How statement and proof provenance work
The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.
- Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
- AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
- AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.
These labels describe origin, not correctness: citations and verification chips remain separate evidence.
Soergel Intersection Forms and Hodge Theory — Examples
1 · Prerequisites
- Abelian Categories
- Adjunctions Units and Counits
- Algebraic Extensions, Extension Degree, and Finite Fields
- Binary Operations, Monoids, Groups and Subgroups
- Bounded Bimodule Complexes and Derived Tensor
- Braided and Symmetric Monoidal Categories
- Bruhat Decomposition and Flags over Finite Fields
- Cardinal Arithmetic, Cofinality and the Alephs
- Categorical Braid Actions and Decategorification
- Categories, Functors and Natural Transformations
- Chain Complexes and Homology
- Chain Homotopy and the Homotopy Category
- Congruences, the Integers Modulo n and the Chinese Remainder Theorem
- Conjugacy in Sₙ, Generation, and the Simplicity of Aₙ
- Construction of the Natural Numbers
- Construction of the Real Numbers via Cauchy Sequences
- Construction of the Real Numbers via Dedekind Cuts
- Continuity, IVT, EVT, and Uniform Continuity
- Cosets, Index and Lagrange's Theorem
- Countability and Uncountability
- Derived Categories
- Derived Functors
- Determinants of Matrices over a Commutative Ring
- Divided Powers, Integral Forms, and Quantum Highest Weights
- Divisibility, Euclidean Domains, Principal Ideal Domains and Unique Factorisation
- Divisibility, Greatest Common Divisors and Bézout's Identity
- Dual Spaces, Bilinear and Quadratic Forms, and Sylvester's Law of Inertia
- Exactness and the Member Calculus
- Ext and Balanced Resolutions
- Finite Counting, Factorials and Binomial Coefficients
- Finite Weyl Invariants, Bruhat Order, and Kostant Harmonics
- Foundations of the Real Numbers for Analysis
- Free Groups and Presentations
- Free Modules, Exact Sequences, Projective and Injective Modules
- Garside Structure, Normal Forms, and the Center
- Gaussian Elimination, Elementary Matrices and Reduced Row Echelon Form
- Graded Bimodules and Tensor Functors
- Group Actions, Orbits, Stabilisers and Cayley's Theorem
- Group Homomorphisms and the Isomorphism Theorems
- Homological Gaussian Elimination
- Ideals, Quotient Rings and the Isomorphism Theorems for Rings
- Kazhdan–Lusztig Bases, Polynomials, and Cells
- Limits and Colimits
- Limits of Real Functions
- Linear Independence, Bases and Dimension
- Linear Transformations, Rank-Nullity and Quotient Spaces
- Long Exact Sequences in Homology
- Mapping Cones Cylinders and Chain Triangles
- Matrices, the Matrix of a Linear Map, and Change of Basis
- Modules, Submodules, Quotient Modules and the Isomorphism Theorems
- Monotone Sequences, Bolzano-Weierstrass, and Cauchy Completeness
- Normal Subgroups and Quotient Groups
- Order, Zorn's Lemma, and the Axiom of Choice
- Ordinal Arithmetic and the First Uncountable Ordinal
- Ordinals, Cardinals, and Transfinite Recursion
- Permutation Statistics, Inversions and Eulerian Numbers
- Polynomial Rings, the Division Algorithm and Roots
- Preadditive and Additive Categories and Biproducts
- Primes, Euclid's Lemma and the Fundamental Theorem of Arithmetic
- Principal Series Representations of GL N over a Finite Field
- Projective and Injective Resolutions
- Rees Modules Artin Rees and Hilbert Samuel Theory
- Reflective Subcategories and the Adjoint Functor Theorems
- Relations, Functions, and Quotients
- Rings, Subrings, Integral Domains and Fields
- Roots, Rational Powers, and Classical Inequalities
- Rouquier Complexes and Categorical Braid Relations
- Simple Field Extensions and the Construction of the Complex Numbers
- Soergel Intersection Forms and Hodge Theory
- Splitting Fields
- Subobject Lattices Generators and the Grothendieck Axioms
- Suprema and Infima
- Symmetric Groups, Cycle Decomposition and the Sign Homomorphism
- Tensor Products of Modules
- The Diagram Lemmas in an Abelian Category
- The Field of Fractions and Localisation
- The ZFC Axioms and the Basic Set Constructions
- Topology of ℝ
- Tor Flatness and Global Dimension
- Triangulated Categories
- Type-A Soergel Bimodules and Hecke Categorification
- Universal Properties, Representables and the Yoneda Lemma
- Vector Spaces, Linear Subspaces, Span and Direct Sums
- Young Diagrams Tableaux and Permutation Modules
2 · Summary
This companion collects finite computations for soergel-intersection-forms-and-hodge-theory. The rank-one example computes the trace form, the reduced module and the Lefschetz map of for ; the rank-two examples compute the character of and the braid decomposition , and identify the coinvariant algebra of with the longest-element bimodule, including its Kazhdan-Lusztig expansion.
The final example works the hard Lefschetz and Hodge-Riemann signs of the coinvariant algebra of in an explicit two-variable model, with the normalized top trace , and then verifies the same signs for every element of the dominant chamber.
3 · Logical flowchart
4 · Definitions, theorems and proofs
The rank-two Bott-Samelson bimodule B_sB_t: character, decomposition and the braid relation
Example
Take and the two simple reflections , . The rank-two Bott-Samelson bimodule has character in the standard basis, so its -multiplicities are at , at and at with shift , and at the identity with shift ; in particular all multiplicities are non-negative and , whose value at is the ordinary real dimension . In the product has length and has the same character , so is already indecomposable; equivalently the only summand is the top one and all multiplicities for vanish. Similarly with characters , whereas ; the two Bott-Samelson products are not isomorphic; the example also verifies the rank-two values of the Kazhdan-Lusztig polynomials used on the companion page.
Verification
Given: The simple reflections , of , the one-letter bimodules and the Bott-Samelson product (The Bott–Samelson bimodule of a word, The real type-A Soergel category by scalar extension), the character with its -multiplicities (Standard graph bimodules, support filtrations and characters, The type-A standard character is multiplicative), the normalized Hecke algebra with its two bases and (Existence and uniqueness of the Kazhdan–Lusztig basis), the rank-two decomposition theorem (Rank-two type-A Soergel bimodule decompositions).
[F1] The character is multiplicative and satisfies ; the coefficient of in is , and the reduced dimension of a layer is one in degree (The type-A standard character is multiplicative, Standard graph bimodules, support filtrations and characters).
[F2] The standard basis of the Hecke algebra multiplies by when ; the Kazhdan-Lusztig basis is the unique basis of bar-invariant elements (Existence and uniqueness of the Kazhdan–Lusztig basis).
[F3] The rank-two decomposition theorem: with no shifts; exchanging and gives (Rank-two type-A Soergel bimodule decompositions).
[F4] The polynomial ring has only nonnegative degrees and degree-zero part . The generator has left basis for , since ; tensor products have the corresponding finite left bases (The real type-A Soergel category by scalar extension).
The character of the two-letter product. By [F1] and multiplicativity, ; since is reduced, [F2] gives , so the standard-basis expansion is . Its coefficients give the multiplicities displayed, and reading the layers through [F1] the reduced degrees are , so the graded dimension is and the ordinary real dimension is .
The Kazhdan-Lusztig element of . Put and ; then . Since , and (the last identity uses ), applying the semilinear involution gives , whose coefficients are at and at ; hence , and by uniqueness [F2], .
Indecomposability of . The unshifted tensor is generated as a bimodule by . To see this, write a middle polynomial as with , which slide to the first slot. The remaining slides to the last slot because fixes . Therefore every tensor is a sum of . After the shift its lowest component is one-dimensional in degree , by its finite left tensor basis and . A degree-zero bimodule endomorphism must send the generator to a scalar multiple of itself, and generation forces that same scalar on all of . Hence its degree-zero endomorphism algebra is , with no nontrivial idempotent, so is indecomposable. The balanced map is surjective onto the graph bimodule, showing that this indecomposable has the top graph of its reduced word; it is the normalized summand . Thus with no lower summands, independently of the Hom formula.
The three-letter braid relation. By multiplicativity and the products computed in steps 1.1 and 2.1, ; expanding with the Hecke relations gives Indeed the expansion is where . The element is bar fixed, with triangular lower coefficients in , so uniqueness in [F2] identifies it with . The ordinary real dimensions of the reductions add to . Exchanging and in [F3] gives , whose character differs by , so these Bott-Samelson products are not isomorphic; the rank-two Kazhdan-Lusztig data are the coefficients displayed in step 2.1. This verifies the example.
Remarks
- The reduction to rank two is essential here: the computation uses the explicit two- and three-letter decompositions and is an illustration only; the design of the page forbids its use as a supplier inside the general induction, whose base is computed on the axiom page.
- The two-letter indecomposability is established by its scalar degree-zero endomorphisms; the triple decomposition is the explicit rank-two theorem. Neither computation uses the general character theorem.
The coinvariant algebra of S_3 as the Soergel bimodule of the longest element
Example
Take and let be the longest element of . Then and its reduced module is , where is the unshifted coinvariant algebra: has the Poincare polynomial , so , and in the centered grading The standard-basis expansion of the Kazhdan-Lusztig basis element is obtained by solving the bar-invariance and triangularity conditions with coefficients in ; its coefficients are non-negative, its value at sums to , and it agrees with the rank-one and rank-two Kazhdan-Lusztig data for . Hence : Soergel's conjecture holds for the longest element of , and the -multiplicities of the coinvariant bimodule are the six coefficients just displayed, one for each element of .
Verification
Given: with , the longest element , the rank-two longest bimodule (The rank-two longest type-A Soergel bimodule), the reduced module and the character (The reduced Soergel module and the dominant Lefschetz operator, Standard graph bimodules, support filtrations and characters), the normalized Hecke algebra with and the Kazhdan-Lusztig basis (Existence and uniqueness of the Kazhdan–Lusztig basis), and multiplicativity of the standard character (The type-A standard character is multiplicative).
[F1] The rank-two longest bimodule is after real scalar extension, and right reduction is tensoring with (The rank-two longest type-A Soergel bimodule, The reduced Soergel module and the dominant Lefschetz operator). The symmetric-polynomial and quotient-basis computations needed here are supplied in step 1.1; no general invariant-ring theorem is assumed.
[F2] In the Hecke algebra for every simple , and the bar involution acts on the standard basis by ; the Kazhdan-Lusztig basis is the unique basis of elements with and (Existence and uniqueness of the Kazhdan–Lusztig basis).
[F3] The coefficient of in is with non-negative integral multiplicities, and for the standard layer the reduced dimension sits in degree ; hence the multiset of monomials of the character, read with the degrees , is the graded dimension of the reduced module (Standard graph bimodules, support filtrations and characters).
[F4] The character is multiplicative with simple value , and the rank-two decomposition is (The type-A standard character is multiplicative, Rank-two type-A Soergel bimodule decompositions).
The bimodule and its reduction. Symmetric polynomials in three variables are generated by : for a homogeneous symmetric polynomial, its lexicographically leading monomial has exponents ; subtracting its coefficient times lowers that leading monomial. The finite set of monomials of the fixed total degree makes this reduction terminate. Thus and [F1] gives the reduced module with . Eliminating gives : the relation is , and subtracting this from gives , with the same identities proving the reverse inclusion of ideals. Over , division by the monic quadratic in gives the free basis ; hence is a real basis and the dimensions are in degrees , centered at after shifting by . The tensor bimodule is generated as a bimodule by and its bottom component is one dimensional, so every degree-zero endomorphism is scalar; it is therefore indecomposable. The balanced surjection onto proves that it has the graph in its support. In the rank-two decomposition it is the summand carrying the graph (the other summand has only the graphs), so its label is .
The Kazhdan-Lusztig element. Put and . The element lies in , since all coefficients belong to . For bar-invariance we expand: from and the reduced products and one computes and , while and . Applying the semilinear involution to gives ; collecting coefficients in the standard basis, the coefficient of becomes , the coefficient of becomes , and the coefficient of becomes . Hence , and by the uniqueness clause of [F2] .
Character and multiplicities. By [F4], , the last equality following by the quadratic relation and reduced multiplication. Step 1.2 gives , so the -multiplicities of are exactly the six coefficients of , one for each of the six elements of . Reading them through [F3], the layer contributes in degree , the two layers contribute in degree with multiplicity two, the two layers contribute in degree with multiplicity two and the layer contributes in degree ; this is the graded dimension of the centered coinvariant algebra computed in step 1.1, and the value at is . The coefficients are the classical normalization of these coefficients. This verifies the example.
Remarks
- The two computations are independent: the bar-invariance determines the Kazhdan-Lusztig element from the Hecke algebra alone, and the graded dimensions of the coinvariant algebra come from the invariant theory; their agreement is the content of Soergel's conjecture at .
- The example is an illustration of the character theorem, not a supplier of it; the base of the simultaneous induction is computed on the axiom page and does not use this page.
Hard Lefschetz fails for a non-dominant degree-two element: the rank-one case
Statement refuted
In the rank-one case the element is a degree-two element of with , so it is not strictly dominant. Its action on the reduced rank-one Soergel module is zero: is -invariant and of positive degree, and acts on the right basis element as the right action by an element of , hence ; a fortiori it cannot map isomorphically to . Therefore for the hard Lefschetz property fails (the map on the degree is the zero map), even though is a symmetric degree-two endomorphism of the reduced module; the same failure occurs for every degree-two element with . Consequently the hard Lefschetz and Hodge-Riemann conclusions do not hold for arbitrary degree-two elements when strict dominance is omitted. This does not assert that every Lefschetz element is strictly dominant.
Facts & Assumptions
Given: The rank-one setting , with , the simple reflection with coordinate root , the generator with right basis of degree and of degree , its reduced module with the induced basis , and the degree-two element (The Soergel bimodule of a simple reflection, The real type-A Soergel category by scalar extension, The reduced Soergel module and the dominant Lefschetz operator).
The left and right actions on satisfy and , and , is the unit of the commutative ring (The trace, the invariant intersection form on Bott-Samelson bimodules and its reduction, The Soergel bimodule of a simple reflection).
The reduction functor kills right multiplication by : , and for (The reduced Soergel module and the dominant Lefschetz operator).
Hard Lefschetz for a degree-two symmetric operator on a graded space with lowest degree requires to be an isomorphism in the case ; definiteness and symmetry conventions as in Primitive subspaces, hard Lefschetz and the Hodge-Riemann signs: the dictionary of hypotheses S, hL and HR and Bilinear forms, and symmetric, skew-symmetric, and alternating bilinear forms.
Counterexample
Non-dominance. The reflection exchanges and , so is -invariant and in the standard form of the coordinate normalization; in particular is not strictly dominant.
The left action of equals its right action. Since , the balanced tensor relation gives in ; hence for every the left multiplication equals the right multiplication : indeed both actions are -linear and the identity holds on the two basis elements (where ) and (where both equal by [F1]).
The operator vanishes on the reduced module. By step 2.1, lies in , since it is the right product of with ; by [F2] its class in is zero, so . Similarly lies in and reduces to zero. Since and span , the induced operator is the zero endomorphism of .
Failure of hard Lefschetz. The reduced module has with bases and . The degree-one Lefschetz map is zero by step 3.1, hence not an isomorphism, and hard Lefschetz fails for in the sense of [F3]. The bottom Lefschetz form is likewise identically zero, so the Hodge-Riemann relations with any nonzero sign also fail.
The general degree-two case. Let satisfy ; then and is a scalar multiple of . If the operator is trivially zero; if the same computation as in steps 2.1-4.1 applies because invariance and positivity of degree are unaffected by the scalar. Hence every degree-two element with vanishing evaluation on yields the zero operator on the reduced rank-one module and fails hard Lefschetz. This proves the claimed failure and shows that allowing arbitrary degree-two elements invalidates the conclusion; no choice principle is used.
Remark
The counterexample isolates the hypothesis at the very bottom of the induction: the strict dominance of is used on this page exactly through the positivity of the scaling factors for reduced prefixes, and at the simple reflection itself the vanishing of the coroot evaluation makes the operator vanish on the whole reduced module. Since is still symmetric of degree two, the failure is not an artifact of a badly normalized form.
The rank-one Soergel bimodule: trace form, reduced module and hard Lefschetz
Example
Take , so and with , and . In the right -basis of degrees and the multiplication relations are The trace is for and the intersection form has the Gram matrix whose determinant is , so the form is non-degenerate, and the induced form on the reduced module is non-degenerate too. The reduced module has the real basis with degrees and , so ; for , which is strictly dominant with , one computes and . Hence is an isomorphism, hard Lefschetz holds, the primitive subspace in degree is all of , and the Lefschetz form is positive on it: , matching the Hodge-Riemann sign . Finally , the Kazhdan-Lusztig basis element of the simple reflection, which is Soergel's conjecture at , and the rational model over gives the same numbers.
Verification
Given: The rank-one data above: with right basis (degree ) and (degree ), the trace and intersection form of the Bott-Samelson bimodule (The trace, the invariant intersection form on Bott-Samelson bimodules and its reduction, The Soergel bimodule of a simple reflection), the reduced module and dominant operator (The reduced Soergel module and the dominant Lefschetz operator, The real type-A Soergel category by scalar extension), and the HR dictionary (Primitive subspaces, hard Lefschetz and the Hodge-Riemann signs: the dictionary of hypotheses S, hL and HR).
[F1] The left action satisfies and ; the products are , , ; the trace is the coefficient of in the right basis, and (The trace, the invariant intersection form on Bott-Samelson bimodules and its reduction).
[F2] The reduction kills right multiples by positive-degree elements; a non-degenerate invariant form descends to a non-degenerate form on the reduction, and the induced form on the reduced module is the one used in the HR dictionary (The reduced Soergel module and the dominant Lefschetz operator, The trace, the invariant intersection form on Bott-Samelson bimodules and its reduction).
[F3] For a graded space with lowest degree hard Lefschetz requires to be isomorphisms, the primitive subspaces are and the standard sign in degree is (Primitive subspaces, hard Lefschetz and the Hodge-Riemann signs: the dictionary of hypotheses S, hL and HR, Hard Lefschetz gives the primitive decomposition, orthogonality and the signature formula).
[F4] The standard character sums the layers with coefficients , where (Standard graph bimodules, support filtrations and characters). The bar-fixed triangular element at a simple reflection is (Existence and uniqueness of the Kazhdan–Lusztig basis).
Products and the trace. The multiplication rules are those of [F1]. By definition the trace extracts the coefficient of in the right basis, so for .
The Gram matrix and non-degeneracy. Using and the products of step 1.1, , and , so the Gram matrix in the basis is ; its determinant is , hence the form is non-degenerate, and by [F2] so is the induced form on the reduced module.
The reduced action of . Let . Since is -anti-invariant, and , so by [F1] we get ; reducing kills the right multiple by [F2], so . Similarly by [F1] and is a right multiple by a positive-degree element, so . Also , so is strictly dominant.
Hard Lefschetz, primitivity and the sign. The reduced module has real basis in degree and in degree , hence both graded pieces are one-dimensional; by step 2.2 the degree-one Lefschetz map sends a basis vector to a basis vector and is an isomorphism, which is hard Lefschetz by [F3]. The primitive subspace is , since maps the one-dimensional space to the higher space in two steps and lands in degree , which is zero; the Lefschetz form on it is , the reduction of the Gram entry , and the standard sign in degree is , so HR holds with the standard sign and the bottom form is positive.
The character. Multiplication is surjective with kernel generated on the right by , of degree . The left-action rule gives , so the kernel is and the quotient is . These two layers give . The top graph layer has generator degree and the unit quotient has generator degree . The rational model has the same formulas after real scalar extension.
Remarks
- This is the base case of the rank-one computations used both in the counterexample on the non-dominant element and in the base of the simultaneous induction; the two-dimensional reduced module has a one-dimensional primitive space in degree , and the sign is the standard sign.
- The trace normalization is the one of the Bott-Samelson intersection form; no freedom remains after the normalizations of the page, since the form is unique up to scalar and the trace fixes the scalar by .
Hard Lefschetz and the Hodge-Riemann signs for the coinvariant algebra of S_3
Example
For identify the centered reduced longest-element module with , where has dimensions in degrees , so that the centered degrees are . For every strictly dominant hard Lefschetz is the pair of isomorphisms and . The primitive spaces used in the Hodge-Riemann relations are of dimension one and of dimension one; their Lefschetz forms are respectively positive and negative definite, and the primitive decomposition consists of one length-four -string and one length-two -string. For an explicit strictly dominant choice let , , , and normalize . Then , with basis ; the -map from degree to degree has the matrix , , and . These computations verify hard Lefschetz and the primitive signs directly in rank two.
Verification
Given: The coinvariant algebra and the identification with the centered reduced module of (The coinvariant algebra of S_3 as the Soergel bimodule of the longest element), the operator of left multiplication by a strictly dominant degree-two element and the trace normalization (The reduced Soergel module and the dominant Lefschetz operator, The trace, the invariant intersection form on Bott-Samelson bimodules and its reduction), and the hard-Lefschetz and Hodge-Riemann dictionary with the sign for lowest degree (Primitive subspaces, hard Lefschetz and the Hodge-Riemann signs: the dictionary of hypotheses S, hL and HR, Hard Lefschetz gives the primitive decomposition, orthogonality and the signature formula).
[F1] The centered module has lowest degree , graded dimensions in degrees and a non-degenerate invariant form normalized by the positive bottom value; for it, hard Lefschetz is together with , and the standard sign in degree is , so the primitive space must be negative definite and positive definite (The coinvariant algebra of S_3 as the Soergel bimodule of the longest element, Primitive subspaces, hard Lefschetz and the Hodge-Riemann signs: the dictionary of hypotheses S, hL and HR).
[F2] In the quotient the relations hold and, after the substitution , are equivalent to and ; the reduction of the intersection form is the standard top trace with , and is spanned by the six monomials of the display (The coinvariant algebra of S_3 as the Soergel bimodule of the longest element; the trace identification is checked in step 2.1).
The two-variable model. Put , and . Then , while and ; hence the coinvariant ideal is generated by and , and since the second relation can be replaced by , giving . The monomials span it: the relations and , together with and , reduce every monomial of degree in to one of them, and their independence follows by monic division by over , which gives the free basis over that ring.
The trace and pairing. The rank-two summand map pulls the Bott–Samelson trace back to , as follows by taking the right-coefficient trace successively in the three slots; this is the explicit trace calculation in the rank-two decomposition proof, also obtained directly from the rank-one trace convention. After right reduction it is the top trace on , since , and . Hence and all degrees other than have zero trace. The complementary-degree pairings are in degrees and in degrees , of determinant ; they prove nondegeneracy directly. The form used below is therefore the actual induced reduced intersection form, without invoking the general Hodge theorem.
The special dominant element. For one computes and using ; the matrix of from the degree- block with basis to the degree- block with basis is therefore , whose determinant is , so is an isomorphism. Moreover and : indeed , ; hence and ; hence and , so is an isomorphism as both sides are one-dimensional.
The signs for the special element. With the normalization , the bottom primitive form is , positive as required in degree ; the middle primitive form is , negative as required in degree , where the standard sign is . The primitive decomposition is read off from the two strings: spans the four degrees with one vector each, and spans degrees and ; together they account for all four graded pieces and their dimensions, so there are exactly one length-four string and one length-two string.
The general dominant chamber. Let be strictly dominant, so ; subtracting a multiple of (which is zero in ) writes its class as with . Then and , so the middle matrix is with determinant , and : indeed and , so satisfies , and expanding gives with , , which simplifies to . Hence (one-dimensional, since for ) carries a negative definite Lefschetz form, the bottom form is positive, and every strictly dominant satisfies hard Lefschetz with the standard signs on both primitive spaces.
Conclusion. Steps 2.1 and 3.1 verify the displayed isomorphisms, the two primitive spaces and their signs for the explicit choice, and step 5.1 shows the same for every strictly dominant element; the primitive decomposition has the two strings claimed. This verifies the example by its quotient and trace calculations.
Remarks
- Define the top trace to be zero off degree and to send to . Its multiplication pairing is non-degenerate: the degree- and degree- pieces pair by , and the degree- and degree- pairing in the displayed bases has matrix , of determinant . The divided-difference normalization agrees, since , and . This is a positive scalar normalization of the invariant form, as its bottom Lefschetz value is positive.
- The example is an illustration in rank two and is not used as a supplier in the simultaneous induction; the general statement for is proved on the axiom page by the induction itself.
5 · Examples, counterexamples and false statements
None yet.
Sources
- Elias-Williamson, The Hodge theory of Soergel bimodules (arXiv:1212.0791v2, final version; Ann. of Math. 180 (2014) 1089-1136)
- Soergel, Kazhdan-Lusztig-Polynome und unzerlegbare Bimoduln ueber Polynomringen (arXiv:math/0403496, published J. Inst. Math. Jussieu 6 (2007) 501-525)
- Riche, La theorie de Hodge des bimodules de Soergel (d'apres Soergel et Elias-Williamson), Seminaire Bourbaki no. 1139 (arXiv:1711.02464, Octobre 2017)