This is a standalone ASCII explainer (for 16-year-olds) of the Kollar-Pardon classification of semialgebraic universal covers: a connected normal projective complex variety X whose ordinary universal cover X~ is biholomorphic to a semialgebraic open subset of a projective variety must have X~ = D x C^m x F, where D is a bounded symmetric domain, F is a simply connected normal projective variety, and m >= 0. The source is the LaTeX article in this directory (main.tex plus sections/*.tex, concatenated into semialgebraic_universal_covers_original.txt). All mathematical content and citations are transcribed; each exercise names the original theorem number at the end for cross-checking.
Exercise: Universal cover. What is the universal cover of a connected space, and why does it matter here?
For a connected space X with a basepoint, the universal cover X~ is a simply connected covering space of X. Its fibre over the basepoint is naturally identified with the fundamental group pi_1(X), and the deck transformations of X~ -> X form a group isomorphic to pi_1(X). When X is a connected normal complex variety, X~ inherits a complex structure, and the covering map is a local biholomorphism. The whole question of this paper is: what can X~ look like as a complex space, if X is a connected normal projective variety?

Exercise: Complex projective variety. What are projective varieties, and where do the standard examples come from?
Complex projective space P^n is the set of complex lines through the origin of C^(n+1); equivalently, the quotient of C^(n+1) minus 0 by scaling. It is a compact complex manifold with a natural Fubini-Study Kahler metric. A projective variety is the common zero locus of finitely many homogeneous polynomials on P^n. A projective variety is normal when its ring of local holomorphic functions is integrally closed in its fraction field. Smooth projective varieties, Fano varieties, and abelian varieties are all normal projective; the open unit disc D subset C is not projective, since it is not compact.

Exercise: Semialgebraic set. What data defines a real semialgebraic set, and what closure properties does it have?
A subset of R^n is semialgebraic if it is a finite Boolean combination of sets {x : f(x) = 0} and {x : g(x) > 0}, where f and g are real polynomials. The collection is closed under images under polynomial maps (this is the Tarski-Seidenberg theorem), so projections of semialgebraic sets are semialgebraic. Every semialgebraic set has finitely many connected components, each again semialgebraic, and admits a compatible triangulation.
Bierstone-Milman 1988, Semianalytic and subanalytic sets: https://www.numdam.org/item/PMIHES_1988__67__5_0.pdf
Coste 1999, An introduction to o-minimal geometry: https://perso.univ-rennes1.fr/michel.coste/polyens/OMIN.pdf

Exercise: Semialgebraic projective-open presentation. What does it mean for a complex space to have a semialgebraic projective-open presentation?
There exist a projective complex variety Y and a real semialgebraic subset W of the underlying real-analytic manifold of Y such that W is open in the ordinary complex topology, and U is biholomorphic to W. The biholomorphism need not be algebraic, and deck transformations of any covering of U are not required to preserve W. Examples are C^n itself as the affine chart {X_0 != 0} inside P^n, the unit ball as its Borel-embedded image in its compact dual, and any product of these with a compact smooth projective variety.
Kollar-Pardon 2012, Algebraic varieties with semialgebraic universal cover, Conjecture 2: https://arxiv.org/abs/1104.2309v2

Exercise: Bounded symmetric domain. What property turns a bounded open subset of C^n into a bounded symmetric domain?
Every point of D is an isolated fixed point of some holomorphic involution of D. Equivalently, D is homogeneous under a real semisimple Lie group acting by biholomorphisms, with a maximal compact stabilizer, so that D is a Hermitian symmetric space of noncompact type with a bounded realization. Standard examples: the unit disc, the unit ball in C^n, the polydisc, the Siegel upper half space. A single point is allowed as the zero-dimensional case.
Borel 1954, Kahlerian coset spaces of semisimple Lie groups: https://doi.org/10.1073/pnas.40.12.1147
Milne 2005, Introduction to Shimura varieties: https://www.jmilne.org/math/xnotes/svi.pdf

Exercise: The Kollar-Pardon classification. State the main theorem, in full.
Let X be a connected normal projective complex variety, and let X~ be its ordinary universal cover. Then X~ is biholomorphic to a semialgebraic open subset of a projective variety if and only if there exist an integer m >= 0, a bounded symmetric domain D, and a simply connected normal projective variety F such that X~ is biholomorphic to D x C^m x F. A point is permitted as either D or F. No algebraicity of the deck transformations is assumed.
Kollar-Pardon 2012: https://arxiv.org/abs/1104.2309v2

Exercise: The easy direction. Why does X~ = D x C^m x F imply X~ is semialgebraic projective-open?
The Borel embedding places D inside its projective compact dual D^v as an open orbit. The adjoint real semisimple group whose identity component preserves D is the identity component of the real points of a real algebraic group; its orbit map onto D^v is therefore semialgebraic by real quantifier elimination. Combining D subset D^v with the standard affine chart C^m subset P^m and the compact projective factor F, the whole product sits semialgebraically and openly in the projective variety D^v x P^m x F.
Borel 1954: https://doi.org/10.1073/pnas.40.12.1147
Milne 2005, Introduction to Shimura varieties: https://www.jmilne.org/math/xnotes/svi.pdf

Exercise: Dimension one. How does the classification reduce to classical uniformization?
A smooth projective curve has universal cover P^1 (genus 0), C (genus 1), or the disc D (genus >= 2). The first case is F = P^1, m = 0, D = point; the second is m = 1, D = point, F = point; the third is D = D, m = 0, F = point. So the classification already covers curves; the content of the theorem is that the three possibilities persist in every dimension, with an extra simply connected projective factor F.

Exercise: Normal, not smooth. Why must the compact projective factor F be allowed to be singular?
Kollar-Pardon explicitly do not replace F by a smooth resolution. A smooth resolution of the compact factor may destroy the structure of the fibration over the transverse base. The normal singularity of F is preserved throughout the proof: the relative Albanese step uses the normal-projective Albanese theorem, and the product-triviality step uses normality of the compact fibres to conclude from finite birational maps that the projection is an isomorphism. Retaining normal singularities is essential to the statement of the theorem.
Kollar-Pardon 2012: https://arxiv.org/abs/1104.2309v2
Claudon-Horing-Kollar 2013, Algebraic varieties with quasi-projective universal cover: https://doi.org/10.1515/crelle.2012.017

Exercise: History. What was known before the full classification?
Iitaka studied the case X~ = C^n: a smooth projective variety covered by C^n should be a finite unramified quotient of an abelian variety. Nakayama proved the affine-space case from smooth abundance of the canonical divisor. Claudon-Horing-Kollar extended the question to quasi-projective universal covers, proving the classification from smooth abundance and proving the Albanese-bundle theorem unconditionally. Kollar-Pardon formulated the semialgebraic case as Conjecture 2, proved a nonabelian-lattice case of it, and proved a fibre-bundle theorem that remains an essential input here.
Nakayama 1999, Projective algebraic varieties whose universal covering is C^n: https://doi.org/10.2969/jmsj/05130643
Claudon-Horing-Kollar 2013: https://doi.org/10.1515/crelle.2012.017
Kollar-Pardon 2012: https://arxiv.org/abs/1104.2309v2

Exercise: Two inputs new in 2026. What new theorems supply the missing hypotheses?
First, the log abundance theorem in characteristic zero, applied with zero boundary, supplies the smooth abundance that Claudon-Horing-Kollar had to assume. Second, the abelianity conjecture for smooth compact Kahler special manifolds: the entire ordinary fundamental group of a compact special Kahler manifold is virtually abelian, with no linearity or residual-finiteness hypothesis. Both are used as black boxes; every further conclusion about the transverse action and its kernel is proved in this paper.
OpenAI 2026, Log abundance in characteristic zero, Theorem 1.1: log_abundance_char_zero_original.txt
OpenAI 2026, The abelianity conjecture for special compact Kahler manifolds, Theorem 1.1: abelianity_special_kahler_original.txt

Exercise: Functorial projective resolution. How is the semialgebraic projective specimen turned into a smooth semialgebraic open set?
Normalize the projective ambient variety along the irreducible component containing X~, and apply functorial resolution of singularities. Compatibility with local analytic isomorphisms identifies the restriction with the pullback of a projective resolution X^+ -> X. The result is a proper modification mu : V -> X~ where V is a smooth semialgebraic open subset of a smooth projective variety Y, and the lifted deck action of Gamma = pi_1(X) on V is free, properly discontinuous, and has compact projective quotient X^+ = V/Gamma.
Bierstone-Milman 2008, Functoriality in resolution of singularities: https://arxiv.org/abs/math/0702375

Exercise: Local pseudoconvexity of the cover. Why is V locally pseudoconvex in Y?
This is the analytic-to-algebraic step. Every locally pseudoconvex open subset of a smooth projective variety is Stein at its boundary in the precise sense: it satisfies the Hartogs extension criterion. Ivashkovich shows that a projective covering p : V -> X^+ extends meromorphically over small Hartogs figures in Y, and any first exit along a straight segment contradicts the evenly covered neighbourhood structure of the covering in X^+. The same argument gives local pseudoconvexity of inverse images along holomorphic local sections used later.
Ivashkovich 2008, Extra extension properties of equidimensional holomorphic mappings, Corollary 2: https://matstud.org.ua/texts/2008/30_2/198-213.pdf

Exercise: Levi form on a real-analytic boundary. What does the Levi form measure, and what is its null distribution?
Let M be a smooth real hypersurface inside a complex manifold, with a real defining function rho and its real one-form theta restricted to M. The Levi form is the Hermitian form on the complex hyperplane TM cap JTM given by dtheta evaluated on complex vectors. On a locally pseudoconvex boundary it is semidefinite (say nonnegative). Its kernel, the null distribution K = {v : dtheta(v, D) = 0}, has constant rank 2*ell when the Levi form has constant rank, and is involutive by the Jacobi identity. Thus K integrates to complex ell-dimensional foliations by "null plaques". A complex submanifold of M has its tangent space inside K, because the Levi form vanishes on it.
(Standard argument in sections/windows.tex, Lemma win:plaques.)

Exercise: Algebraic plaques. Why are the Levi null plaques open subsets of complex algebraic varieties?
In affine coordinates, the real defining polynomial rho has the form v(z)^* C v(z) for a Hermitian matrix C and a vector v of holomorphic monomials. If f is any holomorphic parametrization of a complex plaque with phi(f(zeta), conjugated f(zeta)) = 0, polarization shows v(f(eta))^* C v(f(zeta)) = 0 for independent parameters; hence the complex span W of the vectors v(f(zeta)) is totally C-isotropic. The algebraic set v^{-1}(W) contains the plaque, sits in {rho = 0}, and by the dimension bound (every complex submanifold of M is a plaque) has dimension exactly ell on the component containing the plaque. Bounded monomial degree gives bounded algebraic data.
(Standard argument in sections/windows.tex, Lemma win:plaques.)

Exercise: Contact quotient. Why does the boundary near a regular plaque look like a contact manifold?
Divide M by the null distribution K. The quotient inherits a hyperplane field D/K, and its induced exterior derivative is nondegenerate; this is the standard contact structure of dimension 2r+1 (r = rank of the Levi form). Choosing Darboux coordinates (s, w) with contact form ds - sum_j v_j du_j, one takes w = u as a Lagrangian transverse coordinate, s as a characteristic direction, and v as external parameters. The Darboux chart turns the boundary into a family of quadric surfaces, one for each external parameter.
(Standard argument in sections/windows.tex, Lemma win:period.)

Exercise: Period matrix and the quadratic test. What is the period matrix H(z), and what weighted tangent inequality does it define?
Comparing the fixed real Lagrangian frame of the contact quotient with the infinitesimal evaluations along the plaque gives a symmetric r x r complex matrix H(z). Symmetry follows from the (1,1)-nature of the contact symplectic form. The sign convention makes B = Im H positive definite. The Taylor expansion of the pulled-back defining function after scaling the transverse directions quadratically produces the weighted tangent inequality Im s - (Im w)^T (Im H(z))^{-1} Im w > 0 for points on the inward side. H is a meromorphic matrix on the smooth projective model L of the plaque.
(Standard argument in sections/windows.tex, Lemma win:period.)

Exercise: The first kernel and the whole seed fibre. What does the open kernel of the first affine quadratic family look like?
Given the family in (s, w, z), consider the open kernel of the complexified family in a small ball at a regular point z_0. The component E through the seed ((i, 0), z_0) is semialgebraic, locally pseudoconvex, and invariant under real translations in (s, w) and under positive real scalings of the transverse block. Its whole fibre over z_0 is exactly the Siegel tube {Im s > (Im w)^T A Im w} with A = (Im H(z_0))^{-1}, not just a relatively compact piece. A Bergman-function change of variables biholomorphically maps this tube onto a unit ball, so the whole fibre has a strictly positive Bergman kernel.
(Standard argument in sections/windows.tex, Lemmas win:seed-bergman and win:first-epigraph.)

Exercise: Exact epigraph equation. Why is E equal to the quadratic side over a whole domain V_2 of L?
Near generic points of L, the Levi form is nonsingular and the boundary is one of three types: empty, full, or one of the two strict quadratic sides. Each side is connected as an epigraph or hypograph of a quadratic form, and Bochner's tube theorem plus Berndtsson's subharmonicity of log of the Bergman kernel rule out a complex affine line, hence force strict definiteness. The sign is selected by the seed. Extending across the exceptional divisor of L uses tangency propagation along paths inside the tube slice, and a codimension-two thin-omission step (Vajaitu) extends holomorphically. Result: E = {(s, w, z) : z in V_2, Im s > (Im w)^T (Im H(z))^{-1} Im w}.
Noguchi 2020, A brief proof of Bochner's tube theorem: https://arxiv.org/abs/2007.04597v4
Berndtsson 2006, Subharmonicity properties of the Bergman kernel: https://www.numdam.org/item/10.5802/aif.2223.pdf
Vajaitu 2000, Analyticity of q-concave sets: https://aif.centre-mersenne.org/item/10.5802/aif.1789.pdf

Exercise: First epigraph in the source coordinates. How does the quadratic condition descend to V itself?
The composite map from the original domain V to the first projective model L sends V to the projection V_2 subset L. Near a generic real center and after choosing scales, V corresponds to the condition Im s > (Im w)^T (Im H(z))^{-1} Im w on V_2, for a scalar s and vector w that are algebraic functions of the original coordinates. The map sends the boundary of V to the boundary of the epigraph, and the parameterization is compatible with the real structure. This is the first block of the eventual recursive tower.
(Standard argument in sections/windows.tex, Lemma win:first-epigraph.)

Exercise: Normalizing a varying Siegel matrix. How is a parameter-dependent H replaced by a polynomial matrix?
Given a symmetric H_* holomorphic on a full neighbourhood of the real center x_2 with Im H_* > 0, form T_t = (Im H_*(x_2 + R^2_t p_2))^{1/2} and
H_t(z) = T_t^{-1} ( H_*(x_2 + R^2_t z) - Re H_*(x_2 + R^2_t p_2) ) T_t^{-1}.
Montel's theorem and maximum principle bound H_t in Siegel space in the normalized coordinates, so its limit is a symmetric polynomial matrix tau with tau(p_2) = iI and Im tau > 0 on the marked component P_2 of the open kernel. The positive square root and its inverse have power bounds under the normalization, so they are admissible.
(Standard argument in sections/windows.tex, Lemma win:siegel-normalization.)

Exercise: The recursive tower of matrix inequalities. What is the final form of the limiting domain P, and why is it diffeomorphic to R^(2d)?
The domain P is built inductively: start from a point, add a block (s, w) in C x C^r over the previous tower P_2 by the condition Im M(s, w, z) > 0, where M(s, w, z) = [[s, w^T], [w, tau(z)]] and tau is polynomial with Im tau > 0 on P_2. Schur-complement and the positivity of the "slack" give smooth coordinates, which exhibit P as a Euclidean space of real dimension 2*sum_j (1 + r_j) = 2*complex dimension. The function Phi(Z) = |Z|^2 - sum_j log det Im M_j(Z) is a strictly plurisubharmonic exhaustion, so P is Stein. A one-block-at-a-time Cayley transform gives a bounded realization.
Kobayashi 1967, Invariant distances on complex manifolds: https://www.jstage.jst.go.jp/article/jmath1948/19/4/19_4_460/_pdf
(Standard argument in sections/windows.tex, Proposition win:tower-properties.)

Exercise: Triangular recenterings and open kernels. Why do triangular normalization matrices convert bounded paths into attached-disc boundaries?
After normalizing with R_t(x), the boundary of the scaled domain becomes the graph {(u, Z) : u + G_x(u) Z_in S} in original coordinates. Because the normalization is block triangular, each block of G_x depends only on later block cycles, and the whole construction is polynomial and invertible. As t -> 0, the associated graph converges to a graph of the form u + G_x(u)F(a), which is exactly the boundary of a complex disc with a schedule a(.) on the circle. This is the reduction used by Bishop and Hill-Taiani.
(Standard argument in sections/localization.tex, proof of Theorem loc:theorem; Bishop 1965; Hill-Taiani 1978.)

Exercise: Bishop's reduction. How does Bishop's argument convert the boundary-of-a-disc question into a linear equation on the circle?
Given a real-valued smooth boundary function v(theta) on the circle and an imaginary part w(theta), the boundary of a holomorphic disc equals v + i w exactly when v - mean(v) = -H w, where H is the Hilbert transform (H(cos n theta) = sin n theta). For triangular systems in the tower this becomes a strictly triangular equation in C^alpha; the operator is identity plus nilpotent, so it is invertible, and the disc close-up has full real rank 2*ell in the imaginary perturbation. Small perturbations of the schedule are controlled by the implicit function theorem in C^alpha, by compactness.
Bishop 1965, Differentiable manifolds in complex Euclidean space: https://doi.org/10.1215/S0012-7094-65-03201-1
Hill-Taiani 1978, Families of analytic discs with boundaries on a prescribed CR submanifold: https://www.numdam.org/item/ASNSP_1978_4_5_2_327_0.pdf
Priwaloff 1916, Sur les fonctions conjuguees: https://www.numdam.org/article/BSMF_1916__44__100_0.pdf
Essen 1984, A superharmonic proof of the M. Riesz conjugate function theorem: https://doi.org/10.1007/BF02384381

Exercise: Bounded-path localization. What exactly does Theorem loc:theorem guarantee, and why is it needed?
Local pseudoconvexity (Ivashkovich) alone controls interior points, not the endpoint of a bounded path that exits through a boundary face. The theorem asserts that, under triangular recenterings satisfying the ratio condition R_t(x)^{-1} R_t(x + R_t(x) u) -> G_x(u) with polynomial limits, any bounded path in the scaled domain S_{t,x} starting at the seed p and converging to z has z in the closure of the marked component P. The mechanism is that the boundary of any such path, when pushed through the triangular normalization, becomes the boundary of a family of attached discs. Persistence and the center-differential invertibility then put a ball around the discriminating parameter map in P; the endpoint of the path is a limiting center.
(Standard argument in sections/localization.tex, Theorem loc:theorem.)

Exercise: Power diagonalization. Why can one replace two scales by a single scale equation?
Given a bounded globally subanalytic function f(e, t, v) with iterated limit f_{00}(v) = lim_t->0 lim_e->0 f(e, t, v), there is an integer N_0 such that for every integer N > N_0, lim_t->0 f(t^N, t, v) = f_{00}(v) for each fixed v. The threshold of convergence is a definable function of (v, delta), bounded below by a power of t by the Puiseux-preparation lemma. Composing scales via t^N then preserves the marked component of the open kernel on every fixed bounded box. This reduces the two-scale construction of the tower to a single-scale theorem.
(Standard argument in sections/preliminaries.tex, Lemma prep:power-diagonalization.)

Exercise: Guard functions. What is the guard, and why does it control the exhaustion?
The guard attached to a chart is h(Z) = exp(-alpha * sum_j tr log(I - i M_j(Z))) on the normalized tower P, with alpha a small positive rational. On the resolved chart it satisfies |h(Z)| <= C (1 + |Z|)^{-alpha}, it lies in a fixed sector {|arg h| < theta} with theta < pi/2, and it extends holomorphically to a neighbourhood of any compact parameter region. Every compact subset of P is eventually contained in the domains on which h is defined, so h controls the exhaustion of deck-translated charts. Its purpose is to keep the translated charts inside a uniformly bounded model so that the compactness principle applies.
(Standard argument in sections/quotient.tex, Lemma quo:guard.)
Grauert 1958, Analytische Faserungen ueber holomorph-vollstaendigen Raeumen: https://doi.org/10.1007/BF01351803

Exercise: Incidence trace. What is the incidence trace function, and why is it plurisubharmonic and bounded?
Fix a chart, let H_i be its marked region, and let pi_i : U_i -> H_i be the local parameter submersion. Put u_i(b) = log|h| + a_i |b|^2 with a_i small. The trace Phi_i(y) = sum over parameters b with y in Z_b and b in H_i of max(0, u_i(b) - m_i), counted with finite-map multiplicity, is bounded above and plurisubharmonic on V. It equals a sum of subharmonic functions near regular points of the finite projection, and away from the discriminant the bounded removable extension plus the fact that u_i is strictly plurisubharmonic at each root stop the sum from being constant; hence each root must be locally constant. The trace is used to prove that distinct fibres of the intrinsic quotient are disjoint.
(Standard argument in sections/quotient.tex, Lemma quo:incidence.)
Demailly 2012, Complex Analytic and Differential Geometry, Corollary I.5.24: https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf

Exercise: Whole projective fibres. Why does each parametrized intersection with the cover become a whole fibre, and not a proper piece?
Because pi_i : U_i -> H_i is a submersion with connected fibres, and the trace argument shows any two positive-dimensional pieces over distinct parameters cannot intersect (they would force equality of parameters). The fibre pi_i^{-1}(b) = Z_b cap V. Then by passing through the local evaluation submersion q : Q|_S -> Y, the parameter family is a proper analytic family whose complement to the incidence locus is analytic; and the actual holes are analytic. The image of these holes is precisely the algebraic complement Z_b minus V in the model variety Z_b. The key point is "generic fibrewise algebraic", so Z_b cap V is a Zariski open subset of Z_b, not just an image of one.
(Standard argument in sections/quotient.tex, Lemma quo:incidence.)

Exercise: Cutoff-guard compactness. If h_i satisfies uniform boundary bound delta_i -> 0 and h_i extends by zero, what does the resulting limit satisfy?
Given a complex manifold D, open A_i subset D, holomorphic h_i on A_i with |h_i| <= C and |h_i| <= eta_i uniformly near every boundary point of A_i inside D, extend h_i by zero to D. A subsequence converges uniformly on compact subsets of D to a holomorphic function h. Despite the discrete jumps at each fixed index, an asymptotic-equicontinuity argument (one-variable analysis on complex lines) upgrades the Sobolev bound to Arzela-Ascoli, and the argument principle on level sets of |h_i| excludes values above the level inside the disk of convergence. The conclusion: any limit is nowhere zero, and every compact subset of D lies in A_i for all large i.
(Standard argument in sections/quotient.tex, Lemma quo:cutoff-normal.)

Exercise: Trapping corollary. What does Corollary quo:trapping say about translated charts?
Let D_i exhaust a connected complex manifold D and g_i : D_i -> V blow up to V and satisfy at points x_i -> x that the normalized parameters pi_i(g_i(x_i)) converge to the seed p in the intrinsic tower. Then every compact subset of D is eventually mapped into U_i by g_i, and after a subsequence pi_i o g_i converges normally in D to a holomorphic map with values in P. The same statement holds for maps into a manifold furnished with these projection charts and their guards. The proof applies cutoff-guard compactness to the composite h o pi_i o g_i, then applies bounded-path localization to the limit parameter map.
(Standard argument in sections/quotient.tex, Corollary quo:trapping.)

Exercise: Intrinsic quotient. Why is the quotient of V by the union of fibres a complex manifold?
Each chart U_i is locally biholomorphic to its parameter domain H_i via pi_i. Two charts whose fibres meet equal whole fibres (by the incidence lemma above). So the equivalence "same fibre" defines a Hausdorff, second-countable quotient M which is covered by biholomorphic coordinate patches isomorphic to the H_i, with holomorphic transition maps (compose a section with the other projection). Deck-translations act on M. Fixing deck-translations and applying trapping gives an exhaustion of V by the translated charts, so the quotient map f : V -> M is a submersion.
(Standard argument in sections/quotient.tex, Section quo:section.)

Exercise: Inverse tightness. Why do inverse charts of the quotient close up in a fixed chart?
Let K be a compact subset of the intrinsic domain P. The translated charts U_i contain every compact subset of V. The inverse maps psi_i : P -> M are constructed from chart inverse maps. Failure of K to land in a fixed M_j would allow infinitely many i_nu with parameters escaping every bounded box. Translating back and applying trapping (with the second-scale coordinates chosen to hit the seed) shows that psi_{i_nu}(K) lies in one fixed M_j after all, contradiction. Then Montel plus normal convergence of phi_i and psi_i identifies the two inverse maps as biholomorphisms, and M is biholomorphic to P.
(Standard argument in sections/quotient.tex, Lemma quo:inverse-tightness.)

Exercise: Normal-source descent. Why does the intrinsic quotient of V descend to the normal projective universal cover?
Every compact connected fibre of mu is a compact irreducible complex space, so holomorphic functions are constant on it, and mu_*O_V = O_X~. Thus any holomorphic f : V -> M that is constant on compact fibres factors uniquely through mu, giving a holomorphic f_X : X~ -> M with f = f_X o mu. Equivariance of f under the deck action plus uniqueness makes f_X equivariant. Connectedness of the original fibres and mu^{-1}(f_X^{-1}(m)) = f^{-1}(m) show connectedness of the descended fibres. Normalizing the compactifications gives locally analytic algebraic holes on every fibre.
(Standard argument in sections/quotient.tex, Lemma quo:descent.)

Exercise: Normal and flat compactifications. Why does the normal-source family descend with normal fibres and flatness?
After shrinking the parameter box at a generic real center, the incidence family normalizes to a proper map N -> B with normal fibres, and Greuel's theorem makes the "non-normal" locus a nowhere dense analytic subset. Since the chosen parameter is a generic point of a full-dimensional totally real box, it avoids the closure. Then flatness and normality hold over the full box. A simultaneous resolution of the correspondence gives smooth projective families on every fibre; the map to the normalized family is bimeromorphic on every fibre. Finally, the canonical Whitney stratification of the total space is by definition intrinsic, and after shrinking further the parameter projection is submersive on every stratum, giving topological local triviality (Mather's first isotopy lemma for proper stratified submersions).
Greuel 2017, Equisingular and equinormalizable deformations: https://arxiv.org/abs/1702.05505v2
Teissier 1982, Varietes polaires II, Propositions 3.1-3.2: https://webusers.imj-prg.fr/~bernard.teissier/documents/VarPol2.pdf
Mather 1970, Notes on topological stability, Proposition 11.1: https://webhomes.maths.ed.ac.uk/~v1ranick/papers/matherj.pdf

Exercise: Intrinsic projection theorem. What does the projection theorem assemble, and what does it hand to the next step?
Theorem quo:projection, under condition (i) of Theorem 1 and the boundary construction of Theorem win:construction, constructs a contractible Stein manifold M, an equivariant surjective submersion f : V -> M, and an equivariant holomorphic map f_X : X~ -> M with f = f_X o mu. Both maps have connected fibres, and every fibre has a local projective compactification with analytic holes. The normality of the source fibres is preserved, and f_X is flat. If in addition the image Lambda of the deck action on M is discrete acting freely, then the descended map X -> M/Lambda, its pullback to M, and its finite covering families are flat, topologically locally trivial, and have connected normal projective fibres; fibrewise resolutions are smooth proper projective families; the morphisms become local-system maps on integral cohomology.
(Standard argument in sections/quotient.tex, Theorem quo:projection.)

Exercise: Campana core and special varieties. What does the Campana core of a compact Kahler manifold record?
A smooth compact Kahler manifold Z is special when it has no Bogomolov sheaf: no line bundle L with nonzero morphism L -> Omega_Z^r and Iitaka dimension kappa(Z, L) = r > 0. Campana's core theory provides a projective modification r : Z' -> Z and a holomorphic core map c : Z' -> C whose very general smooth connected fibre is special, and whose orbifold base is of general type unless C is a point. The orbifold Kobayashi-Ochiai theorem then says that a meromorphic map from a punctured manifold into Z, of maximal rank dim C, extends meromorphically across the puncture after composing with the core.
Campana 2004, Orbifolds, special varieties and classification theory, Theorem 8.2: https://arxiv.org/abs/math/0110051v9

Exercise: Special compact Kahler abelianity. What is the abelianity statement for compact special Kahler manifolds?
For every smooth compact Kahler manifold Z that is special in the sense of Campana, the entire ordinary fundamental group pi_1(Z) is virtually abelian: it contains an abelian subgroup of finite index. No linearity, no residual-finiteness, no algebraicity is assumed. This is used as a black box, applied to smooth connected compact special core fibres and to the whole manifold when the core is a point.
OpenAI 2026, The abelianity conjecture for special compact Kaehler manifolds, Theorem 1.1: abelianity_special_kahler_original.txt

Exercise: Liouville lemma for virtually abelian covers. Why is every bounded holomorphic function on a virtually abelian cover of a compact Kahler manifold constant?
Let a : W -> Z be a connected regular cover of a compact connected Kahler manifold, with deck group virtually abelian. Choose an abelian subgroup A of finite index; the quotient Z_A = W/A is compact Kahler. Pull back a Kahler metric to W (complete) and consider the convex set H of strictly positive harmonic functions with h(o) = 1. Harnack and elliptic estimates make H compact; an extreme point h has minimal cone, and by a uniform Harnack chain the translates h(alpha x) are positive constant multiples of h(x) for each alpha in A. So d log h descends to a closed one-form on Z_A with divergence = -|eta|^2; integrating forces eta = 0 and h = 1. Real and imaginary parts of any bounded holomorphic function are harmonic, so both are constant.
Lyons-Sullivan 1984, Function theory, random paths and covering spaces, Theorem 1: https://doi.org/10.4310/jdg/1214438681
Serrin 1964, Local behavior of solutions of quasi-linear equations: https://doi.org/10.1007/BF02391014
Nirenberg 1959, On elliptic partial differential equations: https://www.numdam.org/item/ASNSP_1959_3_13_2_115_0.pdf

Exercise: Countable levels become locally finite parameters. Why is the set of parameters reached by a fixed core level countable and locally finite?
Fix a very general smooth connected special core fibre F_s = c^{-1}(s) in the projective model Z of X^+. Pull F_s back under the cover p to V; each connected component is a regular cover of F_s with group the image of pi_1(F_s) in Gamma, and this image is virtually abelian by the abelianity input. Bounded holomorphic coordinates of the intrinsic map f are constant on every such component by the Liouville lemma. So f sends the connected components of p^{-1}(r(F_s^o)) to a countable set D_s. Local finiteness: apply the proper-meromorphic-fibre lemma to a local incidence chart Q -> B, where the map to C is a meromorphic extension. A proper analytic map from an analytic set meeting a countable-fibre locus is locally constant, because a positive-rank holomorphic map has uncountable image; properness cuts off accumulation.
(Standard argument in sections/groups.tex, Lemmas grp:countable-core-level and grp:finite-level-projection.)

Exercise: Discreteness of the transverse action. Why is Lambda = rho(Gamma) subset Aut(M) discrete and cocompact?
Take the closure H of Lambda inside Aut(M). For every very general core level, D_s is Lambda-invariant (deck action preserves p, equivariance of f), and since D_s is locally finite in M it is also closed. Any orbit of the identity component H^0 is connected and lies inside the discrete set D_s, so it is a single point. Therefore H^0 fixes the dense union of the D_s pointwise, hence M pointwise, hence H^0 is trivial. A closed subgroup of a Lie group with trivial identity component is discrete, so H is discrete, so Lambda is discrete.
The quotient X^+ = V/Gamma is compact, so its image M/Lambda is compact since f is surjective.
(Standard argument in sections/groups.tex, Proposition grp:discrete.)

Exercise: Free finite-index action. Why is there a torsion-free finite-index subgroup of the transverse image acting freely on M?
The group Gamma is finitely generated (a quotient of pi_1(X^+)), and so is Lambda. Its image under the adjoint Ad(Lambda) subset GL(Lie(Aut M)) is a finitely generated linear group over R, hence by Selberg's lemma has a torsion-free finite-index subgroup L. Set Lambda_1 = (Ad|_Lambda)^{-1}(L), still finite index in Lambda. For any m in M, the stabilizer (Lambda_1)_m acts on the tangent space via identity (because the complete holomorphic vector fields of Aut(M) span every complex tangent space at m, and Ad-invariance gives d lambda_m = 1); a finite-order biholomorphism with identity tangent map is the identity near the fixed point and hence globally. Then (Lambda_1)_m is trivial. Torsion-freeness of Lambda_1 itself follows from a finite free resolution of Z over Z[Lambda_1] obtained by lifting a finite triangulation of the compact quotient; a finite cyclic subgroup would force bounded cohomological dimension. The preimage of Lambda_1 in Gamma corresponds to a finite unramified projective cover X' -> X via Riemann existence.
Selberg 1960, On discontinuous groups in higher-dimensional symmetric spaces: https://mathweb.tifr.res.in/Documents/Publications/Studies/Function_Theory.pdf
Wolf 1968, Growth of finitely generated solvable groups: https://math.berkeley.edu/~jawolf/publications.pdf/paper_033.pdf

Exercise: Bergman kernel and positive canonical bundle. Why does M/Lambda have ample canonical bundle?
When Lambda acts freely, put B = M/Lambda. The intrinsic Bergman kernel of square-integrable canonical forms on the bounded realization of M is positive everywhere, so it gives a Hermitian metric on K_M with strictly positive complex Hessian; biholomorphism-invariance makes this descend to a metric on K_B of strictly positive curvature. Kodaira embedding then makes K_B ample and c_1(B) < 0. In dimension one the same positivity-to-ample step is Riemann-Roch; the point case is immediate. So B is a compact aspherical projective manifold with negative first Chern class.
Kodaira 1954, On Kahler varieties of restricted type: https://doi.org/10.1073/pnas.40.5.313

Exercise: Nadel-Frankel splitting. Why is M biholomorphic to a bounded symmetric domain?
The Farb-Weinberger statement of the Nadel-Frankel splitting theorem applies to compact aspherical B with ample canonical bundle and gives a factorization of universal covers: M = D_sym x R, where D_sym is a bounded symmetric domain and Aut(R) is discrete. Bergman-kernel invariance gives a product structure of the Bergman metric, g_M = g_D + g_R. If R had a positive-dimensional Euclidean factor, its parallel vector fields would generate a C^a factor and Liouville would contradict boundedness of M; so R has no Euclidean de Rham factor. Then by uniqueness of de Rham decomposition, every isometry in the identity component preserves the two factors. Aut(R) being discrete, any element of Aut(M)^0 acts on R as the identity. Therefore the complete holomorphic vector fields from Aut(M)^0 have zero R-component. Spanning of tangent spaces over C forces dim R = 0, hence M = D_sym.
Farb-Weinberger 2008, Isometries, rigidity and universal covers, Theorem 1.10: https://annals.math.princeton.edu/wp-content/uploads/annals-v168-n3-p05.pdf
De Rham 1952, Sur la reductibilite d'un espace de Riemann: https://doi.org/10.1007/BF02564308
Kobayashi 1967: https://www.jstage.jst.go.jp/article/jmath1948/19/4/19_4_460/_pdf

Exercise: Theorems for the transverse symmetric factor. What does Theorem grp:discrete-symmetric conclude now that all inputs are in place?
Under the intrinsic-projection hypotheses (contractible Stein M with complete holomorphic vector fields spanning every tangent space) and the intrinsic-quotient inputs (guarded propagation from any chart), the transverse image Lambda is discrete and cocompact, and after the free finite-index passage Lambda acts freely; the domain M is biholomorphic to a bounded symmetric domain D_sym, with a point allowed. The theorem body is the combination of the discreteness proposition (countable-locally-finite D_s plus connected-closure argument), the free finite-index lemma (Selberg + adjoint + torsion-free quotient), the Bergman-Kodaira ampleness, and the Nadel-Frankel decomposition with the R-factor exclusion.
(Standard argument in sections/groups.tex, Theorem grp:discrete-symmetric.)

Exercise: Compactifiable covers. Why is the deck group virtually abelian when the cover is a Zariski-open subset of a smooth projective manifold?
Suppose a : U -> N is a connected regular cover of a smooth connected projective manifold, with deck group K, and U is a Zariski open subset of a smooth projective Ubar. Choose a smooth projective core model r : N' -> N. Use Campana's core extension along the boundary to get a meromorphic h : Uhat -> C and a projective modification tau : Uhat -> Ubar. Very general levels h^{-1}(s) intersect tau^{-1}(U) in finitely many components; components on the retained ordinary part recover connected components of a^{-1}(r(F_s^o)), where F_s = c^{-1}(s) is a connected special fibre. Then the index of the image H = im(pi_1(F_s^o) -> K) is finite (one per component). The quotient H is virtually abelian since F_s is special; and since F_s^o has the same pi_1 as F_s (perturbed loops), H is virtually abelian. An abelian finite-index subgroup of H also has finite index in K, so K is virtually abelian.
(Standard argument in sections/groups.tex, Lemma grp:compactifiable-cover.)

Exercise: Fibre kernel abelianity. Why is the kernel K = ker(Gamma -> Lambda) finitely generated and virtually abelian?
After the free finite-index passage, K is the stabilizer in Gamma of any point m in M. The equivariant submersion f descends to a proper holomorphic submersion g : X^+ -> B = M/Lambda. Its fibre N_b is a smooth connected projective manifold; restricting p gives f^{-1}(m) -> N_b as a connected regular covering with group K. The projection f^{-1}(m) -> N_b is a Zariski open subset of a compactification with analytic boundary, so the compactifiable-cover lemma plus the compact special abelianity conclusion on N_b both give "K is virtually abelian". Surjectivity of pi_1(N_b) -> K (as monodromy of the connected regular cover) plus finite presentability of pi_1 of compact manifolds gives finite generation.
(Standard argument in sections/groups.tex, Theorem grp:kernel-abelian.)

Exercise: Characteristic free abelian subgroup. Why is there a characteristic finite-index subgroup K' of the fibre kernel K that is free abelian?
The kernel K is finitely generated and virtually abelian. Choose any finite-index abelian subgroup and pass to its torsion-free part B, which is free abelian of some rank 2a and finite index d in K. Intersect all subgroups of K of index at most d: there are finitely many such, because a finitely generated group has only finitely many homomorphisms to any fixed finite symmetric group, and their intersection is characteristic of finite index and lies inside B. Since K is normal in Gamma, so is K'. The quotient X = U/K' is then normal with a proper flat map p : X -> M, topologically locally trivial with connected normal projective fibres N_m, and pi_1(N_m) -> pi_1(X) = K' is an isomorphism.
(Standard argument in sections/compact.tex, Lemma cmp:characteristic.)

Exercise: Relative Albanese descent. How does the fibrewise Albanese of the normal projective fibres assemble?
The normal-projective Albanese theorem says that for a normal projective variety N, the induced map H_1(N, Z) -> H_1(Alb N, Z) is surjective with finite kernel, and is an isomorphism whenever pi_1(N) is torsion-free. In our situation pi_1(N_m) = K' is torsion-free, so H^1(N_m, Z) sits as a lattice of rank 2a. Taking the weight-one variation of Hodge structures on the smooth proper projective family p^+ : Y -> M of the simultaneous resolution, let W^{1,0} = W_O cap F^1 H_O where W_Z is a rank 2a sublocal-system and F^1 H_O the Hodge filtration. This gives a smooth family of Albanese tori q : A -> M together with a proper morphism alpha : X -> A over M whose fibre over m is the normal Albanese alpha|_{N_m}. Both constructions are Gamma/K'-equivariant.
Griffiths 1968, Periods of integrals on algebraic manifolds II, Theorem (1.1): https://publications.ias.edu/sites/default/files/periodsofintegralII68.pdf
Claudon 2018, Smooth families of tori and linear Kahler groups: https://afst.centre-mersenne.org/item/10.5802/afst.1576.pdf
(Standard argument in sections/compact.tex, Lemma cmp:albanese.)

Exercise: Kollar-Pardon bundle theorem. What does the Kollar-Pardon fibre-bundle theorem assert, and how is it applied?
Suppose N is a normal projective variety, B a smooth projective variety with contractible universal cover, and g : N -> B a morphism such that the entire pullback of N to the universal cover B~ is biholomorphic to a semialgebraic open subset of a projective variety. Then g is a holomorphic fibre bundle. Applying this to the fibrewise Albanese N_m -> Alb(N_m) (whose target is an abelian variety with universal cover C^a, and whose pullback cover is exactly the universal cover of N_m by the previous exercise) makes the Albanese map a holomorphic fibre bundle with flat and connected, reduced normal projective, in fact simply connected, fibres. The resulting map X -> A is flat by the fibrewise-flatness criterion.
Kollar-Pardon 2012, arXiv v2, Theorem 20: https://arxiv.org/abs/1104.2309v2
(Standard argument in sections/compact.tex, Proposition cmp:affine-base.)

Exercise: The proper map to the affine base. Why does the universal cover of the Albanese family give T isomorphic to M x C^a?
Let T -> A be the universal cover of the family of Albanese tori. The pullback X x_A T is connected and simply connected (because on pi_1 the map alpha is an isomorphism), and it is a proper flat family with simply connected normal projective fibres: this is U, with j : U -> T proper and flat. Because the universal cover of a torus bundle is an affine bundle over M with linear part the relative Hodge bundle E, its translation cocycle lies in H^1(M, O(E)) = 0 by Cartan's theorem B. So T = E as a holomorphic vector bundle; topologically E is trivial (contractible base), and by the Oka principle it is holomorphically trivial, so T = M x C^a.
Cartan 1952, Faisceaux analytiques sur les varietes de Stein, Expose 18: https://www.numdam.org/article/SHC_1951-1952__4__A18_0.pdf
Forstneric-Prezelj 2000, Oka's principle for holomorphic fiber bundles with sprays: https://arxiv.org/abs/math/0305238
Grauert 1958: https://doi.org/10.1007/BF01351803
(Standard argument in sections/compact.tex, Proposition cmp:affine-base.)

Exercise: Rigidity for invariant closed subsets. If a closed invariant subset of a contractible manifold has finite-CW homotopy type, why is it the whole manifold?
Let a countable discrete group act properly and cocompactly by diffeomorphisms on T homeomorphic to R^b, let A subset T be closed, invariant, of finite CW type, and nonempty. A proper cocompact action admits a complete invariant Riemannian metric. Using a locally finite partition of unity subordinated to the orbit-centre balls, one constructs a continuous proper map s : T -> A (successive applications of a contracting homotopy of A, translated into the covering). Composing the resulting proper map s with a proper contraction of R^b (obtained the same way but into a point) gives a proper homotopy from id_T to i o s. Then if b > 0 and x is not in A, the one-point compactification argument (degree of a proper map) gives a contradiction, so A = T. Finite stabilizers only add finite repetitions and do not affect the argument.
Kollar-Pardon 2012, Section 1, Lemmas 11-12 and Proposition 13: https://arxiv.org/abs/1104.2309v2
(Standard argument in sections/compact.tex, Lemma cmp:closed-rigidity.)

Exercise: Chow space of fibres. Why do the fibres of j form a semialgebraic locus in a Chow space?
Choose a semialgebraic realization U subset Z of a projective variety Z, and consider the projective Chow space Chow_{n,d}(Z) of integral n-cycles of degree d, where n is fibre dimension and d is the degree of one fibre. The locus C of integral cycles whose support lies in U is semialgebraic: the integrality condition is the complement of the images of finitely many proper addition maps, and support containment in U is real-semialgebraic by incidence and quantifier elimination. The graph of j inside T x Z thus has an analytic family of fibre cycles, continuous in the analytic cycle space; comparison of Chow and analytic cycle topologies (via continuous bijection from a compact Chow space onto a compact analytic cycle family) makes t -> [j^{-1}(t)] a homeomorphism from T to C.
Barlet 1999, How to use the cycle space in complex geometry: https://library.slmath.org/books/Book37/files/barlet.pdf
Andreotti-Norguet 1967, La convexite holomorphe dans l'espace analytique des cycles: https://www.numdam.org/item/ASNSP_1967_3_21_1_31_0.pdf
(Standard argument in sections/compact.tex, Lemma cmp:chow-base.)

Exercise: Polarized isomorphism classes. Why does the locus of fibres polarized-isomorphic to a fixed fibre form a semialgebraic subset of C?
Fix a fibre F and an ample line bundle L_F, and choose k > 0 so that L_F^k is very ample. Any polarized isomorphism g : F -> j^{-1}(t) has graph of fixed degree D = integral over j^{-1}(t) of (k c_1(L) + c_1(O_Z(1)))^n, which is constant in t by continuity of the analytic cycle family. Introducing the locus P in the projective Chow space of cycles in F x Z satisfying the graph conditions (image in F x U, both projections finite and injective), one shows P is semialgebraic and parametrizes isomorphisms. The universal incidence over P identifies as F x P as a topological space, and the evaluation map on F x P gives a continuous family of line bundles; since Pic(F) -> H^2(F, Z) is injective on a simply connected normal projective F, the locus where this bundle is isomorphic to L_F is a union of connected components of P, hence semialgebraic. Its image in C under the "support of graph" map is the requested polarized-isomorphism class.
Kleiman 2005, The Picard scheme, Theorem 5.4: https://arxiv.org/abs/math/0504020
Serre 1956, Geometrie algebrique et geometrie analytique, Proposition 18: https://numdam.org/item/10.5802/aif.59.pdf
(Standard argument in sections/compact.tex, Lemma cmp:polarized-locus.)

Exercise: Triviality of the compact factor. Why do all compact fibres become the same variety, and why is the family a holomorphic product?
Every polarized isomorphism class is a Gamma-invariant semialgebraic subset of C. Its closure in C is nonempty, closed, invariant, and (by semialgebraic triangulation) has the homotopy type of a finite CW complex; the rigidity lemma (with T = C modeled on the semialgebraic space) then forces that closure to be all of T. So every class is dense. But two disjoint semialgebraic subsets of a semialgebraic space cannot both be dense, because a dense semialgebraic subset contains a relatively open dense subset. So all fibres are isomorphic as polarized varieties.
Now all fibres are isomorphic, so on a small open base B subset T the family j^{-1}(B) -> B admits a holomorphic Hilbert embedding into B x P^q, and the Hilbert-map locus is contained in the orbit of the fixed embedded fibre under G_0 = PGL_{q+1}(C). This orbit is a smooth locally closed complex subvariety, so the Hilbert map factors holomorphically through it. For the algebraic stabilizer G of that orbit, the projection G_0 -> G_0/G has local holomorphic sections; lifting locally gives a holomorphic product B x F, so the family is associated to a holomorphic principal G-bundle. Since T is contractible and paracompact, that bundle is topologically (hence holomorphically, by the Oka principle) trivial. So U = T x F.
Grauert 1960, Ein Theorem der analytischen Garbentheorie, Satz 5: https://www.numdam.org/item/PMIHES_1960__5__5_0.pdf
Douady 1966, Le probleme des modules pour les sous-espaces analytiques compacts: https://www.numdam.org/article/AIF_1966__16_1_1_0.pdf
Grothendieck 1961, Seminaire Bourbaki, Expose 221: https://www.numdam.org/article/SB_1960-1961__6__249_0.pdf
Forstneric-Prezelj 2000: https://arxiv.org/abs/math/0305238
(Standard argument in sections/compact.tex, Proposition cmp:product.)

Exercise: The Kollar-Pardon classification assembled. What is the pattern of the full classification proof?
Combine: (a) Bierstone-Milman functorial resolution to get the smooth semialgebraic open set V subset Y with deck action; (b) Ivashkovich's local pseudoconvexity of V; (c) boundary windows: algebraic plaques, contact quotient with Darboux coordinates, period matrix, first quadratic epigraph, recursive tower of matrix inequalities, bounded-path localization with attached discs; (d) the intrinsic quotient theorem making M = P and the descended f_X : X~ -> M equivariant; (e) Campana core plus compact special abelianity of OpenAI 2026 plus Liouville making the transverse image discrete and cocompact; (f) Selberg + adjoint making it free after finite cover; (g) Bergman + Kodaira + Nadel-Frankel making M a bounded symmetric domain; (h) the compactifiable-cover lemma making the fibre kernel virtually abelian; (i) relative Albanese + Kollar-Pardon bundle theorem making U -> T = M x C^a proper flat with simply connected compact normal projective fibres; (j) Chow-space rigidity plus polarized-isomorphism classes plus Oka principle making the compact fibres all isomorphic and the family holomorphically trivial. Assembling gives X~ = D x C^m x F.
(Standard argument in main.tex, Section sec:assembly.)

Exercise: The converse direction. What goes wrong if you try to prove the classification without the easy direction?
The easy direction is not obvious -- it is where the three factors actually produce a semialgebraic projective-open presentation. If X~ = D x C^m x F, the bounded symmetric factor D embeds in its projective compact dual D^v by the Borel embedding; the affine C^m embeds in P^m; and the compact F already sits in a projective variety. The product D x C^m x F sits therefore as a semialgebraic open subset of a projective variety. The content is only the fact that D's Borel embedding is semialgebraic -- which is what real quantifier elimination on the adjoint group orbit provides. Without this step, the classification would only give "X~ = D x C^m x F implies X~ is a bounded domain modulo a compact factor", which is not the same statement.
Borel 1954: https://doi.org/10.1073/pnas.40.12.1147
Milne 2005: https://www.jmilne.org/math/xnotes/svi.pdf
Falbel-Guilloux-Will 2024 (author version 2024-03-27), Hilbert metric for bounded symmetric domains, Section 2.1.1: https://www-fourier.ujf-grenoble.fr/~will/Hilbert-BoundedSymmetric3.pdf
(Standard argument in sections/compact.tex, Proposition cmp:converse.)

Exercise: Remmert reduction. Why is U holomorphically convex, and what is its Remmert reduction?
Under the classification, write U = X~ = B x F with B = D x C^m the Stein base and F connected compact normal projective. Projection p : U -> B is proper and surjective. For every open V subset B and every h in O_U(V x F), fibrewise constancy (holomorphic functions on a connected compact complex space are constant) gives h(b, z) = h(b, z_0) for some fixed z_0 in F. So the restriction along the holomorphic section b -> (b, z_0) inverts pullback, giving O_B -> p_* O_U an isomorphism. For every compact K subset U:
  hat{K}_{O(U)} = p^{-1}( hat{p(K)}_{O(B)} ),
which is compact because B is Stein and p is proper. Thus U is holomorphically convex, and p is its Remmert reduction. U is Stein iff F is a point: if F has two distinct points, global holomorphic functions cannot separate them; the point case gives U = B Stein directly.
Demailly 2012, Complex Analytic and Differential Geometry, Corollary II.5.8: https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf
(Standard argument in main.tex, Corollary cor:remmert-semialgebraic.)

Exercise: Singer vanishing. What are the L^2 Betti numbers of a smooth aspherical projective X when U is semialgebraic projective-open?
Let X be a smooth closed aspherical projective complex n-fold whose ordinary universal cover X~ = U is semialgebraic projective-open. Its L^2 Betti numbers satisfy b_j^(2)(X) = 0 for all j != n, and if m > 0 in the classification X~ = D x C^m x F, then b_j^(2)(X) = 0 for all j >= 0. The proof has two cases:
m > 0: take the characteristic free abelian subgroup K' = Z^{2m} normal in Gamma from the previous construction. This is an infinite normal amenable subgroup of Gamma, so Cheeger-Gromov's L^2 vanishing theorem kills every group L^2 Betti number; finite-cover proportionality transfers this to X~.
m = 0: then U = D, complete symmetric of noncompact type. Olbrich's L^2 cohomology of the universal symmetric space has harmonic kernels in dimension n and only n, and the deck group acts freely and cocompactly by Bergman isometries, so the cocompact L^2 de Rham identification gives the stated vanishing.
Cheeger-Gromov 1986, L_2-cohomology and group cohomology, Theorem 0.3(3): https://www.ihes.fr/~gromov/wp-content/uploads/2018/08/549.pdf
Olbrich 2002, L^2-invariants of locally symmetric spaces, Proposition 1.2 and Section 4: https://arxiv.org/abs/math/0009039
(Standard argument in main.tex, Corollary cor:singer-semialgebraic.)

Exercise: Quasi-projective covers. What does the classification say when X~ is quasi-projective rather than semialgebraic projective-open?
For a connected normal projective X over C with ordinary universal cover U, the following are equivalent:
(i) U is biholomorphic to a quasi-projective variety;
(ii) some finite etale Galois cover X' -> X admits a morphism to an abelian variety A which is a locally trivial holomorphic fibre bundle with simply connected projective fibre;
(iii) U is biholomorphic to C^m x F for some m >= 0 and some simply connected projective variety F.
More generally, for a connected infinite etale Galois cover V -> X of complex spaces with V quasi-projective, there are a finite etale Galois X' -> X, a bundle morphism alpha : X' -> A to an abelian variety, and an etale cover A~ -> A with no positive-dimensional compact analytic subvarieties, such that V -> X' is biholomorphic as a covering to X' x_A A~ -> X'. No algebraicity of the deck action is assumed.
Claudon-Horing-Kollar 2013, Algebraic varieties with quasi-projective universal cover: https://doi.org/10.1515/crelle.2012.017
(Standard argument in sections/quasiprojective.tex, Corollary cor:quasiprojective.)

Exercise: How does the quasi-projective corollary follow from smooth abundance?
For every smooth projective complex variety Y with nef K_Y, the log-abundance theorem in characteristic zero applies in zero boundary, so K_Y is semiample. That is exactly the smooth-abundance premise demanded by Claudon-Horing-Kollar, so their Theorem 1.1 and Corollary 1.5 give the two quasi-projective statements. The finite etale Galois cover X' -> X is described as a locally trivial bundle over A, not necessarily a global product, and the infinite-cover statement only asserts the pullback direction V = X' x_A A~ -> X'.
OpenAI 2026, Log abundance in characteristic zero, Theorem 1.1: log_abundance_char_zero_original.txt
(Standard argument in sections/quasiprojective.tex, proof of Corollary cor:quasiprojective.)

Exercise: Affine space uniformization. What is the classification when X~ = C^n?
If X is a smooth connected projective n-fold whose ordinary universal cover is biholomorphic to C^n, then there is a finite etale Galois cover A -> X where A is an abelian variety. Proof: the simply connected projective factor F of the bundle in Corollary cor:quasiprojective(ii) lifts to a holomorphic embedding in C^n. Every coordinate function is constant on the connected compact complex space F, so F is a point; then the bundle morphism X' -> A is an isomorphism. This is the Iitaka uniformization statement, derived from smooth abundance via Claudon-Horing-Kollar.
Nakayama 1999, Projective algebraic varieties whose universal covering is C^n: https://doi.org/10.2969/jmsj/05130643
Claudon-Horing-Kollar 2013: https://doi.org/10.1515/crelle.2012.017
(Standard argument in sections/quasiprojective.tex, Corollary cor:affine-uniformization.)

Exercise: Bibliographic map. What are the references for each main tool of the proof, in order of use?
Pi_1 / cover -- Hatcher (Algebraic Topology). Projective and normal varieties -- Hartshorne, Kollar. Semialgebraic -- Bierstone-Milman 1988, Coste 1999. Main conjecture -- Kollar-Pardon 2012, arXiv 1104.2309v2. Functorial resolution -- Bierstone-Milman 2008. Local pseudoconvexity -- Ivashkovich 2008. Levi form & attached discs -- Bishop 1965, Hill-Taiani 1978. Puiseux preparation -- Lion-Rolin 1997, Bierstone-Milman 1988. Thin omission -- Vajaitu 2000. Bochner tube & Bergman subharmonicity -- Noguchi 2020, Berndtsson 2006. Guard / Oka principle -- Grauert 1958, Forstneric-Prezelj 2000. Guard boundary compactness -- Arzela-Ascoli, Montel (textbook). Campana core -- Campana 2004 (math/0110051v9). Compact special abelianity -- OpenAI 2026. Liouville -- Lyons-Sullivan 1984. Selberg -- Selberg 1960. Bergman & Kodaira -- Kodaira 1954. Nadel-Frankel -- Farb-Weinberger 2008. Albanese & Hodge variation -- Griffiths 1968, Claudon 2018. Normal Albanese -- Claudon-Horing-Kollar 2013, Lemma 4.7. Bundle theorem -- Kollar-Pardon 2012 arXiv v2 Thm 20. Cycle spaces -- Barlet 1999, Andreotti-Norguet 1967. Cartan B & Oka principle -- Cartan 1952 (Semin. Cartan Expose 18), Grauert 1958. Picard injectivity -- Kleiman 2005, Serre GAGA 1956. Douady & Hilbert -- Douady 1966, Grothendieck 1961 (Bourbaki 221). Cheeger-Gromov -- Cheeger-Gromov 1986. Olbrich -- Olbrich 2002. Quasi-projective -- Claudon-Horing-Kollar 2013. Log abundance -- OpenAI 2026. Affine unif -- Nakayama 1999.

Exercise: What is the answer to "what is a semialgebraic projective-open universal cover" in one sentence?
A connected normal projective X whose ordinary universal cover X~ is biholomorphic to a D x C^m x F factor (D bounded symmetric, F simply connected normal projective, m >= 0) -- equivalently, X~ sits semialgebraically as an open subset of a projective variety. This is the Kollar-Pardon classification, conjectured in 2012 and proved for normal projective X in 2026 using log abundance and the abelianity of compact special Kahler fundamental groups.

Exercise: What are the exact hypotheses in the two algebraic inputs one cannot drop?
The bounded symmetric domain D must be of noncompact type (Hermitian symmetric, semisimple adjoint group) to make the Borel embedding semialgebraic -- compact symmetric spaces would fail boundedness. The simply connected normal projective factor F may not be replaced by a Fano, Calabi-Yau, or rationally connected subcase: any simply connected normal projective is allowed, and its singularities may be essential to the bundle structure over D x C^m. Similarly, the deck action on X~ is not required to be algebraic, because the classification concerns the biholomorphic type of X~ and not the specific deck action realizing it.
(Standard argument in sections/compact.tex, Lemma cmp:characteristic; Kollar-Pardon 2012: https://arxiv.org/abs/1104.2309v2)

Exercise: Open kernel. How is the "open kernel" of a scaled family of open sets defined, and why is the neighbourhood quantifier essential?
Let U_t be open subsets of a fixed manifold M for 0 < t < t_0. The open kernel is the set of points z such that some neighbourhood W of z satisfies W subset U_t for every sufficiently small t. Equivalently, in terms of closed complements F_t = M minus U_t, the open kernel is M minus limsup F_t. The neighbourhood quantifier is essential: pointwise eventual membership undercounts. What open kernels capture is the exact scaling limit of the boundary, and by the Hartogs-figure criterion an open kernel of locally pseudoconvex open sets is locally pseudoconvex.
Fritzsche-Grauert 2002, From Holomorphic Functions to Complex Manifolds, II.3.4 and II.3.7: https://doi.org/10.1007/978-1-4683-9273-6
(Standard argument in sections/preliminaries.tex, Definition prep:kernel.)

Exercise: Generic real parameters. What is meant by the phrase "at generic real parameters", and why does it suffice to discard a countable union of smaller-dimensional sets?
An assertion at generic real parameters in an open real box means it holds outside a locally finite union of subanalytic subsets of smaller dimension. Countably many fixed compact boxes or positive accuracies are tested one at a time, and each may be granted its own exceptional set. Then any finite application of the assertion can be restricted to a smaller neighbourhood of a chosen real parameter on which the finitely many data and their inverses are uniformly controlled. The countable accumulation of exceptions is allowed because the parameters are tested one box at a time; no uniform neighbourhood for all of a countable list is ever needed.
Bierstone-Milman 1988, Theorems 0.1, 0.2 and 6.4: https://www.numdam.org/item/PMIHES_1988__67__5_0.pdf
Lion-Rolin 1997, Theoreme de preparation, Theorem 1: https://aif.centre-mersenne.org/item/10.5802/aif.1583.pdf
(Standard argument in sections/preliminaries.tex, Definition prep:generic.)

Exercise: Power bounds and Puiseux preparation. What does the Puiseux-preparation lemma guarantee, and how is it used?
If f(v, t) is globally subanalytic for v in A and 0 < t < eps(v), and if f(v, t) > 0, then there is an integer N such that, for every fixed v, t^N < f(v, t) < t^{-N} for all sufficiently small t. The threshold may depend on v. For finitely many such functions one may use a common N. Additionally, near a generic parameter in a full-dimensional box, a finite collection of such functions admits joint convergent Puiseux expansions with a common denominator q and real-analytic coefficients, so after multiplying by a suitable power of t, they are jointly analytic in (v, t^{1/q}) through t = 0.
Lion-Rolin 1997: https://aif.centre-mersenne.org/item/10.5802/aif.1583.pdf
Bierstone-Milman 1988: https://www.numdam.org/item/PMIHES_1988__67__5_0.pdf
(Standard argument in sections/preliminaries.tex, Lemma prep:puiseux.)

Exercise: Finite-jet specialization. Why does the open kernel of a polynomial test family depend only on finitely many Taylor coefficients of the scaling map?
In bounded projective charts, suppose a family of sets is described by finitely many Boolean combinations of polynomial equations and inequalities in the fibre variable y, with real-analytic coefficients in the real parameter b. Substitute b = x + R_t h, where R_t tends to zero with convergent Puiseux expansions and bounded inverse powers. Then the upper limits of these sets, and the open kernels of their complements, are semialgebraic in (h, y) on the central fibre: on any fixed bounded h-box, sufficiently many Taylor coefficients of the data at x are all that matter, because the subanalytic distance inequality for a closed set gives a Puiseux power bound on distance in terms of the residual slack of equations and inequalities.
Bierstone-Milman 1988, Theorems 0.1 and 6.4: https://www.numdam.org/item/PMIHES_1988__67__5_0.pdf
(Standard argument in sections/preliminaries.tex, Lemma prep:finite-jet-specialization.)

Exercise: Generic convergence of distance tests. Why is the limit of the distance-to-complement of a subanalytic family continuous at generic real centres?
Give a bounded globally subanalytic family d_t(x, z) on K, uniformly Lipschitz in z, and define the extension by the pointwise limit at t = 0. Then there is a relatively closed subanalytic subset E subset U of dimension less than dim U such that the extension is jointly continuous at every (x, z, 0) with x outside E. Uniformity on compact x subsets is a finite-net argument plus definable boundedness; the exceptional set is the set of x where the near-boundary shape can shift, and one shows that this set has no interior via a finite-quantifier subanalytic encoding.
Bierstone-Milman 1988, Theorem 6.4: https://www.numdam.org/item/PMIHES_1988__67__5_0.pdf
(Standard argument in sections/preliminaries.tex, Lemma prep:generic-distance-convergence.)

Exercise: Power diagonalization (revisited). Why can any two-variable bounded subanalytic double limit be captured on the diagonal with a sufficiently high power?
Let f(e, t, v) be bounded globally subanalytic with iterated limit f_{00}(v) = lim_t->0 lim_e->0 f(e, t, v). Then for every integer N > N_0 (a fixed N_0 depending on f), lim_t->0 f(t^N, t, v) = f_{00}(v) for each fixed v. The idea: the e-threshold of convergence is a definable function of (v, delta), bound below by a power of t by Lemma prep:puiseux, and choosing N above that power makes the diagonal fall inside the threshold. The parameter v may include box sizes and positive accuracies; finitely many tests share one N.
Bierstone-Milman 1988: https://www.numdam.org/item/PMIHES_1988__67__5_0.pdf
(Standard argument in sections/preliminaries.tex, Lemma prep:power-diagonalization.)

Exercise: Polynomial maximum over the holes. Why does the maximum of |g| over a compact hole fibre produce a plurisubharmonic function?
Let B be a complex manifold and J subset B x C be closed with J -> B proper, locally bounded in the line coordinate, and (B x C) minus J locally pseudoconvex. For a monic polynomial g(b, t) of positive degree in t with holomorphic coefficients, the function psi_g(b) = log max over t in J_b of |g(b, t)| is plurisubharmonic, with value -infinity if the fibre is empty or the maximum is zero. The proof is that the map (b, t) -> (b, g(b, t)) is finite and proper, and its image of J is a proper over-graph, so the inverse of the "radius of the largest vertical disc" is plurisubharmonic by the Hartogs-radius criterion.
Demailly 2012, Complex Analytic and Differential Geometry, Section I.7.2: https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf
(Standard argument in sections/preliminaries.tex, Lemma holes:polynomial-maximum.)

Exercise: Boundary trapping. How does a divisor that extends across a real boundary segment trap the projective holes?
Let a proper holomorphic submersion be given near a relatively compact parameter interval I on the boundary of a one-sided complex disc. Suppose G is locally pseudoconvex on that side and every limit of its fibre holes as the transverse parameter approaches I belongs to a relative divisor D that extends holomorphically across I and is proper in each fibre. Then after restriction near an interior point of I, the holes on that side belong directly to D. The proof uses Weierstrass cylinders around D to reduce to a subharmonic maximum on a small disc slice, and the limit hypothesis confines the holes to a strictly smaller vertical disc. The same argument with fixed additional parameters gives a locally uniform version.
Demailly 2012: https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf
(Standard argument in sections/preliminaries.tex, Lemma holes:boundary-trapping.)

Exercise: Interior trapping. How does a nonpluripolar parameter slice propagate the same divergence containment?
Suppose G is open and locally pseudoconvex over a whole connected base box, subject to the preceding local confinement hypotheses. If a relative divisor D is holomorphic in the box and contains the holes over a nonpluripolar parameter slice inside the box, then it contains the holes nearby; in particular for a real-analytic maximally totally real slice. The proof reduces to the same Weierstrass cylinders: for fixed remaining fibre coordinates, psi_g is subharmonic and equals -infinity on a nonpluripolar set, so by the one-variable uniqueness principle applied successively to the real intervals of a real coordinate box, it is identically -infinity.
Demailly 2012: https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf
(Standard argument in sections/preliminaries.tex, Lemma holes:interior-trapping.)

Exercise: Tangency propagation. How does a divisor containing all near-boundary holes force it to contain the whole connected projective fibre minus a point?
Suppose q : Q -> Delta is a holomorphic submersion, G subset Q is open and locally pseudoconvex, A is a divisor in Q_0, and Q_0 minus A is connected. Suppose that on one side of a regular embedded real-analytic arc through 0, every limit of holes belongs to A, and G_0 is nonempty. Then Q_0 minus A subset G_0. The proof proceeds through finitely many submersion boxes along a compact path: in each box, choose a transverse disc internally tangent to the arc; the Hartogs-radius maximum principle forces a strict inequality at the tangent point to spread to a strict inequality everywhere in the box, and successive boxes cover the path.
Demailly 2012, Section I.7.2: https://www-fourier.univ-grenoble-alpes.fr/~demailly/manuscripts/agbook.pdf
(Standard argument in sections/preliminaries.tex, Lemma holes:tangency-propagation.)

Exercise: Thin pseudoconcave set. Why does a codimension at-least-two subanalytic set with Stein complement have to be analytic?
Let M be a complex manifold of dimension n and A subset M a closed subanalytic set of real dimension at most 2n - 2. If M minus A is locally pseudoconvex, then A is either empty or a pure complex-codimension-one analytic set. The dimension hypothesis supplies finite Hausdorff (2n-2)-measure for the analyticity theorem. For n > 1, application of Vajaitu's thin-pseudoconcave-set theorem with q = 1 gives analyticity; for n = 1, a zero-dimensional closed subanalytic set is locally finite and hence a divisor. This is the step that converts "we know the omission is small" into "the omission is genuinely an analytic hole".
Vajaitu 2000, Analyticity of q-concave sets, Theorem 1: https://aif.centre-mersenne.org/item/10.5802/aif.1789.pdf
Hirschowitz 1973, Entre les hypersurfaces et les ensembles pseudoconcaves, Theorem 1.4: https://www.numdam.org/article/ASNSP_1973_3_27_4_873_0.pdf
(Standard argument in sections/preliminaries.tex, Theorem holes:thin-analyticity.)

Exercise: Bounded algebraic containers. How does one bound the degree of a Zariski container of a bounded-data semialgebraic family of fibres?
For a smooth proper projective family with connected fibres and a bounded-data semialgebraic family of subsets S_b of the fibres, there is a uniform degree bound D (depending only on the family description) such that if any S_b is contained in a proper complex algebraic subset of its fibre, then it is contained in an effective divisor of degree at most D. The set of parameters admitting such a container is subanalytic. The proof uses real-algebraic cell decomposition, subanalytic parameter-dependence of Nash pieces, and a compactness argument on the projective Chow variety: containers can be selected subanalytically in the parameter.
(Standard argument in sections/preliminaries.tex, Lemma holes:bounded-containers.)

Exercise: Projective-fibre propagation. Why is a locally-pseudoconvex, bounded-data semialgebraic fibrewise family G automatically "all but a divisor" on each fibre?
Let q : Q -> B be a smooth, proper, projective family with connected smooth fibres, and let G subset Q be open, subanalytic, locally pseudoconvex, and fibrewise semialgebraic with bounded data. Let S = q(G) with a connected component S_0. Suppose G_b contains a Zariski open dense subset of Q_b for every b in a nonempty open subset of S_0. Then J = q^{-1}(S_0) minus G is an analytic divisor (possibly empty), and its intersection with each fibre is proper, so every G_b contains a Zariski open dense subset of Q_b. The proof separates generic hypersurface walls using tangency propagation and bounded containers, then applies thin-pseudoconcave-set analyticity on the codimension-two residual set.
(Standard argument in sections/preliminaries.tex, Proposition holes:projective-propagation.)

Exercise: Container propagation. Once a relative divisor contains the fibres over some open piece of a parameter box, why does it contain them everywhere?
In the situation of Proposition holes:projective-propagation, let D be a relative analytic divisor defined on the full parameter box and proper in every fibre. If J subset D over a nonempty open subregion of S_0, then J subset D over all of S_0. The proof is the analytic identity theorem applied to the irreducible components of J: every component H of J dominates S_0 (otherwise dimension of H minus dimension of the generic fibre = 1, so H has full-dimensional generic fibre = Q_b, contradicting G_b nonempty); therefore H meets the open subregion; the local equations of D vanish on a nonempty open subset of H and hence on H by analytic identity. Serial application to every component proves the statement.
(Standard argument in sections/preliminaries.tex, Corollary holes:container-propagation.)

Exercise: Section test. How does a section avoiding a divisor detect the projection domain of a pseudoconvex family?
If a full-box container D is available (as in container propagation), a local holomorphic section sigma of q that avoids D lies in G at every parameter of S_0. Near a boundary point of S_0 that lies in the parameter box, S_0 is locally pseudoconvex wherever such a section exists. The proof notes that sigma^{-1}(G) is locally pseudoconvex, and the local component of S_0 near sigma coincides with a local component of sigma^{-1}(G) by path-lifting in the submersion boxes.
(Standard argument in sections/preliminaries.tex, Corollary holes:section-test.)

Exercise: The projective-windows theorem. What does Theorem win:construction assert, as one statement?
Given a connected smooth projective Y and a nonempty connected semialgebraic open V subset Y that is locally pseudoconvex at its boundary, there exist, locally at generic real centers in totally real parameter boxes: a complex box B subset C^d, a smooth proper projective family q : Q -> B with connected (dim Y - d)-dimensional fibres, and a full-rank holomorphic evaluation map ev : Q -> Y, algebraic with bounded data. Letting G = ev^{-1}(V), there is a component S of q(G) and a relative proper algebraic divisor D subset Q, proper in every fibre, such that Q_b minus D_b subset G_b for b in S. The actual holes Q|_S minus G are analytic. There is a fixed interior seed p, real block-diagonal matrices R_t(x) with bounded polynomial inverses, and a marked component P of the open kernel of S_{t,x} = R_t(x)^{-1}(S - x). P is a recursive tower (point, then add (s, w) in C x C^r over the previous tower with Im M(s, w, z) > 0). The scaling ratios R_t(x)^{-1} R_t(x + R_t(x) u) converge polynomially to a block-diagonal G_x(u) that fixes the origin and gives an automorphism of P. Finally, endpoints of any bounded path in S_{t,x} stay in the closure of P.
(Standard argument in sections/windows.tex, Theorem win:construction.)

Exercise: Controlled replacement paths. What does loc:paths supply, and why is the complex extension of the paths important?
Given a bounded subanalytic family of open sets D_{x,t} subset R^n on an open base U subset R^k, and two bounded subanalytic sections v, w in the same component of D_{x,t}, there are locally (at a generic x_0) a neighbourhood U', a threshold eps' > 0, and paths F_t : U' x [0, 1] -> R^n with endpoints v(x, t) and w(x, t), entirely in D_{x,t}, with uniform bounds: for each fixed t, F_t extends holomorphically to a fixed complex neighbourhood U'_C of U', and there is a uniform Lipschitz bound on both F_t and its complex neighbourhood, driving the Cauchy estimates for any fixed number of derivatives. The complex extension is what makes the attached-disc argument possible: the defining function of D_t can then be expanded in a Taylor series around the segment in the complex direction and the various blocks arranged into a triangular system.
(Standard argument in sections/localization.tex, Lemma loc:paths.)

Exercise: Semidefinite outer tests. What does loc:matrix-outer assert, and how is it applied?
Suppose additionally that the marked kernel is described recursively by strict positive-definiteness tests for finitely many continuous Hermitian matrix functions M_j(Z), and that every point of the closure of P satisfies M_j(Z) >= 0. Then for each fixed finite L, at a generic fixed center, the infimum over Z in the accessible component of P at scale t and radius L of the smallest eigenvalue of M_j(Z) is >= -o(1) as t -> 0, simultaneously for all j, uniformly over the accessible component in that fixed box. The point is: the closure membership gives semidefinite tests, and an upper-bound on the least eigenvalue follows from the accessible component being inside P. The proof uses bounded-compactness plus the localization theorem: if a negative minimum appeared, a bounded subsequence would have a limit point in the closure of P contradicting continuity.
(Standard argument in sections/localization.tex, Corollary loc:matrix-outer.)

Exercise: Finite Taylor extraction. Why does a Puiseux-bounded matrix expression have a polynomial holomorphic limit on every bounded box?
Suppose A_t(x) and A_t(x)^{-1} have joint Puiseux expansions with polynomial bounds in t and R_t(x) has the same property tending to zero. If f is holomorphic on a fixed complex neighbourhood, then any expression obtained by multiplying f(x + R_t z) on both sides by such Puiseux matrices and subtracting a bounded constant term has a polynomial holomorphic limit on every bounded z-box, provided it is bounded on one nonempty complex open z set. Convergence is in the C^k topology for every fixed k. The proof Taylor-expands f through a large order, uses the positivity and Puiseux bounds to kill all negative powers of v = t^{1/q}, and uses boundedness on one nonempty open complex z set to force the lowest-order coefficient to vanish identically as a polynomial.
(Standard argument in sections/windows.tex, Lemma win:finite-taylor.)

Exercise: Relative divisor container. How does one build a relative divisor containing a proper algebraic subset of every fibre of a projective family?
Let q : Q -> B be a smooth projective family with connected fibres, and let A subset Q be a relative algebraic subset with analytic coefficient data, proper in every fibre. Near a generic parameter, there is a relative effective divisor D whose restriction to every fibre is proper and contains A. Restrict near a parameter where A is flat over the base; the ideal sheaf I_A is also flat; in a relative projective embedding, choose a twist I_A(k) generated with vanishing higher cohomology, then use Grauert's constant-cohomology theorem to make the direct image locally free and base change well. Choose a section of the twisted ideal whose central restriction is not identically zero; its zero divisor contains A and is proper in every nearby fibre.
Grauert 1960, Ein Theorem der analytischen Garbentheorie, Saetze 3 and 5: https://www.numdam.org/item/PMIHES_1960__5__5_0.pdf
Grauert 1963, Berichtigung, p. 36: https://www.numdam.org/item/PMIHES_1963__16__35_0.pdf
(Standard argument in sections/preliminaries.tex, Lemma holes:relative-divisor-container.)

Exercise: The proof-overview picture. What does the diagram in figures/proof-overview.tex record?
The top row shows V (smooth resolution of X~) --mu--> U = X~ (normal universal cover) --f_X--> M ~= P (intrinsic quotient). An additional arrow f: V -> M skips directly to M by "projective windows and guards". The bottom row shows T ~= M x C^m (affine base) with arrow j : V -> T "uses K virtually abelian, compact fibre F", and D (bounded symmetric domain) with projection T -> M. A dashed arrow from M to M sums up "core and abelianity; discrete action and splitting". In Prose: the proof composes (i) boundary-window construction of M = P from V, with (ii) transverse symmetry giving M = D, with (iii) relative Albanese giving V -> T, with (iv) product triviality giving V = D x C^m x F.
(Standard argument in figures/proof-overview.tex.)

Exercise: Two recent results the paper does NOT use, but which locate it in the literature. What are they?
Rogov proves a metric form of the Kollar-Pardon statement for smooth aspherical projective varieties with definable lifted Kahler metric: that metric hypothesis supplies more information than a semialgebraic presentation, and the theorem is not enough for the current classification. Correa-Kollar-Schreieder-Wang construct examples with finite-CW-type universal covers whose Albanese maps are not differentiable fibre bundles, showing that finite-CW-type alone cannot substitute for the entire semialgebraic projective-open pullback in the Kollar-Pardon bundle theorem. The paper's application of that bundle theorem retains the entire semialgebraic pullback.
Rogov 2024, On the metric Kollar-Pardon problem, Theorem C: https://arxiv.org/abs/2401.07983v1
Correa-Kollar-Schreieder-Wang 2026, Invisible singularities, Corollary 30: https://arxiv.org/abs/2608.10973v1
(Standard argument in sections/history.tex.)

Exercise: When is X~ already a topological product, but not the biholomorphic one? What does the compact-Kahler compactification splitter say?
If a smooth projective X has universal cover X~ = P x C^m where P is compact Kahler, the symplectic form on P forced by the product plus holonomy makes P algebraically simply a Kuga-Satake situation; the harder question is when P is projective and the bundle is globally biholomorphic. Claudon-Horing (with Campana) prove that for a compactifiable Kahler cover, the fibration is a fibre bundle and the Albanese map splits after a finite cover. Log-abundance upgrades this to the quasi-projective case. The semialgebraic theorem of this paper is stronger: it drops the compactifiability of X~, keeping only the semialgebraic projective-open presentation.
(Standard argument in sections/history.tex.)

Exercise: What is the single most delicate technical step, in the paper's own view?
The most delicate step is Theorem loc:theorem (localization): scalar endpoints of bounded paths in a near-boundary scaling limit are controlled using attached discs above a sequence of triangular recenterings. The complication is that the "limit of scaled domains" (open kernel) is smaller than the pointwise limit, because the paths could oscillate inside the boundary face; the triangular recenterings and their Gram matrices G_x are exactly what force path endpoints to remain in the closure of the marked component. The whole tower P is built by applying this step at each stage; the guard-decay / incidence-trace construction cannot be done without this accessibility assertion.
(Standard argument in sections/localization.tex, Theorem loc:theorem.)