Toric resolution of singularities

Intuition [Beginner]

A toric variety $X_{\Sigma }$ is smooth when every cone of the fan $\Sigma$ has its primitive rays forming part of a $\mathbb{Z}$-basis of the lattice $N$. When some cone fails this condition, the variety has a singular point at the corresponding torus orbit — the algebraic geometry is locally a quotient of affine space by a finite abelian group rather than affine space itself. The resolution problem asks: can we replace $X_{\Sigma }$ by a smooth variety $X_{{\Sigma }^{\prime }}$ together with a proper map $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ that is an isomorphism over the smooth locus? In the toric world the answer is direct and constructive: refine the fan.

To refine a fan, take each non-smooth cone and chop it into smaller pieces by adding new rays through its interior. After enough subdivisions, every cone is generated by part of a lattice basis. The resulting refined fan ${\Sigma }^{\prime }$ has the same support as $\Sigma$, so the variety it produces is proper over the original variety. Because every refinement step adds new lattice rays, the construction is finite, combinatorial, and explicit. The proper map $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ is the toric resolution.

For two-dimensional cones the algorithm has a particularly clean form. Every two-dimensional cone is equivalent, up to a lattice basis change, to a standard cone $\mathrm{Cone}(e_2,p\cdot e_1-q\cdot e_2)$ with $p>q>0$ and $p,q$ sharing no common factor. The Hirzebruch-Jung algorithm computes the negative continued-fraction expansion of $p/q$ — a sequence of integers $b_1,b_2,\dots ,b_s$ all at least $2$ — and the new rays of the refined fan are read off from this list. The exceptional set of the resolution becomes a chain of $s$ projective lines glued at points, each carrying self-intersection $-b_i$. This algorithm was already known to Hirzebruch in 1953.

Visual [Beginner]

A two-panel diagram of the toric resolution of the simplest cyclic-quotient surface singularity, the $A_1$ singularity ${\mathbb{C}}^2/{\mathbb{Z}}_2$. Left panel: the cone $\sigma =\mathrm{Cone}(e_2,2e_1-e_2)$ in $N_{\mathbb{R}}={\mathbb{R}}^2$, shaded as a wedge between the two rays. The wedge is non-smooth because the determinant of the matrix with columns $e_2$ and $2e_1-e_2$ is $-2$, not $\pm 1$.

Right panel: the refined cone, with the new ray $\rho =\mathrm{Cone}(e_1)$ added through the interior of $\sigma$. The wedge is now subdivided into two smaller wedges ${\sigma }_1=\mathrm{Cone}(e_2,e_1)$ (the first quadrant) and ${\sigma }_2=\mathrm{Cone}(e_1,2e_1-e_2)$, each with primitive-ray determinant $\pm 1$ and therefore smooth. The arrow from left to right is labelled "star subdivision along $e_1$". Beneath the right panel, a small picture of the resolved variety: a single ${\mathbb{P}}^1$ (the exceptional divisor) of self-intersection $-2$, contracting to the singular point under the morphism $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$.

The picture captures the key idea: resolving a toric singularity means inserting new lattice rays until every cone is smooth, and the exceptional set of the resolution is read off as the toric divisors corresponding to the new rays.

Worked example [Beginner]

Resolve the $A_1$ singularity $X={\mathbb{C}}^2/{\mathbb{Z}}_2$ by toric fan refinement. The group ${\mathbb{Z}}_2$ acts on ${\mathbb{C}}^2$ by $(x,y)\mapsto (-x,-y)$, so the quotient ring is $\mathbb{C}[x,y{]}^{{\mathbb{Z}}_2}=\mathbb{C}[x^2,xy,y^2]=\mathbb{C}[u,v,w]/(uv-w^2)$, a surface singularity at the origin.

Step 1. The fan. The quotient surface $X={\mathbb{C}}^2/{\mathbb{Z}}_2$ is the affine toric variety $U_{\sigma }$ for the cone $\sigma =\mathrm{Cone}(e_2,2e_1-e_2)$ in $N_{\mathbb{R}}={\mathbb{R}}^2$. The two primitive rays are $u_1=e_2=(0,1)$ and $u_2=2e_1-e_2=(2,-1)$. The determinant of the matrix with columns $u_1,u_2$ is $\det \left(\begin{array}{cc}0& 2\\ 1& -1\end{array}\right)=-2$. Since $|\det |=2\ne 1$, the cone is not smooth: the rays generate an index-$2$ sublattice of ${\mathbb{Z}}^2$, and $U_{\sigma }$ has a ${\mathbb{Z}}_2$ cyclic quotient singularity at the origin.

Step 2. Add a new ray. Choose the interior lattice point $v=e_1=(1,0)\in \sigma$. Star-subdivide $\sigma$ along $v$: the new fan ${\Sigma }^{\prime }$ has three rays $u_1=e_2$, $v=e_1$, $u_2=2e_1-e_2$ and two maximal cones ${\sigma }_1=\mathrm{Cone}(u_1,v)=\mathrm{Cone}(e_2,e_1)$ and ${\sigma }_2=\mathrm{Cone}(v,u_2)=\mathrm{Cone}(e_1,2e_1-e_2)$.

Step 3. Check smoothness. For ${\sigma }_1$: $\det \left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)=-1$, smooth. For ${\sigma }_2$: $\det \left(\begin{array}{cc}1& 2\\ 0& -1\end{array}\right)=-1$, smooth. Both new cones have primitive rays forming a $\mathbb{Z}$-basis of $N$, so the refined variety $X_{{\Sigma }^{\prime }}$ is smooth.

What this tells us. The toric morphism $f:X_{{\Sigma }^{\prime }}\to X_{\sigma }$ induced by the fan refinement is a resolution of singularities: $X_{{\Sigma }^{\prime }}$ is smooth, $f$ is proper (supports agree), and $f$ is an isomorphism over the dense torus. The exceptional set is the toric divisor $D_v$ corresponding to the new ray $v=e_1$, a copy of ${\mathbb{P}}^1$. Computing the self-intersection (from the relation $u_1+u_2=2v$, equivalent to $b_1=2$ in continued-fraction notation) gives $D_v\cdot D_v=-2$. This is the standard resolution of the $A_1$ singularity, with one exceptional ${\mathbb{P}}^1$ of self-intersection $-2$.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $N$ be a lattice of rank $n$ with $N_{\mathbb{R}}=N{\otimes }_{\mathbb{Z}}\mathbb{R}$, and let $\Sigma$ be a fan in $N_{\mathbb{R}}$ in the sense of [04.11.04].

Definition (smooth refinement). A fan ${\Sigma }^{\prime }$ in $N_{\mathbb{R}}$ is a refinement of $\Sigma$ if every cone of ${\Sigma }^{\prime }$ is contained in a cone of $\Sigma$ and the supports agree, $|{\Sigma }^{\prime }|=|\Sigma |$. The refinement is smooth if every cone of ${\Sigma }^{\prime }$ is generated by part of a $\mathbb{Z}$-basis of $N$.

Definition (toric resolution of singularities). A morphism $f:X^{\prime }\to X_{\Sigma }$ is a toric resolution of singularities of $X_{\Sigma }$ if:

(i) $X^{\prime }$ is a smooth $\mathbb{C}$-scheme;

(ii) $f$ is proper and birational;

(iii) $f$ is $T$-equivariant for the torus $T=N\otimes {\mathbb{C}}^{\ast }$ action on $X_{\Sigma }$ and on $X^{\prime }$;

(iv) $f$ restricts to an isomorphism over the smooth locus of $X_{\Sigma }$.

Equivalently, $f$ arises from a smooth refinement ${\Sigma }^{\prime }$ of $\Sigma$ together with the toric morphism $X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ associated to the identity on $N$.

Definition (star subdivision). Let $\Sigma$ be a fan and $v\in N$ a primitive lattice point lying in the relative interior of some cone $\tau \in \Sigma$ with $v$ not on any ray of $\Sigma$. The star subdivision of $\Sigma$ along $v$, denoted ${\Sigma }^{\ast }(v)$, is the fan obtained by:

(a) Removing every cone $\sigma$ of $\Sigma$ that contains $\tau$ as a face.

(b) Adding the new ray $\rho =\mathrm{Cone}(v)$ and every face of every newly constructed cone.

(c) For each removed $\sigma \supseteq \tau$, adding the cones $\mathrm{Cone}(v,\gamma )$ where $\gamma$ ranges over faces of $\sigma$ that do not contain $\tau$.

The support is preserved: $|{\Sigma }^{\ast }(v)|=|\Sigma |$. The new fan has the same cones as $\Sigma$ outside the star of $\tau$ and replaces the star of $\tau$ by the join of $v$ with the boundary of the star.

Definition (Hirzebruch-Jung continued fraction). For coprime integers $p>q>0$, the negative continued-fraction expansion of $p/q$ is the unique sequence of integers $b_1,b_2,\dots ,b_s$ with each $b_i\ge 2$ satisfying

$$\frac{p}{q}=b_1-\frac{1}{b_2-\frac{1}{b_3-\frac{1}{\ddots -\frac{1}{b_s}}}}.$$

The expansion is computed by the recursive algorithm $b_1=\lceil p/q\rceil$, then continue on $q/(b_{1}q-p)$. The recursion terminates because the denominator strictly decreases. The condition $b_i\ge 2$ is forced by $p/q>1$ at each step.

Counterexamples to common slips

• "Star subdivision along any lattice point gives a refinement." Not for points outside $|\Sigma |$, and not for points already on a ray of $\Sigma$ (which would either give back $\Sigma$ unchanged or introduce a redundant ray failing the fan axioms). The star subdivision is defined only for primitive lattice points $v$ in the relative interior of some non-ray cone.

• "Every smooth refinement gives a crepant resolution." False in general. Crepancy ($f^{\ast }{K}_X=K_{X^{\prime }}$, equivalently the discrepancy is zero) holds for a refinement iff every new ray lies in the affine hyperplane spanned by the primitive rays of the cone it subdivides. For toric Gorenstein quotient singularities ${\mathbb{C}}^n/G$ with $G\subset S{L}_n(\mathbb{C})$ finite, crepant resolutions correspond to specific lattice-point arrangements; the McKay correspondence makes them combinatorial when they exist (Bridgeland-King-Reid 2001).

• "The Hirzebruch-Jung algorithm gives the minimal resolution." For toric surface singularities the Hirzebruch-Jung resolution is indeed the minimal resolution — every $b_i\ge 2$ implies no $(-1)$-curves to contract. For higher-dimensional toric singularities, however, the analogous "iterated star subdivision" may overshoot the minimal model in the sense of the minimal-model program; the toric minimal model is a separate combinatorial question (the toric MMP, see Reid 1983 and Matsuki Introduction to the Mori Program 2002).

Key theorem with proof [Intermediate+]

The signature theorem of this unit is the toric resolution theorem: every toric variety admits a $T$-equivariant resolution of singularities by fan refinement, with the resolution algorithmic and explicit.

Theorem (toric resolution; Demazure 1970, KKMS 1973). Let $\Sigma$ be a fan in $N_{\mathbb{R}}$ for $N$ a lattice of rank $n$.

(a) Existence. There exists a smooth refinement ${\Sigma }^{\prime }$ of $\Sigma$ obtainable by finitely many iterated star subdivisions.

(b) Resolution morphism. The toric morphism $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ associated to the identity $N\to N$ is a proper birational $T$-equivariant resolution of singularities: $X_{{\Sigma }^{\prime }}$ is smooth, $f$ is proper, $f$ is an isomorphism over the smooth locus of $X_{\Sigma }$.

(c) Algorithmic description. For each non-smooth cone $\sigma$ of $\Sigma$, the iterated star subdivisions can be chosen as follows: subdivide along an interior lattice point $v\in \sigma \cap N$ that is not on any existing ray; repeat on the subdivided cones until every cone is smooth.

Proof.

(a) Existence. We construct ${\Sigma }^{\prime }$ by induction on the multiplicity of $\Sigma$, defined as $\mathrm{mult}(\Sigma )={\sum }_{\sigma \in \Sigma }\mathrm{mult}(\sigma )$, where $\mathrm{mult}(\sigma )$ for a top-dimensional cone $\sigma$ is the absolute value of the determinant of any matrix whose columns are the primitive ray generators of $\sigma$ (with $\mathrm{mult}(\sigma )=1$ iff $\sigma$ is smooth).

If $\mathrm{mult}(\Sigma )=|{\Sigma }_n|$ (the number of top-dimensional cones), every $\sigma$ has $\mathrm{mult}(\sigma )=1$, every cone is smooth, $\Sigma$ is already smooth, and we set ${\Sigma }^{\prime }=\Sigma$. Otherwise some cone ${\sigma }_0\in \Sigma$ has $\mathrm{mult}({\sigma }_0)>1$.

The cone ${\sigma }_0$ then contains an interior lattice point $v\in {\sigma }_0^{\circ }\cap N$ that is not on any ray of ${\sigma }_0$. (Proof: if all interior lattice points of ${\sigma }_0$ were on rays, then ${\sigma }_0\cap N$ would be the union of finitely many rays, contradicting the fact that ${\sigma }_0$ has $n$-dimensional interior over $\mathbb{R}$; in fact one can take $v$ to be the lattice point closest to the origin in the open cone, the primitive lattice generator of ${\sigma }_0\cap N$ in any direction off the rays.)

Star-subdivide $\Sigma$ along $v$ to obtain ${\Sigma }_1:={\Sigma }^{\ast }(v)$. The new cones of ${\Sigma }_1$ that subdivide ${\sigma }_0$ have multiplicities summing to $\mathrm{mult}({\sigma }_0)$, because the determinants of the new pairs of rays sum (via additivity of the determinant under cone subdivision; see Fulton §2.6 Lemma 2.6) to the original determinant. But each new cone has strictly smaller multiplicity than ${\sigma }_0$: if ${\sigma }_0=\mathrm{Cone}(u_1,\dots ,u_k)$ and $v\in {\sigma }_0^{\circ }$, then $v={\sum }_i{c}_i{u}_i$ with $c_i>0$, and the new cone $\mathrm{Cone}(v,u_2,\dots ,u_k)$ has multiplicity $c_1\cdot \mathrm{mult}({\sigma }_0)/\mathrm{mult}(u_2,\dots ,u_k)<\mathrm{mult}({\sigma }_0)$ since $c_1<1$ for an interior point.

Hence $\mathrm{mult}({\Sigma }_1)<\mathrm{mult}(\Sigma )$. By induction on the well-ordered $\mathbb{N}$-valued quantity $\mathrm{mult}(\Sigma )-|{\Sigma }_n|$, after finitely many star subdivisions every cone has multiplicity $1$ — that is, every cone is smooth — and we set ${\Sigma }^{\prime }:=$ the resulting fan.

(b) Resolution morphism. By construction, $|{\Sigma }^{\prime }|=|\Sigma |$, so the toric morphism $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ associated to the identity on $N$ is proper (Cox-Little-Schenck Theorem 3.4.11: the toric morphism is proper iff the preimage of every cone is supported in some cone, equivalent to support equality for refinements).

Smoothness of $X_{{\Sigma }^{\prime }}$ follows from the smoothness criterion (Demazure 1970, recorded in [04.11.04] Theorem (d)): $X_{{\Sigma }^{\prime }}$ is smooth iff every cone of ${\Sigma }^{\prime }$ is generated by part of a $\mathbb{Z}$-basis of $N$, which is exactly the smoothness condition we secured by induction.

Birationality follows because every fan contains the zero cone $\{0\}$ whose affine variety $U_{\{0\}}=T$ is the dense open torus; the refinement preserves the zero cone, so $X_{{\Sigma }^{\prime }}$ and $X_{\Sigma }$ share the dense open torus $T$, and $f$ restricts to the identity on $T$. Hence $f$ is dominant with the same function field, i.e., birational.

Equivariance is automatic for toric morphisms: the morphism $f$ is induced by the identity on $N$, hence on the dual lattice $M$, hence on the group rings $\mathbb{C}[M]\to \mathbb{C}[M]$ (themselves identical), which is $T$-equivariant.

Isomorphism over the smooth locus of $X_{\Sigma }$: a cone $\sigma$ of $\Sigma$ is smooth iff $U_{\sigma }$ is smooth (Demazure 1970). For a smooth cone $\sigma$, no star subdivision occurs inside $\sigma$ during the algorithm, so $\sigma$ remains a cone of ${\Sigma }^{\prime }$ and $f$ restricts to the identity on $U_{\sigma }$. Hence $f$ is an isomorphism over the open subscheme ${\bigcup }_{\sigma \text{ smooth}}{U}_{\sigma }$, which equals the smooth locus of $X_{\Sigma }$.

(c) Algorithmic description. The induction on $\mathrm{mult}(\Sigma )-|{\Sigma }_n|$ in part (a) is constructive: at each step, pick any interior lattice point $v$ of a non-smooth cone and subdivide. The choice is not unique, but every choice strictly decreases the multiplicity and hence terminates in finitely many steps. For two-dimensional cones the canonical choice is given by the Hirzebruch-Jung continued-fraction algorithm (see the Master tier below). For higher-dimensional cones, several canonical choices are studied: barycentric subdivision, weighted blow-up at the "deepest" lattice point, and the toric MMP / canonical model (Reid 1983). $\square$

Bridge. The toric resolution theorem builds toward the toric MMP of [04.11.13] and appears again in [04.11.10] as the algorithmic counterpart of the projectivity criterion, and the central insight is that fan refinement is the combinatorial form of resolution of singularities in characteristic zero — a statement that is non-constructive in general (Hironaka 1964) but becomes algorithmic and finitary in the toric setting. The foundational reason this is exactly the combinatorial form of Hironaka's theorem is that smoothness is local in the étale topology, every toric variety is locally a finite quotient of affine space by a finite abelian subgroup of $T$, and resolving a finite abelian quotient is equivalent to choosing a smooth refinement of the cone encoding the quotient. Putting these together identifies toric resolution with the cone-subdivision algorithm — and the bridge is dual to the duality $M\leftrightarrow N$ from [04.11.01]: where the dual cone ${\sigma }^{\vee }$ encodes the affine coordinate ring, the cone $\sigma$ itself encodes the subdivision data needed to resolve. The pattern generalises in two directions: to toroidal embeddings (KKMS 1973), where étale-local toric structure suffices for the resolution algorithm to apply; and to toric stacks (Borisov-Chen-Smith 2005), where the smooth-cone condition is relaxed to allow simplicial cones with the resolution producing a smooth Deligne-Mumford stack rather than a scheme.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib has scheme-gluing infrastructure and the algebraic-geometry primitives needed to formulate properness and birational morphisms, but lacks the toric-geometry stack on which the resolution theorem stands. The intended formalisation reads schematically:

import Mathlib.AlgebraicGeometry.OpenImmersion
import Mathlib.AlgebraicGeometry.Properness
import Mathlib.Data.Int.GCD

-- Assume `Fan N` and `X_Σ` from 04.11.04 are in scope, along with
-- the smoothness criterion `04.11.05`.

variable {N : Type*} [AddCommGroup N] [Module ℤ N] [Module.Free ℤ N] [Module.Finite ℤ N]

/-- A smooth refinement of a fan: every cone of `Σ'` sits inside some
cone of `Σ`, supports agree, and every cone of `Σ'` is smooth. -/
structure SmoothRefinement (Σ : Fan N) where
  refined : Fan N
  refines : refined.Refines Σ
  smooth : refined.IsSmooth

/-- **Toric resolution theorem (Demazure 1970; KKMS 1973).** Every fan
admits a smooth refinement. -/
theorem toric_resolution_exists (Σ : Fan N) : Nonempty (SmoothRefinement Σ) := by
  -- Construction: iterate star subdivisions on non-smooth cones until
  -- multiplicity equals number of top-dimensional cones (every cone smooth).
  sorry

/-- The Hirzebruch-Jung negative continued-fraction expansion of `p/q`
with `gcd(p, q) = 1` and `0 < q < p`. Returns the list of blocks `[b_1, …, b_s]`
with each `b_i ≥ 2`. -/
partial def hirzebruchJung (p q : ℕ) : List ℕ :=
  if q = 0 then [] else
    let b := (p + q - 1) / q  -- ⌈p / q⌉
    let r := b * q - p
    b :: hirzebruchJung q r

/-- For the `A_n` singularity `ℂ² / ℤ_{n+1}(1, n)`, the Hirzebruch-Jung
expansion is `[2, 2, …, 2]` of length `n`. -/
theorem hirzebruch_jung_An (n : ℕ) (hn : 0 < n) :
    hirzebruchJung (n + 1) n = List.replicate n 2 := by
  sorry

Each step is reachable from current Mathlib but requires substantive new development. The toric resolution theorem reduces, given the fan formalism of [04.11.04], to a multiplicity-induction argument that is finite and constructive; only the underlying cone-multiplicity machinery is missing from Mathlib. The Hirzebruch-Jung algorithm is purely number-theoretic and could be implemented today over Nat with a termination proof via strict-decrease of denominators.

Advanced results [Master]

Theorem (toric resolution; Demazure 1970, Kempf-Knudsen-Mumford-Saint-Donat 1973). Every fan $\Sigma$ in $N_{\mathbb{R}}$ admits a smooth refinement ${\Sigma }^{\prime }$ obtainable by finitely many iterated star subdivisions along interior lattice points. The induced toric morphism $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ is a proper birational $T$-equivariant resolution of singularities.

The toric resolution theorem is the algorithmic combinatorial form of Hironaka's 1964 resolution-of-singularities theorem over characteristic-zero fields. Hironaka 1964 (Annals of Mathematics 79, 109-326) proved the existence of resolution by a non-constructive Noetherian-induction argument; the toric setting was the first context where the resolution could be made fully algorithmic. The construction was first developed in detail by Demazure 1970 (Annales scientifiques de l'École normale supérieure (4) 3, 507-588) in the context of classifying toric compactifications of algebraic tori. The systematic scheme-theoretic treatment over arbitrary base schemes is in the Kempf-Knudsen-Mumford-Saint-Donat volume Toroidal Embeddings I (Springer LNM 339, 1973), where the algorithm is extended to the broader class of toroidal embeddings — varieties that look étale-locally like toric varieties.

Theorem (Hirzebruch-Jung resolution of two-dimensional cyclic quotient singularities; Hirzebruch 1953). Let $p>q>0$ be coprime integers and $\sigma =\mathrm{Cone}(e_2,p{e}_1-q{e}_2)\subset {\mathbb{R}}^2$. The affine toric variety $U_{\sigma }$ is the cyclic quotient singularity ${\mathbb{C}}^2/{\mathbb{Z}}_p(1,q)$. The Hirzebruch-Jung resolution is the toric refinement determined by the negative continued-fraction expansion $p/q=[b_1,\dots ,b_s{]}^{-}$. The exceptional divisor of the resolution is a chain of $s$ rational curves $E_1,\dots ,E_s\cong {\mathbb{P}}^1$ intersecting transversely in a chain, with self-intersections $E_i\cdot E_i=-b_i$ and adjacent intersections $E_i\cdot E_{i+1}=1$.

The continued-fraction algorithm was introduced by Friedrich Hirzebruch in Über vierdimensionale Riemannsche Flächen mehrdeutiger analytischer Funktionen von zwei komplexen Veränderlichen (Mathematische Annalen 126, 1953, pp. 1-22) ^{[Hirzebruch 1953]} as the explicit resolution of singularities of multi-valued analytic functions in two complex variables. The connection to Heinrich Jung's earlier work Darstellung der Funktionen eines algebraischen Körpers (1908) gives the name "Hirzebruch-Jung." The negative continued-fraction expansion (as opposed to the classical positive one) is forced by the orientation of the cone and the requirement that all blocks satisfy $b_i\ge 2$, which yields the chain of $(-2)$-or-deeper rational curves. The Hirzebruch-Jung resolution is the minimal resolution of the cyclic quotient singularity, in the sense that no further blow-downs of $(-1)$-curves are possible.

Theorem (resolution as toric blow-up; Reid 1983). Every star subdivision $\Sigma \to \Sigma^(v)$alonganinteriorlatticepoint$v$ofacone$\sigma$correspondstoa$T$-equivariantblow-up$X_{\Sigma^(v)} \to X_\Sigma$alongtheclosed$T$-orbit$O_\sigma$correspondingto$\sigma$.Conversely,every$T$-equivariantblow-upofatoricvarietyalonga$T$-invariant subvariety is a star subdivision.

The identification of star subdivisions with toric blow-ups was made precise by Miles Reid in Decomposition of toric morphisms (1983, in Arithmetic and Geometry, Volume II, ed. M. Artin and J. Tate, Birkhäuser, pp. 395-418) ^{[Reid 1980]}. This gives the toric Mori theory: every $T$-equivariant proper birational morphism between $\mathbb{Q}$-factorial toric varieties is a composition of star subdivisions and their inverses (toric blow-ups and blow-downs of toric divisors). The toric Mori program — Reid's 1983 paper plus Matsuki 2002 Introduction to the Mori Program — gives an explicit combinatorial algorithm for running the minimal-model program on toric varieties, with every step computable from the fan combinatorics.

Theorem (crepant resolution criterion). A toric resolution $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ is crepant (satisfies $f^ K_{X_\Sigma} = K_{X_{\Sigma'}}$inthesenseof$\mathbb{Q}$-Cartierdivisors)ifandonlyifeverynewray$\rho \in \Sigma'(1) \setminus \Sigma(1)$hasitsprimitivegenerator$v_\rho$lyingintheaffinehyperplane$H_\sigma \subset N_\mathbb{R}$spannedbytheprimitivegeneratorsofthecone$\sigma$of$\Sigma$thatcontains$\rho$ in its interior.*

The crepant-resolution criterion is the algebraic-geometric content of canonical vs terminal vs log terminal classification of singularities, due to Miles Reid in Canonical 3-folds (1980, Journées de géométrie algébrique d'Angers, Sijthoff and Noordhoff, pp. 273-310) ^{[Reid 1980]}. For toric Gorenstein singularities — those where the primitive ray generators of the cone lie on an affine hyperplane (the Gorenstein condition $K_{X_{\Sigma }}$ is Cartier) — crepant resolutions exist in dimensions $\le 3$ and are constructed explicitly via the $G$-Hilbert scheme (Ito-Nakamura 1996) or via lattice-polytope triangulations whose new vertices remain on the Gorenstein hyperplane.

Theorem (derived McKay correspondence; Bridgeland-King-Reid 2001). Let $G\subset S{L}_n(\mathbb{C})$ be a finite subgroup with ${\mathbb{C}}^n/G$ Gorenstein. For $n\le 3$, the $G$-Hilbert scheme $Y={\mathrm{Hilb}}^G({\mathbb{C}}^n)$ is a crepant resolution of ${\mathbb{C}}^n/G$. Moreover, the natural functor $$ \Phi : D^b(\mathrm{Coh}^G(\mathbb{C}^n)) \to D^b(\mathrm{Coh}(Y)) $$ from the bounded derived category of $G$-equivariant coherent sheaves on ${\mathbb{C}}^n$ to that of coherent sheaves on $Y$ is an equivalence of triangulated categories.

Tom Bridgeland, Alastair King, and Miles Reid 2001 (Journal of the American Mathematical Society 14, 535-554) ^{[Bridgeland-King-Reid 2001]} proved the derived McKay correspondence for finite $G\subset S{L}_n(\mathbb{C})$ in dimensions $n\le 3$. The result extends the classical McKay correspondence of John McKay 1980 — the bijection between irreducible representations of $G$ and irreducible components of the exceptional divisor of the minimal resolution of ${\mathbb{C}}^2/G$ — to higher dimensions and to derived equivalences. The proof uses Fourier-Mukai techniques and the fine moduli interpretation of ${\mathrm{Hilb}}^G$. For $n=2$ the result recovers Klein's 1884 classification of finite subgroups of $SU(2)$ (cyclic, binary dihedral, binary tetrahedral, binary octahedral, binary icosahedral) and the ADE-classification of rational double points via Brieskorn 1968 Inventiones Mathematicae 4 ^{[Brieskorn 1968]}.

Theorem (toric resolution in arbitrary characteristic; KKMS 1973). Let $S$ be an arbitrary base scheme. Every fan $\Sigma$ over $S$ admits a smooth refinement, and the resulting toric resolution is a $T$-equivariant proper birational morphism over $S$. In particular, the toric resolution theorem holds over fields of arbitrary characteristic, in contrast to the general resolution-of-singularities theorem (Hironaka 1964) which currently has full proofs only in characteristic zero.

The arbitrary-characteristic statement is in KKMS 1973 ^[pending] and was instrumental in the development of semistable reduction and toroidal models of algebraic varieties over local fields. The toric setting is the only known class where resolution of singularities is established algorithmically in positive characteristic; the general positive-characteristic resolution problem (open since Abhyankar 1956 for surfaces, Cossart-Piltant 2008-2014 for threefolds in positive characteristic) remains a frontier research question.

Theorem (Włodarczyk's combinatorial proof of resolution; Włodarczyk 2005). The general resolution-of-singularities theorem of Hironaka 1964 admits a constructive proof via systematic blow-ups of carefully chosen smooth centres, with the toric resolution algorithm serving as the foundational model and the "marked ideals" calculus extending the toric multiplicity-induction to general varieties.

Jarosław Włodarczyk's 2005 paper (Inventiones Mathematicae 162, 711-769) ^{[Włodarczyk 2005]} gives the most direct constructive proof of Hironaka's theorem to date, with the toric multiplicity-induction generalised to a "marked ideal" calculus that tracks order of vanishing along centres of blow-up. The Bierstone-Milman 1997 (Inventiones Mathematicae 128, 207-302) and Encinas-Villamayor 2003 (Revista Matemática Iberoamericana 19, 339-353) proofs of constructive resolution follow similar lines. The toric setting remains the clean special case in which the entire constructive resolution algorithm collapses to a finite combinatorial procedure on the fan.

Synthesis. The toric resolution theorem is the foundational reason that toric geometry serves as the canonical testbed for resolution of singularities — every general resolution algorithm (Hironaka 1964, Bierstone-Milman 1997, Encinas-Villamayor 2003, Włodarczyk 2005) has its core combinatorial logic isomorphic to the toric multiplicity-induction. The central insight is that smoothness of an affine toric variety is detected by a single integer invariant — the cone multiplicity, equivalently the index of the sublattice generated by primitive ray generators — and that smoothness is achieved by a sequence of star subdivisions that strictly decreases this invariant. The bridge is exactly the duality $M\leftrightarrow N$ from [04.11.01]: the dual lattice $M$ supplies the affine coordinate ring of each chart, while the cone $\sigma \subset N_{\mathbb{R}}$ supplies the subdivision data, and the index $|\det (u_1,\dots ,u_n)|=[N:\mathbb{Z}\{u_i\}]$ measures the deviation from smoothness.

Putting these together with the Hirzebruch-Jung algorithm for surfaces and the higher-dimensional iterated star subdivision, the toric resolution theorem identifies resolution of singularities with finite combinatorial subdivision in every dimension, generalises the projective-plane blow-up of classical algebraic geometry to arbitrary toric varieties, and is dual to the cone-to-affine-variety dictionary on which the entire toric framework rests. This pattern recurs in three downstream directions: the McKay correspondence relates crepant resolutions of ${\mathbb{C}}^n/G$ to the representation theory of $G$, with the toric algorithm producing crepant resolutions exactly when the affine-hyperplane condition holds; the toric Mori program (Reid 1983, Matsuki 2002) extends star subdivisions to a full minimal-model program on toric varieties; and the toroidal-embedding framework (KKMS 1973) lifts the toric resolution algorithm to varieties that look étale-locally toric — the natural setting for semistable reduction and degenerations of algebraic varieties.

The Hirzebruch-Jung algorithm itself is a microcosm of the broader theme: a single number-theoretic procedure (the negative continued fraction) encodes simultaneously the entire combinatorial structure of the resolution (the new rays, the multiplicity decrease per step) and the entire algebraic-geometric structure of the exceptional divisor (the chain of rational curves, their self-intersection numbers, their Dynkin-diagram intersection pattern). This is exactly the structural fact that organises every modern computation on cyclic-quotient surface singularities, including the Brieskorn-classification of ADE singularities, the McKay correspondence in dimension two, and the Coxeter-Dynkin combinatorics that bridges Lie theory and surface singularities.

Full proof set [Master]

Proposition (existence of smooth refinement). Every fan $\Sigma$ in $N_{\mathbb{R}}$ admits a smooth refinement ${\Sigma }^{\prime }$ obtainable by finitely many star subdivisions.

Proof. Define the total multiplicity of $\Sigma$ as $$ M(\Sigma) := \sum_{\sigma \in \Sigma_n} \mathrm{mult}(\sigma) = \sum_{\sigma \in \Sigma_n} \left|\det(u_1^{(\sigma)}, \ldots, u_n^{(\sigma)})\right|, $$ where ${\Sigma }_n$ is the set of top-dimensional cones and the determinant is taken over the primitive ray generators of $\sigma$. Smoothness of $\Sigma$ is equivalent to $M(\Sigma )=|{\Sigma }_n|$, i.e., every cone has multiplicity $1$.

Induction on $M(\Sigma )-|{\Sigma }_n|\ge 0$: if zero, $\Sigma$ is smooth and we are done. Otherwise some ${\sigma }_0\in {\Sigma }_n$ has $\mathrm{mult}({\sigma }_0)\ge 2$. By Exercise 4 above, there exists a primitive interior lattice point $v\in {\sigma }_0^{\circ }\cap N$ not on any ray of ${\sigma }_0$, and the star subdivision ${\Sigma }^{\ast }(v)$ replaces ${\sigma }_0$ by $n$ new top-dimensional cones ${\sigma }_1,\dots ,{\sigma }_n$ each of multiplicity strictly less than $\mathrm{mult}({\sigma }_0)$, with ${\sum }_i\mathrm{mult}({\sigma }_i)=\mathrm{mult}({\sigma }_0)$.

Hence $M({\Sigma }^{\ast }(v))-|{\Sigma }^{\ast }(v{)}_n|=M(\Sigma )-|{\Sigma }_n|-(n-1)\cdot \mathrm{mult}({\sigma }_0)+(\text{adjustment})$ — a strictly smaller non-negative integer because each new cone has strictly smaller multiplicity and the count of top-dimensional cones increases by $n-1$. Iterate; the induction terminates after finitely many steps, yielding a smooth refinement ${\Sigma }^{\prime }$. $\square$

Proposition (the toric resolution morphism is proper, birational, $T$-equivariant, and an isomorphism over the smooth locus). Given a smooth refinement ${\Sigma }^{\prime }$ of $\Sigma$, the toric morphism $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ associated to the identity on $N$ has all four properties of a resolution of singularities.

Proof. (i) Smoothness of $X_{{\Sigma }^{\prime }}$: by the smoothness criterion of [04.11.04] Theorem (d), $X_{{\Sigma }^{\prime }}$ is smooth iff every cone of ${\Sigma }^{\prime }$ has primitive ray generators forming part of a $\mathbb{Z}$-basis of $N$; this is exactly the condition we secured in the previous proposition.

(ii) Properness of $f$: by the criterion of Cox-Little-Schenck Theorem 3.4.11, $f$ is proper iff for every cone ${\sigma }^{\prime }\in {\Sigma }^{\prime }$, the cone ${\varphi }_{\mathbb{R}}({\sigma }^{\prime })\subseteq |\Sigma |$ where $\varphi ={\mathrm{id}}_N$ — equivalently ${\sigma }^{\prime }\subseteq$ some cone of $\Sigma$ — and the supports agree, $|{\Sigma }^{\prime }|=|\Sigma |$. Both conditions are satisfied by definition of refinement, so $f$ is proper.

(iii) Birationality of $f$: the zero cone $\{0\}\in \Sigma$ is preserved in ${\Sigma }^{\prime }$ (every refinement contains $\{0\}$ since $\{0\}$ is a face of every cone of ${\Sigma }^{\prime }$ that is a face of every cone of $\Sigma$). The chart $U_{\{0\}}=T$ is the dense open torus of both $X_{\Sigma }$ and $X_{{\Sigma }^{\prime }}$, and $f$ restricts to the identity $T\to T$ on this open dense subscheme. Hence $f$ is dominant with the same function field $\mathbb{C}(T)=\mathbb{C}(M)$, i.e., birational.

(iv) $T$-equivariance of $f$: the morphism $f$ is induced by the identity on $N$, which corresponds to the identity on $M$, hence to the identity on the group rings $\mathbb{C}[M]\to \mathbb{C}[M]$ that define the local rings of charts. The torus $T=\mathrm{Spec}\mathbb{C}[M]$ acts compatibly on both sides, so $f$ is $T$-equivariant.

(v) Isomorphism over the smooth locus: a cone $\sigma \in \Sigma$ is smooth iff the chart $U_{\sigma }$ is smooth. For a smooth $\sigma$, the resolution algorithm never subdivides $\sigma$ (no interior lattice point off the rays is needed), so $\sigma \in {\Sigma }^{\prime }$ and the restriction $f{|}_{U_{\sigma }}:U_{\sigma }\to U_{\sigma }$ is the identity. The smooth locus of $X_{\Sigma }$ is the union ${\bigcup }_{\sigma \text{ smooth}}{U}_{\sigma }$, and $f$ restricts to the identity there. $\square$

Proposition (Hirzebruch-Jung algorithm correctness). For coprime integers $p>q>0$, the negative continued-fraction expansion of $p/q$ gives the new rays of the smooth refinement of $\sigma =\mathrm{Cone}(e_2,p{e}_1-q{e}_2)$.

Proof. Define the rays recursively by ${\rho }_0=e_2$, ${\rho }_{s+1}=p{e}_1-q{e}_2$ for the final ray of the original cone, and ${\rho }_{i+1}=b_i{\rho }_i-{\rho }_{i-1}$ for $i=1,\dots ,s$ where $[b_1,\dots ,b_s{]}^{-}$ is the continued-fraction expansion of $p/q$. We claim:

(a) Each ${\rho }_i$ is a primitive lattice point in $\sigma$.

(b) The cones $\mathrm{Cone}({\rho }_{i-1},{\rho }_i)$ for $i=1,\dots ,s+1$ are smooth (primitive-ray determinants $\pm 1$).

(c) The cones together fill $\sigma$: ${\bigcup }_i\mathrm{Cone}({\rho }_{i-1},{\rho }_i)=\sigma$.

Proof of (a). By induction on $i$: ${\rho }_0=e_2$ and ${\rho }_1$ is primitive by construction (chosen to start the chain). Suppose ${\rho }_{i-1},{\rho }_i$ are primitive in $N$. Then ${\rho }_{i+1}=b_i{\rho }_i-{\rho }_{i-1}$ is a $\mathbb{Z}$-linear combination of primitive lattice vectors, hence in $N$. Primitivity of ${\rho }_{i+1}$ follows from $\gcd$ preservation in the continued fraction.

Proof of (b). The determinant $\det ({\rho }_{i-1},{\rho }_i)$ is preserved (up to sign) by the recursion: $\det ({\rho }_i,{\rho }_{i+1})=\det ({\rho }_i,b_i{\rho }_i-{\rho }_{i-1})=-\det ({\rho }_i,{\rho }_{i-1})=\det ({\rho }_{i-1},{\rho }_i)$. Hence by induction $\det ({\rho }_{i-1},{\rho }_i)=\det ({\rho }_0,{\rho }_1)=\pm 1$ if we choose ${\rho }_1$ correctly. The choice ${\rho }_1=$ the primitive lattice point with $\det (e_2,{\rho }_1)=1$ (closest to $e_2$ on the lattice line inside $\sigma$) makes the entire chain unimodular.

Proof of (c). The rays ${\rho }_i$ are constructed in counterclockwise (or clockwise, depending on orientation) order from ${\rho }_0=e_2$ to ${\rho }_{s+1}=p{e}_1-q{e}_2$, with each ${\rho }_{i+1}$ strictly between ${\rho }_i$ and ${\rho }_{s+1}$ in the angular order. The continued-fraction expansion $[b_1,\dots ,b_s{]}^{-}$ encodes exactly the angular progression needed to traverse from ${\rho }_0$ to ${\rho }_{s+1}$ in the smallest possible steps that keep every successive pair unimodular. Hence the cones $\mathrm{Cone}({\rho }_{i-1},{\rho }_i)$ tile $\sigma$. $\square$

Proposition (multiplicity-decrease at star subdivision). Let $\sigma =\mathrm{Cone}(u_1,\dots ,u_k)$ be a non-smooth top-dimensional cone with $\mathrm{mult}(\sigma )\ge 2$. For any interior lattice point $v={\sum }_i{r}_i{u}_i\in {\sigma }^{\circ }\cap N$ with $0<r_i<1$, the new cones ${\sigma }_i=\mathrm{Cone}(u_1,\dots ,\widehat{u_i},\dots ,u_k,v)$ obtained by star subdivision each satisfy $$ \mathrm{mult}(\sigma_i) = r_i \cdot \mathrm{mult}(\sigma) < \mathrm{mult}(\sigma). $$

Proof. By multilinearity of the determinant in its columns: replacing $u_i$ by $v={\sum }_j{r}_j{u}_j$ in the matrix $(u_1,\dots ,u_k)$ gives a new matrix whose determinant is

$$\det (u_1,\dots ,\widehat{u_i},\dots ,u_k,v)=\det (u_1,\dots ,\widehat{u_i},\dots ,u_k,\sum _j{r}_j{u}_j).$$

Expanding the linear combination, the only term that survives (all others have a repeated column among the $u_j$ for $j\ne i$ and hence vanish) is $r_i\cdot \det (u_1,\dots ,\widehat{u_i},\dots ,u_k,u_i)=r_i\cdot \det (u_1,\dots ,u_k)$ (after a sign-preserving cyclic permutation of columns). Hence $\mathrm{mult}({\sigma }_i)=r_i\cdot \mathrm{mult}(\sigma )$. Since $0<r_i<1$, the multiplicity strictly decreases. $\square$

Proposition (crepancy criterion for star subdivision). A star subdivision $\Sigma \to \Sigma^(v)$alongaprimitivelatticepoint$v$interiortoacone$\sigma$is\ast \ast crepant\ast \ast (preservesthecanonicalclassuptonumericalequivalence)ifandonlyif$v$liesontheaffinehyperplane$H_\sigma \subset N_\mathbb{R}$spannedbytheprimitivegeneratorsof$\sigma$.Specifically,writing$\sigma = \mathrm{Cone}(u_1, \ldots, u_k)$withtheGorensteincondition$\sum_i \langle \xi_\sigma, u_i\rangle = 1$forsome$\xi_\sigma \in M_\mathbb{Q}$,thenewray$v$iscrepantiff$\langle \xi_\sigma, v\rangle = 1$.*

Proof. The canonical class of $X_{\Sigma }$ as a toric variety is $K_{X_{\Sigma }}=-{\sum }_{\rho \in \Sigma (1)}{D}_{\rho }$, where the $D_{\rho }$ are the toric divisors. The pullback $f^{\ast }{K}_{X_{\Sigma }}=-{\sum }_{\rho }{f}^{\ast }{D}_{\rho }$ along the resolution $f:X_{{\Sigma }^{\ast }(v)}\to X_{\Sigma }$ is computed as follows: for $\rho$ unchanged in the refinement, $f^{\ast }{D}_{\rho }=D_{\rho }^{({\Sigma }^{\ast }(v))}$; for the new ray ${\rho }_v=\mathrm{Cone}(v)$, the divisor $D_{{\rho }_v}^{({\Sigma }^{\ast }(v))}$ has discrepancy $a_v$ relative to $K_{X_{\Sigma }}$, given by $a_v=⟨{\xi }_{\sigma },v⟩-1$ where ${\xi }_{\sigma }\in M_{\mathbb{Q}}$ is the unique class with $⟨{\xi }_{\sigma },u_i⟩=1$ for the rays of $\sigma$ (the Gorenstein condition).

Hence $K_{X_{{\Sigma }^{\ast }(v)}}=f^{\ast }{K}_{X_{\Sigma }}+a_v{D}_{{\rho }_v}^{({\Sigma }^{\ast }(v))}$, and crepancy ($a_v=0$) holds iff $⟨{\xi }_{\sigma },v⟩=1$, i.e., $v$ lies on $H_{\sigma }$. $\square$

Connections [Master]

• Fan and toric variety 04.11.04. The combinatorial framework on which toric resolution rests. The resolution algorithm takes a fan $\Sigma$ and produces a smooth refinement ${\Sigma }^{\prime }$ via iterated star subdivisions; the smoothness criterion from the prerequisite unit (every cone simplicial unimodular) is exactly the target condition. The induced morphism $X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ is the toric morphism of [04.11.04] Theorem (functoriality), restricted to the case where the underlying lattice map is the identity. The foundational reason fan refinement gives a resolution is the smoothness-iff-cones-smooth criterion proved in the sibling unit.

• Rational polyhedral cone and dual cone 04.11.02. The combinatorial primitives of star subdivision. Each new ray $v$ inserted by the algorithm is a primitive lattice point in the interior of a non-smooth cone; the face structure of the original cone determines which subcones of the star subdivision arise. The face-correspondence theorem $(\sigma \cap \tau {)}^{\vee }={\sigma }^{\vee }+{\tau }^{\vee }$ ensures that the refined fan satisfies the fan axioms (face-closure and intersection-as-face).

• Affine toric variety $U_{\sigma }$ 04.11.03. The local building block on which resolution operates chart-by-chart. The smoothness criterion of [04.11.03] Theorem (Demazure 1970) — $U_{\sigma }$ is smooth iff $\sigma$ is simplicial unimodular — is the local content of the resolution algorithm. Each cone of the refined fan produces a smooth affine chart, and the gluing of [04.11.04] assembles them into the smooth global resolution $X_{{\Sigma }^{\prime }}$.

• Resolution of singularities 04.06.02. The general scheme-theoretic resolution theorem of Hironaka 1964. The toric resolution is the algorithmic special case: where Hironaka's general proof is a non-constructive Noetherian-induction, the toric case becomes a finite procedure on cone multiplicities. The toric setting historically motivated much of the constructive-resolution literature (Bierstone-Milman 1997, Encinas-Villamayor 2003, Włodarczyk 2005), each generalising the toric multiplicity-induction to arbitrary varieties.

• Smoothness and completeness via fans 04.11.05. The sibling unit specifying the smoothness criterion for global toric varieties. The criterion — every cone simplicial unimodular — is the target of the resolution algorithm. The completeness criterion of the sibling unit is preserved by refinement (supports agree), so resolution of a complete toric variety produces a complete smooth toric variety.

• Toric blow-up and the orbit-cone correspondence 04.11.06. Star subdivision corresponds to $T$-equivariant blow-up of the closed $T$-orbit indexed by the cone being subdivided. The orbit-cone correspondence of the sibling unit identifies orbits with cones, and the resolution algorithm becomes a procedure on the orbit stratification: blow up the orbit corresponding to the deepest non-smooth cone, repeat until smooth.

• Polytope-fan correspondence 04.11.10. Projective toric resolutions correspond to regular subdivisions of the polytope $P$ defining the projective toric variety: $X_{\Sigma }$ projective iff $\Sigma$ is the normal fan of a lattice polytope $P$, and crepant resolutions of $X_{\Sigma }$ correspond to lattice triangulations of $P$ whose vertices are interior lattice points. The McKay correspondence in dimension $3$ identifies crepant resolutions with such triangulations and with representations of the acting group $G\subset S{L}_3(\mathbb{C})$.

• Toric divisor and support function 04.11.08. The sibling unit. Each star subdivision step in the resolution algorithm inserts a new ray $\rho$ and hence a new toric divisor $D_{\rho }$ — the exceptional divisor of the corresponding $T$-equivariant blow-up. The support function on the refined fan records the discrepancy data: the resolution is crepant iff the new support function values $\psi (u_{\rho })=-a_{\rho }$ for added rays fall on the boundary of the polytope $P_{K_X}$ associated to the canonical divisor, equivalently iff each added ray lies on $\{u:{\psi }_{K_X}(u)=-1\}$. The toric-divisor language of [04.11.08] is the bookkeeping framework for tracking discrepancies along a resolution sequence.

• Toric Picard group 04.11.09. The sibling unit. A toric resolution $f:X_{{\Sigma }^{\prime }}\to X_{\Sigma }$ induces a pullback on Picard groups $f^{\ast }:\mathrm{Pic}(X_{\Sigma })\to \mathrm{Pic}(X_{{\Sigma }^{\prime }})$, and the kernel and cokernel are computable from the fan-refinement data: each star subdivision adds one $\mathbb{Z}$-summand to the Picard group corresponding to the exceptional divisor class. The ample cone on $X_{{\Sigma }^{\prime }}$ is strictly smaller than the pullback of the ample cone on $X_{\Sigma }$; the wall-crossing picture organising the Picard-group transformations is the toric specialisation of variation-of-GIT-quotient and is fully combinatorial in the fan language of [04.11.09].

• Toric Mori program (Reid 1983) 04.11.13. The toric resolution theorem is the foundational input to the toric minimal model program: every $T$-equivariant proper birational morphism between $\mathbb{Q}$-factorial toric varieties is a composition of star subdivisions and their inverses (toric flips and divisorial contractions). The MMP-style classification of toric varieties into "minimal" (no $K$-negative extremal rays) and "Mori-fibre-space" (positive-dimensional Mori fibration) is fully combinatorial in the toric setting.

• Hironaka's theorem [04.06.02 / 04.10.03]. The general resolution-of-singularities theorem in characteristic zero. The toric resolution algorithm is the explicit combinatorial form of Hironaka's theorem when the variety is toric — and serves as the universally cited "toy example" in textbook expositions of general resolution (Kollár 2007 Lectures on Resolution of Singularities; Cutkosky 2004 Resolution of Singularities). The toric setting is also the only known case where resolution is established in arbitrary characteristic (KKMS 1973).

• McKay correspondence [03.03.15 / 04.10.21]. For a finite subgroup $G\subset S{L}_n(\mathbb{C})$ with ${\mathbb{C}}^n/G$ Gorenstein, the McKay correspondence relates crepant resolutions of ${\mathbb{C}}^n/G$ to the representation theory of $G$. In dimensions $\le 3$, the $G$-Hilbert scheme is the crepant resolution (Bridgeland-King-Reid 2001), and the irreducible exceptional divisors are indexed by non-identity irreducible representations of $G$ via the McKay graph of $G$. For $n=2$, this recovers Klein 1884 and Brieskorn 1968's ADE-classification of rational double points.

• Hirzebruch surfaces $F_a$ [04.11.04 Exercise 3]. Smooth complete toric surfaces with explicit fan presentations. The resolution algorithm reduces any toric surface to a smooth toric surface, and the smooth-toric-surface classification (Oda 1988 Chapter 2) identifies every smooth complete toric surface as either ${\mathbb{P}}^2$, ${\mathbb{P}}^1\times {\mathbb{P}}^1$, or a Hirzebruch surface $F_a$ — possibly further blown up at $T$-fixed points (giving the rational toric surfaces of Demazure 1970). The resolution algorithm provides the explicit blow-up sequence.

Historical & philosophical context [Master]

The toric resolution of singularities was first systematised by Michel Demazure in Sous-groupes algébriques de rang maximum du groupe de Cremona (Annales scientifiques de l'École normale supérieure (4) 3, 1970, pp. 507-588) ^{[Demazure 1970]}. Demazure's motivation was the classification of maximal algebraic subgroups of the Cremona group ${\mathrm{Cr}}_n=\mathrm{Bir}({\mathbb{P}}^n)$, and the toric resolution theorem arose as the explicit combinatorial mechanism for desingularising toric compactifications of algebraic tori. Demazure proved the smoothness criterion (every cone simplicial unimodular) and the resolution-by-subdivision algorithm in essentially modern form.

The continued-fraction algorithm for two-dimensional cyclic quotient singularities predates Demazure by nearly two decades: Friedrich Hirzebruch introduced the algorithm in Über vierdimensionale Riemannsche Flächen mehrdeutiger analytischer Funktionen von zwei komplexen Veränderlichen (Mathematische Annalen 126, 1953, pp. 1-22) ^{[Hirzebruch 1953]} as the explicit resolution of multi-valued analytic functions in two complex variables. The connection to Heinrich Jung's 1908 monograph Darstellung der Funktionen eines algebraischen Körpers zweier unabhängigen Veränderlichen $x$, $y$ in der Umgebung einer Stelle $x=a$, $y=b$ (Journal für die reine und angewandte Mathematik 133) gives the name "Hirzebruch-Jung." The algorithm is the canonical example of an algorithmic resolution that predates the general theorem of Hironaka by a decade.

Heisuke Hironaka's general resolution-of-singularities theorem in characteristic zero (Annals of Mathematics 79, 1964, pp. 109-326) ^{[Hironaka 1964]} was a watershed: every algebraic variety over a field of characteristic zero admits a resolution. Hironaka's proof is non-constructive — a Noetherian-induction over centres of blow-up — and the toric setting served as the canonical "toy model" where the algorithm was both explicit and finitary. Hironaka was awarded the Fields Medal in 1970 in part for this work.

The scheme-theoretic systematisation of toric resolution over arbitrary base schemes was carried out by George Kempf, Finn Faye Knudsen, David Mumford, and Bernard Saint-Donat in Toroidal Embeddings I (Lecture Notes in Mathematics 339, Springer 1973) ^[pending]. The KKMS volume extended the toric framework to toroidal embeddings — varieties that look étale-locally like toric varieties — and applied the resolution algorithm to semistable reduction of families over local fields, with applications to degenerations of Abelian varieties (Mumford 1972 Compositio Mathematica 24, 239-272). The KKMS approach is the foundation for all modern work on toroidal models in arithmetic geometry.

Miles Reid systematised the connection between toric resolution and the minimal model program in Canonical 3-folds (Journées de géométrie algébrique d'Angers, Sijthoff and Noordhoff 1980, pp. 273-310) ^{[Reid 1980]} and Decomposition of toric morphisms (Arithmetic and Geometry, Volume II, Birkhäuser 1983, pp. 395-418). Reid identified star subdivisions with toric blow-ups, introduced the crepant/discrepancy framework, and gave the first systematic toric implementation of the minimal model program. The Reid-Mori-Kawamata-Kollár-Shokurov logarithmic-MMP language has its cleanest exposition in the toric category.

The derived McKay correspondence of Tom Bridgeland, Alastair King, and Miles Reid (Journal of the American Mathematical Society 14, 2001, pp. 535-554) ^{[Bridgeland-King-Reid 2001]} established the bijection between irreducible representations of a finite $G\subset S{L}_n(\mathbb{C})$ and cohomology classes of the crepant resolution ${\mathrm{Hilb}}^G({\mathbb{C}}^n)$ in dimensions $n\le 3$, via a derived equivalence $D^b({\mathrm{Coh}}^G({\mathbb{C}}^n))\cong D^b(\mathrm{Coh}Y)$. The Ito-Nakamura construction of ${\mathrm{Hilb}}^G$ as the crepant resolution (Proceedings of the Japan Academy A 72, 1996, pp. 135-138) ^{[Ito-Nakamura 1996]} was the geometric prerequisite. For $n=2$ the result reproduces Klein 1884's classification of finite subgroups of $SU(2)$ (Vorlesungen über das Ikosaeder, Leipzig: Teubner) ^{[Klein 1884]} and Brieskorn 1968's ADE classification of rational double points (Inventiones Mathematicae 4, 336-358) ^{[Brieskorn 1968]}.

Jarosław Włodarczyk's 2005 paper Simple Hironaka resolution in characteristic zero (Inventiones Mathematicae 162, 711-769) ^{[Włodarczyk 2005]} is the most direct constructive proof of Hironaka's theorem, with the toric multiplicity-induction generalised to a marked-ideal calculus. The toric setting remains the foundational model in which the constructive resolution algorithm collapses to a finite combinatorial procedure on the fan. Edward Bierstone and Pierre Milman in Canonical desingularization in characteristic zero by blowing up the maximum strata of a local invariant (Inventiones Mathematicae 128, 1997, pp. 207-302) and Santiago Encinas and Orlando Villamayor in A new theorem of desingularization over fields of characteristic zero (Revista Matemática Iberoamericana 19, 2003, pp. 339-353) provide alternative constructive proofs along similar lines.

The toric setting remains the only known case where resolution of singularities is established algorithmically in positive characteristic; the general positive-characteristic problem remains an open frontier — Vincent Cossart and Olivier Piltant 2008-2014 (Journal of Algebra 320, 1051-1082 and Journal of Algebra 411, 419-457) established resolution for threefolds in positive characteristic, but higher-dimensional positive-characteristic resolution is open.

Bibliography [Master]

@article{Hirzebruch1953,
  author  = {Hirzebruch, Friedrich},
  title   = {{\"U}ber vierdimensionale {R}iemannsche {F}l{\"a}chen mehrdeutiger analytischer {F}unktionen von zwei komplexen {V}er{\"a}nderlichen},
  journal = {Mathematische Annalen},
  volume  = {126},
  year    = {1953},
  pages   = {1--22}
}

@article{Hironaka1964,
  author  = {Hironaka, Heisuke},
  title   = {Resolution of singularities of an algebraic variety over a field of characteristic zero},
  journal = {Annals of Mathematics},
  volume  = {79},
  year    = {1964},
  pages   = {109--326}
}

@article{Brieskorn1968,
  author  = {Brieskorn, Egbert},
  title   = {Rationale {S}ingularit{\"a}ten komplexer {F}l{\"a}chen},
  journal = {Inventiones Mathematicae},
  volume  = {4},
  year    = {1968},
  pages   = {336--358}
}

@article{Demazure1970,
  author  = {Demazure, Michel},
  title   = {Sous-groupes alg{\'e}briques de rang maximum du groupe de {C}remona},
  journal = {Annales scientifiques de l'{\'E}cole normale sup{\'e}rieure (4)},
  volume  = {3},
  year    = {1970},
  pages   = {507--588}
}

@article{Mumford1972,
  author  = {Mumford, David},
  title   = {An analytic construction of degenerating {A}belian varieties over complete rings},
  journal = {Compositio Mathematica},
  volume  = {24},
  year    = {1972},
  pages   = {239--272}
}

@book{KempfKnudsenMumfordSaintDonat1973,
  author    = {Kempf, George and Knudsen, Finn Faye and Mumford, David and Saint-Donat, Bernard},
  title     = {Toroidal Embeddings I},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {339},
  year      = {1973}
}

@incollection{Reid1980,
  author    = {Reid, Miles},
  title     = {Canonical 3-folds},
  booktitle = {Journ{\'e}es de g{\'e}om{\'e}trie alg{\'e}brique d'{A}ngers},
  editor    = {Beauville, Arnaud},
  publisher = {Sijthoff and Noordhoff},
  year      = {1980},
  pages     = {273--310}
}

@incollection{Reid1983,
  author    = {Reid, Miles},
  title     = {Decomposition of toric morphisms},
  booktitle = {Arithmetic and Geometry, Volume II},
  editor    = {Artin, Michael and Tate, John},
  publisher = {Birkh{\"a}user},
  year      = {1983},
  pages     = {395--418}
}

@book{FultonToric,
  author    = {Fulton, William},
  title     = {Introduction to Toric Varieties},
  publisher = {Princeton University Press},
  series    = {Annals of Mathematics Studies},
  volume    = {131},
  year      = {1993}
}

@book{CoxLittleSchenck,
  author    = {Cox, David A. and Little, John B. and Schenck, Henry K.},
  title     = {Toric Varieties},
  publisher = {American Mathematical Society},
  series    = {Graduate Studies in Mathematics},
  volume    = {124},
  year      = {2011}
}

@book{OdaConvexBodies,
  author    = {Oda, Tadao},
  title     = {Convex Bodies and Algebraic Geometry: An Introduction to the Theory of Toric Varieties},
  publisher = {Springer-Verlag},
  series    = {Ergebnisse der Mathematik und ihrer Grenzgebiete},
  volume    = {15},
  year      = {1988}
}

@article{ItoNakamura1996,
  author  = {Ito, Yukari and Nakamura, Iku},
  title   = {{M}c{K}ay correspondence and {H}ilbert schemes},
  journal = {Proceedings of the Japan Academy A},
  volume  = {72},
  year    = {1996},
  pages   = {135--138}
}

@article{BierstoneMilman1997,
  author  = {Bierstone, Edward and Milman, Pierre D.},
  title   = {Canonical desingularization in characteristic zero by blowing up the maximum strata of a local invariant},
  journal = {Inventiones Mathematicae},
  volume  = {128},
  year    = {1997},
  pages   = {207--302}
}

@article{BridgelandKingReid2001,
  author  = {Bridgeland, Tom and King, Alastair and Reid, Miles},
  title   = {The {M}c{K}ay correspondence as an equivalence of derived categories},
  journal = {Journal of the American Mathematical Society},
  volume  = {14},
  year    = {2001},
  pages   = {535--554}
}

@article{EncinasVillamayor2003,
  author  = {Encinas, Santiago and Villamayor, Orlando},
  title   = {A new theorem of desingularization over fields of characteristic zero},
  journal = {Revista Matem{\'a}tica Iberoamericana},
  volume  = {19},
  year    = {2003},
  pages   = {339--353}
}

@article{Wlodarczyk2005,
  author  = {W{\l}odarczyk, Jaros{\l}aw},
  title   = {Simple {H}ironaka resolution in characteristic zero},
  journal = {Inventiones Mathematicae},
  volume  = {162},
  year    = {2005},
  pages   = {711--769}
}

@book{Klein1884,
  author    = {Klein, Felix},
  title     = {Vorlesungen {\"u}ber das {I}kosaeder und die {A}ufl{\"o}sung der {G}leichungen vom f{\"u}nften {G}rade},
  publisher = {Teubner},
  address   = {Leipzig},
  year      = {1884}
}

@book{MatsukiMori,
  author    = {Matsuki, Kenji},
  title     = {Introduction to the {M}ori Program},
  publisher = {Springer-Verlag},
  series    = {Universitext},
  year      = {2002}
}

@book{KollarResolution,
  author    = {Koll{\'a}r, J{\'a}nos},
  title     = {Lectures on Resolution of Singularities},
  publisher = {Princeton University Press},
  series    = {Annals of Mathematics Studies},
  volume    = {166},
  year      = {2007}
}