). Inducing them back gives (3)⊕2(2,1)⊕(1,1,1). Adding V supplies the third standard summand and reproduces EFV.

The dimensions check as 8=6+2. This calculation connects the categorical relation to the box-addition and box-removal rules in the earlier Young-tableau guide.

A general dimension check is useful but weaker

For an Sₙ representation of dimension d, EF has dimension (n+1)d because induction multiplies dimension by the index n+1 and restriction preserves it. FE has dimension nd for n≥1, so the identity predicts (n+1)d=nd+d.

This verifies that the proposed decomposition is dimensionally possible. It does not prove the natural isomorphism. Different representations can have the same dimension, and unrelated functors can induce the same map on dimensions.

The examples above therefore use three levels of checking: dimensions, irreducible multiplicities, and the general bimodule-based functor decomposition. Keeping those levels separate avoids presenting a numerical coincidence as a categorical proof.

Natural transformations carry information between functors

A natural transformation η:F→G assigns a morphism η_X:F(X)→G(X) to every object X, compatibly with all morphisms X→Y. It is not merely one matrix chosen on one test object.

Induction and restriction for finite groups have adjunction maps. Units and counits can be composed to produce natural transformations, and their triangular identities express compatibility. Source [2] organizes such maps diagrammatically.

Passing to K₀ usually remembers the induced operators but forgets these maps between functors. Higher representation theory retains them because they can distinguish actions that look identical at the level of Grothendieck groups.

An explicit natural endomorphism of induction

In C[Sₙ₊₁], let Jₙ₊₁=Σᵢ₌₁ⁿ(i,n+1), the Jucys–Murphy element. Conjugation by Sₙ permutes these transpositions, so Jₙ₊₁ commutes with C[Sₙ].

Induction is C[Sₙ₊₁]⊗_C[Sₙ]V. Right multiplication by Jₙ₊₁ on the first factor defines T(a⊗v)=aJₙ₊₁⊗v. Commutation with C[Sₙ] makes this well-defined under the tensor relation ah⊗v=a⊗hv.

The map is left Sₙ₊₁-equivariant and natural in V. It therefore supplies an endomorphism of the induction functor, not merely a number attached to induction. Such endomorphisms are part of the additional structure explored in source [2].

What a 2-representation adds

A 2-category has objects, 1-morphisms between objects, and 2-morphisms between 1-morphisms. Categories, functors and natural transformations provide the basic example. Functors compose, natural transformations compose vertically, and they can also be combined horizontally through functor composition.

A 2-representation assigns categorical objects and maps to an algebraic 2-category in a way respecting these compositions and relations. It therefore remembers not just operators on K₀ but specified transformations between the corresponding functors.

The numeral 2 refers to this additional morphism level. It does not say that the original representation has vector-space dimension two. A 2-representation can act on categories containing objects of many dimensions or with no finite-dimensional underlying vector-space model at all.

Even a categorical group action needs coherence

Suppose each element g of a group is assigned an equivalence F_g of a category. To represent multiplication categorically, provide natural isomorphisms F_gF_h≅F_gh and a unit identification.

For three group elements, the two routes from F_gF_hF_k to F_ghk must agree. If they do not, the assignment has failed to define a coherent action even if the induced K₀ operators satisfy the group law.

This is the same structural issue encountered with associators in fusion categories. Equality after forgetting maps is weaker than a coherent isomorphism before forgetting them.

Quantum coefficients can become shifted identity functors

A quantum integer such as [3]_q=q²+1+q⁻² has nonnegative Laurent coefficients. In a graded categorification, that expression can be represented by the direct sum Id⟨2⟩⊕Id⊕Id⟨−2⟩.

On a suitable positive weight component of a categorical sl₂ action, a relation can compare EF with FE plus a direct sum of shifted identity functors whose class is the relevant quantum integer. Negative weight conventions reverse which side receives those summands.

This explains how graded functor decompositions can lift quantum-group coefficients. It is not a full construction of a categorical sl₂ action: additional natural transformations and their defining relations must be supplied. Source [1] explains graded abelian lifts, while source [2] relates diagrammatics to broader higher-representation constructions.

Why subtraction must be interpreted carefully

At the K₀ level one may write EF−FE=I. In an additive category there is no ordinary negative functor −FE that can simply be placed beside EF. The lifted statement is the positive direct-sum isomorphism EF≅FE⊕Id.

In derived settings, subtraction can be encoded through shifts and distinguished triangles. That is a different structure from a direct sum of actual functors. One must state whether the construction is additive, abelian, homotopy-theoretic or triangulated.

Formal signs are useful because they compress exact relationships. They become misleading only when the compressed sign is treated as if it named an ordinary object with negative multiplicity.

Category O connects the algebraic and geometric routes

Projective and translation functors on suitable blocks of category O provide important categorifications of Weyl-group and Hecke-algebra actions. Graded versions retain powers of the Hecke parameter through grading shifts. These examples are developed in source [1].

The distinction between simple, Verma and projective objects produces different useful bases after taking K₀. Changing from one basis to another encodes multiplicities, while the full category retains extensions and morphisms that those matrices alone do not recover.

Geometric models using sheaves and cohomology supply additional ways to understand the grading and functors. The algebraic and geometric constructions need comparison theorems; they do not become equivalent merely because their Grothendieck groups have the same rank.

What counts as a successful categorification?

First specify the target algebra, invariant or representation. Then identify a category and a precise decategorification, such as exact K₀, split K₀ or graded Euler characteristic. Construct the lifted operators on objects and maps, and prove that they respect the relevant relations.

For a stronger higher action, specify the natural transformations and verify their coherence. Finally explain what new information survives upstairs: extensions, gradings, homology groups, morphism spaces, diagrammatic operations or functorial behaviour.

A construction can satisfy a weak categorification definition while providing little extra insight. That does not make it false, but it limits what has been achieved. The two-component vector-space example is valuable for learning the mechanism; richer non-semisimple or geometric examples reveal why the subject extends beyond matrix imitation.

Common failures that small examples expose

Confusing exact and split K₀ erases or preserves extension data unintentionally. Omitting exactness can make the proposed map on classes ill-defined. Confusing internal and homological shifts changes q-powers into signs. Forgetting naturality leaves an objectwise coincidence rather than a functor isomorphism.

Another error is to claim categorical equivalence from equal Grothendieck groups. Vec and the dual-number module category both have exact K₀≅Z, but one is semisimple and the other contains non-split extensions. Their matching rank does not make their objects and maps equivalent.

Finally, a categorical construction does not automatically simplify computation. It can introduce larger chain complexes and more maps. Its value may be stronger invariants or conceptual structure rather than speed. The computational benefit must be evaluated for the actual task.

Practice questions

1. Find the exact K₀ class of a seven-dimensional complex vector space. 2. For an A=C[ε]/(ε²)-module of dimension 6 with rank ε=2, find its split and exact classes. 3. Give the graded dimension of C in degree −2 and three copies in degree 1. 4. Apply the internal shift ⟨2⟩ to that answer.

5. Recompute H₁ of the displayed three-term complex. 6. Give two complexes with Euler characteristic zero but different homology. 7. For F(V₀,V₁)=(V₁,V₀⊕V₁), compute F³ on dimensions (1,2). 8. Explain why F²≅Id⊕F is stronger than a dimension equality.

9. For an S₄ representation of dimension 3, find the dimensions of EF and FE. 10. Does EF−FE=I identify an sl₂ action? 11. What is the graded class of Id⟨2⟩⊕Id⊕Id⟨−2⟩? 12. Explain why matching K₀ actions does not prove two categorical actions equivalent.

Worked answers

1–4. Additive and graded classes

1. The class is 7[C]. Under K₀(Vec)≅Z it corresponds to the integer 7. The group also contains negative integers, but those are formal differences rather than dimensions of actual vector spaces.

2. Rank ε counts the size-two Jordan blocks, so there are two copies of A. They use four dimensions, leaving two copies of k. The split class is 2[A]+2[k], while exact K₀ replaces each [A] by 2[k], giving 6[k].

3. The graded dimension is q⁻²+3q. At q=1 it becomes 4, the ordinary dimension. The two degree locations remain visible in the Laurent polynomial.

4. The shift multiplies by q², so the result is 1+3q³. This is an internal grading shift and introduces no alternating homological sign.

5–8. Homology and functor lifts

5. ker d₁ consists of vectors (x,0,z), and im d₂ consists of (x,0,0). Their quotient is represented by the z-coordinate and is one-dimensional. Taking the kernel alone would incorrectly give dimension two.

6. Use C→C with identity differential and C→C with zero differential. Both have one-dimensional chain spaces in degrees 1 and 0 and Euler characteristic zero. The first has zero homology; the second has one copy of C in each degree.

7. The successive dimension pairs are (1,2)→(2,3)→(3,5)→(5,8). Thus F³ gives (5,8). Each step follows directly from the functor’s two output components.

8. A functor isomorphism identifies the output objects compatibly with every morphism of the input category. Equal dimensions merely say that some vector-space isomorphisms could exist object by object; they do not supply a natural family.

9–12. Relations and coherence

9. At n=4 and d=3, EF has dimension 5·3=15. FE has dimension 4·3=12. Adding the original three-dimensional representation makes the dimensions agree. Mackey decomposition supplies the stronger representation and functor statement.

10. No. The displayed relation is the rank-one Weyl or Heisenberg commutator. An sl₂ action needs a weight operator h and the additional relations [h,e]=2e, [h,f]=−2f and [e,f]=h, or their appropriate categorified versions.

11. Its class is (q²+1+q⁻²)Id, the quantum integer [3]_q times the identity operator. The three direct summands are honest shifted functors; none has negative multiplicity.

12. K₀ forgets morphism spaces, natural transformations and extension information. Two actions can induce the same operators on classes while differing in those forgotten structures. An equivalence requires compatible functors and transformations, not just equal matrices after decategorification.

A teaching route that keeps the return map visible

Start with vector-space dimension, then immediately compare A and k⊕k in the dual-number example. The learner can see both the equality of exact classes and the different ranks of ε. This gives a concrete meaning to information lost under decategorification.

Next compute the small chain complex and compare it with a complex having the same Euler characteristic but different homology. Require the learner to name the kernel, image and quotient separately. Only then introduce the word categorification of a polynomial invariant.

Finish with induction and restriction. First check dimensions, then irreducible types, then explain why the bimodule decomposition is natural. The progression trains a general habit: distinguish a numerical check, an object isomorphism and a coherent functorial structure.

Sources and further study

[1] Mikhail Khovanov, Volodymyr Mazorchuk and Catharina Stroppel, A Brief Review of Abelian Categorifications, for exact Grothendieck groups, functorial lifts, graded versions and category O examples. [2] Mikhail Khovanov, Heisenberg Algebra and a Graphical Calculus, especially induction, restriction, Mackey decomposition and natural transformations. [3] Khovanov, A Categorification of the Jones Polynomial. [4] Khovanov and Robert Lipshitz, Categorical Lifting of the Jones Polynomial: A Survey. The dual-number, chain-complex and two-component functor calculations above are developed explicitly to show what each return map preserves and forgets.

For earlier representation tools, use Induction and Restriction, Symmetric Group Representations and Quantum Groups. Return to the BTT Mathematics Learning Hub.