Graev Construction in Mathematical Structures

• Graev construction is a collection of canonical procedures that extend local metric, combinatorial, or Whittaker data into global structures with invariant and extremal properties.
• In geometric group theory and ultrametric settings, it provides two-sided invariant metrics on free groups, amalgams, and HNN extensions using combinatorial methods like noncrossing involutions.
• In representation theory and convexity, it yields Gelfand–Graev representations and lattice polytopes that underpin explicit Hecke module constructions and applications to Einstein metrics and cosmological polytopes.

The name Graev construction appears in several mathematically distinct settings. In geometric group theory it denotes Graev’s canonical extension of a pointed metric or ultrametric to a two-sided invariant metric on a free group and, by analogous formulas, to free products with amalgamation and certain HNN extensions (Slutsky, 2011, Shlossberg, 2014, Slutsky, 2012). In combinatorial convexity it denotes the assignment of a lattice polytope $P(T)$ to a symmetric ternary relation $T$, with applications to bounds on invariant Einstein metrics and to cosmological polytopes (Lavrov, 8 Sep 2025). In representation theory the closely related Gelfand–Graev constructions produce Whittaker-type modules, Hecke algebra models, and Weyl-group actions attached to reductive groups, their covers, and finite analogues (Bakic et al., 2020, Ginzburg et al., 2018, Taylor, 2014). This range suggests a common formal role: the passage from local combinatorial, metric, or Whittaker data to a canonical global object.

1. Metric extension on free groups and amalgams

In the classical metric setting, the input is a pointed metric space $(X,d,e)$. One forms a formal inverse copy $X^{-1}$, extends the metric by

$$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$

and considers words over $X\cup X^{-1}$. If $u,v$ are words of the same length $n$, one sets

$\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$

For $f,g\in F(X)$, the Graev metric is then

$T$0

Graev’s theorem asserts that this is a genuine two-sided invariant metric on $T$1, that it extends the given metric on $T$2, and that it is free in the category of groups equipped with two-sided invariant metrics and Lipschitz homomorphisms (Slutsky, 2011).

A particularly effective reformulation uses matches. If $T$3 is a reduced word representing $T$4, and $T$5 is an involution on the index set without crossing pairs, then one defines the matched trivial word $T$6, and obtains

$T$7

This replaces the infimum over arbitrary word pairs by a finite combinatorial search over noncrossing involutions. The formula is central in proofs of maximality and in later generalizations.

The same pattern extends to free products with amalgamation. If $T$8 are groups with two-sided invariant metrics and $T$9 is a common closed subgroup on which the metrics agree, one first defines the amalgam metric on the union by

$(X,d,e)$0

Using the same wordwise infimum construction with $(X,d,e)$1, one obtains a canonical two-sided invariant metric on the abstract amalgam $(X,d,e)$2, extending the original metrics. Positivity is proved by reduced forms and the estimate

$(X,d,e)$3

for reduced $(X,d,e)$4 representing $(X,d,e)$5 (Slutsky, 2011).

The HNN-extension case has the same character but requires a diameter bound. For a two-sided invariant metric group $(X,d,e)$6, an isometric isomorphism $(X,d,e)$7 of closed subgroups, and

$(X,d,e)$8

the construction proceeds through two auxiliary free products and an amalgamation. If $(X,d,e)$9, one obtains a two-sided invariant metric on $X^{-1}$0 extending $X^{-1}$1 and prescribing $X^{-1}$2. The same source also records the necessary boundedness condition

$X^{-1}$3

for any such extension (Slutsky, 2011).

The significance of the classical Graev construction lies in its maximality. Any two-sided invariant metric on the relevant free object that extends the initial metric is bounded above by the Graev metric. In that sense the construction is not merely canonical; it is extremal among all invariant extensions.

2. Ultrametric versions and non-Archimedean free objects

The ultrametric analogue replaces sums by maxima. For an ultrametric space $X^{-1}$4, one extends $X^{-1}$5 to

$X^{-1}$6

by the rules

$X^{-1}$7

If

$X^{-1}$8

have the same length, then

$X^{-1}$9

and the Graev ultrametric on $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$0 is defined by

$$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$1

Equivalently,

$$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$2

where $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$3 ranges over matches. This $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$4 is a two-sided invariant ultrametric extending the original $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$5 on $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$6 (Shlossberg, 2014).

The non-Archimedean maximality statement is completely parallel to the classical one: $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$7 is maximal among all two-sided invariant ultrametrics on $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$8 agreeing with $$d(x^{-1},y^{-1})=d(x,y), \qquad d(x,y^{-1})=d(x,e)+d(e,y),$$9 on $X\cup X^{-1}$0, and under the additional inverse-compatibility hypothesis it is maximal among those extending $X\cup X^{-1}$1 on $X\cup X^{-1}$2. The proof uses induction on word length together with the match formula, with the ultrametric inequality replacing the ordinary triangle inequality (Shlossberg, 2014).

For free products of ultrametric groups with a common closed subgroup $X\cup X^{-1}$3, the amalgam metric itself is ultrametric: $X\cup X^{-1}$4 For words $X\cup X^{-1}$5 of equal length, one defines

$X\cup X^{-1}$6

and then

$X\cup X^{-1}$7

The resulting $X\cup X^{-1}$8 is a two-sided invariant ultrametric on $X\cup X^{-1}$9 extending the ultrametric on $u,v$0. Non-degeneracy is proved through a sequence of reductions—multipliable pairs, evaluation forests, simple pairs, symmetrization, and reduced minimal-length pairs—organized by the combinatorics of a maximal evaluation forest (Slutsky, 2012).

The same paper extends the construction to HNN extensions of ultrametric groups. If $u,v$1 is two-sided invariant ultrametric, $u,v$2 are closed, $u,v$3 is $u,v$4-isometric, and $u,v$5, then the HNN extension

$u,v$6

admits a two-sided invariant ultrametric extending $u,v$7 and satisfying $u,v$8 (Slutsky, 2012).

These constructions connect directly with Polish-group theory. Using scaled spaces $u,v$9 and the associated free groups $n$0, one forms a Polish free product

$n$1

where $n$2 is the metric amalgam of the scaled spaces and $n$3 is the closed normal subgroup generated by the kernels of the quotient maps. The resulting Polish group contains topological copies of $n$4 and $n$5 and satisfies the stated universal property. Moreover, any two Polish groups $n$6 embed in a Polish group $n$7 so that the subgroup generated by them is naturally isomorphic to the abstract free product $n$8; if $n$9 admit two-sided invariant ultrametrics, then $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$0 can be chosen two-sided invariant ultrametric (Slutsky, 2012).

A further identification occurs when $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$1. In that case the Graev ultrametric $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$2 coincides with the Savchenko–Zarichnyi ultrametric $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$3 defined via ball partitions and induced quotient maps of free groups: $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$4 The same work shows that the free non-Archimedean balanced group on $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$5 is metrizable by $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$6 (Shlossberg, 2014).

3. Graev polytopes from symmetric ternary relations

In the combinatorial-convex setting, one starts with a finite set $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$7 and a symmetric ternary relation

$\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$8

where $\rho(u,v)=\sum_{i=1}^n d\bigl(u(i),v(i)\bigr).$9 denotes the unordered triple $f,g\in F(X)$0. For each triple in $f,g\in F(X)$1, one introduces the three vectors

$f,g\in F(X)$2

and defines the Graev polytope

$f,g\in F(X)$3

This is a lattice polytope contained in the affine hyperplane

$f,g\in F(X)$4

so $f,g\in F(X)$5 (Lavrov, 8 Sep 2025).

The dual-cone description is explicit. If $f,g\in F(X)$6 is a dual variable, then

$f,g\in F(X)$7

Each facet of $f,g\in F(X)$8 corresponds to an extremal ray of $f,g\in F(X)$9, equivalently to a selection of one tight inequality in each triple. More generally, faces are in bijection with collections of corners

$T$00

on which the corresponding inequalities are imposed as equalities (Lavrov, 8 Sep 2025).

The normalized volume $T$01 is taken with respect to the affine lattice $T$02, normalized so that a unimodular simplex has volume $T$03. Computed examples include

$T$04

For a root system $T$05, one writes $T$06. The paper states that in many cases closed-form product formulas are known, while for $T$07 with $T$08 the general formula remains open (Lavrov, 8 Sep 2025).

A basic example takes

$T$09

The six vertices are

$T$10

all lying in the hyperplane $T$11. The resulting polytope is combinatorially a hexagon of dimension $T$12, and its facets arise by selecting one tight inequality from each triple (Lavrov, 8 Sep 2025).

Two applications organize much of the theory. First, for a compact homogeneous space $T$13 whose isotropy representation splits into $T$14 pairwise non-equivalent irreducible summands, the Einstein equations reduce to Laurent polynomials whose Newton polytope is precisely $T$15. By the Bernstein–Kushnirenko theorem,

$T$16

where $T$17 is the number of isolated complex solutions up to homothety; when the inequality is strict, the missing solutions are accounted for by faces of $T$18 and corresponding Inönü–Wigner contractions (Lavrov, 8 Sep 2025).

Second, if $T$19 is an undirected graph and

$T$20

then the cosmological polytope $T$21 satisfies

$T$22

In particular, the facets of $T$23 are in bijection with connected subgraphs of $T$24. The paper describes this as a combinatorial unification: the Einstein-metric bound and the cosmological-polytope formalism are produced by the same underlying Graev construction (Lavrov, 8 Sep 2025).

4. Gelfand–Graev representations and Hecke-algebra realizations

For $T$25-adic groups, the Gelfand–Graev construction begins with a Borel subgroup $T$26 and a fixed nondegenerate Whittaker character $T$27. For

$T$28

the Gelfand–Graev representation is the compact induction

$T$29

with the right-translation action. It is the “universal $T$30-generic” representation: every irreducible $T$31-generic representation of $T$32 appears with multiplicity one as a $T$33-quotient of $T$34 (Bakic et al., 2020).

Bernstein theory places this object in a single Bernstein block $T$35. Choosing a pro-generator

$T$36

the functor

$T$37

is an equivalence $T$38. For the Gelfand–Graev representation, the corresponding $T$39-module is

$T$40

Heiermann’s presentation identifies $T$41 as an affine Hecke algebra of type $T$42 or, in special cases, $T$43, generated by $T$44 together with $T$45 subject to quadratic, braid, and cross-relations. The paper proves that $T$46 as an $T$47-module and that, in the unequal-parameter case,

$T$48

while in the zero-parameter affine $T$49 case,

$T$50

Thus the universal Whittaker model becomes an explicit induced Hecke module (Bakic et al., 2020).

An analogous program exists for central covers of $T$51. For an $T$52-fold Kazhdan–Patterson cover or Savin cover

$T$53

with $T$54, the unipotent radical splits uniquely. Fixing the standard nondegenerate character

$T$55

one forms

$T$56

For a simple type $T$57, the Hecke algebra $T$58 satisfies

$T$59

with $T$60 a commutative Laurent polynomial algebra and $T$61 a finite Hecke algebra of type $T$62. The associated module

$T$63

decomposes as

$T$64

For a genuine discrete series $T$65 of inertial class represented by $T$66, this yields the Whittaker-dimension formula

$T$67

and, in the Savin case,

$T$68

when $T$69 (Zou, 11 Feb 2025).

At pro-$T$70 Iwahori level, the same theme reappears for Brylinski–Deligne covers of split reductive groups. The genuine pro-$T$71 Iwahori–Hecke algebra $T$72 admits Iwahori–Matsumoto and Bernstein presentations, and for the genuine Gelfand–Graev module

$T$73

one has

$T$74

as right $T$75-modules. Passing to Iwahori-fixed vectors,

$T$76

where the summands are indexed by $T$77-orbits of torus characters, and for splitting orbits $T$78 one has

$T$79

with $T$80 a sign character of the parabolic Hecke subalgebra (Gao et al., 2022).

These results show that, in the local theory, the Gelfand–Graev construction is not only a source of Whittaker models but also a mechanism for converting generic representation theory into explicit Hecke-module combinatorics.

5. Finite reductive groups and generalized Gelfand–Graev representations

For finite reductive groups, the construction takes the form of Kawanaka’s generalized Gelfand–Graev representations. Let $T$81 be a connected reductive algebraic group over $T$82, let $T$83 be a Frobenius endomorphism, and write $T$84. Fix an $T$85-stable Borel subgroup $T$86 with unipotent radical $T$87. Under the assumptions used by Taylor, an $T$88-equivariant Springer isomorphism $T$89 and a $T$90-invariant form $T$91 lead, for $T$92, to the linear functional

$T$93

and the character

$T$94

After passing to the intermediate subgroup $T$95, one obtains

$T$96

equivalently

$T$97

Taylor proves a decomposition formula

$T$98

where each summand is expressed in terms of change-of-basis coefficients $T$99, Fourier-transform scalars $(X,d,e)$00, relative Weyl-group characters $(X,d,e)$01, and characteristic functions $(X,d,e)$02 of IC-extensions. From this decomposition, every irreducible character $(X,d,e)$03 has a unique maximal unipotent class $(X,d,e)$04 for which $(X,d,e)$05, giving uniqueness of wave-front sets when $(X,d,e)$06 is good (Taylor, 2014).

Geck studies what remains when the good-prime hypothesis is dropped. A weighted Dynkin diagram

$(X,d,e)$07

defines a cocharacter $(X,d,e)$08, a filtration

$(X,d,e)$09

and a candidate character built from a linear map $(X,d,e)$10. When $(X,d,e)$11, the alternating form

$(X,d,e)$12

must be nondegenerate, after which one chooses an $(X,d,e)$13-stable Lagrangian $(X,d,e)$14, lifts it to $(X,d,e)$15, extends the character, and defines

$(X,d,e)$16

If $(X,d,e)$17 is good, Kawanaka showed that every $(X,d,e)$18 works. In bad characteristic, Geck exhibits diagrams for which the nondegeneracy condition fails and formulates Conjecture 4.4: for every diagram $(X,d,e)$19 occurring as a unipotent support of some irreducible $(X,d,e)$20-character, there should exist a map $(X,d,e)$21 satisfying the required conditions. The same paper proposes an integral determinant criterion for special unipotent classes (Geck, 2018).

In type $(X,d,e)$22, the generalized Gelfand–Graev construction admits a fully combinatorial model. For $(X,d,e)$23 and a partition $(X,d,e)$24, Andrews–Thiem define a subgroup $(X,d,e)$25 from a centered Ferrers diagram and a linear character

$(X,d,e)$26

where the relation $(X,d,e)$27 comes from the set-partition $(X,d,e)$28 of the row-reading tableau. The induced module

$(X,d,e)$29

coincides, up to scalar, with Kawanaka’s $(X,d,e)$30. The character vanishes outside dominance-above classes,

$(X,d,e)$31

and under the characteristic map one has

$(X,d,e)$32

Using the expansion

$(X,d,e)$33

the paper deduces

$(X,d,e)$34

so the Kostka–Foulkes polynomials appear as multiplicities of unipotent characters (Andrews et al., 2015).

The finite theory also includes explicit endomorphism algebras. For $(X,d,e)$35 and $(X,d,e)$36, a refined Bruhat decomposition parametrizes intersections of the form

$(X,d,e)$37

by distinguished subexpressions and parameters. This leads to explicit formulas for the structure constants of

$(X,d,e)$38

for the Gelfand–Graev module $(X,d,e)$39, reducing them to $(X,d,e)$40-conditions, Gauss sums, and Kloosterman-type sums (Paolini et al., 2018).

6. Weyl-group actions, Hamiltonian reduction, and related analytic constructions

A different branch of the Graev tradition concerns actions of the Weyl group on algebras and symplectic varieties. In unpublished work described by Ginzburg–Kazhdan, for each simple reflection $(X,d,e)$41 one constructs an analytic intertwiner

$(X,d,e)$42

using oscillatory integrals and Fourier transforms; in rank $(X,d,e)$43 it is essentially the partial Fourier transform on $(X,d,e)$44. Analytic estimates show that the $(X,d,e)$45 satisfy the Coxeter relations, producing a $(X,d,e)$46-action on $(X,d,e)$47 by algebra automorphisms. The same paper gives a purely algebraic construction via quantum Hamiltonian reduction: $(X,d,e)$48 Since $(X,d,e)$49 acts on $(X,d,e)$50 and fixes $(X,d,e)$51, this isomorphism transports the Weyl-group action to $(X,d,e)$52. The quasi-classical counterpart is the Poisson isomorphism

$(X,d,e)$53

which yields a Poisson $(X,d,e)$54-action on $(X,d,e)$55 and hence an algebraic $(X,d,e)$56-action by symplectomorphisms on the affinization $(X,d,e)$57 (Ginzburg et al., 2018).

Wang gives a quiver-theoretic realization of this quasi-classical action. For a doubled quiver with dimension vectors $(X,d,e)$58, the DKS-type variety

$(X,d,e)$59

carries automorphisms $(X,d,e)$60 defined by correspondence varieties $(X,d,e)$61 and principal-bundle isomorphisms on invariant coordinate rings. When the dimension-balance condition

$(X,d,e)$62

holds for an ADE quiver, the simple reflections satisfy the Coxeter relations and generate a Weyl-group action on $(X,d,e)$63. In type $(X,d,e)$64, Wang proves that under the isomorphism

$(X,d,e)$65

the reflection $(X,d,e)$66 matches the quasi-classical Gelfand–Graev action of Ginzburg–Kazhdan; concretely it is given by conjugation by an explicit elementary block matrix $(X,d,e)$67 depending on the eigenvalue coordinates of $(X,d,e)$68 (Wang, 2019).

A separate analytic usage of the Graev name is the Gel'fand–Graev–Radon transform on Euclidean $(X,d,e)$69. For a bulk field $(X,d,e)$70, the transform integrates over horospheres in the embedding space, and under the conformal scaling ansatz on the light cone the inverse transform reconstructs the unique normalizable bulk solution: $(X,d,e)$71 The paper interprets this as an HKLL-type bulk reconstruction formula and relates loop corrections of bulk correlators to higher-point boundary correlators (Bhowmick et al., 2017).

Taken together, these developments show that “Graev construction” no longer denotes a single object. It denotes a family of canonical procedures—metric, ultrametric, polyhedral, Whittaker-theoretic, Hamiltonian, and integral-geometric—whose common feature is the rigid promotion of local data to a global object with strong universal, extremal, or symmetry properties.