Affine Hecke algebras convert a part of p-adic representation theory into explicit algebra. A reductive group acts on a representation, an Iwahori subgroup selects fixed vectors, and convolution operators act on that smaller space. The algebra combines reflection generators with a commuting lattice of translation parameters.
This guide develops affine Hecke algebra through calculations: a finite-flag model derives the quadratic relation, polynomial projectors separate its eigenspaces, and a two-dimensional Bernstein module reveals exactly when reducibility occurs. The examples show why eigenvalues, central characters and dimensions are not always enough to identify a representation.
The central proposition is that a representation consists of compatible actions, not one operator considered in isolation. A reflection generator may be diagonalizable while the full affine module is irreducible. At special parameter ratios, one invariant line can appear without a complementary invariant line. Both phenomena can be verified with two-by-two matrices.
Scope. We use complex coefficients. The detailed GL₂-type calculation assumes q>1 and nonzero complex parameters a,b. The normalization is (T−q)(T+1)=0 and TX₁T=qX₂. Other conventions may invert or rescale lattice variables. This is advanced mathematical enrichment rather than a school examination requirement.
Your 50-second route
New to p-adic groups? Start with Iwahori subgroups and the counting model. Comfortable with matrices? Go to the two-dimensional module. Investigating reducibility? Use the invariant-line test. Comparing Hecke theories? Read scope and normalization. Finish with worked practice.
Open the learning map
1–6: local fields, convolution and the quadratic relation
7–11: affine Weyl groups and Bernstein generators
12–16: construct and verify a matrix module
17–22: invariant lines, extensions and central characters
23–27: spherical sectors, normalization and geometry
Practice and worked answers
Sources and companion articles
Teaching guide
Prerequisites and adjacent owners are Hecke Algebras, Kazhdan–Lusztig Theory, Automorphic Representations and q-Schur Algebras. This page owns the Iwahori and explicit affine-module route.
1. What the local field contributes
Let F be a non-Archimedean local field, O its ring of integers, and ϖ a uniformizer. The residue field O/ϖO is finite; write its cardinality as q. Examples include finite extensions of the p-adic numbers and Laurent-series fields over finite fields. The finite residue field supplies the numerical parameter in basic p-adic Hecke examples.
For GL₂(F), let K=GL₂(O). Reduce entries modulo ϖ and require the resulting matrix to be upper triangular. The inverse image of that triangular subgroup is an Iwahori subgroup I. Concretely, an element of I is an invertible integral matrix whose lower-left entry belongs to ϖO.
K preserves a standard lattice, while I preserves additional flag data after reduction. Therefore I is smaller and its fixed-vector space can be larger. The foundational Iwahori–Matsumoto work identifies the Hecke-algebra structure arising from this subgroup geometry. [1]
2. Fixed vectors compress a specified representation sector
For a smooth complex representation V of G=GL₂(F), define V^I={v:iv=v for every i∈I}. Smoothness means each vector is fixed by some open subgroup. Admissibility requires fixed-vector spaces for compact open subgroups to be finite-dimensional. An infinite-dimensional admissible representation can therefore have a finite-dimensional Iwahori-fixed space.
This compression is selective. A representation without I-fixed vectors maps to zero. The classical Hecke correspondence concerns the appropriate Iwahori-generated sector, not the entire category of all smooth representations. Borel’s theorem provides a primary reference for the required categorical boundary. [2]
Studying a representation known to be generated by its I-fixed vectors is different from deciding whether an arbitrary representation belongs to that sector. The Hecke algebra addresses the first problem in its proper setting; it does not make the second problem disappear.
3. Convolution and measure normalization
Let H(G,I) consist of compactly supported, I-bi-invariant complex functions on G. Normalize Haar measure by vol(I)=1. Convolution is (f*h)(g)=∫_G f(x)h(x⁻¹g)dx. Compact support and local constancy allow the relevant calculations to be expressed through finitely many cosets.
The characteristic function 1_I is the identity. On an I-fixed vector v, the action of f is the averaged representation operator ∫f(g)g·v dg. Without vol(I)=1, the same characteristic function acts by vol(I), so the visible multiplication constants change.
This product is not pointwise multiplication of functions. Functions supported on two double cosets can convolve to a combination supported on several double cosets. The coefficients count ways of composing relative positions, explaining why the group geometry produces a deformed reflection algebra.
4. A finite flag model derives the quadratic relation
The projective line over F_q has q+1 points. On the vector space with basis v_L indexed by those points, define Tv_L=Σ_(M≠L)v_M. It sends a point to every different point and gives an elementary rank-one flag model.
Apply T twice. Returning from L to L allows any of q intermediate points. Ending at a fixed N≠L requires an intermediate point different from both L and N, giving q−1 choices. Hence T²=qI+(q−1)T, equivalently (T−qI)(T+I)=0.
At q=3 there are four points. T is the four-by-four matrix with zeros on the diagonal and ones elsewhere. T² has diagonal entries three and off-diagonal entries two. This verifies T²=3I+2T by direct matrix multiplication. The finite model explains the coefficients; it is not the whole infinite affine Hecke algebra.
5. Two eigenspaces do not determine a full module decomposition
The vector v_all=Σ_Lv_L satisfies Tv_all=qv_all. If v=Σc_Lv_L has coefficient sum zero, the coefficient of v_N in Tv is Σ_(L≠N)c_L=−c_N. Thus T acts by −1 on the q-dimensional zero-sum subspace.
The eigenvalues q and −1 follow from the quadratic relation. Their multiplicities one and q belong to this particular permutation model, not to every module of the algebra. A relation determines possible eigenvalues without fixing how often each appears.
Our later two-dimensional affine module has one line for each T-eigenvalue. Those lines need not be preserved by X₁ and X₂. Decomposing the action of one generator does not decompose the combined action of the whole algebra.
6. Exact projectors and powers
For q≠−1, put e_+=(T+1)/(q+1) and e_-=(q−T)/(q+1). The quadratic relation gives e_+²=e_+, e_-²=e_-, e_+e_-=0 and e_++e_-=1. Therefore T=q e_+−e_- and T^n=q^n e_++(−1)^n e_-.
Eliminating the projectors gives T^n=((q^n−(−1)^n)/(q+1))T+((q^n+q(−1)^n)/(q+1))I. For q=3, n=3, this is T³=7T+6I. Repeated multiplication from T²=2T+3I gives the same answer.
The relation reduces every power to a fixed two-element basis. But these projectors belong to the finite reflection subalgebra. They need not be central in the affine algebra, so their images need not be affine submodules. This qualification will be demonstrated explicitly rather than left abstract.
7. Affine reflections generate translations
On the real line, s₁(x)=−x and s₀(x)=2−x are reflections of order two. Their product s₀s₁ sends x to x+2 and has infinite order. The resulting infinite dihedral group is the simplest affine Weyl model.
Intervals between successive reflecting points are alcoves in this picture. Crossing a wall applies a reflection; repeated products move through infinitely many translated positions. The local reflection rule is simple even though the global group is infinite.
This explains why affine Hecke algebras remain infinite-dimensional while using quadratic relations resembling those of finite Hecke algebras. Deforming the local relation does not remove the infinitely many translation elements.
8. Affine and extended affine Weyl groups
A root datum specifies a finite root system together with a lattice. The affine Weyl group uses the appropriate root or coroot translation lattice; the extended affine group can use a larger lattice. Because conventions may use dual root data, naming the lattice is safer than relying on the word affine alone.
For GL₂, the extended translation lattice can be represented by Z², with S₂ swapping coordinates. The affine root lattice consists of pairs with sum zero. Translation by (1,1) is central in the extended picture and is absent from the sum-zero root lattice.
Solleveld describes these root-data distinctions explicitly. [3] The additional central direction explains why the GL₂ laboratory has two invertible lattice variables, not only one ratio of root coordinates.
9. Bernstein generators expose the torus
The Bernstein presentation separates a commuting Laurent-polynomial subalgebra from a finite Hecke subalgebra. Lattice characters multiply as θ_xθ_y=θ_(x+y), while cross-relations specify their interaction with reflection generators. [3]
For this GL₂-type example, let A=C[X₁^±1,X₂^±1]. Adjoin T with T²=(q−1)T+q, X₁X₂=X₂X₁ and TX₁T=qX₂. Both X variables are invertible. These conventions are the complete starting data for the matrix calculations below.
The algebra is free of rank two as a right A-module with basis 1,T. This rank-one Bernstein basis property means that inducing a one-dimensional A-character gives a two-dimensional module. The module is finite-dimensional even though the algebra contains infinitely many Laurent monomials.
10. Derive the cross-relations in the needed order
The quadratic relation gives T⁻¹=q⁻¹T+(q⁻¹−1). Multiplying TX₁T=qX₂ on the right by T⁻¹ yields TX₁=X₂T−(q−1)X₂. Multiplying on the left by T⁻¹ yields X₁T=TX₂−(q−1)X₂.
Rearrange to obtain TX₂=X₁T+(q−1)X₂ and X₂T=TX₁+(q−1)X₂. These are the forms that move a lattice variable past T and return an expression to the right-A basis 1,T.
Signs are easily lost because equivalent equations put T on different sides. Check any recalled cross-relation by returning to TX₁T=qX₂. Resemblance to a formula from another convention is not a substitute for that verification.
11. Central combinations
Adding TX₁=X₂T−(q−1)X₂ and TX₂=X₁T+(q−1)X₂ cancels the corrections, so T commutes with X₁+X₂. It also commutes with X₁X₂. The lattice variables themselves commute with these symmetric expressions, making both expressions central.
In the standard nondegenerate affine Hecke setting, the center is the Weyl-invariant Laurent-polynomial subalgebra. [3] In this example, sum and product encode the unordered pair of nonzero torus parameters.
The center records less than a module’s entire structure. It cannot remember the order of an inducing pair, or how composition factors are joined in a nonsplit extension. The next calculations show this loss explicitly.
12. Construct an induced module
Choose a,b∈C×. Let C_(a,b) be the one-dimensional A-module with X₁=a and X₂=b. Form M(a,b)=H⊗_A C_(a,b). Its basis is v₁=1⊗1 and v₂=T⊗1.
The defining character gives X₁v₁=av₁ and X₂v₁=bv₁. Also Tv₁=v₂. The quadratic relation gives Tv₂=qv₁+(q−1)v₂. Thus the T action is already known.
The module is generated by v₁ but is not assumed irreducible. Induction enforces the prescribed A-character on its generating vector. It does not guarantee that the resulting representation has no proper invariant subspace.
13. Write every generator matrix
The cross-relations give X₁v₂=bv₂−(q−1)bv₁ and X₂v₂=av₂+(q−1)bv₁. In the basis (v₁,v₂), with columns recording images, the matrices are:
T = [ 0 q ]
[ 1 q−1 ]
X₁ = [ a −(q−1)b ]
[ 0 b ]
X₂ = [ b (q−1)b ]
[ 0 a ].
Reading rows as images would transpose these matrices and change the eigenvector calculation. Fix the column convention before multiplying, extracting invariant lines or identifying quotient actions.
14. Verify the algebra relations
Direct multiplication gives T²=[[q,q(q−1)],[q−1,q+(q−1)²]], equal to (q−1)T+qI. In X₁X₂, the off-diagonal terms cancel and the product is abI. The reverse product is the same. Since ab≠0, the lattice matrices are invertible.
Multiplying TX₁T gives qX₂. We have therefore verified the quadratic relation, commuting lattice variables and the cross-relation in the exact normalization chosen. These are compatible actions, not merely matrices with suggestive labels.
A dimension check would only show that the matrices act on a two-dimensional space. Relation checks establish a representation of the algebra. Size and structure are separate, even in this small example.
15. Compute the central character
The matrices satisfy X₁+X₂=(a+b)I and X₁X₂=abI. Thus symmetric central functions act through the unordered pair {a,b}. Swapping the parameters preserves both scalars.
The polynomial z²−(a+b)z+ab has roots a,b. Its coefficients recover the unordered pair but not the order. This elementary observation explains why Weyl orbits of torus parameters enter the center.
It does not imply that every module with that central character is isomorphic. At q=2, M(2,1) and M(1,2) will have the same central scalars and the same composition factors, yet different invariant submodules. Central data cannot determine every extension.
16. Decompose T, then test the other generators
The vectors v_+=v₁+v₂ and v_-=−qv₁+v₂ have T-eigenvalues q and −1. They are independent because q+1≠0. Thus T is diagonalizable for every choice of a,b in our range.
A line invariant under the full affine algebra must be T-stable, hence must be one of these two eigenlines. It must also be preserved by X₁ and X₂. Those further conditions provide the entire reducibility test for this two-dimensional module.
Instead of guessing an arbitrary invariant line, we now have two explicit candidates. This is the benefit of using a simple generator first while remembering that it is only the first stage of the classification.
17. The q-eigenline is invariant exactly when a=qb
Apply X₁ to v_+=(1,1). The result is (a−(q−1)b,b). It is a multiple of (1,1) exactly when a−(q−1)b=b, giving a=qb. The same condition makes X₂ preserve the line.
On that line, X₁ acts by b, X₂ acts by a and T acts by q. The lattice pair on the invariant line is therefore (b,a), reversed from the original inducing character (a,b).
An inducing character need not be the character of an invariant submodule. Applying the matrices resolves that issue directly. A remembered labeling convention from principal-series notation can otherwise put the correct character on the wrong part of an extension.
18. The minus-one eigenline is invariant exactly when a=b/q
Apply X₁ to v_-=(-q,1). The result is (−qa−(q−1)b,b). Equating it to b(-q,1) gives qa=b, or a=b/q. X₂ preserves the line under the same condition.
The line has T-eigenvalue −1 and lattice eigenvalues (b,a). The other T-eigenline is not invariant, since q>1 and b≠0 prevent a=qb and a=b/q from holding simultaneously.
Thus each reducible case has exactly one proper nonzero invariant line. It cannot split as a direct sum of two one-dimensional affine submodules. Reducibility and complete reducibility are different properties.
19. A complete criterion for the displayed family
Every proper nonzero subspace of a two-dimensional complex space is a line. An affine invariant line must be a T-eigenline, and we have tested both. Therefore M(a,b) is irreducible exactly when a/b is neither q nor q⁻¹, with q>1 and ab≠0.
This is a proof for this GL₂-type family. Higher-rank and unequal-parameter affine Hecke algebras require their own criteria involving root data and parabolic induction. Solleveld’s treatment separates those cases. [3]
The mechanism is transferable: a parameter relation makes several generator actions preserve the same line. Reducibility is not caused by one eigenvalue alone; it is caused by compatibility with a common proper subspace.
20. Equal torus parameters can still give an irreducible module
Take q=2 and a=b=1. Then X₁=[[1,−1],[0,1]] is a nontrivial Jordan matrix, not diagonalizable. Yet a/b=1 is neither two nor one-half, so the full affine module is irreducible.
The unique X₁-eigenline is spanned by v₁, but Tv₁=v₂ leaves that line. Thus an eigenline for the lattice algebra is not a submodule for the full Hecke algebra.
This disproves two shortcuts. An irreducible representation need not make every individual generator diagonalizable. Conversely, diagonalizing one generator does not prove a decomposition into irreducible submodules. The combined action is what matters.
21. Same center, opposite extensions
At q=2, M(2,1) has a unique invariant line with T=2 and lattice eigenvalues (1,2). Its quotient has T=−1 and lattice eigenvalues (2,1). In M(1,2), the unique invariant line instead has T=−1 and lattice eigenvalues (2,1), while the quotient has T=2 and lattice eigenvalues (1,2).
Both modules have X₁+X₂=3I and X₁X₂=2I. They have the same two composition factors attached in opposite directions. They are not isomorphic: an isomorphism must preserve the unique proper submodule, but T acts on those submodules by different scalars.
Neither extension splits. A complement would make the other T-eigenline an affine submodule, contrary to the explicit tests. We have thus exhibited extension data invisible to central characters and composition-factor lists.
22. Traces also forget extension direction
With an invariant subspace, every representing matrix can be put into block upper-triangular form. Its trace is the sum of the traces on the submodule and quotient. Off-diagonal extension information does not contribute.
The two resonant modules therefore have the same ordinary trace character even though they are not isomorphic. Their composition factors agree, so all trace sums agree. Characters can be complete in a semisimple setting and incomplete when extensions survive.
This repeats the lesson of Categorification and Grothendieck Groups: an additive invariant can be powerful while forgetting how an object was assembled. Here the matrices show exactly what was lost.
23. A spherical projector selects a corner algebra
The projector e_+=(T+1)/(q+1) selects the q-eigenspace of the finite reflection generator. In the p-adic interpretation, finite Hecke symmetrization relates Iwahori invariants to invariants under a larger compact subgroup. The algebra acting naturally on the selected sector is the corner e_+He_+.
In M(1,1) at q=2, the e_+-image is one-dimensional but not X₁-stable. Projecting first does not turn every affine operator into an endomorphism of the line. Sandwiching h as e_+he_+ does.
This distinguishes spherical and Iwahori Hecke algebras, and illuminates spherical corners in other representation families. A corner is related to its parent but is not identical to it because both arise from the same representation problem.
24. Parameter changes can invalidate earlier arguments
In a split p-adic equal-parameter example, q is a positive residue-field cardinality. Abstract algebras allow other complex specializations. At q=−1, the roots of the quadratic relation coincide and our projector denominators vanish.
The finite rank-one algebra becomes C[T]/((T+1)²), containing a nonzero nilpotent class T+1. This differs from q>1, where the finite reflection algebra splits into two eigensectors. The value −1 is an abstract specialization, not a residue-field size.
Even at q>1, the full affine algebra admits nonsplit modules, as our resonant example proves. Semisimplicity of a finite subalgebra does not make the infinite affine representation category semisimple. Finite and affine deformation arguments must be kept separate. [3]
25. What higher rank adds
Higher rank introduces several reflection generators, braid relations and more torus variables. Principal-series modules are induced from torus characters, but reducibility can involve several roots. Parabolic subalgebras allow induction from smaller affine Hecke systems.
Classification uses tempered representations, discrete series, induction parameters and intertwining operators. Solleveld presents these structures with their root-data and parameter conditions. [3] Roche’s work constructs types and Hecke algebras for principal-series sectors of split reductive p-adic groups. [4]
The rank-one laboratory still matters. It explains how a root ratio creates an invariant subspace, why the center misses extensions and why induction is separate from irreducibility. These mechanisms survive even when the classification requires more geometry and combinatorics.
26. Geometry explains distinguished structures
Canonical Hecke bases and intersection-cohomological constructions meet in Kazhdan–Lusztig theory. Affine flag geometry extends the finite flag framework, while geometric classification uses additional dual-group data. The original papers distinguish the basis construction from the Deligne–Langlands classification theorem. [5]
Our matrix module was constructed algebraically, not by counting components of an unspecified variety. Geometry provides a wider explanation for distinguished bases and classification; the explicit matrices provide a checkable local model. Neither description should be used to pretend that a proof in the other has already been supplied.
Continue geometrically through Affine Springer Fibres and Soergel Bimodules. For a different enlargement with two affine directions, use Double Affine Hecke Algebras. Affine and double affine are distinct constructions.
27. Verify an unfamiliar module systematically
Record the field, q-parameters and generator normalization. Identify the commuting torus algebra and its character. Declare a matrix convention. Verify quadratic, braid and cross-relations, then check invertibility of lattice generators. Test candidate subspaces against every generator.
Compute central scalars and compare them with inducing parameters. At reducibility, identify the actual submodule and quotient and test for a complement. Do not replace a nonsplit extension by a direct sum. Finally identify exactly which p-adic sector, if any, the module models.
The result is a verified representation, not only a list of possible eigenvalues. Compatible actions are the structure; dimensions, central values and characters measure selected parts of that structure.
Practice: algebra first, module second
1. At q=4, find T² in the finite projective-line model. 2. Calculate T⁴ at q=3. 3. Derive T⁻¹ for nonzero q. 4. Compute X₁+X₂ and X₁X₂ in M(a,b).
5. At q=3,a=6,b=2, identify the invariant line and its three generator eigenvalues. 6. Do the same for a=2,b=6. 7. Is M(1,1) irreducible at q=3? 8. Why do M(6,2) and M(2,6) share a central character but fail to be isomorphic? 9. Is e_+M automatically X₁-stable? 10. Why can Iwahori invariants fail to see a smooth representation?
Answers 1–4: exact relation checks
1. T²=4I+3T. Four intermediate points return to the start; three reach a fixed different endpoint. These coefficients count different conditions.
2. T⁴=20T+21I. Since T³=7T+6I and T²=2T+3I, multiply once more to get 7(2T+3I)+6T. This agrees with the projector formula.
3. From T²−(q−1)T=qI, divide by nonzero q to obtain T⁻¹=q⁻¹T+(q⁻¹−1)I. The hypothesis q≠0 is used explicitly.
4. The sum is (a+b)I and the product is abI. Both are symmetric in a,b, so neither remembers their order.
Answers 5–8: invariant lines and extension direction
5. Since a=qb, the invariant line is C(v₁+v₂). On it T acts by three, X₁ by two and X₂ by six. The lattice eigenvalues are reversed from the inducing pair.
6. Since a=b/q, the line is C(−3v₁+v₂). Its eigenvalues are T=−1, X₁=6 and X₂=2. The other T-eigenline is not affine invariant.
7. Yes. The ratio one is neither three nor one-third. Neither T-eigenline is preserved by the full algebra, so no proper nonzero submodule exists.
8. Both central sums are eight and products twelve. Their unique invariant lines have T-eigenvalues three and −1. An isomorphism would preserve the unique submodule and intertwine T, which is impossible. Central equality has not recovered extension direction.
Answers 9–10: the boundary of compression
9. No. The projector selects a T-eigenspace, but X₁ can move it. The corner e_+He_+ acts on the selected space; unrestricted H need not. The irreducible equal-parameter example gives a concrete counterexample.
10. A smooth representation can have V^I=0. The Iwahori Hecke algebra models the appropriate I-generated sector, not every smooth representation. Admissibility ensures finite-dimensional fixed spaces when present but does not guarantee they are nonzero for a chosen I.
Sources and reading boundaries
[1] Nagayoshi Iwahori and Hideya Matsumoto, On Some Bruhat Decomposition and the Structure of the Hecke Rings of p-Adic Chevalley Groups. Primary source for the Iwahori–Matsumoto structure and its p-adic origin.
[2] Armand Borel, Admissible Representations of a Semi-Simple Group over a Local Field with Vectors Fixed under an Iwahori Subgroup. Read the subgroup, coefficient and category hypotheses before applying a fixed-vector equivalence.
[3] Maarten Solleveld, Affine Hecke Algebras and Their Representations. Detailed treatment of root data, presentations, centers, rank-one examples and classification. The present matrix module is derived in its declared GL₂ normalization.
[4] Alan Roche, Types and Hecke Algebras for Principal Series Representations of Split Reductive p-Adic Groups. Primary reference for Hecke models of specified representation sectors.
[5] David Kazhdan and George Lusztig, Representations of Coxeter Groups and Hecke Algebras, and Proof of the Deligne–Langlands Conjecture for Hecke Algebras. Canonical-basis construction and geometric classification are distinct results.
Representation Mathematics — Batch 17
Elliptic Hall Algebra: rank-degree geometry and Hall counting
Quantum Toroidal Algebras: currents, Fock modules and spectral parameters
Quiver Hecke Superalgebras: parity, odd nilHecke operators and categorification
Return to the BTT Mathematics Learning Hub
Teaching guide: one generator is not the whole representation
Begin with the q+1-point adjacency model and derive the quadratic relation by counting two-step paths. Construct the projectors and explain their image spaces. This establishes the local algebra without requiring p-adic analysis.
Next derive the matrices of M(a,b) from the cross-relations, then verify those relations before classifying the module. Find the two T-eigenlines and test each against X₁ and X₂. The exceptional ratios should emerge as conditions for joint invariance.
Finish with the nonsplit resonant pair and the equal-parameter irreducible example. Ask what the center, trace character, composition-factor list and full module each retain. Answering with these matrices demonstrates understanding of representation structure rather than only terminology.