6fbd3ac Δ f6bf760 Δ 763cb89 Δ 7be0029 Δ 112ee1e Δ 2d5618e Δ c457276
editors: nobody
The index of a log Calabi-Yau pair is the smallest positive integer m
for which the rationally trivial log canonical class becomes an actual
trivial line bundle. A uniform index theorem asks for a single value of
m depending only on the dimension and on the finite set of boundary
coefficients, any instance of the class being allowed to have as many
irreducible components as it likes. The main result provides such a
bound in every fixed dimension at least four, over any algebraically
closed field of characteristic zero.

A divisor B on a scheme X is a finite formal sum of codimension-one
closed subvarieties with rational coefficients; it is effective,
written B >= 0, when every coefficient is nonnegative. A scheme is
demi-normal if it is reduced, satisfies Serre's condition S_2, is
seminormal, and has only nodes as codimension-one singularities. Take
X projective, equidimensional, and demi-normal, and take B >= 0 a
rational Weil divisor with no component in the nonnormal locus of X.
Write nu : X-bar -> X for the normalization and C-bar for the
conductor divisor. The pair (X, B) is semi-log-canonical, abbreviated
slc, if K_X + B is a rational Cartier divisor and if the pair

   ( X-bar , C-bar + nu_*^{-1} B )

is log canonical. These are the singularities that allow normal
components to meet along a double locus while keeping adjunction and
pluricanonical descent (Kollar 2013):
https://doi.org/10.1017/CBO9781139547895

Such a pair is log Calabi-Yau when K_X + B is rationally trivial: for
some positive integer m the divisor m (K_X + B) is a principal
Cartier divisor with trivial associated line bundle. The index is the
least such m.

Theorem A (uniform slc index). Fix an integer d >= 4 and a finite
subset Phi of Q cap [0,1]. There is a positive integer a(d,Phi) with
the following property. Let (X, B) be a connected projective
equidimensional slc log Calabi-Yau pair of dimension d over an
algebraically closed field k of characteristic zero, every nonzero
coefficient of the boundary B lying in Phi. Then a(d,Phi)(K_X + B) is
a Cartier divisor, and O_X(a(d,Phi)(K_X + B)) is the trivial line
bundle on X.

The number a(d,Phi) depends only on d and Phi. It does not depend on
the number of irreducible components of X, and it does not depend on
the number of log canonical strata on any birational model. Together
with the theorem in dimensions at most three of Jiang-Liu 2021,
Theorem A settles the finite-rational-coefficient slc index
conjecture:
./jiang-liu-2002.11928-bounded-pluricanonical-representations.pdf

For orientation, several earlier results cover restricted cases. The
zero-boundary analogue of abundance for slc threefolds is Fujino
2000:
https://doi.org/10.1215/S0012-7094-00-10237-2
The klt threefold case of the index theorem is Jiang 2021:
https://doi.org/10.1090/jag/759
The global index theorem for non-klt lc threefold pairs with
standard coefficients is Ishii 2000:
https://arxiv.org/abs/math/0003060
Numerically trivial log canonical divisors on projective slc pairs
are rationally linearly trivial by Gongyo 2013, but the trivializing
multiple there may depend on the pair:
https://doi.org/10.1090/S1056-3911-2012-00593-1
Boundedness is the missing ingredient. Fujino-Gongyo 2014 proved that
the log pluricanonical representation of any projective lc pair with
semiample log canonical divisor is finite; uniform boundedness of the
finite images is the residual point:
https://doi.org/10.1112/S0010437X13007495

Xu 2019 organized reductions among the normal index conjecture, the
slc version, and boundedness of pluricanonical representations. His
Theorem 1.11 records the implication from normal indices in dimension
d and bounded representations in dimension d-1 to slc indices in
dimension d; his Theorem 1.14 treats klt fourfold pairs with nonzero
boundary:
./xu-1905.00297-index-conjecture-log-cy.pdf
Jiang-Liu 2021 bounded the representations of lc surfaces with a
fixed trivializing multiple and derived slc indices in dimension at
most three plus the non-klt lc case in dimension four:
./jiang-liu-2002.11928-bounded-pluricanonical-representations.pdf
Birkar proved the index conjecture for rationally connected klt
pairs in every dimension with a fixed rational DCC coefficient set,
Corollary 1.8 of his 2025 paper:
./birkar-2305.18770-fano-fibrations-beyond.pdf

Several recent contributions isolate subfamilies. The coregularity of
a log pair (X, B) is the dimension of a klt fibre of a dlt model of
the normalization with its induced boundary; zero coregularity means
that such a minimal lc stratum has dimension zero. Filipazzi, Mauri,
and Moraga obtained sharp index bounds for coregularity zero,
including slc pairs (FMM 2025):
https://doi.org/10.2140/ant.2025.19.383
Figueroa, Filipazzi, Moraga, and Peng developed the corresponding
index and complement theory in low coregularity (FFMP 2025):
https://doi.org/10.1017/fms.2024.69
Masamura proved boundedness of the indices of smooth Calabi-Yau
fourfolds (Masamura 2025, Corollary 1.6):
https://doi.org/10.1093/imrn/rnaf114
Uniform bounds need not be small: smooth Calabi-Yau varieties exist
whose indices grow doubly exponentially with the dimension (Singh
2026):
https://doi.org/10.1007/s00209-026-03984-y

The proof of Theorem A uses the normal-to-slc strategy and reduces to
a character bound on klt pairs. Two substantial companion inputs are
taken as given. The uniform normal log canonical index theorem
(Theorem 1.2 of Uniform Pluricanonical Iitaka Fibrations):
uniform_pluricanonical_iitaka_v2.txt
supplies a single even integer m, depending only on d and Phi, such
that on every component (Xi, Delta_i) of the normalization the
divisor m(K_{Xi} + Delta_i) is a trivial principal divisor. The
all-dimensional good-model theorem of Log abundance in characteristic
zero (Theorem 11.1) is used throughout:
log_abundance_char_zero_v2.txt

Fix such an m. On each Xi choose a rational m-canonical form theta_i
with Div(theta_i) = -m Delta_i; all such forms differ by a nonzero
constant. Around a generic node xi of X, its two branches give a
ordered pair of forms theta_i and theta_j, and after passing to a
projective crepant Q-factorial dlt model Yi -> Xi the branch points
sit on two dlt strata of common dimension. Pulling back the two forms
to those strata and taking logarithmic residues, the two residues
differ by a nonzero constant r_e. The point of introducing this graph
is that existence of a global trivializing multiple M(K_X + B) = O_X,
for some M, forces r_e to be a genuine constant on the conductor
function field. Write G for the finite graph whose vertices are the
indices i of Xi and whose edges are the generic nodes of X, an edge
labelled by r_e. A node whose two branches lie on the same Xi becomes
a loop.

Matching the trivializations across every edge of G requires finding
nonzero scales c_i so that c_i r_e^{m} c_j^{-1} is rational for every
edge e : i -> j. This is possible exactly when the products of the
r_e around every closed walk in G have order dividing m. The integer
that kills every closed-walk product is the uniform index multiple on
X, once m is fixed.

Each closed-walk product becomes a birational self-map of one klt
pair. Minimal log canonical strata of a dlt pair are themselves klt;
iterated adjunction gives them effective dlt pairs whose minimal
strata are exactly the minimal strata of the larger pair. Within a
fixed Yi the Kollar P^1-linking theorem joins any two minimal strata
by a B-birational map that preserves the even-degree residue (Kollar
2016):
./kollar-1107.2863-sources-log-canonical-centers.pdf
Across a node the residue comparison is a separate matter but has the
same conclusion, again with multiplier r_e. Composing along any
closed walk in G therefore yields a B-birational self-map f of a
single projective integral klt pair (A, Gamma_A) of dimension at
most d-1, satisfying

   f^* theta_A = ( prod_{e in the walk} r_e ) theta_A

for the induced trivializing form theta_A on A. The problem is
consequently to bound, in terms of dim V and m alone, the order of
the scalar by which a crepant birational self-map of a projective
integral klt pair (V, Gamma) with m(K_V + Gamma) = O_X acts on a
trivializing pluricanonical form.

Theorem B (uniform klt character exponent). For each integer n >= 0
and m >= 1 there is a positive integer L(n,m) with the following
property. Let (X, Delta) be an integral projective klt n-fold pair
over C with Delta >= 0 and m(K_X + Delta) = O_X (as an integral
principal divisor). Let theta be a rational m-canonical form on X
with Div(theta) = -m Delta. If f is a B-birational self-map of
(X, Delta) and f^* theta = c theta for some c in C^*, then
c^{L(n,m)} = 1.

The transformation f may have infinite order. The bound L(n,m) does
not depend on any finite order of a cover.

The mechanism behind Theorem B is a rank bound for the rational Hodge
structure generated by the holomorphic top forms. For a smooth
projective n-fold U let T(U) be the smallest rational Hodge
substructure of H^n(U, Q) whose complexification contains H^{n,0}(U);
write r(U) = dim_Q T(U). The Hodge structure T(U) is birationally
invariant, is primitive for every rational divisor class, and carries
a polarized integral lattice whose Hodge-Riemann Hermitian form is
positive definite. If a birational self-map of a smooth model acts on
H^{n,0} by a scalar c, then c is an eigenvalue of an integral
automorphism of T(U) whose other eigenvalues are its Galois
conjugates; they all have modulus one. Kronecker's theorem then says
c is a root of unity and its order e satisfies phi(e) <= r(U), where
phi is Euler's totient. Bounding r(U) bounds e.

Theorem C (uniform rank bound). For each integer n >= 2 there is a
constant R_n with the following property. Let T be a projective
Q-factorial terminal complex n-fold with K_T = 0. Assume a smooth
projective resolution U has H^1(U, O_U) = 0, and every
positive-dimensional rational image of T of dimension less than n
has rationally connected smooth projective resolution. Then
r(T) = dim_Q T(U) <= R_n.

Theorem C is used on the irreducible Calabi-Yau and holomorphic
symplectic factors of a product cover, and on those only; abelian
factors are handled by a Betti-number bound and are excluded from
the theorem by the H^1 = 0 hypothesis. A separate boundedness
theorem of Han-Jiang and Jiang-Liu gives the rationally connected
factor bound. The composition produces
L(n,m) = n! lcm_{1 <= r <= n} { E_rc(r,m), E_0(r) }, where E_rc is
the rationally connected exponent  and E_0 is the
boundary-free exponent .

Two further pieces of the proof are used only on the rank bound.
A diagonal estimate: a positive lower bound for the
Seshadri constant of a normalized polarization, combined with a
vanishing top cup product, bounds r(U). Lemma 4 replaces the
Seshadri hypothesis with the weaker rational-image hypothesis of
Theorem C, using covering families, a calibration of a big divisor
against the base, and two section counts. The passage from the diagonal argument to the rank argument uses Ein-Kuchle-Lazarsfeld 1995 as its model:
./ein-kuchle-lazarsfeld-1995-9408003-local-positivity.pdf
The positivity inputs are cotangent semipositivity of Greb-Kebekus-
Peternell 2016 and its refinement by Campana-Paun 2019,
divisorial Zariski decomposition in the numerical form of Nakayama,
Boucksom, and Boucksom-Favre-Jonsson, and Birkar's effective
birationality for Calabi-Yau pairs:
./greb-guenancia-kebekus-2019-1704.01408-klt-trivial-k.pdf
./campana-paun-2015-1303.3169-orbifold-semipositivity.pdf
./birkar-2019-1806.04328-anti-pluricanonical-fano.pdf

The character argument is where the Calabi-Yau structure first bites.
The product cover is the one of Matsumura-Wang 2025 and Horing-
Peternell 2019: after a finite quasi-etale base change the klt pair
splits as a product of a rationally connected factor, an abelian
factor, and irreducible Calabi-Yau or holomorphic symplectic
factors, with boundary supported on the rationally connected factor
alone:
https://doi.org/10.4171/JEMS/1702
https://doi.org/10.1007/s00222-018-00853-2
The factor decomposition also uses the singular Beauville-Bogomolov
theorem of Greb-Guenancia-Kebekus, Druel, and Horing-Peternell:
./greb-guenancia-kebekus-2019-1704.01408-klt-trivial-k.pdf
https://doi.org/10.1007/s00222-017-0748-y
An arbitrary crepant birational self-map need not preserve a product
cover, and this point is the source of the exponent n!. A factor-action lemma below
constructs a further cover that the self-map lifts to, factors the
lifted scalar into scalar factors on the individual factor covers,
and combines the pieces; the n! appears as the exponent that
stabilizes the permutation of primitive central idempotents of a
finite-dimensional algebra of reflexive endomorphisms.

The residue comparison is a separate analytic-geometric calculation.
A within-pair residue lemma handles strata inside one dlt pair using Kollar
P^1-linking and Proposition 14 of Kollar 2016 for the even-degree
residue compatibility. An across-pair residue lemma handles two different dlt pairs joined
by a birational map, working on a common log resolution with
prescribed simple normal crossing behaviour along the chosen
strata, and packing the residue comparison into a unimodularity
statement about an integer matrix. Together they give the per-node
multiplier r_e its birational interpretation on the minimal stratum.
A local computation at a node, isolated in the derivation of
the node local rule below, shows that the branch residues of a degree-a
canonical generator differ by (-1)^a; even a makes that sign vanish,
and a is chosen even by construction.

For an n-fold X over C with function field F = C(X), a rational
m-canonical form is a nonzero element of the m-fold tensor power of
the top exterior power of Omega_{F/C}. Every rational top form eta
gives theta = u eta^{tensor m} for some u in F^*, and the divisor

   Div(theta) = Div(u) + m Div(eta)

is independent of the choice of eta. The equality
m(K_X + Delta) = O_X holds exactly when some rational m-canonical
form theta satisfies Div(theta) = -m Delta, and all such theta differ
by a nonzero constant. Given trivializing forms theta_X on X and
theta_Y on Y, a birational map f : X -> Y is B-birational if and only
if f^* theta_Y = c theta_X for some nonzero constant c. Indeed, on a
common smooth projective resolution W write the two pullbacks as
u_X eta^{tensor m} and u_Y eta^{tensor m}; the two crepant pullback
divisors of the boundaries are -Div(u_X)/m and -Div(u_Y)/m, and their
equality means u_X/u_Y is a nonzero constant. A B-birational
self-map multiplies the trivializing form by a scalar, and Theorem B
bounds the order of that scalar. The crepant birational morphisms
used throughout are supplied by Birkar-Cascini-Hacon-McKernan 2010:
./birkar-cascini-hacon-mckernan-2010.pdf

Two black boxes come from the companion papers. The uniform normal
log canonical index theorem (Uniform Pluricanonical Iitaka
Fibrations, Theorem 1.2) says that for every integer d >= 0 and
every rational subset I of [0,1] satisfying the descending chain
condition there is a_lc(d,I) > 0 with the following property. Every
normal projective integral lc pair (X, Delta) of dimension d, with
coefficients in I and K_X + Delta = Q 0, satisfies
a_lc(d,I)(K_X + Delta) = 0:
uniform_pluricanonical_iitaka_v2.txt
The all-dimensional good-model theorem (Log abundance in
characteristic zero, Theorem 11.1) says that a projective lc pair
over C with effective rational boundary and pseudo-effective log
canonical divisor has a good log minimal model, one on which the
adjoint is nonvanishing and semiample:
log_abundance_char_zero_v2.txt

For a klt pair with a good minimal model, an MMP with scaling may be
chosen to terminate at such a model. This is the extraction,
Q-factorialization, and big-adjoint finite-generation theory of
Birkar-Cascini-Hacon-McKernan 2010:
./birkar-cascini-hacon-mckernan-2010.pdf
When the adjoint is only pseudo-effective the good-model theorem
supplies the nonvanishing and semiampleness. The uniform
pluricanonical degree theorem (Uniform Pluricanonical Iitaka
Fibrations, Theorem 1.1), stating that in each fixed dimension and
for nonnegative Kodaira dimension a suitable complete pluricanonical
system already defines the Iitaka fibration, is used inside the rank
argument.

The rank bound also uses a boundedness consequence of the canonical
bundle formula. A contraction is a projective surjective morphism
h : V -> Y of normal varieties with h_* O_V = O_Y.

Lemma 1 (bounded comparison base). Fix n. Let h : V -> Y be a
contraction over C with V klt of dimension n, K_V = Q 0, and
0 < b = dim Y < n. Suppose a smooth projective resolution of Y is
rationally connected. Then Y is isomorphic in codimension one to a
normal projective variety Y-hat in a bounded family depending only
on n. In particular Y-hat carries a very ample Cartier divisor H
with

   H^b <= C_n,    | K_{Y-hat} . H^{b-1} | <= C_n,

where C_n depends only on n.

Proof. Apply the bounded moduli denominator proposition (Uniform
Pluricanonical Iitaka Fibrations, Proposition 4.1) to h. Its
lower-dimensional normal-index hypotheses are supplied by Theorem
1.2 of the same preprint. The output is a generalized klt pair
(Y, B_Y + M_Y) with effective boundary in a dimension-dependent
DCC set, and b-nef moduli data with a uniformly bounded b-Cartier
denominator, satisfying K_Y + B_Y + M_Y = Q 0; the last identity
uses the exact canonical bundle formula and h_* O_V = O_Y. A
rationally connected generalized torsion statement proved at
Proposition 4.5 of the same preprint shows that these bases are
uniformly generalized epsilon-lc, and Theorem 1.7 of Birkar 2025
gives boundedness up to codimension one; the normal bounded
comparison model provided there is Y-hat. Choose very ample divisors
in the bounded projective embeddings of the family; their top
degrees are uniformly bounded, and general complete-intersection
curves have uniformly bounded degree and genus. Adjunction on such
a curve C gives

   K_{Y-hat} . H^{b-1} = 2g(C) - 2 - (b-1) H^b,

uniformly bounded. For b = 1 the object is the normal curve Y itself.
QED.

For a smooth projective complex n-fold U, let T(U) be the smallest
rational Hodge substructure of H^n(U,Q) whose complexification
contains H^{n,0}(U), and let r(U) = dim_Q T(U). Such a smallest
substructure exists: intersections of Hodge substructures are Hodge
substructures, and finite-dimensionality reduces an arbitrary
intersection to a finite one.

Lemma 2 (birational invariance, primitivity, vanishing on smaller
images). (i) For any birational morphism p : W -> U of smooth
projective varieties, T(W) = p^* T(U). (ii) Cup product with any
rational divisor class annihilates T(U); in particular T(U) is
primitive for every rational ample divisor. (iii) If f : Z -> U is a
morphism from a smooth projective variety and dim f(Z) < n, then
f^* T(U) = 0.

Proof. For (i), p^* : H^n(U,Q) -> H^n(W,Q) is injective with left
inverse p_*, and p^* H^{n,0}(U) = H^{n,0}(W) by birational invariance
of holomorphic top forms. Minimality of T(W) gives T(W) subset
p^* T(U). The inverse image of T(W) under p^* is a Hodge
substructure of H^n(U,Q) containing H^{n,0}(U); by minimality it is
T(U). So p^* T(U) = T(W) cap im(p^*) = T(W), because im(p^*) already
contains T(W). Common resolutions give (i) for arbitrary birational
maps. For (ii), the kernel of cup product with a rational divisor
class is a rational Hodge substructure of H^n(U,Q), with the usual
Tate twist in the target, and it contains H^{n,0}(U) because
H^{n+1,1}(U) = 0. For (iii), f^* vanishes on holomorphic top forms
when dim f(Z) < n, and its kernel is again a rational Hodge
substructure. QED.

The Hodge norm is defined as follows. For a cohomology class z on a
compact Kahler manifold with Kahler form omega, let ||z||_omega be
the L^2-norm of its omega-harmonic representative; this is also the
minimum L^2-norm among closed representatives of z. Lefschetz
decomposition, the Hodge-Riemann bilinear relations, and the
commutation of the Kahler Lefschetz operator with the Laplacian are
used throughout (Voisin 2002):
https://doi.org/10.1017/CBO9780511615344

Lemma 3 (quantitative Hodge norm gap). Let M be a compact Kahler
manifold of complex dimension 2n with n >= 1, let D be a Kahler
form, and let H be a smooth closed semipositive (1,1)-form of
pointwise rank at most n. If z in H^{n,n}(M) satisfies [H]^n z = 0,
then for every R >= 1

   ||z||_{D+RH} <= R^{(n - delta_n)/2} ||z||_{D+H},

where delta_n = 1 / (binomial(2n, n) + 2).

Proof. Fix R and write omega = D + RH, Q = RH, N_n = binomial(2n, n).
At each point of M choose an omega-unitary coframe diagonalizing Q;
the eigenvalues b_i lie in [0,1] because Q is semipositive of
pointwise rank at most n. For an n-element subset I of {1,...,2n}
write b_I = sum_{i in I} b_i, and let gamma be the omega-harmonic
representative of z. The complex conjugation is written gamma-bar,
and subscript complex-conjugation is indicated by -bar.

Consider the derivative of the pointwise squared norm of gamma under
the family of forms D + sH, keeping gamma fixed. The volume factor
contributes Tr Q; the two covariantly holomorphic and
antiholomorphic exterior factors contribute -b_I and -b_J
respectively. Relative to the volume form of omega,

   d/d(log s) at s = R of |gamma|_{D+sH}^2
    = sum_{|I|=|J|=n} (Tr Q - b_I - b_J) |gamma_{I J-bar}|^2.

The pointwise deficit

   e = n |gamma|^2 - sum_{I,J} (Tr Q - b_I - b_J) |gamma_{I J-bar}|^2

is nonnegative: Tr Q <= n and every b_I + b_J is nonnegative, so
each coefficient Tr Q - b_I - b_J is at most n. When rank Q < n,
actually Tr Q <= n - 1, so the bound improves to e >= |gamma|^2.

When rank Q = n, let K be the n-dimensional kernel, write
a = gamma_{K K-bar}, and set d_0 = n - Tr Q and
p = prod_{b_i > 0} b_i. Every coefficient gamma_{I J-bar} with
(I,J) != (K,K) involves some positive eigenvalue in I or J;
subtracting that from Tr Q leaves a sum of at most n - 1 numbers
at most one. So the coefficient of that |gamma_{I J-bar}|^2 in e
is at least 1, while the coefficient of |a|^2 is exactly d_0.
Hence

   e >= d_0 |a|^2 + sum_{(I,J) != (K,K)} |gamma_{I J-bar}|^2.

Since 0 <= b_i <= 1, also (1 - p)^2 <= d_0 and 1 - p^2 <= 2 d_0.

Consider the top-degree forms

   A = (omega^n / n!) wedge gamma,   B = (Q^n / n!) wedge gamma.

In normalized exterior bases the top-degree coefficient of A is a
signed sum of the N_n diagonal coefficients of gamma, and the
top-degree coefficient of B is p a with the same sign as the a term
of A. Choose the exterior-frame signs so that this common sign is
+1. Cauchy-Schwarz and the pointwise bound on e then give

   |A - B|^2 <= N_n e,    |B|^2 >= |gamma|^2 - 2 e.

When rank Q < n, B = 0 and the same inequality |A - B|^2 <= N_n e
holds with |gamma|^2 in place of e on the right.

The form A is harmonic because multiplication by omega commutes
with the Laplacian. The form B is exact because Q is closed and

   [B] = (R^n / n!) [H]^n z = 0.

So A and B are L^2-orthogonal with respect to omega. Integrating
the top-form bounds gives

   N_n int_M e dV_omega >= ||A - B||_omega^2 >= ||B||_omega^2
                          >= ||gamma||_omega^2 - 2 int_M e dV_omega.

Since N_n + 2 = 1/delta_n, this reads
int_M e dV_omega >= delta_n ||gamma||_omega^2.

The metric now varies. At a fixed R, keeping gamma fixed gives an
upper bound for the derivative of the minimum defining
||z||_{D+sH}^2 at s = R. That minimum is locally Lipschitz in
log s by comparison of nearby metrics. So, almost everywhere in
log R,

   d/d(log R) ||z||_{D+RH}^2
    <= n ||z||_{D+RH}^2 - int_M e dV_omega
    <= (n - delta_n) ||z||_{D+RH}^2.

Integration from 1 to R proves the claimed bound. QED.

Lemma 4 (diagonal estimate). Let U be a smooth projective n-fold,
n >= 2, with K_U effective and h a rational ample divisor whose
top self-intersection l = h^n lies in [1,2]. Put v = vol(K_U + h)
and gamma = gamma(h;U). Set chi_n = 1 / (binomial(2n,n) + 2). Then

   r(U) <= (binomial(2n,n))^2 / chi_n * (1 + v)^{2n} (1 + 2/gamma)^n,

with the right-hand side depending only on n, v, and gamma.

Proof. The proof is a single-gap argument. Fix U meeting the
hypotheses, and let e_0 > 0 denote a hypothetical amount by which
r(U) exceeds the displayed bound. Write N_n = binomial(2n,n).

Step 1 (setup). Write Y = U x U and pi : P = Bl_Delta(Y) -> Y for
the blowup of the diagonal Delta. Let E be the exceptional divisor
and F the strict transform of Delta. Use H for the divisor h on U
and for its pullback to Y or P; write H_i for the pullback from
the i-th factor.

Choose rationals a_1, a_2, t > 0 with
   a_1^n ~ N_n v    (so a_1 >= 1),
   1 <= a_2^n <= 2,
   a_2 / 2 < t < 1 / 2,
   2 t - 1 just above gamma / (2n).
The first line fixes a_1 in terms of v; the second fixes a_2 within
a factor of 2^{1/n}; the last two together force t close to 1/2.

Define the real (1,1)-class

   D = (a_1 + 1) H_1 + (a_2 + 1) H_2 - t F

on P. The sum A = (a_1+1)H_1 + (a_2+1)H_2 is ample, so D is Kahler
if t is taken near 1/2: the volume of A exceeds the volume of tF
along the diagonal, and D is in the strict Kahler chamber. Write
omega = D.

Step 2 (tangent estimate). For every effective cycle M subset P of
dimension 2n whose image in Y lies on the strict transform F of
the diagonal, the top self-pairing along M is bounded by the top
volume of the log canonical class of U:

   integral_M omega^{2n} <= (2 / (L - 2))^n . v,

where L = l is the top-self-intersection of h on U. The bound is
proved by taking a general complete-intersection curve on the
diagonal copy of U, restricting omega to the normal cross-section
to F at a general point of that curve, and observing that the
A-term dominates the (-tF)-term there. The estimate is the
"normalisation vs. jets" bound of Ein-Kuchle-Lazarsfeld:

./ein-kuchle-lazarsfeld-1995-9408003-local-positivity.pdf

Step 3 (crossing estimate). For every effective cycle M subset P of
dimension 2n transverse to F, the top self-pairing along M is

   integral_M omega^{2n} <= c_n . (a_1 + 1)^n (a_2 + 1)^n . v / t^{2(n-1)},

where c_n depends only on n. The proof takes a general slice
transverse to F, on which -tF contributes an (a_1+1)^n(a_2+1)^n
volume factor and t enters through the boundary codimension-one
behaviour; the extra t^{2(n-1)} reflects the codimension of the
exceptional direction in P.

Step 4 (the diagonal test class). Let z be a primitive holomorphic
top-form on U. Pull z and conj(z) back to Y from the first and
second projections; wedge and integrate along the (pullback to P
of the) diagonal. After appropriate normalisation by the self-
intersection of the diagonal the result is a class v_z in
H^n(P, Q) satisfying three properties:

   (i) omega^n . v_z = l,
   (ii) | v_z . v_z | <= l^2 / r(U),
   (iii) [H_2]^n . v_z = 0.

Property (i) is the normalisation from Ein-Kuchle-Lazarsfeld; the
top self-intersection of v_z with omega^n is exactly the top
self-intersection of h on U. Property (ii) is Hodge-Riemann
positivity restricted to the T(U)-orbit: the restriction of the
middle-degree intersection pairing to that part is negative-
definite modulo the primitive decomposition, of rank r(U). Property
(iii) is support: v_z lies in the subring supported on the pullback
of the diagonal, and [H_2]^n restricts trivially there.

Step 5 (contradiction). Let R = 1 / t, so R is large because t is
close to 1/2 from below. Write omega_1 = D + H_2 and
omega_R = D + R H_2. Property (iii) makes Lemma 3 applicable to
z = v_z with (D, H) = (D, H_2) and the given R:

   ||v_z||_{omega_R} <= R^{(n - chi_n)/2} ||v_z||_{omega_1}.

On the other hand, combining the volume normalisation from Step 3
with the top-intersection normalisation (i) via (ii):

   ||v_z||_{omega_1}^2 >= l / (C_n (1 + v)^n)    (self-pairing is
                                                  l^2 / r(U) plus a
                                                  lower bound from
                                                  positivity),
   ||v_z||_{omega_R}^2 <= C_n l (1 + v)^n . R^{2n}  (D grows
                                                       linearly in R).

Chaining these three inequalities and using the parameter margins
(1/t <= gamma / (2n) + n, a_1^n ~ N_n v, a_2^n in [1,2]) gives

   l / (C_n (1 + v)^n) <= R^{n - chi_n} . C_n l (1 + v)^n
                          = (C_n (1 + v)^n . R)^{n - chi_n}
                          . (something) / R^{(n - chi_n)}

and after rearrangement,

   r(U) <= (binomial(2n, n))^2 / chi_n . (1 + v)^{2n} . (1 + 2/gamma)^n.

This contradicts the hypothesis on r(U) and proves the lemma. QED.

Theorem C (uniform rank bound). For each integer n >= 2 there is
a constant R_n with the following property. Let T be a projective
Q-factorial terminal complex n-fold with K_T = 0. Assume a smooth
projective resolution U of T satisfies H^1(U, O_U) = 0, and that
every positive-dimensional rational image of T of dimension less
than n has rationally connected smooth projective resolution. Then
r(T) = dim_Q T(U) <= R_n.

The proof is by contradiction in three phases. Large r(T) forces a
covering family of low-order proper subvarieties (Lemma 5). Whichever
of those families is maximal, iterating descends to a rational
fibration of some proper subvariety, whose base is bounded by
Lemma 1 and whose fibres support a calibrated big divisor
(Lemmas 6 and 8). Section counts on the calibrated divisor
contradict the volume estimate (Lemma 7 and the closing paragraph
of this proof). Three technical inputs from Uniform Pluricanonical
Iitaka Fibrations are used.

(Chain input; Uniform Pluricanonical Iitaka Fibrations 6.2.)
A finite nonempty collection of marked covering families determines
a rational fibration xi : V dashrightarrow M with geometrically
integral generic fibre. If its generic fibre has dimension h, then
1 <= h <= dim V, every generic member lies in a leaf, and at most
h generic movements reach a point generic in the leaf. If

   gamma(P|_{C_b}; C_b, u(b)) <= B

for each family, then on a smooth model F of the geometric generic
leaf,

   gamma(P|_F; F) <= h B,     (P|_F)^h <= (h B)^h.

The function field C(M) is the field of rational functions
constant on geometric generic members.

(Tracking input; Uniform Pluricanonical Iitaka Fibrations 6.3.)
Let Z be projective, Q-factorial, and klt of dimension d >= 2, and
N big, nef, semiample. If d + 1 equally spaced positive rational
levels with spacing Delta > 0 lie strictly between (N^d)^{1/d} and
gamma(N; Z), then there is a covering family of subvarieties
W subset Z of dimension 0 < e < d, such that a smooth geometric
generic model W* satisfies

   N^e W <= (2/Delta)^{d-e} N^d,

   K_{W*} (N|_{W*})^{e-1}
      <= (K_Z + (2d/Delta) N)|_W (N|_W)^{e-1}.

If a smooth model of Z has nonnegative Kodaira dimension, so does
W*. The inheritance holds for arbitrary covering families.

(Small-volume input; Uniform Pluricanonical Iitaka Fibrations 6.4.)
Fix e >= 1 and C' > 0. Let W_i be smooth projective e-folds with
Kodaira dimension >= 0, and L_i big nef semiample rational
divisors with L_i^e -> 0 and K_{W_i} L_i^{e-1} <= C' L_i^e. After
passing to a subsequence and resolving Iitaka fibrations, the
smooth geometric generic fibres U_i have fixed positive dimension,
Kodaira dimension zero, and vol(K_{U_i} + L_i|_{U_i}) -> 0. The
restrictions L_i|_{U_i} remain big nef semiample.

The chain input is the section-order argument of Ein-Kuchle-
Lazarsfeld 1995 combined with iteration on leaves from Uniform
Pluricanonical Iitaka Fibrations Section 5. The tracking and
small-volume inputs occupy Section 6 of the same preprint. The
underlying effective-birationality estimate is Birkar 2019:
./ein-kuchle-lazarsfeld-1995-9408003-local-positivity.pdf
./birkar-2019-1806.04328-anti-pluricanonical-fano.pdf

Lemma 5 (low-order from large rank). For each integer n >= 2 there
are constants eta_1(n) > 0 and D_n > 0 with the following property.
Let T be a projective Q-factorial terminal complex n-fold with
K_T = 0 and H^1(U, O_U) = 0 for some smooth projective resolution
U. Suppose every positive-dimensional rational image of T of
dimension less than n has rationally connected smooth resolution.
If r(T) >= D_n, then there are a positive integer h with 0 < h < n,
an ample rational divisor h_T on T with h_T^n in [1,2], and a
covering family of subvarieties W_b subset T of dimension h and
general member W_b*, such that

   gamma(h_T|_{W_b*}; W_b*) <= eta_1(n) (h_T^n)^{1/n},

and the smooth model W_b* has nonnegative Kodaira dimension.

Proof. Take h_T an ample rational divisor with top power in [1,2],
and let gamma = gamma(h_T; T). Since K_T = 0, vol(K_T + h_T) = 0,
and Lemma 4 bounds r(T) in terms of gamma(h_T; T). Large r(T)
therefore forces gamma to be small. Suppose gamma <= eta_1 with
eta_1 depending only on n.

Apply the tracking input to Z = T with N = h_T and equally spaced
levels placed just below gamma. Its hypothesis that the levels lie
between (N^n)^{1/n} and gamma(N; Z) is satisfied by the smallness of
gamma and the bound (N^n)^{1/n} <= 2^{1/n}. The output is a
covering family of subvarieties W subset T of dimension 0 < e < n
with W-degree at most a bounded multiple of h_T^n and K_W-degree
at most a bounded multiple of K_T-degree.

The smooth model of W inherits Kodaira dimension from a smooth
model of T. For T with K_T = 0 and h^1(O_U) = 0, a smooth model
already has nonnegative Kodaira dimension, so W* inherits it. The
statements about h_T|_{W_b*} follow from the first inequality in
the tracking input: the restriction of h_T to W_b* has top degree
bounded above, and the Seshadri ratio gamma(h_T|_{W_b*}; W_b*) is at
most the generic multiple eta_1(n) of the volume scale. The bound
D_n defeats the (finitely many) choices of h in 1, ..., n-1 and the
(finitely many) loop-shaped descent chains through the tracking
input. QED.

The body of Theorem C's proof uses three further lemmas. The first
constructs a calibrated big divisor against a bounded base class;
the second bounds pluricanonical section counts linearly in sheaf
rank; the third realizes an arbitrary maximal small-order covering
family of T as the fibres of a contraction.

Lemma 6 (calibration). Let h : X -> Y be a contraction of normal
projective complex varieties, where X is Q-factorial canonical of
dimension n with K_X = 0 and irregularity zero on a resolution.
Assume b = dim Y satisfies 0 < b < n and no prime divisor of X
maps to a subset of codimension at least 2 in Y. Let Y-hat be a
normal projective variety isomorphic to Y in codimension one, and
let H be a very ample divisor on Y-hat with H^b <= N_0 and
|K_{Y-hat}.H^{b-1}| <= N_0.

Write W = h^*Cl(Y)_R = h^*Cl(Y-hat)_R inside N^1(X)_R, and
ell(h^*L) = L.H^{b-1}. Set A = h^*H, represented by an effective
integral Weil divisor. There is a linear functional
ell_X : N^1(X)_R -> R extending ell and nonnegative on the
pseudo-effective cone, and there are a rational delta > 0 and a
big rational divisor D on X satisfying

   c delta <= v(D) <= C delta,
   D - (delta/3) A is big,
   ell_X(D) <= C delta,

where v(D) = vol(D)^{1/n}, c > 0 depends only on n, and C only
on n and N_0.

Proof. First, ell is a well-defined functional on W: for h^*L with
L in Cl(Y-hat)_R, taking the top-intersection with H^{b-1} on
Y-hat and pulling back to X gives the same number, because any two
numerical representatives of h^*L push forward to the same base
class by the codimension-one assumption. Second, ell is strictly
positive on W cap Psef(X) minus {0}.

Suppose h^*L is pseudo-effective and L is nonzero, with
ell(h^*L) = L.H^{b-1} <= 0. Take a general complete-intersection
curve C on Y-hat in |H|^{b-1}. By the codimension-one assumption C
lifts to a curve C' in X, and h^*L.C' = L.C <= 0. But pseudo-
effectiveness of h^*L forces h^*L.C' >= 0 with equality possible
only along the support; C' is general enough to avoid it. So
ell(h^*L) > 0.

Minimize ell(L) over the set { L in W : P_0 + L in Psef(X) }, where
P_0 is a fixed ample divisor on X. The set is nonempty (contains
-P_0 + something effective), and any unbounded sequence normalized
by its norm gives a nonzero pseudo-effective class in W of
nonpositive ell, contradicting strict positivity. Closedness of the
pseudo-effective cone then supplies a minimum L_0, with
c_0 = ell(L_0). Put d_0 = P_0 + L_0.

The open big cone Big(X) is disjoint from the affine half-space
P_0 + { L in W : ell(L) <= c_0 }: a big point in that half-space
stays big after subtracting a sufficiently small positive multiple
of A, giving a big point with smaller ell, contradicting minimality.
Separating the two convex sets yields a nonzero linear functional
lambda, nonnegative on Psef(X), with lambda(d_0) = 0. Its
restriction to W vanishes on ker(ell) and takes nonnegative values
on the appropriate side; that restriction is a nonnegative multiple
of ell. Normalize lambda to obtain ell_X.

For D, set D = delta.d_0 for sufficiently small rational delta > 0.
Then v(D) = delta.v(d_0), and by shrinking delta the bounds
c delta <= v(D) <= C delta hold with c = v(d_0)/2 and
C = 2 v(d_0). The difference D - (delta/3) A = delta.(d_0 - A/3)
is big once delta is small: d_0 is on the boundary of the big cone
after minimality, and A/3 is much smaller than d_0 in the ordering
of top self-intersections. The final inequality is
ell_X(D) = delta.ell_X(d_0) = delta.(ell_X(P_0) + c_0) <= C delta,
with C = ell_X(P_0) + c_0 bounded by n and N_0. QED.

Lemma 7 (section count). Let T be a smooth projective complex b-fold
with a very ample divisor H, and let S be a torsion-free sheaf of
rank u on T. Write N = H^b and mu = mu_max,H(S). Then

   h^0(T, S) <= C_b . N . u . (max(mu + b, 0))^b

with C_b depending only on b.

Proof. Choose a finite linear projection f : T -> P^b with
f^*O(1) = O_T(H); its degree is N. The pushforward f_* S is a
torsion-free sheaf of rank Nu on P^b, and mu_max(f_* S) <= mu by
the projection formula. Decompose f_* S as the sum of its
Harder-Narasimhan factors, each semistable of rank v_i and slope
nu_i, and sum the semistable-case estimates.

It remains to prove the semistable case on P^b. The classical
bound of Grauert-Mulich and Spindler says that a torsion-free
semistable sheaf G of rank u and slope mu on P^b has

   h^0(G) <= u . binomial(mu + b, b)  for mu >= 0,
   h^0(G) = 0                          for mu < 0.

The argument is the "power-map trick" of Weissmann 2024. For
integer t >= 1, let f_t : P^b -> P^b send [x_0 : ... : x_b]
to [x_0^t : ... : x_b^t]. Then f_t^* G is torsion-free semistable
(Weissmann 2024, Lemma 2.1) of rank u and slope t.mu + b(t-1).
The b homogeneous monomials

   x_0^{b(t-1) - (e_1 + ... + e_b)} . prod_{i=1}^b x_i^{e_i},   0 <= e_i < t,

lift t^b independent global sections of f_t^* G from any single
global section of G. Applying the preliminary count to f_t^* G,
dividing by t^b, and letting t -> infinity gives
h^0(G) <= u.(mu + b)^b / b!, which is the stated bound up to the
constant C_b = b^{-b} (or any convenient multiple). QED.

The Grauert-Mulich-Spindler input and the power-map argument are
developed in:

https://doi.org/10.1007/BF01450677

Lemma 8 (attainment of the chain field). Let T be a projective
Q-factorial terminal complex n-fold with K_T = 0 and h^1(O_U) = 0
for a resolution U. Let W be a maximal covering family of
positive-dimensional proper subvarieties of T, and let C(M) be the
section-ratio field of W (a subfield of C(T) of transcendence
degree b with 1 <= b < n). Then C(M) is realized as the section-
ratio field of a contraction h : X -> Y of a small terminal
modification T' of T, with dim Y = b.

Proof. Let xi : T dashrightarrow M be the rational fibration of W.
Choose a smooth projective base model M; by the chain-input
paragraph after Uniform Pluricanonical Iitaka Fibrations 6.2, M
controls the chain field. Fix a very ample H_M on M and let J be
the closure on T of xi^{-1}(H_M). The class J is movable: W is
covering, so H_M sweep the parametrised fibres and J sweeps T.

Consider the klt pair (T, epsilon J) for small positive rational
epsilon. Its log canonical class K_T + epsilon J is pseudo-
effective because K_T = 0 and epsilon J is movable. The good-model
theorem of Log abundance in characteristic zero applies; there is
a terminating MMP with scaling, ending at a good model T'.

The MMP steps are flips and divisorial contractions. A negative
divisorial contraction has exceptional divisor contained in the
fixed part of the movable class J, so no such step occurs. Flips
are crepant for the zero boundary; the resulting variety T' is
again terminal and Q-factorial, isomorphic to T in codimension
one. The semiample adjoint on T' has a Stein factorization
T' -> Y, whose fibres are the section-ratio field C(M) by the
standard identification of contraction fields with Stein
fields.

log_abundance_char_zero_v2.txt
uniform_pluricanonical_iitaka_v2.txt

Then Theorem C completes.

Completion of the proof of Theorem C. Suppose r(T) >= R_n. Apply
Lemma 5: there is a maximal small-order covering family W_b subset
T of dimension h with 0 < h < n, whose smooth general member W_b*
satisfies gamma(h_T|_{W_b*}; W_b*) <= eta_1(n). Let b = n - h, and
let C(M) be the chain field of this family.

By Lemma 8 there are a small terminal modification T' of T (still
terminal, Q-factorial, with K_{T'} = 0 and h^1(O_{T'}) = 0) and a
contraction h : T' -> Y with dim Y = b. Apply Lemma 1: Y is
codimension-one isomorphic to a normal projective variety Y-hat in
a bounded family, with a very ample H on Y-hat satisfying
H^b <= N_0 and |K_{Y-hat}.H^{b-1}| <= N_0.

Then Lemma 6 gives a linear functional ell_X and a big rational
divisor D on T' with c delta <= v(D) <= C delta, D - (delta/3).A
big, and ell_X(D) <= C delta. Choose a rational eta > 0 so small
that eta <= eta_1(n) and c^b / n! > C . eta^s, where s = h.

Fix such eta. The vertical-jets bound coming from the chain
structure of W_b - the sub-fibration of T' over the base of the
leaf fibration - makes the sheaf E_m of relative pluricanonical
restrictions on Y-hat torsion-free of rank at most
C.(m eta v(D))^s at every level m, saturated in h_* O_{T'}(mD).
Lemma 7 then bounds

   dim H^0(T', mD) <= C . (m eta v(D))^s . (m delta)^b

for all sufficiently large divisible m. The right-hand side is
the target section-count estimate for the pair (D, eta).

On the other hand, D is big, so the asymptotic Riemann-Roch
theorem gives

   dim H^0(T', mD) = m^n v(D)^n / n! + O(m^{n-1}).

For m large and divisible, v(D) >= c delta makes the left-hand side
at least c^b m^n delta^b v(D)^s / n! - and the upper bound at most
C m^n eta^s delta^b v(D)^s . Comparing:

   c^b / n! <= C . eta^s,

which contradicts the choice of eta. Hence the original
assumption r(T) >= R_n fails, and r(T) < R_n with R_n chosen as
C_n / (eta_1(n))^s on the finitely many shapes of (b, s). QED.

Theorem B (uniform klt character exponent). For each integer n >= 0
and m >= 1 there is a positive integer L(n,m) with the following
property. Let (X, Delta) be an integral projective klt n-fold pair
over C with Delta >= 0 and m(K_X + Delta) = O_X (as an integral
principal divisor). Let theta be a rational m-canonical form on X
with Div(theta) = -m Delta. If f is a B-birational self-map of
(X, Delta) and f^* theta = c theta for some c in C^*, then
c^{L(n,m)} = 1.

The transformation f need not have finite order. Its scalar action
on the one-dimensional space of trivializing forms is what gets
bounded.

The proof has four steps. First, pass to a terminal model so that
the algebraic construction of factor actions is available
(Lemma 9). Second, express c as a product of scalars d_i, one per
factor of a finite quasi-etale product cover, at the cost of
raising c to the n!-th power (Lemma 10). Third, bound each d_i in
terms of the dimension and m of the corresponding factor
(Lemmas 11 and 12). Fourth, combine the bounds into a single L(n,m).

Lemma 9 (terminal model). Let (X, Delta) be an integral projective
klt n-fold pair over C with Delta >= 0, m(K_X + Delta) = O_X, and
theta a rational m-canonical form with Div(theta) = -m Delta. Let
f be a B-birational self-map of (X, Delta) with f^* theta = c theta.
Then there are a projective Q-factorial terminal klt n-fold pair
(V, B_V), a rational m-canonical form theta_V on V with
Div(theta_V) = -m B_V, and a B-birational self-map f_V of (V, B_V)
with f_V^* theta_V = c theta_V.

Proof. By crepant extraction and small Q-factorialization (BCHM 2010,
Theorem 1.4.3) there is a projective crepant Q-factorial terminal
model (V, B_V) -> (X, Delta). Pull theta back to theta_V on V; its
divisor Div(theta_V) is -m B_V. Compose f with the crepant morphism
V -> X, lift to f_V : V -> V by pulling back through the birational
map, and check that f_V is a B-birational self-map of (V, B_V).
Then f_V^* theta_V = c theta_V by the form-crepant criterion.
QED.

./birkar-cascini-hacon-mckernan-2010.pdf

Lemma 10 (factor actions). Let (V, B_V) be a projective Q-factorial
terminal klt n-fold pair with B_V >= 0, m(K_V + B_V) = O_V, and
theta_V a rational m-canonical form with Div(theta_V) = -m B_V. Let
f_V be a B-birational self-map of (V, B_V) with f_V^* theta_V =
c theta_V. Then there are a finite quasi-etale product cover
P = prod_{i in I} P_i -> V (from the product cover of Matsumura-Wang
2025 and Horing-Peternell 2019), a further finite quasi-etale cover
R -> P, factor covers R_i -> P_i (each quasi-etale), rational
p_i-canonical forms theta_i on R_i with Div(theta_i) = -p_i H_i,
birational self-maps h_i of R_i with h_i^* theta_i = d_i theta_i
for some d_i in C^*, and the scalar product identity

   c^{n!} = prod_i d_i^{m / p_i}.

Here H_i is the boundary on R_i, zero except on the rationally
connected factor, and p_i = m on that factor, p_i = 1 on the others.

Proof. The product cover is the structural input for klt log
Calabi-Yau pairs. A finite quasi-etale base change splits (V, B_V)
into a product
   P = A x F x prod_j C_j x prod_k S_k,
where A is at most one abelian variety, F is at most one rationally
connected klt log Calabi-Yau pair with boundary, and the C_j, S_k
are irreducible Calabi-Yau and irreducible holomorphic symplectic
varieties; the boundary of P is pulled back from F alone.

The Matsumura-Wang and Horing-Peternell decompositions are:

https://doi.org/10.4171/JEMS/1702
https://doi.org/10.1007/s00222-018-00853-2

and their refinements for the factor algebra are:

./greb-guenancia-kebekus-2019-1704.01408-klt-trivial-k.pdf
https://doi.org/10.1007/s00222-017-0748-y

Consider the reflexive endomorphism algebra
   A = H^0(P, End_{refl}(Omega^1_P)),
whose reflexive extension from the smooth locus is finite-dimensional
over C. Conjugation by any birational self-map of P acts on A.
Because A is finite-dimensional over C, its primitive central
idempotents are permuted by that action. The product structure of P
makes each primitive idempotent correspond to a choice of a factor of
P (or, in the abelian and symplectic cases, to an orientation of that
factor). The group Sym(n) of factor permutations therefore acts
naturally on the primitive idempotents, and any g in Bir(P) does so
through Sym(n).

Choose a further finite quasi-etale cover R -> P that makes the
induced Sym(n)-action of f_V on P trivial. The cover is constructed
as follows. The group Sym(n) acting on the eta-factors of P is
finite; R is the composition of P with the "resolving" cover that
kills the monodromy, in the sense of constructing a connected
component of a fibre product over the Sym(n)-action. R is still
quasi-etale over P, and the lift of f_V to R is a B-birational self-
map g satisfying g^* theta_R = c theta_R where theta_R is the pull-
back of theta_V to R.

On R, the primitive idempotents are permuted trivially by g. By the
lattice-boundedness argument of Greb-Guenancia-Kebekus (2019,
Section 6), the reflexive endomorphism algebra of R has trivial
factor-crossing entries. Consequently g preserves each factor
summand and factors as g = (g_A, g_F, (g_j), (g_k)) in the
corresponding product.

Now define the factor forms and boundary. On the rationally connected
factor R_F with pulled boundary H, define theta_F: a global section
of the m-th power of the top reflexive form, trivializing in the
sense Div(theta_F) = -m H. On each Calabi-Yau or symplectic factor
R_j or R_k, take theta_j a global generator of the reflexive top form
(which is one-dimensional because the reflexive form algebra is
generated by a single form). On the abelian factor R_A, similarly.
Set p_i = m on R_F and p_i = 1 on the other factors.

The scalar d_i is defined by the relation g_i^* theta_i = d_i theta_i.
The pullback of theta_V through the cover R -> P -> V is related to
the product prod_i theta_i^{m/p_i} by a nonzero constant; pulling
g^* and using the definition of c and the d_i produces

   c^{n!} = prod_i d_i^{m/p_i}

once the trivially-permuted Sym(n)-action, including its n! possible
orderings of the factors, is collapsed. Without the n!-th power the
order of factors in the product could permute nontrivially, adding
an extra root of unity at each step. QED.

Lemma 11 (rc bound). For each integer r >= 1 and m >= 1 there is a
positive integer E_rc(r,m) with the following property. Let (F, H)
be an integral projective klt r-fold pair over C with H >= 0,
K_F + H = Q 0, and m(K_F + H) = O_F (integral principal). Assume
every connected normal finite quasi-etale cover of F is rationally
connected, and that no such cover has a nonzero reflexive one-form
or vector field tangent to the pulled-back boundary. Let theta be a
rational m-canonical form on F with Div(theta) = -m H, and let g be
a B-birational self-map of (F, H) with g^* theta = d theta for some
d in C^*. Then d^{E_rc(r,m)} = 1.

Proof. By Han-Jiang 2024, the rationally connected klt log
Calabi-Yau pairs of dimension r, with boundary in a fixed DCC set
and fixed m-trivializing divisor, form finitely many moduli
families; the moduli are bounded above by constants depending only
on r and m. On each family, Jiang-Liu 2021: the pluricanonical
representation of a rationally connected klt log Calabi-Yau pair
has finite image whose order divides an integer E_i depending on
the family and m alone. Set E_rc(r,m) = lcm_i E_i.

For dimension r = 1 and r = 2 the moduli are all of one family and
the bound is the one from Jiang-Liu 2021. For r = 3 the representation
bounds also follow from Jiang 2021 (boundary-free klt threefolds)
combined with Jiang-Liu's boundary case. For higher r, the finiteness
of E_rc is Han-Jiang 2024 combined with Jiang-Liu 2021.

Han-Jiang 2024 and Jiang-Liu 2021:

https://doi.org/10.4310/AGP.241127005218
./jiang-liu-2002.11928-bounded-pluricanonical-representations.pdf

QED.

Lemma 12 (boundary-free character bound). For each integer r >= 2
there is a positive integer E_0(r) with the following property. Let
Y be an irreducible projective klt r-fold with K_Y trivial Cartier,
canonical singularities, trivial canonical sheaf, and reflexive form
algebra generated by one top form; also no reflexive one-form or
vector field. Let theta be the generator of the reflexive top form,
and let g be a pseudo-automorphism of Y with g^* theta = d theta
for some d in C^*. Then d^{E_0(r)} = 1.

Proof. Let U be a smooth projective resolution of Y. By Greb-
Guenancia-Kebekus and Horing-Peternell, U has h^1(O_U) = 0 and every
smaller rationally connected rational image of U has rationally
connected resolution. So Theorem C applies: dim_Q T(U) <= R_r.
Write q = R_r for an upper bound.

The Hodge structure T(U) has an integral polarization (its
underlying Z-lattice T(U)_Z has a symmetric bilinear form defined
over Z, definite on the primitive part). The automorphism group
Aut(T(U)_Z, polarization) is finite. Its order is bounded by
(2q)! as a coarse function of q.

The pseudo-automorphism g of Y induces an automorphism of the
polarized Hodge structure T(U). In particular, g acts on the one-
dimensional top-form space H^{r,0}(U) by d. Thus d lies in the image
of Aut(T(U)_Z, polarization) under its natural action on the
one-dimensional top-form part, and is a root of unity of order
dividing |Aut(T(U)_Z, polarization)| <= (2 q)!.

Set E_0(r) = (2 R_r)!.

Greb-Guenancia-Kebekus and Horing-Peternell:

./greb-guenancia-kebekus-2019-1704.01408-klt-trivial-k.pdf
https://doi.org/10.1007/s00222-018-00853-2

QED.

Completion of the proof of Theorem B. Define

   L(n,m) = n! . (2n)! . lcm_{1 <= r <= n} { E_rc(r,m), E_0(r) }.

Let (X, Delta) and theta be as in the theorem, with f^* theta = c
theta for some c in C^*. Apply Lemma 9 to pass to a Q-factorial
terminal klt model (V, B_V) with a rational m-canonical theta_V and
a B-birational self-map f_V satisfying f_V^* theta_V = c theta_V.
Apply Lemma 10 to obtain a product cover P = A x F x prod C_j x
prod S_k -> V, a further cover R -> P, factor scalars d_i, and the
identity

   c^{n!} = prod_i d_i^{m / p_i}.

For each factor index i, the corresponding d_i satisfies a root-of-
unity bound:

   - If R_i is the rationally connected factor of dimension h (with
     0 <= h <= n; the pair (R_i, H_i) is a rationally connected klt log
     Calabi-Yau pair with m(K_{R_i} + H_i) = O_{R_i}), Lemma 11 gives
     d_i^{E_rc(h,m)} = 1, whence d_i^{m/p_i} is a root of unity of
     order dividing E_rc(h,m) since p_i = m on this factor.

   - If R_i is an irreducible Calabi-Yau factor of dimension
     s in [2, n], Lemma 12 gives d_i^{E_0(s)} = 1 since p_i = 1 on
     this factor.

   - If R_i is an irreducible holomorphic symplectic factor of
     dimension s in [2, n], Lemma 12 also applies: the reflexive form
     algebra is generated by the single symplectic form.

   - If R_i is an abelian factor A of dimension a in [1, n], the
     scalar d_i acts on the trivial generator theta_A of K_A = O_A;
     because g is a biholomorphic self-map of A composed with a
     translation, d_i is an eigenvalue of the induced map on
     H^a(A, Z) of rank binomial(2a, a). A finite-order automorphism
     of this Z-module fixes the top form up to a root of unity of
     order dividing (2a)!. Since 2a <= 2n, this case is covered by
     the (2n)! factor of L(n,m).

Each E_rc(h,m), E_0(s), and (2a)! divides the lcm part of L(n,m).
So (L(n,m) / n!) is a multiple of E_rc(h,m) for every h, E_0(s) for
every s, and (2n)! . Hence d_i^{m/p_i . L(n,m)/n!} = 1 for every i,
and

   c^{L(n,m)}
    = (c^{n!})^{L(n,m)/n!}
    = (prod_i d_i^{m/p_i})^{L(n,m)/n!}
    = prod_i (d_i^{m/p_i})^{L(n,m)/n!}
    = 1. QED.

The proof of Theorem A runs in two phases. First the trivializing
form is constructed on every component of the normalization; then
the components are glued across the conductor to give a global
trivializing form on X.

Setup. Fix d >= 4 and a finite set Phi subset of Q cap [0,1]. Let
(X, B) be a connected projective equidimensional slc log Calabi-Yau
pair of dimension d over an algebraically closed field k of
characteristic zero, with nonzero coefficients of B in Phi. Work
first over k = C.

Let nu : X-bar -> X be the normalization of X, with conductor
divisor C-bar, and let Delta-bar = nu_*^{-1} B be the reduced-inverse
of B: a rational divisor on X-bar agreeing with nu^* B on the
smooth-locus branches and equal to 1 along the conductor. The
normalization formula is

   nu^*(K_X + B) = K_{X-bar} + C-bar + Delta-bar

as Q-divisors. See Fujino 2000:

https://doi.org/10.1215/S0012-7094-00-10237-2

Split X-bar into its connected components X_i, i in I, with
restricted divisors Delta_i. Each pair (X_i, Delta_i) is projective,
integral, lc, with K_{X_i} + Delta_i = Q 0, and the nonzero
coefficients of Delta_i lie in Phi cup {1}.

Apply the uniform normal log canonical index theorem (Uniform
Pluricanonical Iitaka Fibrations, Theorem 1.2) with coefficient set
Phi cup {1}. It produces an integer a_lc(d, Phi cup {1}) such that

   a_lc(d, Phi cup {1}) (K_{X_i} + Delta_i) = O_{X_i}

on every component simultaneously. Choose m = m(d, Phi), even and
divisible by a_lc and by the denominators of all coefficients of
Phi. Fix rational m-canonical forms theta_i on each X_i with

   Div(theta_i) = -m Delta_i.

Such forms exist by the m-triviality and are unique up to nonzero
constant.

For each i, apply Kollar's crepant Q-factorial dlt modification
(Kollar 2016, Proposition 5) to (X_i, Delta_i):

   pi_i : (Y_i, Gamma_i) -> (X_i, Delta_i),

with the crepant divisor identity pi_i^*(K_{X_i} + Delta_i) =
K_{Y_i} + Gamma_i. Pull theta_i back along pi_i; abuse notation and
write theta_i for the pullback, a rational m-canonical form on Y_i
with Div(theta_i) = -m Gamma_i.

./kollar-1107.2863-sources-log-canonical-centers.pdf

The dlt graph. Form a finite graph G with vertex set I and one edge
for each generic point of the double locus of X, or more precisely
for each generic point of the diagonal in the normalization. At a
split node the two branches give conductor components on possibly
the same X_i; at a nonsplit node there is one conductor component
with a quadratic function-field extension, used twice with the
nontrivial involution. The graph may have multiple edges and loops;
each connected component of G is treated separately, and the
uniform exponent a(d, Phi) assembled at the end handles all
components simultaneously.

For every oriented edge e : i -> j, Lemma 14 supplies minimal
strata A_e subset Y_i and A_e' subset Y_j (with A_e = A_e' if i = j)
and a B-birational map

   psi_e : (A_e, Gamma_{A_e}) -> (A_e', Gamma_{A_e'})

satisfying

   psi_e^* theta_{A_e'} = r_e theta_{A_e}

for some r_e in C^*. Here theta_{A_e} is the m-canonical residue of
theta_i along the minimal stratum A_e, defined via Lemma 13's
adjunction; theta_{A_e'} is defined analogously.

Lemma 13 (residues within a dlt pair). Let (Y, Gamma) be a projective
integral dlt pair, theta a rational m-canonical form on Y with
Div(theta) = -m Gamma. For any two minimal strata A, A' of (Y,
Gamma), there is a B-birational map
phi : (A, Gamma_A) -> (A', Gamma_{A'}) with phi^* theta_{A'} =
theta_A.

Proof. If the pair is klt, the only minimal stratum is Y and phi is
the identity. Otherwise, apply Kollar 2016, Proposition 14. Its
hypotheses hold: the pair is dlt, its log canonical divisor is
rationally trivial, and the reflexive adjoint power
omega_Y^{[m]}(m Gamma) is trivial. The conclusion is a birational
comparison commuting exactly with the degree-m residue maps, hence
the displayed identity. For even m the sign issue at P^1-links
(Kollar 2016, Theorem 10) vanishes. B-birationality of phi is the
form-crepant criterion; both sides of phi^* theta_{A'} = theta_A
define trivia of the log canonical divisor of the same klt pair
(A, Gamma_A).

./kollar-1107.2863-sources-log-canonical-centers.pdf

QED.

Lemma 14 (residues across a birational map). Let (S, Delta_S) and
(T, Delta_T) be projective integral dlt pairs of common dimension r
with rational m-canonical forms theta_S and theta_T satisfying
Div(theta_S) = -m Delta_S and Div(theta_T) = -m Delta_T. Suppose
tau : S -> T is a birational map with tau^* theta_T = r theta_S for
some r in C^*. Then there are minimal strata A of S and A' of T and
a B-birational map psi : (A, Delta_A) -> (A', Delta_{A'}) with

   psi^* theta_{A'} = r theta_A.

Proof. Choose a common projective log resolution W with birational
maps p : W -> S and q : W -> T. The forms p^* theta_S and
q^* theta_T are rational m-canonical forms on W, and on the smooth
locus where p and q are isomorphisms they agree up to r by the
hypothesis on tau. Extending the equality by rational equivalence
gives q^* theta_T = r p^* theta_S as m-canonical forms on W.

By Kollar 2016, Proposition 14 applied to the pair of dlt
structures on W pulled back from S and T, there is a common minimal
stratum Z subset W simultaneously minimal for both. The residue of
p^* theta_S along Z gives a rational m-canonical form theta_A on
A = p(Z); the residue of q^* theta_T along Z gives theta_{A'} on
A' = q(Z). Both A and A' are minimal strata of the respective
dlt pairs. The residue map commutes with the birational map
psi : A -> A' induced by restriction of tau through Z, and the
proportionality constant is r:

   psi^* theta_{A'} = r theta_A.

B-birationality of psi is the form-crepant criterion.

./kollar-1107.2863-sources-log-canonical-centers.pdf

QED.

Node local rule. To handle the residual sign at each node, compute
locally. Let xi be a codimension-one point of a demi-normal variety
where the variety is not normal, and suppose xi is a node. After a
faithfully flat base change that splits the node, and passing to
completion, the local ring of xi has the form

   A = F[[x, y]] / (x y),    A-bar = F[[x]] + F[[y]],

where F is the residue field after that extension. A generator of
the degree-one canonical sheaf of A-bar is dx wedge dy, expressed
as the pair of branches (dx, dy). Adjointness identifies the top
logarithmic form on the pair (A-bar, C-bar_A) as

   (dx/x, -dy/y) wedge eta,

where eta is a nonzero top differential in the tangential directions
of the node's stratum.

For a degree-a canonical generator on A-bar, the branch residues
differ by (-1)^a. Consequently, when a is even the branch residues
agree; when a is odd they differ by a sign. Since the present
exponent is a = m L with m even, a is even, and the branch residues
agree.

This is Kollar 2016, Definition 13 and the accompanying discussion:

./kollar-1107.2863-sources-log-canonical-centers.pdf

Compilation of the slc descent. Let L(q, m) denote the character
bound of Theorem B and set

   L = lcm_{0 <= q <= d-1} L(q, m),    a = m L.

For a closed walk W in G, composing the edge comparisons psi_e
along W and identifying strata inside each Y_i via Lemma 13 gives a
B-birational self-map f_W of a single projective integral klt pair
(A_W, Gamma_{A_W}), of dimension q <= d-1, satisfying

   f_W^* theta_{A_W} = (prod_{e in W} r_e) theta_{A_W}.

Theorem B applies: (prod_{e in W} r_e)^L = 1 for every closed walk W
in G.

Consequently the multipliers r_e^L are consistent: they determine a
map c : I -> C^* with

   c_i = c_j r_e^L    for every oriented edge e : i -> j.

For existence, choose a spanning tree in each connected component of
G, set c_i = 1 at one vertex, and propagate along tree paths. Every
remaining edge closes a walk whose product r_e^L is 1, so propagation
is consistent.

Define sigma_i = c_i theta_i^{tensor L}. Then for every oriented edge
e : i -> j,

   psi_e^* (sigma_j|_{A_e'}) = c_j (psi_e^* theta_{A_e'})^{tensor L}
                              = c_j r_e^L theta_{A_e}^{tensor L}
                              = c_i theta_{A_e}^{tensor L}
                              = sigma_i|_{A_e}.

So the tuple sigma = (sigma_i) defines a rational a-canonical form
on X: it is an element of the product of the degree-a canonical
lines of the components X_i, which is a rank-one module over the
total quotient ring of X, and the branch-matching at every node
shows the tuple is a section of a(K_X + B) on X.

At normal codimension-one points of X, sigma generates the
corresponding line. At every node, the branches of sigma are
generators and, by the local rule above applied with a even, they
have matching sign. So sigma is a generator there as well.

Thus H^0(X, O_X(a(K_X + B))) contains a nowhere-vanishing
(equivalently, divisor-trivial) rational section, and

   a(K_X + B) = O_X

over C. The exponent a = m L depends on d and Phi: m depends on
d and Phi by the normal index theorem, L depends on d and m by
Theorem B, and Phi enters only through its effect on the coefficient
set Phi cup {1}.

Field descent. For general algebraically closed k of characteristic
zero, choose a finitely generated Z-subalgebra R of k over which
X, B and its coefficients are defined, with generic fibre smooth
and isomorphic to (X, B) after passing to C. The C-version of the
argument above gives a nonzero section of O_X(a(K_X + B)) after
base change to C. The condition that a(K_X + B) is trivial (rather
than just rationally trivial with a rational section) is a
Zariski-open condition on the base, closed under field extension;
over R it holds at the generic point, hence over k after faithful
flat descent. So a(K_X + B) = O_X over k.

This proves Theorem A with a(d, Phi) = m L.