The group of birational selfmaps of a conic fibration is uncountable Thanks: EF thanks the Institut Universitaire de France for the excellent working conditions. She warmly thanks Jérémy Blanc and Thomas Dedieu for many useful conversations. Enrica Floris Abstract . We prove that the group of birational selfmaps of a variety birational to a conic bundle is uncountable. 1. Introduction The group of birational self-maps of a projective variety is a very rich birational invariant. If X is a rationally connected variety, group-theoretic properties of \Bir(X) should mirror the geometry of X . One striking instance of this, is the recent result in [ RUvS25 ] : if a variety X is such that \Bir(X)\cong\Bir(\mathbb{P}^{n}) , then X is a rational n -dimensional variety, see [ CER+25 ] for more results in this direction. In this note, we focus on groups of birational selfmaps of varieties birational to conic fibrations. The aim of this note is to prove the following result. Theorem 1.1 . Let X,Y be projective varieties, let f\colon X\to Y be a fibration whose general fibre is a rational curve. Then \Bir(X/Y) is uncountable. Theorem 1.1 implies in particular that the group \Bir(X) is uncountable. Conic bundles are one of the outcomes of the MMP applied to a variety with non-pseudoeffective canonical divisor and there are several examples of rationally connected varieties which are birational to a conic bundle but are not rational, for example the cubic threefold and the stably rational, but not rational, threefold of [ BCTSSD ] . By a result of Rosenlicht (see [ FFZ25 , Proposition 1.9] and [ BFT21 , Proposition 2.5.1] ), if X is a rationally connected and non-rational variety of dimension 3, then \Autz(X) is trivial. It might be tempting to conjecture that in this case \Bir(X) is discrete, but by Theorem 1.1 this is not true and we have the following. Corollary 1.2 . Let X be a rationally connected projective variety of dimension 3. Assume that X is birational to a conic bundle and X is not rational. Then \Bir(X) is uncountable and for every variety Y birational to X we have \Autz(Y)=\{id_{Y}\} . The core of the proof of Theorem 1.1 follows essentially as in [ BLZ21 ] , where infinitely many involutions of a conic bundle were produced, and those are distinguished by their indeterminacy locus \Gamma . Our key observation is that we can find uncountably many such \Gamma . 2. Proof of Theorem 1.1 We start by proving some preparatory lemmas. We refer to [ KM98 ] for the basic notions on the MMP. The following lemma is a higher-dimensional version of the initial discussion in [ Ser96 ] . We prove that a (birationally) isotrivial fibration whose fibres are of general type is essentially a product. The result is standard, but we include a proof here for lack of an appropriate reference. Lemma 2.1 . Let f\colon Z\to C be a fibration from a smooth variety Z to a curve C . Assume that the general fibres of f are of general type and birational to each other. Let F be the canonical model of a general fibre of f . Then there is a curve B and a group G acting faithfully on F and B such that C\cong B/G and Z is birational to (F\times B)/G over C . Proof. We perform an MMP relative over C , which terminates with a good minimal model by [ BCHM10 ] . The canonical divisor on the endproduct of the MMP is thus relatively semiample and induces a contraction to a fibration f^{\prime}\colon Z^{\prime}\to X where K_{Z^{\prime}} is relatively ample. After passing to Z^{\prime} and by uniqueness of the canonical model , we can assume that f is isotrivial and the general fibre is isomorphic to F . By [ Kol87 , Theorem 2.11] there is a finite map \tau\colon\widetilde{C}\to C such that Z\times_{C}\widetilde{C} is birational to F\times\widetilde{C} over \widetilde{C} . Since the canonical divisor of the general fibre is ample, the map Z\times_{C}\widetilde{C}\dasharrow F\times\widetilde{C} is an isomorphism over an open set \widetilde{C}_{0}\subseteq\widetilde{C} . Let C_{0}\subseteq C be an open set such that \tau^{-1}C_{0}\subseteq\widetilde{C}_{0} and such that \tau\colon\widetilde{C}_{0}\to C_{0} is étale. Therefore, if \Delta\subseteq C_{0} is a disc, the preimage f^{-1}\Delta is isomorphic to F\times\Delta . Moreover, if \overline{C}_{0} is the universal cover of C_{0} , the base change Z_{0}\times_{C_{0}}\overline{C}_{0} is isomorphic to F\times\overline{C_{0}} , where Z_{0}=f^{-1}(C_{0}) . Indeed, there is a factorization \overline{C}_{0}\to\widetilde{C}_{0}\to C_{0} and an isomorphism Z_{0}\times_{C_{0}}\overline{C}_{0}\cong(Z_{0}\times_{C_{0}}\widetilde{C_{0}})\times_{\widetilde{C_{0}}}\overline{C_{0}}\cong(F\times\widetilde{C}_{0})\times_{\widetilde{C_{0}}}\overline{C_{0}}\cong F\times\overline{C}_{0}. Moreover, there is an action of \pi_{1}(C_{0}) on F\times\overline{C}_{0} such that Z_{0}\cong(F\times\overline{C}_{0})/\pi_{1}(C_{0}) and the induced action on F is given by the monodromy. With the local trivializations, we define a monodromy homomorphism \rho\colon\pi_{1}(C_{0})\to\Aut(F) . Let G be the image of \rho . The group G is finite because F is of general type. The subgroup \ker\rho has finite index in \pi_{1}(C_{0}) and therefore corresponds to an étale cover B_{0}\to C_{0} where \pi_{1}(B_{0}) identifies with \ker\rho (see for example [ Mil25 , Riemann existence theorem] ). We denote by Z_{B_{0}} the base-change Z_{0}\times_{C_{0}}B_{0} . There is an action of G\cong\pi_{1}(C_{0})/\pi_{1}(B_{0}) on Z_{B_{0}} such that Z_{0}\cong Z_{B_{0}}/G . Moreover, Z_{B_{0}}\cong(F\times\overline{C}_{0})/\pi_{1}(B_{0})\cong(F\times\overline{C}_{0})/\ker\rho\cong F\times(\overline{C}_{0}/\ker\rho)\cong F\times B_{0}. We conclude that Z_{0}\cong(F\times B_{0})/G . The curve B_{0} compactifies to a projective curve B and, thus, the variety Z is birational to (F\times B)/G . The group G acts faithfully on F because G\subseteq\Aut(F) and on B by the construction of B as finite cover. ∎ In the next lemma, we prove that general complete intersections of codimension 2 inside a conic bundle are not birational. Lemma 2.2 . Let Y be a smooth variety of dimension at least two embedded in a projective space \mathbb{P}^{m} . We denote by \mathcal{O}_{Y}(n) the restriction \mathcal{O}_{\mathbb{P}^{m}}(n)|_{Y} . Let a,b be positive integers and \mathcal{E}=\mathcal{O}_{Y}\oplus\mathcal{O}_{Y}(a)\oplus\mathcal{O}_{Y}(b) . Let X\subseteq\mathbb{P}_{Y}(\mathcal{E}) be such that the general fiber of the restriction g=f|_{X} of the projection f\colon\mathbb{P}_{Y}(\mathcal{E})\to Y is a rational curve. Let A be a very ample divisor on Y . Then there are two integers \bar{d}_{1},\bar{d}_{2} such that for every d_{i}\geq\bar{d}_{i} if H is such that \mathcal{O}_{X}(H)\sim\mathcal{O}(1)\otimes g^{*}\mathcal{O}_{Y}(d_{1}A) and H_{1}\in|H| is a general section, two general sections \Gamma,\Gamma^{\prime}\in|(H+d_{2}g^{*}A)|_{H_{1}}| are not birational. Proof. Let \mu\colon W\to X be a resolution of singularities and denote by E the divisor of discrepancies, such that K_{W}=\mu^{*}K_{X}+E. The divisor \mu^{*}(H+kf^{*}A) is base point free and big for every k>0 . Therefore, the general member of |\mu^{*}(H+kf^{*}A)| is smooth and connected by the Bertini theorem for irreductibility [ Jou83 ] , the Seidenberg theorem [ Sei50 ] and the generalized Seidenberg theorem [ BS95 , Theorem 1.7.1] and its proof. We fix H_{1}\in|H| and denote its pullback, which coincides with its strict transform, by \widetilde{H}_{1} . We notice that the restriction of g\circ\mu to \widetilde{H}_{1} is generically finite. By adjunction, we get K_{X}\sim(\mathcal{O}(-1)\otimes g^{*}\mathcal{O}_{Y}(D))|_{X} for a suitable divisor D on Y , and K_{\widetilde{H}_{1}}=\mu^{*}(K_{X}\otimes\mathcal{O}_{X}(1)\otimes g^{*}\mathcal{O}_{Y}(kA)\otimes\mathcal{O}_{X}(E))|_{\widetilde{H}_{1}}\sim\left(\mu^{*}g^{*}\mathcal{O}_{Y}(D+kA)\otimes\mathcal{O}_{X}(E)\right)|_{\widetilde{H}_{1}}. This divisor is big if k is big enough. Let \bar{d}_{1} be such that K_{\widetilde{H}_{1}} is big. Let d\geq 0 and H_{2,d}\in|H+dg^{*}A| and \Gamma_{d}=H_{1}\cap H_{2,d} . Then the restriction g|_{\Gamma_{d}} is birational onto its image. We denote by \widetilde{H}_{2,d} the strict transform of H_{2,d} in W , which coincides with its pullback and by \widetilde{\Gamma}_{d} the intersection \widetilde{H}_{1}\cap\widetilde{H}_{2,d} . Let d_{0} be a positive integer such that for every d\geq d_{0} the Kodaira dimension of \mu^{*}(K_{X}+2H)+E+d\mu^{*}g^{*}(A) equals \dim(Y) . By the adjunction formula, if d\geq d_{0} , the divisor K_{\widetilde{\Gamma}_{d}}\sim\mu^{*}(K_{X}+2H)|_{\widetilde{\Gamma}_{d}}+E|_{\widetilde{\Gamma}_{d}}+d\mu^{*}g^{*}(A)|_{\widetilde{\Gamma}_{d}} is big. Assume by contradiction that two general elements in |\mu^{*}(H+df^{*}A)|_{\widetilde{H}_{1}}| are birational to each other. Let \widetilde{\Gamma}_{d},\widetilde{\Gamma}^{\prime}_{d}\in|\mu^{*}(H+df^{*}A)|_{\widetilde{H}_{1}}| be two general elements and let Z\to\widetilde{H}_{1} be a resolution of the pencil. Let \overline{\Gamma} be the canonical model of \widetilde{\Gamma}_{d} . By Lemma 2.1 , there is a group G and a curve C such that Z is birationally equivalent to (\overline{\Gamma}\times C)/G and G acts on \overline{\Gamma}\times C diagonally and on \overline{\Gamma} and C faithfully. Since \widetilde{H}_{1} , and therefore Z , is of general type, both \Gamma and C are of general type. Since G acts faithfully on both \overline{\Gamma} and C , the projection morphism \pi\colon\overline{\Gamma}\times C\to(\overline{\Gamma}\times C)/G is étale in codimension 1. Hence, we have K_{\Gamma\times C}=\pi^{*}K_{(\Gamma\times C)/G} and K_{(\Gamma\times C)/G} is ample. This implies that (\Gamma\times C)/G is the canonical model of \widetilde{H}_{1} . Therefore, there is a map h\colon\widetilde{H}_{1}\dasharrow(\Gamma\times C)/G which is a composition of divisorial contractions, flips and a final morphism from the endproduct of the MMP to the canonical model. If \Delta is a big divisor on \widetilde{H}_{1} , then h_{*}\Delta is big. On the other hand, we have h_{*}\widetilde{\Gamma}_{d}=[\Gamma\times\{x\}] for some x for a general member of the pencil. This is a contradiction. ∎ Lemma 2.3 . Let X,Y be projective varieties and X\to Y be a fibration whose general fibre is a rational curve. Let a,b be positive integers. Then there exist a positive integer m , a variety X^{\prime} birational to X and Y^{\prime} smooth birational to Y and a diagram \textstyle{X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\alpha} \textstyle{X^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\gamma} \textstyle{\mathbb{P}_{\mathbb{P}^{m}}(\mathcal{O}\oplus\mathcal{O}(a)\oplus\mathcal{O}(b))\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{Y\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\beta} \textstyle{Y^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\delta} \textstyle{\mathbb{P}^{m}} where \alpha,\beta are birational and \gamma,\delta are embeddings. Proof. Let Y^{\prime}\to Y be a desingularisation, X^{\prime\prime} the normalisation of X\times_{Y}Y^{\prime} and f\colon X^{\prime\prime}\to Y^{\prime} the induced fibration. The variety Y is projective, thus it is embedded in a projective space \mathbb{P}^{m} . The sheaf f_{*}\omega_{X^{\prime\prime}/Y^{\prime}} is a rank 3 vector bundle over an open set U_{1} of Y^{\prime} . Moreover, if X_{1}=f^{-1}U_{1} , then we have an embedding of X_{1} in the projectivisation of f_{*}\omega_{X^{\prime\prime}/Y^{\prime}} \textstyle{X_{1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{\mathbb{P}_{U_{1}}(f_{*}\omega_{X_{1}/U_{1}})\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{U_{1}} After possibly shrinking U_{1} , we can assume that f_{*}\omega_{X_{1}/U_{1}} is a trivial vector bundle. Let U_{2} be a trivializing open set for \mathcal{O}(a)|_{Y^{\prime}} and \mathcal{O}(b)|_{Y^{\prime}} . Set U=U_{1}\cap U_{2} and X_{0}=f^{-1}U . We fix two isomorphisms \varphi\colon\mathbb{P}_{U}(f_{*}\omega_{X/U})\to\mathbb{P}^{2}\times U and \psi\colon\mathbb{P}_{U}(\mathcal{O}_{U}\oplus\mathcal{O}(a)|_{U}\oplus\mathcal{O}(b)|_{U})\to\mathbb{P}^{2}\times U . We have thus \textstyle{X_{0}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{i} \textstyle{\mathbb{P}_{U}(f_{*}\omega_{X/U})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\psi^{-1}\circ\varphi} \textstyle{\mathbb{P}_{U}(\mathcal{O}_{U}\oplus\mathcal{O}(a)|_{U}\oplus\mathcal{O}(b)|_{U})\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{U} Finally, we set X^{\prime} to be the Zariski closure of \psi^{-1}\circ\varphi\circ i(X_{0}) in \mathbb{P}_{Y^{\prime}}(\mathcal{O}_{Y^{\prime}}\oplus\mathcal{O}(a)|_{Y^{\prime}}\oplus\mathcal{O}(b)|_{Y^{\prime}}) . ∎ The following theorem is a refined version of [ BLZ21 , Proposition 6.9] and is the main result of this note. Theorem 2.4 . Let X,Y be projective varieties, let f\colon X\to Y be a fibration whose general fibre is a rational curve. Then there exists an uncountable set I and a group homomorphism \Bir(X/Y)\to\bigoplus_{I}\mathbb{Z}/2. Proof. By Lemma 2.3 , we can assume that X\to Y decomposable in the sense of [ BLZ21 , Definition 6.8] . We follow the proof of [ BLZ21 , Proposition 6.9] where the group homomorphism is defined. We only need to prove that the index set I is uncountable. By Lemma 2.2 there are two integers \bar{d}_{1},\bar{d}_{2} such that for every d_{i}\geq\bar{d}_{i} if H is such that \mathcal{O}_{X}(H)\sim\mathcal{O}(1)\otimes f^{*}\mathcal{O}_{Y}(d_{1}A) and H_{1}\in|H| is a general section, two general sections \Gamma,\Gamma^{\prime}\in|(H+d_{2}f^{*}A)|_{H_{1}}| are not birational. We fix such d_{1},d_{2} . In the proof of [ BLZ21 , Proposition 6.9] , instead of chosing two sections in the same linear system |H_{0}+dF| , we chose H_{1}\in|H_{0}+d_{1}F| and H_{2}\in|H_{0}+d_{2}F| with we chose thus d\geq m_{0} . Therefore, if \Gamma,\Gamma^{\prime} are two complete intersections of the form H_{1}\cap H_{2} with H_{2}\in|H_{0}+d_{2}F| general, then \Gamma,\Gamma^{\prime} are not birational and the models (X/Y,f(\Gamma)) , (X/Y,f(\Gamma^{\prime})) are not in the same equivalence class in the sense of [ BLZ21 , Definition 2.12] . Moreover, their classes have non-vanishing image in the direct sum \displaystyle\oplus_{M(X/Y)}\mathbb{Z}/2 . ∎ Proof of Theorem 1.1 . Since the group \Bir(X/Y) has a quotient of uncountable cardinality, it is uncountable. ∎