Fundamental Groups and Coregularity: Klt Pairs with Trivial and Nef Log-Anticanonical Class

For projective klt Calabi-Yau pairs with standard coefficients, Propositions 2.20 and 2.21 give abelian subgroups of their regional orbifold fundamental groups of indices at most 60 in dimension one and 3840 in dimension two. These verify the hypotheses of Proposition 5.1 in all dimensions 1\le k\le2; no higher-dimensional Calabi-Yau index bound is needed when c_0=2. Apply Proposition 5.1 with c_0=2. For an explicit recursive choice of bounds, enlarge J to be integer-valued and nondecreasing, put b_0=1, and set

where the inner maximum is omitted for d=1. The induction gives index at most b_d for a solvable subgroup. Its normal core has the same derived-length bound and index at most b_d!. Thus one may take i(c,d)=b_d! for c\le2.

By replacing F with its image \phi(F) we can assume that F is a subgroup of G with f elements. Since H is virtually abelian, there exists an abelian group A \leqslant H of rank n and of finite index in H. After replacing A by a subgroup of finite index, we may assume that A\simeq\mathbb Z^n. Let A be generated by \{a_i\}_{\{1 \leq i\leq n\}}. Let g_i be elements in G mapping to a_i via \psi. For any x in F and g_i, we have that g_i x g_i^{-1} gets mapped to the identity e\in H via \psi, hence g_i x g_i^{-1} \in F. Thus conjugation by elements y \in \langle g_i \rangle gives a group action on F. Since F is a finite set, there exists a uniform constant m such that m-iterations of all of these actions are the identity, in particular y^{m} x y^{-m} =x for any x \in F and y \in \langle g_i \rangle.

On the other hand, for any y \in \langle g_i \rangle, there exists x\in F, such that g_i^{m} y g_i^{-m}=x y, since it maps to the same element of A as y. Therefore

We conclude that \langle g_i^{fm}\rangle_i is an abelian subgroup of rank n of G. This subgroup has finite index in G, as it surjects onto \langle a_i^{fm}\rangle_i, a finite index subgroup of H, via a homomorphism with finite kernel.

Let H' be the solvable subgroup of H of derived length at most n and index at most i. By letting G' be the preimage of H' under \psi, we obtain the exact sequence

where [G:G']=[H:H']\le i.

By taking a quotient if needed, we may assume that F is a subgroup of G'. Let Z be the centralizer of F in G'. The index of Z in G' is at most |{\rm Aut}(F)|, which is at most (|F|-1)!.

Since F\cap Z is abelian and \psi(Z) \leqslant H' is solvable of derived length at most n, the group Z is solvable of derived length at most n+1. Hence G is virtually solvable of derived length at most n+1 and index at most i|{\rm Aut}(F)|\leq i(|F|-1)!.

For H\le F put m(H)=|H|\,|C_F(H)|. For subgroups H,K\le F, counting the products HK and C_F(H)C_F(K) gives

Thus the subgroups maximizing m are closed with respect to intersection. Their unique smallest member A is characteristic. The centralizer of a maximizing subgroup is also maximizing, since H\subseteq C_F(C_F(H)). Hence A\subseteq C_F(A), so A is abelian. If B is the given abelian subgroup, then

This proves the assertion.

Replace F by its image; the same index bound persists. By Lemma 2.6, choose a characteristic abelian subgroup A\le F with [F:A]\le J^2. It is normal in G. Apply Lemma 2.5 to F/A\to G/A\to H\to1, and lift the resulting solvable subgroup through the abelian kernel A.

Pull back the identity

On a common resolution the resulting boundary is (1-t)B_V+tP_V, where K_V+P_V=f^*(K_X+P). Its log discrepancies are therefore the stated convex combinations. For an lc place they equal tA_{X,P}(E), which belongs to (0,1) when t>0 is sufficiently small.

Take a log resolution f:V\to X on which \mathbf M descends and all the prescribed divisors occur. Write K_V+B_V+\mathbf M_V=f^*(K_X+B+\mathbf M_X). The divisor \mathbf M_V+\epsilon f^*H is nef and big. By Kodaira’s lemma write it as H_\delta+\delta E, up to \mathbb Q-linear equivalence, where E\ge0 is fixed and H_\delta is ample for sufficiently small rational \delta>0. Choose a general effective representative G_\delta\sim_{\mathbb Q}H_\delta with sufficiently small coefficients. We may arrange that (V,B_V+G_\delta+\delta E) is sub-klt and that G_\delta contains none of the prescribed divisors. Their boundary coefficients tend to those of B_V as \delta\to0 and therefore remain in (0,1). Set \Delta=B+f_*(G_\delta+\delta E)\ge B. Then K_X+\Delta\sim_{\mathbb Q}K_X+B+\mathbf M_X+\epsilon H is \mathbb Q-Cartier, and

Indeed, the difference is exceptional and \mathbb Q-linearly trivial over X, so it vanishes by the negativity lemma. The equality proves both that (X,\Delta) is klt and that A_{X,\Delta}(E_i)=1-\operatorname{coeff}_{E_i}(B_V+G_\delta+\delta E) belongs to (0,1). Negative components of B_V are exceptional and disappear under f_*; the added divisor is effective.

Choose t>0 sufficiently small in Lemma 2.8 so that all the prescribed generalized log discrepancies belong to (0,1). Apply Lemma 2.9 to this generalized klt pair, using an ample divisor on X, to obtain an ordinary klt boundary \Delta with 0<A_{X,\Delta}(E_i)<1 for every i. Now [BCHM10, Corollary 1.4.3] extracts exactly these divisors on a \mathbb Q-factorial model. Its discrepancy hypotheses hold because (X,\Delta) is klt and all the specified discrepancies are smaller than one.

Let C have proper image T, and suppose T is not a generalized lc center of the base. Near the generic point of T, choose a general effective Cartier divisor containing T but containing no generalized lc center of the base. Such a choice is possible: centers properly contained in T can be discarded on this open set, and centers properly containing T need not lie in the divisor. On a common log resolution this divisor has zero order along every discrepancy-zero divisor, since it contains no generalized lc center. A sufficiently small positive multiple can therefore be added while preserving generalized lc singularities. Adding its pullback upstairs makes the discrepancy of any lc place centered on C negative. The discriminant changes by exactly that divisor and the moduli part is unchanged. This contradicts the lc equivalence in [Fil20, Proposition 4.16].

Push any generalized lc Calabi-Yau completion on Y to X. Its boundary still dominates D_X, and its nef b-divisor is unchanged. The completion is crepant and has the same coregularity. Taking the minimum proves the assertion. This is used only when the indicated boundary is effective.

The case \dim X=0 is immediate. Pass to a small \mathbb Q-factorialization; completions pull back and push forward crepantly with effective boundaries, so absolute coregularity is unchanged. For any completion (X,B,\mathbf M) we have -(K_X+P)\sim_{\mathbb Q}B-P+\mathbf M_X. The trace of a b-nef divisor is pseudo-effective. The pseudo-effective cone contains no line, so K_X+P\equiv0 and B-P+\mathbf M_X\equiv0. Since both summands are pseudo-effective, they are numerically trivial. The effective divisor B-P has zero intersection with an ample complete-intersection curve, hence B=P; also \mathbf M_X\equiv0. On a model f:V\to X where \mathbf M descends, E=f^*\mathbf M_X-\mathbf M_V\ge0 is exceptional by negativity. If E\ne0, choose an ample divisor H on V and put n=\dim V. Then

contradicting nefness. Thus we get E=0.

The nef part descends, and the completion has the ordinary klt discrepancies. Its dual complex is empty. This holds for every completion, proving the coregularity claim. Finally, numerical-triviality abundance [Gon13, Theorem 1.2] gives K_X+P\sim_{\mathbb Q}0.

Put L=-(K_V+C+\mathbf M_V) and write L\sim_{\mathbb Q}A+E with A ample and E\ge0. Choose \epsilon>0 small enough that (V,C+\epsilon E,\mathbf M) is generalized klt. Ordinary approximation with the ample summand \epsilon A gives a klt boundary \Delta\ge C+\epsilon E\ge R with -(K_V+\Delta)\sim_{\mathbb Q}(1-\epsilon)L nef and big. An additional small Kodaira perturbation makes this divisor ample, increases \Delta and preserves klt singularities. Thus (V,R) is of log Fano type. Finiteness follows from [Bra21, Theorem 2], and the uniform index bound from [BFMS20, Theorem 3].

On a smooth curve the nef part has nonnegative degree. A completion therefore forces either an elliptic curve with zero boundary and degree-zero nef part, or \mathbb P^1 with \deg D_{\mathrm{st}}\le2. In coregularity zero the completion has a coefficient-one point; the klt boundary has strictly smaller degree, so \deg D_{\mathrm{st}}<2.

Write D_{\mathrm{st}}=\sum_{i=1}^r(1-1/m_i)p_i, with m_i\ge2. The orbifold Euler characteristic on \mathbb P^1 is 2-\deg D_{\mathrm{st}}. If it is positive and r\ge3, the orbifold is spherical: its signature (m_1,\ldots,m_r), the list of orders at the points, is (2,2,m), (2,3,3), (2,3,4) or (2,3,5). The corresponding groups are dihedral, tetrahedral, octahedral or icosahedral. With at most two cone points, the presentation directly gives a finite cyclic group (possibly trivial). In all these cases it is well known that there is an abelian subgroup of index at most 60.

If \deg D_{\mathrm{st}}=2, the orbifold Euler characteristic is zero, and the Euclidean signatures are (2,2,2,2), (2,3,6), (2,4,4) and (3,3,3); see [GLM23, Proposition 5.3]. The corresponding orbifold fundamental groups contain their translation subgroups \mathbb Z^2 as normal subgroups of indices 2,6,4,3, respectively. In the elliptic-curve case the boundary is zero, so the regional orbifold fundamental group is the ordinary fundamental group of the elliptic curve, namely \mathbb Z^2.

By [GLM23, Theorem B], \pi_1^{\rm reg}(X,D) is finite of order at most 3840, or has a subgroup \mathbb Z^4 of index at most 96.

On the target, the complement of the common isomorphic open set has codimension at least two. On the source, for the contracted divisors, we get loop relations, including the prescribed orbifold relations. The boundary coefficients agree on uncontracted divisors. For a small map both complements have codimension at least two, so no additional relations occur.

By Lemma 2.10, take a projective extraction h:Y_0\to X of exactly the divisors extracted by g. The map Y_0\dashrightarrow Y extracts no divisors. On the complement of the exceptional locus, the fundamental groups agree with the corresponding fundamental group on X after removing codimension-two subsets. The image in \pi_1^{\mathrm{reg}}(X,P) of a loop around each E_i has finite order: it is the image of a regional local orbifold fundamental group of the klt pair (X,P), which is finite by [Bra21, Theorem 1]. Choose q_i divisible by this order. Imposing the q_i-th power relations gives a surjection from \pi_1^{\mathrm{reg}}(Y_0,R_{Y_0}). Lemma 3.1 then gives the surjection from Y.

Let B_{F_0} and P_{F_0} be defined by the equalities

For any 0<\alpha<1, the divisor -K_{F_0}-(1-\alpha)B_{F_0}-\alpha P_{F_0}-(1-\alpha)\mathbf M_{F_0} is big and nef. Indeed,

The sub-pair (F_0,P_{F_0}) is sub-klt. Since \phi only extracts glc places of (F,B_F,\mathbf M), the pair (F_0,B_{F_0},\mathbf M) is generalized log canonical with B_{F_0}\ge P_{F_0}. Hence for \alpha close enough to zero, the pair (F_0,(1-\alpha)B_{F_0}+\alpha P_{F_0},(1-\alpha)\mathbf M) is gklt by pulling back the interpolation of Lemma 2.8, is weakly Fano, and has (1-\alpha)B_{F_0}+\alpha P_{F_0}\ge R_{F_0}. Note that every exceptional coefficient of B_{F_0} is one, so any fixed exceptional coefficients smaller than one are allowed. Therefore by Theorem 2.19 we get the finiteness claim and the stated uniform index bound.

We use the first stage of the construction in [Mor21a, Lemma 2.36]. Start with a log resolution p:Z_0\to Z dominating W', on which \mathbf M descends. Write K_{Z_0}+B_0+\mathbf M_{Z_0}=p^*(K_Z+B+\mathbf M_Z), and let B_0^+ be the strict transform of B, plus the reduced p-exceptional divisor. Now let

The relative MMP over W' in that construction contracts \operatorname{Supp}E and produces a generalized dlt model of Fano type over W'. At each step its MMP divisor is \mathbb Q-linearly equivalent to the effective transform E_i. A divisorial contraction cannot contract a divisor outside \operatorname{Supp}E_i: a contracted curve through a general point of that divisor is not contained in \operatorname{Supp}E_i, and hence has nonnegative intersection with E_i, contrary to negativity of the contracted ray. Thus every original divisor of Z survives, and every remaining divisor exceptional over Z has discrepancy zero.

We do not perform the additional modifications arranging the inverse image of the non-klt locus as in [Mor21a, Lemma 2.36].

Finally, restricting to a sufficiently general fibre gives the last assertion. This can be checked on the same resolution, where horizontal exceptional divisors restrict to exceptional divisors and keep their discrepancy coefficients. All remaining exceptional and indeterminacy loci meet a sufficiently general fibre in codimension at least two.

Remove from the base its singular locus and the images of those prime divisors on Z whose images in W have codimension at least two. Denote the resulting big open subset by V and put U=f^{-1}(V). By [Nor83, Lemma 1.5.A], we can choose a smaller nonempty open set V^\circ\subset V, with divisorial complement, such that

is a topological fibre bundle. By generic flatness and generic smoothness, we may assume its connected fibre is F_{\mathrm{reg}}\setminus\operatorname{Supp}(R|_F), and that V^\circ avoids the images of the vertical boundary components.

Now, the loops of horizontal boundary components are represented, up to conjugation, by loops in a general fibre. Quotienting the homotopy sequence by their prescribed orbifold relations therefore gives an exact sequence

where H=\pi_1(U^\circ)/N_{\mathrm{hor}} and Q=\pi_1(V^\circ).

As noted by [Nor83, Fact 1.2], reinserting a divisor T_{E,j} with its orbifold coefficient is nothing else than imposing the relation \gamma_{E,j}^{n_{E,j}}=1. At a general smooth point, the image of its loop is conjugate to \delta_E^{m_{E,j}}, where \delta_E is a loop around E. Thus, these relations map to powers \delta_E^{m_{E,j}n_{E,j}} and their normal closure is generated by \delta_E^{k_E}: each exponent is divisible by k_E, and Bézout’s identity expresses k_E as an integral linear combination of these exponents. All components of U_{\mathrm{reg}}\setminus U^\circ reinserted at this step, other than the horizontal boundary components already treated, dominate divisors of V; any removed subsets of codimension at least two in U_{\mathrm{reg}} do not change the fundamental group. We obtain a commutative diagram

The vertical maps are surjective by the definition of the orbifold fundamental groups of pairs.

Let \alpha\in\pi_1^{\mathrm{reg}}(U,R|_U) such that \phi_2(\alpha)=1. We take \alpha' a preimage of \alpha by \phi_X. Then \psi_2(\alpha')\in\ker\phi_C, so it is a product of conjugates of \delta_E^{k_E} and their inverses. Loops around a divisor are well-defined up to conjugation. As \psi_2 is surjective, we may choose conjugates \eta_{E,j} of the loops around T_{E,j} such that \psi_2(\eta_{E,j})=\delta_E^{m_{E,j}}. Choose integers a_{E,j} satisfying \sum_j a_{E,j}m_{E,j}n_{E,j}=k_E. Then

Therefore, taking a suitable product \gamma' of conjugates of these \lambda_E and their inverses, we have \psi_2(\gamma')=\psi_2(\alpha') and \phi_X(\gamma')=1. Hence, by exactness of the top row, there exists \beta\in\pi_1^{\mathrm{reg}}(F,R|_F) such that \psi_1(\beta)=\alpha'\gamma'^{-1}. Thus

which proves \operatorname{im}\phi_1\supseteq\ker\phi_2. The reverse inclusion follows from commutativity, and \phi_2 is surjective because \psi_2 and \phi_C are surjective. Hence the bottom sequence is also exact. Finally, adding the "missing divisors" in Z_{\mathrm{reg}} possibly imposes further loop relations, yielding \pi_1^{\mathrm{reg}}(U,R|_U)\twoheadrightarrow\pi_1^{\mathrm{reg}}(Z,R).

Writing b_{E,j}=\operatorname{coeff}_{T_{E,j}}B, the generalized lc threshold satisfies

Thus R_W\le B_W. All coefficients of R_W are smaller than one. Lowering every coefficient-one component of a generalized dlt pair makes the pair generalized klt since each lc place has center in the coefficient-one boundary and its discrepancy strictly increases. Comparing the discrepancies, we get that (W,R_W) is klt.

Proposition 4.3 proves virtual abelianity and the rank bound 2c for every projective klt Calabi-Yau pair of dimension c with standard coefficients. Thus the hypothesis of Theorem 4.2 holds for every c. Taking the normal core of a finite-index free abelian subgroup gives the stated form.

Preparation. If the assumption holds in dimension c, then it also holds in every dimension k\le c: apply the hypothesis to the product of the given k-dimensional Calabi-Yau pair (V,C) with standard coefficients and E^{c-k}, where E is an elliptic curve, with boundary \operatorname{pr}_V^*C.

If G is the regional orbifold fundamental group of the given k-dimensional pair, then that of the product is G\times\mathbb Z^{2(c-k)}. Virtual abelianity and the bound 2c imply that G is virtually abelian of rank at most 2k. By Lemma 2.16 the product has absolute coregularity c.

We perform an induction on n=\dim X (simultaneously for all absolute coregularities at most c). Choose a generalized lc completion (X,B,\mathbf M) of coregularity a=\widehat{\operatorname{coreg}}(X,D)\le c. First take a small \mathbb Q-factorialization. The regional fundamental group and the completion’s coregularity are unchanged. Put P=D_{\mathrm{st}}, then (X,P) is klt, P\le B, and \pi_1^{\mathrm{reg}}(X,P)=\pi_1^{\mathrm{reg}}(X,D).
Case 1: K_X+P is pseudo-effective. Lemma 2.16 shows that (X,P) is a klt Calabi-Yau pair with standard coefficients and forces a=n. The assumption in dimensions n\le c proves the claim.
Case 2: K_X+P is not pseudo-effective. Run a (K_X+P)-MMP with scaling to a Mori fibre space f:Z\to W. The completion is carried crepantly through the MMP, since its log canonical divisor is \mathbb Q-linearly trivial. It remains generalized lc of coregularity a. We have \pi_1^{\mathrm{reg}}(Z,P_Z)\twoheadrightarrow\pi_1^{\mathrm{reg}}(X,P) by Lemma 3.1. If W is a point, (Z,P_Z) is log Fano, so \pi_1^{\mathrm{reg}}(Z,P_Z) is finite by Theorem 2.19.

Otherwise apply the canonical bundle formula, take a generalized dlt modification W'\to W of the base, and use Lemma 3.4 to obtain f':Z'\to W'. Choose the boundary R_{Z'} with standard coefficients using Lemma 3.2, so that

It satisfies R_{Z'}\le B_{Z'} and has all coefficients smaller than one. Since the completion on Z' is generalized dlt, the ordinary pair (Z',R_{Z'}) is klt.

Let R_{W'} be the gcd boundary from Lemma 3.5. By Lemma 3.6, the pair (W',R_{W'}) is klt and its generalized lc completion has coregularity at most a, by Remark 2.14. Thus its absolute coregularity is a'\le a. Since \dim W'<n, by induction we get virtual abelianity of rank at most 2a'\le2a for \pi_1^{\mathrm{reg}}(W',R_{W'}).

Let F' and F be general fibres of f' and f. We next verify that \pi_1^{\mathrm{reg}}(F',R_{Z'}|_{F'}) is finite, where R_{Z'} is the boundary just constructed using Lemma 3.2. The pair (F,P_Z|_F) is log Fano. Every prime divisor of F survives on F', and every additional divisor on F' is an lc place of the completion due to Lemma 3.4. If necessary, first take a small \mathbb Q-factorialization \widehat F\to F so that the log anticanonical divisor becomes nef and big, and the completion pulls back crepantly. By Lemma 2.10, there is an extraction F_0\to\widehat F of exactly these places. Hence F_0\dashrightarrow F' is small. The strict transform R_{F_0} of R_{Z'}|_{F'} on F_0 is the strict transform of P_Z|_F plus exceptional divisors with standard coefficients smaller than one. Lemma 3.3 gives that \pi_1^{\mathrm{reg}}(F_0,R_{F_0}) is finite. The small map identifies it with \pi_1^{\mathrm{reg}}(F',R_{Z'}|_{F'}) by Lemma 3.1.

Apply Lemma 3.5 to f' and let U=f'^{-1}(V) be the resulting open set. There is an exact sequence

By Lemma 2.3, the middle group is virtually abelian of rank at most 2a. The surjections

yield the desired statement.

Choose a canonical Weil divisor K_X and a rational function g\in\mathbb C(X)^* such that s(K_X+D)=\operatorname{div}_X(g). Let t be an s-th root of g in an algebraic closure of \mathbb C(X), and set L=\mathbb C(X)(t). This is a cyclic Galois extension of degree e dividing s, and t^e\in\mathbb C(X). If e<s, then \operatorname{div}_X(t^e)=e(K_X+D), contrary to the minimality of s. Thus [L:\mathbb C(X)]=s.

Let Y be the normalization of X in L. At a prime divisor P on X, the ramification index of this root extension is

The coefficient of K_X at P is an integer. Also m_i divides s, because s(K_X+D) is integral. At D_i, writing that integer as a_i, we obtain

At a prime divisor P outside \operatorname{Supp}D, the order of vanishing \operatorname{ord}_P(g) is divisible by s, so there is no other divisorial ramification.

Choose the canonical divisor on Y using the pullback of the rational top form defining K_X. At a divisor over D_i, the ramification coefficient is m_i-1=m_i(1-1/m_i). The canonical divisor formula therefore gives

Only the sum K_X+D is pulled back as a \mathbb Q-Cartier divisor; the equality is checked in codimension one. Thus K_Y is Cartier and linearly trivial. Finite crepant pullback preserves klt singularities by [KM98, Proposition 5.20], and since K_Y is Cartier, Y is in fact canonical.

Remove a codimension-two subset of X so that the remaining open set U is smooth and the branch divisors are smooth and disjoint. Locally along D_i\cap U, the normalized cover is a disjoint union of maps (u,z_2,\ldots,z_n)\mapsto(u^{m_i},z_2,\ldots,z_n). Away from the boundary it is étale. Thus p^{-1}(U) is smooth and the cover is an orbifold covering with the exact prescribed ramification. The complement of p^{-1}(U) in Y has codimension at least two, because p is finite. Removing these codimension-two subsets from the regular loci changes neither the fundamental group nor the orbifold loop relations. This proves the claimed exact sequence. Now apply [FGSW26, Theorem 1.3] to Y.

For any connected finite quasi-étale cover p:X'\to X, let D' be the codimension-one pullback of D. Then (X',D') is klt, and its regional orbifold fundamental group is a finite-index subgroup of \pi_1^{\mathrm{reg}}(X,D). It therefore has the same virtual rank. Let \rho:V\to X' be a projective resolution which is an isomorphism over X'_{\mathrm{reg}}. Killing the boundary loops and then filling in the exceptional divisors gives surjections

The last group is isomorphic to \mathbb Z^{2q(V)}: Hodge decomposition and Hodge symmetry give b_1(V)=h^{1,0}(V)+h^{0,1}(V)=2q(V); see [Huy05, Corollary 3.2.12]. The klt pair (X',D') has rational singularities [KM98, Theorem 5.22], so q(V)=q(X'). Hence 2q(X')\le\operatorname{rank}\pi_1^{\mathrm{reg}}(X,D)\le2\widehat{\operatorname{coreg}}(X,D). Taking the supremum of q(X') over all connected finite quasi-étale covers of X gives \widetilde q(X)\le\widehat{\operatorname{coreg}}(X,D).

By [Gon13, Theorem 1.2], we have K_X+D\sim_{\mathbb Q}0, and (X,D,0) is a completion. Apply Theorem 4.1 for the rank bound. When D has standard coefficients, Proposition 4.3 gives the stated cyclic cover, exact sequence, and rank 2\widetilde q(Y). For D=0, the exact rank 2\widetilde q(X) follows directly from [FGSW26, Theorem 1.3].

Set B=D and \mathbf M=\overline{-(K_X+D)}. The equality K_X+D+\mathbf M_X=0 defines a generalized klt, hence generalized lc, Calabi-Yau completion. The nef part descends to X, so the generalized discrepancies are exactly those of (X,D). Theorem 4.1 gives the sharp coregularity bound, and \widehat{\operatorname{coreg}}(X,D)\le\dim X gives the stated dimension bound.

For a finite generating set, write B_G(R) for the ball of radius R in G. A group with a finite-index subgroup \mathbb Z^r has word growth comparable to (1+R)^r: passing to a finite-index subgroup changes word lengths by at most multiplicative and additive constants, and balls in \mathbb Z^r have this growth. Apply Theorem 4.7. The obvious surjection \pi_1^{\mathrm{reg}}(X,D)\twoheadrightarrow\pi_1(X_{\mathrm{reg}}) gives that the latter group is also virtually abelian of rank at most 2c.

Choose a finite-index subgroup \mathbb Z^r of \pi_1(X_{\mathrm{reg}}). Its connected finite topological cover is algebraic. Normalize X in its function field to obtain a finite quasi-étale cover X_0\to X.

The preimage of X_{\mathrm{reg}} is a smooth open subset of X_0 whose complement has codimension at least two. Thus \pi_1((X_0)_{\mathrm{reg}})=\mathbb Z^r. The pulled-back pair is moreover klt. By the local finiteness theorem [Bra21], its regional fundamental groups are finite, hence have trivial image in \mathbb Z^r. The kernel of the surjection \pi_1((X_0)_{\mathrm{reg}})\to\pi_1(X_0) is normally generated by these local images, so this map is an isomorphism.

For a projective log resolution V\to X_0 that is an isomorphism over (X_0)_{\mathrm{reg}}, the kernel of \pi_1((X_0)_{\mathrm{reg}})\to\pi_1(V) is normally generated by loops around the exceptional divisors. Each loop lies in the image of a regional fundamental group and is therefore trivial. Consequently \pi_1(V)=\mathbb Z^r. Rational singularities and Hodge theory now give r=b_1(V)=2q(V)=2q(X_0), as in the proof of Corollary 4.4.

For any connected finite quasi-étale cover W\to X, \pi_1(W_{\mathrm{reg}}) is a finite-index subgroup of \pi_1(X_{\mathrm{reg}}). Its surjection onto the first homology modulo torsion of a resolution of W gives 2q(W)\le r. Together with the equality for X_0, this proves r=2\widetilde q(X). Finally, the isomorphism \pi_1((X_0)_{\mathrm{reg}})\simeq\pi_1(X_0) implies that every finite étale cover of (X_0)_{\mathrm{reg}} extends to a finite étale cover of X_0. The uniqueness of normalization finally shows that a finite quasi-étale cover of X_0 is this étale extension.

Virtual abelianity is obvious from theorem 4.7 and the surjection to \pi_1(X_{\mathrm{reg}}). Apply Lemma 4.9 to obtain r=2q and X_0\to X. The asserted finiteness criterion is clear as well.

Apply [MW25, Theorem 1.1] to the pulled-back nef pair on X_0. After a connected finite quasi-étale cover, its MRC fibration is a locally constant fibration of pairs over a projective klt variety Y with K_Y\equiv0. By [FGSW26, Theorem 1.3], there is a connected finite quasi-étale cover A\times Z\to Y, where A is abelian,+Z is projective canonical, K_Z\sim0, and \pi_1(Z_{\mathrm{reg}})=1. Base change gives a connected finite quasi-étale cover X'\to X_0 and a locally constant fibration of pairs

with fiber (F,D_F). These properties of the base change follow from the local product description: the fibers are connected and normal, and inverse images of the exceptional codimension-two subset of the base still have codimension at least two. The pair (F,D_F) is klt, F is rationally connected, and -(K_F+D_F) is nef by restriction.

By Lemma 4.9, the composite cover X'\to X_0 is étale. In particular, \pi_1(X'_{\mathrm{reg}})\to\pi_1(X') is an isomorphism and both groups are \mathbb Z^{2q}. A rationally connected projective variety of klt type is simply connected; see [MW25, Theorem 2.2]. Thus \pi_1(F)=1. Also \pi_1(Z)=1, since \pi_1(Z_{\mathrm{reg}})\twoheadrightarrow\pi_1(Z). The homotopy sequence of the locally trivial fibration identifies \pi_1(X') with \pi_1(A\times Z), so \dim A=q. Pulling back to the universal cover \mathbb C^q\times Z of the base trivializes the pair fibration, and its simply connected total space is

Finally, the restriction of this universal cover to X'_{\mathrm{reg}} is connected and simply connected, because \pi_1(X'_{\mathrm{reg}})\to\pi_1(X') is an isomorphism. Its total space is \mathbb C^q\times Z_{\mathrm{reg}}\times F_{\mathrm{reg}}. Taking fundamental groups of this product gives \pi_1(F_{\mathrm{reg}})=1.

Repeat the induction and constructions from the proof of Theorem 4.2, writing P=D_{\mathrm{st}} on a small \mathbb Q-factorialization of X. If K_X+P is pseudo-effective, Lemma 2.16 shows that (X,P) is Calabi-Yau and d equals its absolute coregularity, so d\le c_0. The hypothesis of this proposition, with k=d, gives the required solvable subgroup of \pi_1^{\mathrm{reg}}(X,P)=\pi_1^{\mathrm{reg}}(X,D) with index at most I_d. If the (K_X+P)-MMP ends in a Mori fibre space Z\to W with W a point, (Z,P_Z) is log Fano. Theorem 2.19 shows that \pi_1^{\mathrm{reg}}(Z,P_Z) is finite and has an abelian subgroup of index at most J(d). Its image in \pi_1^{\mathrm{reg}}(X,D) has the same index bound. For a base of dimension 1\le w<d, the proof of Theorem 4.2 constructs the exact sequence

where U\subset Z' is the indicated open set, F' is a general fibre of Z'\to W', and \pi_1^{\mathrm{reg}}(U,R_{Z'}|_U) surjects onto \pi_1^{\mathrm{reg}}(X,D). The group \pi_1^{\mathrm{reg}}(F',R_{Z'}|_{F'}) is finite and has an abelian subgroup of index at most J(d-w), by Lemma 3.3 and the comparison with F_0 in the proof of Theorem 4.2. By induction, \pi_1^{\mathrm{reg}}(W',R_{W'}) has a solvable subgroup of derived length at most 2w-1 and index at most B(w,c_0). Lemma 2.7 therefore gives a solvable subgroup of \pi_1^{\mathrm{reg}}(U,R_{Z'}|_U) of derived length at most 2w+1\le2d-1 and index at most B(w,c_0)(J(d-w)^2)!. For any surjection \rho:G\to Q and subgroup S\le G, the image \rho(S) has derived length at most that of S and [Q:\rho(S)]\le[G:S]. Thus the same bounds hold for the image subgroup in \pi_1^{\mathrm{reg}}(X,D). These estimates define B(d,c_0) recursively. For d=1, Proposition 2.20 gives an abelian subgroup of index at most 60 for every curve pair under consideration. A connected zero-dimensional normal variety is a point and has trivial fundamental group.

Every connected finite quasi-étale cover X'\to X, with the pulled-back boundary, is again klt Calabi-Yau. By [Amb05, Theorem 4.8], its Albanese map is surjective, so q(X')\le\dim X. The supremum defining \widetilde q(X) is therefore a maximum. Choose X'\to X with q(X')=\widetilde q(X). Ambro’s theorem also provides a finite étale base change A\to\operatorname{Alb}(X') after which the pair splits as

The composite Y\to X is finite quasi-étale, and \dim A=q(X')=\widetilde q(X). The pair (Z,D_Z), obtained by restricting (Y,D_Y) to a fibre of the projection Y\to A, is klt Calabi-Yau. If Z had a finite quasi-étale cover Z' with q(Z')>0, then A\times Z'\to X would be a finite quasi-étale cover with irregularity larger than \widetilde q(X), a contradiction. Thus \widetilde q(Z)=0.