--- title: Newton Stratification in Algebraic Varieties url: https://www.emergentmind.com/topics/newton-stratification type: topic --- # Newton Stratification in Algebraic Varieties Newton stratification is the decomposition of a space into locally closed loci indexed by Newton data: in group-theoretic terms by \(\sigma\)-conjugacy classes \([b]\in B(G)\) and their dominant Newton points \(\nu(b)\), and in the classical \(\GL_n\) or abelian-variety setting by Newton polygons. It occurs in loop groups, Iwahori double cosets, the reduction modulo \(p\) of Shimura varieties, moduli spaces such as \(\mathcal A_g\) and \(\mathcal M_g\), and the \(B_{\mathrm dR}^+\)-Grassmannian and \(p\)-adic flag varieties. Across these settings, the Newton map is paired with the Kottwitz invariant, and the strata are organized by a dominance partial order on rational coweights or on polygons [1511.03156]. ## 1. Basic definitions and index sets For a connected reductive group \(G\) over a local field, the fundamental index set is \[ B(G)=\{[b]\}, \] the set of \(\sigma\)-conjugacy classes in \(G(\breve F)\). Kottwitz classifies \(B(G)\) by two invariants: the dominant Newton point \(\nu(b)\) and the Kottwitz point \(\kappa(b)\). In the formulation used for Shimura varieties of Hodge type, one has \[ \nu=\nu_G:B(G)\longrightarrow X_\ast(T)_{\Q}^{\dom}/W\simeq N(G),\qquad \kappa=\kappa_G:B(G)\longrightarrow \pi_1(G)_\Gamma, \] and the partial order is \[ [b']\le [b]\quad\Longleftrightarrow\quad \nu([b'])\le \nu([b])\ \text{and}\ \kappa([b'])=\kappa([b]) . \] For every dominant \(\mu\), the subset \(B(G,\mu)\) is finite [1511.03156]. In the abelian-variety setting, the same data are encoded by Newton polygons. If \(\mathcal A\) is a principally polarized abelian variety of dimension \(g\), its \(p\)-divisible group \(\mathcal A[p^\infty]\) determines a convex piecewise-linear function \[ f:[0,2g]\to[0,g], \] with \(f(0)=0\), \(f(2g)=g\), linear slope sequence \(m_1,\dots,m_{2g}\), and symmetry \(f(2g-x)=g-f(x)\). The slopes are precisely the Newton slope multiset of \(\mathcal A[p^\infty]\) [2006.04927]. Different ambient spaces package the same Newton data differently. | Setting | Strata indexed by | Ambient space | |---|---|---| | Iwahori double coset | \([b]\in B(G)_x\) | \(IxI\subset G(\breve F)\) | | Shimura variety special fiber | \([b]\in B(G,\mu)\) | \(S=\mathscr S_K(G,X)\otimes\overline{\F}_p\) | | Moduli of ppav | Newton polygon \(P\) | \(\mathcal A_g\) | | \(B_{\mathrm dR}^+\)-Grassmannian / flag variety | \(b'\in B(G)\) | \(Gr_{G,\mu}\), \(\mathcal F(G,\mu)\) | This common formalism makes it possible to compare apparently different geometric situations through the same partially ordered Newton data. ## 2. Loop groups, affine Bruhat cells, and Iwahori-level Newton strata In the loop-group setting one fixes a split connected reductive group \(G\) over \(\F_q\), a split maximal torus \(T\subset G\), and a Borel \(B\supset T\). The loop group and positive loops are \[ LG(R)=G(R((z))),\qquad K(R)=G(R[[z]]) . \] For \(b\in LG(S)\) over an \(\F_q\)-scheme \(S\), the Newton map \[ \nu_b:S(\bar k)\to X_\ast(T)_\Q^+ \] defines locally closed Newton strata \[ N_\nu(b)=\{x\in S\mid \nu_b(x)=\nu\}, \] and the loci \(N_{\le \nu}(b)=\bigcup_{\nu'\le \nu}N_{\nu'}(b)\) are closed by the Grothendieck–Rapoport–Richartz specialization theorem. A stronger purity statement isolates a single break point: if \(S\) is integral and \(j\) is a break index of the generic Newton point \(\nu\), then \[ U_j:=\{x\in S\mid \operatorname{pr}_{(j)}(\nu_b(x))=\operatorname{pr}_{(j)}(\nu)\} \] is affine, so \(S\setminus U_j\) is either empty or pure of codimension \(1\) [1005.5725]. At Iwahori level one fixes a \(\sigma\)-stable alcove and corresponding Iwahori subgroup \(I\subset G(\breve F)\). The affine Bruhat decomposition is \[ G(\breve F)=\bigsqcup_{x\in \widetilde W}IxI . \] For fixed \(x\in\widetilde W\), the Newton stratification of the double coset is \[ IxI=\bigsqcup_{[b]\in B(G)_x}\mathcal N_{[b],x},\qquad \mathcal N_{[b],x}=[b]\cap IxI . \] Inside the finite set \(B(G)_x\) there is a unique maximal class \([b_x]\), whose Newton point \(\nu_x\) is the generic Newton point. Milićević and Viehmann isolate a condition called cordiality: \[ \ell(x)-\ell(\eta(x))=\langle 2\rho,\nu_x\rangle-\mathrm{def}(b_x). \] If \(x\) is cordial, then \(B(G)_x\) is saturated in \(B(G)\), every Newton stratum in \(IxI\) is equidimensional, and \[ \overline{\mathcal N_{[b],x}} = \bigcup_{\substack{[b']\in B(G)_x\\ [b']\le [b]}}\mathcal N_{[b'],x}. \] Under superregularity, cordiality can be checked by shortest paths in the quantum Bruhat graph [1902.02415]. The Iwahori case is not governed uniformly by these favorable properties. Trentinn and Viehmann prove that there exist Newton strata whose closures cannot be expressed as a union of strata, and that this failure is implied by non-equidimensional affine Deligne–Lusztig varieties. They also give an explicit example for a group of type \(A_4\) [2106.07284]. A common misconception is therefore that Grothendieck-style closure relations extend verbatim from hyperspecial or parahoric level to arbitrary Iwahori double cosets; the Iwahori case admits genuine counterexamples. ## 3. Levi subgroups and \((J,w,\delta)\)-alcove elements A particularly sharp comparison theorem concerns Newton strata in \(IxI\) when \(x\) is a normalized \((J,w,\delta)\)-alcove element. Fix a set of simple relative roots \(\Delta\), a subset \(J\subset \Delta\) with \(\delta(J)=J\), and \(w\in W\). For the standard Levi subgroup \(M=M_J\), an element \(x\in \widetilde W\) is a \((J,w,\delta)\)-alcove element if \[ \tilde x=w^{-1}x\delta(w)\in \widetilde W_M \] and, for every positive root \(\alpha\in \Phi^+\setminus \Phi_J\), \[ U_{w\alpha}\cap xIx^{-1}\subseteq U_{w\alpha}\cap I . \] When \(G\) is split and \(\delta=\mathrm{id}\), these are the classical \(P\)-alcove elements [2305.00683]. For normalized \(x\), the main result is a canonical bijection \[ B(M)_{\tilde x}\longrightarrow B(G)_x \] sending \([b]_M\) to the unique \([b]_G\) with \([b]_M\subseteq [b]_G\). Moreover, \[ \nu_M(b)=\nu_G(b) \quad\text{in }X_\ast(T)_{\Gamma_0}\otimes \Q, \] so the Newton stratifications on \[ (I\cap M)\tilde x(I\cap M)\qquad\text{and}\qquad IxI \] coincide under the identification. Equivalently, one obtains a canonical isomorphism of affine Deligne–Lusztig varieties \[ X_{\tilde x}^M(b)\xrightarrow{\ \simeq\ }X_x^G(b) . \] The proof proceeds first by a minimal-length argument in the \(\delta\)-conjugacy class, where the Newton point is read off from \[ x\,\delta(x)\cdots \delta^{n-1}(x)=\mu, \] and then by induction using the Deligne–Lusztig reduction sequence \[ x\xrightarrow{s}s\,x\,\delta(s) \] to lower length [2305.00683]. The theorem sharpens earlier containment statements: for normalized \((J,w,\delta)\)-alcove elements, the Newton stratification of the Iwahori double coset in \(G\) is identical to that of the corresponding double coset in the Levi subgroup. In \(\GL_n\), if \(J\) corresponds to a Levi of type \(\GL_{n_1}\times \GL_{n_2}\) and the Newton point has two slopes \(\lambda_1>\lambda_2\) with multiplicities \(n_1,n_2\), then \(\tilde x\) is the block-diagonal translation by \((\lambda_1,\dots,\lambda_1,\lambda_2,\dots,\lambda_2)\), and different refinements of slopes inside a block do not create new strata in the larger group [2305.00683]. ## 4. Shimura varieties, central leaves, and almost-product structures For a Shimura variety of Hodge type with hyperspecial level structure at \(p\), let \[ S=\mathscr S_K(G,X)\otimes_{O_{E,(p)}}\overline{\F}_p . \] Each geometric point \(x\in S(k)\) determines a \(p\)-divisible group with \(G\)-structure and hence a class \([b_x]\in B(G)\). If \(\mu\) is the Hodge cocharacter, the admissible set is \[ B(G,\mu)=\{[b]\in B(G)\mid \kappa([b])=\mu^\natural,\ \nu([b])\le \mu\}, \] and Mazur’s inequality asserts \([b_x]\in B(G,\mu)\). This yields the Newton stratification \[ S=\bigsqcup_{[b]\in B(G,\mu)}S_{[b]}, \qquad S_{[b]}(k)=\{x\in S(k)\mid [b_x]=[b]\}, \] with closed unions \[ S_{\le [b]}=\bigcup_{[b']\le [b]}S_{[b']} . \] The set \(B(G,\mu)\) is finite and totally ordered from the basic class to the ordinary class \([\mu(p)]\) [1511.03156]. The geometric structure of these strata is especially rigid. Hamacher’s dimension theorem gives \[ \dim S_{[b]}=\langle \rho,\mu+\nu([b])\rangle-\tfrac12\,\mathrm{def}(b), \] equivalently \[ \operatorname{codim}_S(S_{[b]})=\langle \rho,\mu-\nu([b])\rangle+\tfrac12\,\mathrm{def}(b). \] In the PEL context, the Newton stratification satisfies strong purity and exact closure: \[ \overline{S_{[b]}} = \bigcup_{[b']\le [b]}S_{[b']} . \] The same dimension formula and closure statement hold for Newton strata in the universal deformation space of a Barsotti–Tate group with PEL structure [1312.0490; 1511.03156]. A finer internal decomposition is given by central leaves. Fix a completely slope-divisible Barsotti–Tate group with crystalline Tate tensors \((X_0,\lambda_0,t_0)\) in class \([b]\). The central leaf \[ C(X_0,t_0)=\{x\in \mathscr S_0^b(\overline{\F}_p)\mid (A_x[p^\infty],t_x)\cong (X_0,t_0)\} \] is a closed smooth subscheme of \(\mathscr S_0^b\) of dimension \[ \dim C=\langle 2\rho,\bar\nu(b)\rangle . \] Let \(\mathcal M_{G,b}\) be the Hodge-type Rapoport–Zink space and \(\Ig_\infty\) the infinite-level Igusa tower over \(C\). Hamacher proves that there is a canonical diagram \[ \Ig_\infty\times_{\overline{\F}_p}\mathcal M_{G,b}\xrightarrow{\ \pi_\infty\ }\mathscr S_0^b{}^{\mathrm{perf}}, \] where \(\pi_\infty\) is a surjective \(J_b\)-equivariant pro-étale morphism and exhibits \(\Ig_\infty\times\mathcal M_{G,b}\) as a \(J_b\)-torsor over \(\mathscr S_0^b{}^{\mathrm{perf}}\). After a pro-étale trivialization one gets the almost-product isomorphism \[ \mathscr S_0^b{}^{\mathrm{perf}}\cong \Ig_\infty\times_{\overline{\F}_p}\mathcal M_{G,b} \] [1605.05540]. ## 5. Abelian varieties, Newton polygons, and curves On the moduli space \(\mathcal A_g\) of principally polarized abelian varieties of dimension \(g\), Newton stratification is formulated directly in terms of polygons. For a fixed Newton polygon \(P\) of height \(2g\) with integer breakpoints, one defines \[ W_P=\{x\in \mathcal A_g: NP_x\succeq P\},\qquad W_P^0=\{x\in \mathcal A_g: NP_x=P\}. \] Then \(W_P\) is closed, \(W_P^0\) is open in \(W_P\), and if \[ \Omega(P)=\{(x,y)\in \Z^2:0\le x\le g,\ y2\) and \[ \frac{2g}{3}-\frac{2p(p-1)}{3}\ge 2k(p-1), \] then there exists a smooth proper curve \(C_{g,k}\) of genus \(g\) in characteristic \(p\) with \[ NP(C_{g,k})=\{0,1\}^{g-k(p-1)}\sqcup \{\tfrac12\}^{2k(p-1)}. \] The same paper constructs families \(C_g\) whose scaled Newton polygons satisfy \[ sNP(C_g)\succeq P(x^2/4,2), \] so asymptotically the polygons lie above the parabola \(y=x^2/4\) in the unit square [2006.04927]. For curves themselves, Newton strata refine the \(p\)-rank stratification of \(\mathcal M_g\). If \(M_g[\nu]\) denotes the locus of smooth curves whose Jacobian has Newton polygon \(\nu\), then \[ M_g^f=\bigsqcup_{\nu:f(\nu)=f}M_g[\nu]. \] Achter and Pries prove that for every prime \(p\) and every genus \(g\ge 4\), there exists a smooth projective curve whose Jacobian has Newton polygon \[ \{\,0^{\oplus(g-4)},\tfrac14^{\oplus 2},\tfrac34^{\oplus 2},1^{\oplus(g-4)}\,\}, \] and they formulate generic polygons \(\nu_g^f\) for the \(p\)-rank \(f\) strata, together with inductive results on the generic Newton polygon in fixed \(p\)-rank [1311.5846]. Recent explicit classifications show how Newton strata intersect Ekedahl–Oort strata in small dimensions. For the \(\mathsf{GU}(3,2)\) Shimura variety, exactly four Newton polygons occur, with the supersingular, intermediate, \(p\)-rank \(2\), and \(\mu\)-ordinary strata of dimensions \(3,4,5,6\), respectively, together with explicit closure relations and a complete list of which Ekedahl–Oort strata intersect which Newton strata [2510.01090]. For \(\mathcal A_5\), the Newton polygons are listed by \(p\)-rank, and the intersections \(N_\nu\cap S_\varphi\) with the \(32\) Ekedahl–Oort types are described using Oort’s minimality, first-slope criteria, and direct-sum constructions [2509.19878]. ## 6. \(p\)-adic period domains and the \(B_{\mathrm dR}^+\)-Grassmannian On the Fargues–Fontaine side, the \(B_{\mathrm dR}^+\)-Grassmannian \(Gr_G\) classifies \(G\)-torsors on \(\Spec B_{\mathrm dR}^+\) together with a trivialization over \(B_{\mathrm dR}\). For \(C\) algebraically closed complete, one has \[ Gr_G(C)=G(B_{\mathrm dR}(C))/G(B_{\mathrm dR}^+(C)). \] Via Beauville–Laszlo gluing, a point \(x\in Gr_{G,\mu}(C)\) modifies the \(G\)-bundle \(\mathcal E_b\) corresponding to \(b\in B(G)\), producing a new class \([b']\in B(G)\). This defines Newton strata \[ Gr_{G,\mu,b}^{[b']}=\{x\in Gr_{G,\mu}\mid [\mathcal E_{b,x}]=[b']\}. \] Viehmann proves that under the identification \(|Bun_G|\simeq B(G)\), the closure relations on \(|Bun_G|\) coincide with the opposite of the usual partial order on \(B(G)\), and also proves Chen’s conjecture that every non-Hodge–Newton decomposable Newton stratum in a minuscule affine Schubert cell intersects the weakly admissible locus [2101.07510]. For \(p\)-adic flag varieties \(\mathcal F(G,\mu)\) with \(b\) basic and \(\mu\) minuscule, there is again a Newton stratification. The admissible locus is \[ \mathcal F(G,\mu,b)^a=\{x\mid \mathcal E_{b,x}\text{ is the trivial }G\text{-bundle on }X\}, \] and one has \[ \mathcal F(G,\mu,b)^a=\mathcal F(G,\mu,b)[1], \] the unique open Newton stratum. Shen proves that the following are equivalent: the weakly admissible locus is maximal as a union of Newton strata; the pair \((G,\mu)\) is weakly fully Hodge–Newton decomposable; and the Newton stratification is finer than the Harder–Narasimhan stratification [2203.12293]. For \(\GL_n\), Hong classifies the nonempty Newton strata in a minuscule Schubert cell. If the Harder–Narasimhan polygon of \(b\) has the property that any two distinct slopes differ by \(>1\), then \[ Gr_{\GL_n,\mu,b}^{b'}\neq \emptyset \] if and only if the Newton polygons satisfy Mazur-inequality and slope-wise dominance, \[ \nu(b)+\mu\ge \nu(b'),\qquad \nu(b')+id_{[0,n]}\ge \nu(b)\ge \nu(b'), \] together with common break-points: every break-point abscissa of \(\nu(b)\) is also a break-point of \(\nu(b')\) [2209.08374]. An analogous explicit classification is proved for \(\GSp_{2n}\): under the same slope-gap hypothesis, nonemptiness is characterized by polygon inequalities and breakpoint matching [2211.05807]. ## 7. Purity, closure, and the geometry of the Newton poset Purity and closure are the recurrent structural questions of Newton stratification. In moduli of Hodge-type or PEL-type Shimura varieties, one has strong purity and exact closure relations. In \(\mathcal A_g\), the closure order is precisely the order by lying above on Newton polygons. In the loop-group setting, single break-point purity gives codimension-one control on the jumping of individual Newton coordinates [1005.5725; 1511.03156; 2006.04927]. The poset-theoretic behavior can nevertheless change dramatically with level structure. In Iwahori double cosets, cordiality implies saturation of the Newton poset \(B(G)_x\), equidimensionality of Newton strata, and Grothendieck-style closure. Without cordiality, some closures are not unions of Newton strata, and non-equidimensional affine Deligne–Lusztig varieties provide the mechanism for this failure [1902.02415; 2106.07284]. The comparison theorem for normalized \((J,w,\delta)\)-alcove elements shows that this pathology is not universal even at Iwahori level. For those elements, the Newton stratification in \(G\) is canonically identical to the corresponding Newton stratification in a Levi subgroup, and the associated affine Deligne–Lusztig varieties are canonically isomorphic [2305.00683]. This suggests that Hodge–Newton decomposability, Levi reduction, and the geometry of the Newton poset are tightly linked, but the ambient level structure remains decisive for whether closure relations behave as in the hyperspecial case. Source: https://www.emergentmind.com/topics/newton-stratification