Global smoothing of singular Fano and Calabi-Yau varieties Thanks: AT was partially supported by a Simons Dissertation Fellowship in Mathematics Anda Tenie Abstract . We study the problem of smoothing Fano and Calabi-Yau varieties with isolated rational lci singularities. For Fano varieties we show that any such Y Y admits a deformation to a Fano variety with only 1 1 -rational singularities, and if all the singularities of Y Y are not 1 1 -rational, then Y Y is smoothable. For Calabi-Yau varieties, we show first that any such Y Y deforms to a Calabi-Yau variety with only 1 1 -Du Bois singularities. Moreover, if all the singularities of Y Y are not 1 1 -Du Bois then Y Y is smoothable. When allowing 1 1 -liminal singularities, we give a global criterion in terms of the Hodge-Du Bois numbers of Y Y which ensures Y Y is smoothable. These theorems recover and generalize results for threefolds of Friedman, Namikawa, Namikawa-Steenbrink, Gross, and Friedman-Laza. In higher dimensions, our results provide alternative smoothing conditions and also extend the work of Friedman-Laza from the case of hypersurface singularities to lci singularities. 1. Introduction In this paper, we study the problem of smoothing a Fano or Calabi-Yau variety Y satisfying certain singularity conditions. The question of smoothing such varieties has been studied extensively in the case of threefolds [ 11 ] , [ 26 ] , [ 27 ] , [ 25 ] , [ 15 ] , [ 28 ] and, more recently, in higher dimensions [ 10 ] , [ 5 ] , [ 6 ] , [ 9 ] . The construction of smoothings of singular Fano or Calabi-Yau varieties is based on a delicate interplay between local and global phenomena. Even when global deformations are unobstructed and each singularity admits a local smoothing, there is no guarantee that these local directions glue to a single global deformation that smooths all singular points. Definition 1.1 . (i) We say Y is a singular Fano variety if Y is a projective algebraic variety over \mathbb{C} of dimension n\geq 3 with isolated rational lci (local complete intersection) singularities such that \omega_{Y}^{-1} is ample. (ii) By a singular Calabi-Yau variety Y we will mean that Y is a projective algebraic variety over \mathbb{C} (or, more generally, a certain compact complex analytic space; see Remark 4.1 ) of dimension n\geq 3 with isolated rational lci singularities such that \omega_{Y}\cong\mathcal{O}_{Y} and H^{1}(\mathcal{O}_{Y})=0. Definition 1.2 . We say that Y is smoothable if there exists a flat analytic deformation \pi:\mathcal{Y}\to\Delta over the unit disk with special fiber \mathcal{Y}_{0}\simeq Y and smooth general fiber. Remark 1.3 . When Y is Calabi-Yau and projective, to get projective smoothings one should also assume, for instance, that H^{2}(\mathcal{O}_{Y})=0, although this is not necessary if one only needs analytic smoothings. To show that certain singular varieties are smoothable, we impose local criteria on the singularities of Y as well as a global condition on the variety Y (in the Calabi-Yau case). Our approach is inspired by the important work of Friedman-Laza, which initiated the study of smoothing criteria in terms of the recently introduced k -rational and k -Du Bois singularities. These notions emerged from recent advances in Hodge theory and refine the classical notions of rational and Du Bois singularities. For more details, one can consult Section 2.2 and, in particular, we refer to Definition 2.6 for the notions of 1-rational and 1-Du Bois singularities which will be of particular importance in our local conditions. Our global criterion in the Calabi-Yau case is expressed in terms of the Hodge-Du Bois numbers of Y, which are singular analogues of the Hodge numbers of a smooth projective variety, and are defined by \underline{h}^{p,q}(Y):=\dim_{\mathbb{C}}\mathbb{H}^{q}(Y,\underline{\Omega}^{p}_{Y}), where \underline{\Omega}^{p}_{Y} denotes the p -th graded piece of the Du Bois complex of Y (see Section 2.1 ). 1.1. Statement of main results We begin by stating our results, and then discuss their consequences and relation to prior work. We start with the case when Y is a Calabi-Yau variety. Assuming the singularities are not 1-Du Bois, we find that no global conditions are necessary to obtain a smoothing: Theorem 1.4 . (Theorem 4.5 ) Suppose Y is a singular Calabi-Yau variety with rational, isolated lci singularities. Then Y can be deformed to a Calabi-Yau variety whose singularities are 1-Du Bois. Moreover, if one assumes in addition that all the singularities of Y are not 1-Du Bois, then Y is smoothable, and every small smoothing is Calabi-Yau. If one allows 1-liminal singularities (i.e. singularities that are 1-Du Bois but not 1-rational) we find an appropriate global condition which guarantees the existence of a smoothing in terms of the Hodge-Du Bois numbers of Y : Theorem 1.5 . (Theorem 4.7 ) Suppose Y is a singular Calabi-Yau variety of dimension n with rational, isolated lci singularities. Assume that we have the following equality of Hodge-Du Bois numbers \underline{h}^{n-1,2}(Y)=\underline{h}^{1,n-2}(Y). Then Y can be deformed to a Calabi-Yau variety whose singularities are 1-rational. Moreover, if one assumes in addition that all the singularities of Y are not 1-rational, then Y is smoothable, and every small smoothing is Calabi-Yau. See the discussion after Corollary 1.18 , which explains that the equality of Hodge-Du Bois numbers is a natural higher-dimensional generalization of prior smoothing conditions given by Namikawa-Steenbrink [ 25 ] for threefolds. Using the methods of this paper, we are also able to remove a technical assumption in a theorem of Friedman-Laza which deals with the case of hypersurface singularities (see Proposition 1.12 ). We now turn to the case when Y is a Fano variety. We show: Theorem 1.6 . (Theorem 3.8 ) Suppose Y is a singular Fano variety with rational, isolated lci singularities. Then Y can be deformed to a Fano variety whose singularities are 1-rational. Moreover, if one assumes in addition that all the singularities of Y are not 1-rational, then Y is smoothable, and every small smoothing is Fano. Remark 1.7 . If one further assumes that Y has hypersurface singularities in the context of all of the smoothing theorems above, by using Proposition 2.7 we can show along the lines of [ 10 ] the stronger result that there exists a flat analytic deformation over a disk \mathcal{Y}\to\Delta of Y with smooth total space \mathcal{Y} . Of course, this implies Y is smoothable as the general fiber of the family is smooth. We also show the following unobstructedness result, which has a simple proof: Theorem 1.8 . (Theorem 3.1 ) Let Y be a Fano variety with isolated lci singularities. Then \mathbb{T}^{i}_{Y}=0 for i\geq 2 and, in particular, deformations of Y are unobstructed. For an extension outside of the lci case, see Theorem 3.4 . 1.2. Related work For a history of the problem of smoothing Fano and Calabi-Yau threefolds we refer the reader to [ 10 ] . We mention here some theorems which are relevant to the results of this paper. For Calabi-Yau threefolds, Namikawa-Steenbrink show that \mathbb{Q} -factoriality is an appropriate global assumption that guarantees the existence of smoothings: Theorem 1.9 ( [ 25 ] ) . Suppose Y is a projective threefold such that \omega_{Y}\cong\mathcal{O}_{Y} and h^{1}(\mathcal{O}_{Y})=0 and Y has rational isolated hypersurface singularities. Assume moreover that Y is \mathbb{Q} -factorial. Then Y is smoothable. Still for threefolds, in the case of lci singularities, Gross proves Y is smoothable if it has a crepant resolution and no ODPs (ordinary double points). More precisely, he shows: Theorem 1.10 ( [ 15 ] ) . Let Y be a compact analytic threefold with isolated rational lci singularities, with \omega_{Y}\cong\mathcal{O}_{Y}, and admitting a Kähler resolution \widehat{Y}\to Y such that \omega_{\widehat{Y}}\cong\mathcal{O}_{\widehat{Y}}. Then there is a deformation of Y which smooths all singular points of Y except possibly the ordinary double points of Y. In particular, if Y has no ordinary double points, then Y is smoothable. In higher dimensions, there has been recent progress on smoothing Calabi-Yau varieties with isolated rational hypersurface singularities. Indeed, Friedman-Laza [ 10 ] give smoothing results for Calabi-Yau varieties Y whose singularities are either strongly 1-irrational or are 1-Du Bois but not 1-rational (also known as 1-liminal). Note that the strongly 1-irrational condition is stricter than the not 1-Du Bois condition, and it involves another technical assumption on the splitting of a certain short exact sequence (see [ 10 , Definition 2.6.(iv)] ). Denote by Z^{\prime} the union of the 1-liminal singularities and by Z^{\prime\prime} the union of the strongly 1-irrational ones. Let L be the union of the links of each singular 1-liminal point. If x is a 1 -liminal hypersurface singularity, then by [ 3 , Section 4.11] (see also [ 8 , Corollary 6.14] ), \dim_{\mathbb{C}}\Gr_{F}^{n-1}H^{n}(L_{x})=1 , and let \varepsilon_{x} be a generator. Theorem 1.11 . ( [ 10 , Theorem 5.8.] ) Let Y^{\prime}\to Y be a resolution over the points of Z^{\prime\prime} , so that the singular locus of Y^{\prime} is Z^{\prime} . Let \varphi\colon H^{n}(L)\to H_{n-1}(Y^{\prime}) be the composition H^{n}(L)\cong H_{n-1}(L)\to H_{n-1}(Y^{\prime}). Finally, suppose that, for every x\in Z^{\prime\prime} , there exists a nonzero a_{x}\in\mathbb{C} such that (1) \sum_{x\in Z^{\prime\prime}}a_{x}\varphi(\varepsilon_{x})=0\in H_{n-1}(Y^{\prime}). Then there exists a first-order global smoothing \theta . If the deformations of Y are unobstructed, then there is a smoothing \mathcal{Y}\to\Delta with smooth total space \mathcal{Y} . In fact, we can strengthen this result as follows: Proposition 1.12 . In Theorem 1.11 , one may replace the hypothesis that the points of Z^{\prime} are strongly 1 -irrational with the weaker hypothesis that the points of Z^{\prime} are not 1 -Du Bois . Equivalently, the assumption on the splitting of a certain short exact sequence appearing in the definition of strongly 1 -irrational singularities is unnecessary. Remark 1.13 . By the following recent result of Friedman, deformations of Y are unobstructed so Theorem 1.11 and Proposition 1.12 do indeed give smoothings. Theorem 1.14 . ( [ 12 , Theorem 1.6] ) Let Y be a singular Calabi-Yau variety of dimension n\geq 3 , such that the singularities of Y are isolated, lci, and Du Bois. Then the deformations of Y are unobstructed. Let us now mention relevant previous work in the Fano case. For Fano threefolds, Namikawa [ 27 ] showed that if Y has Gorenstein terminal singularities then Y is smoothable. This was recently generalized by Friedman-Laza to Fano varieties with isolated rational hypersurface singularities: Theorem 1.15 . ( [ 10 , Corollary 4.11] ) Suppose that Y is a Fano variety with only isolated rational hypersurface singularities, such that one of the following holds: (i) \dim Y=3 ; (ii) The singularities of Y are 1 -irrational, there exists an element H\in|\omega_{Y}^{-1}| with H\cap Y_{\text{sing}}=\emptyset , and H^{3}(Y,T^{0}_{Y})=H^{n-3}(H,\Omega^{1}_{H})=0 . (iii) \dim Y\geq 4 , the singularities of Y are 1 -liminal, there exists an element H\in|\omega_{Y}^{-1}| with H\cap Y_{\text{sing}}=\emptyset , and H^{n-3}(H,\Omega^{1}_{H})=0 . Then Y is smoothable. 1.3. Consequences of our results and relations to prior work In dimension three, having 1-rational singularities simply means being smooth. Moreover, 1-liminal singularities are precisely ODPs. Hence, our Theorem 1.4 says: Corollary 1.16 . Let Y be a singular Calabi-Yau threefold with rational, isolated, lci singularities. Then Y can be deformed to a Calabi-Yau with only ODP singularities. Moreover, if Y has no ODP singularities, then Y is smoothable. Remark 1.17 . Note this recovers the theorem of Gross (Theorem 1.10 ) and, moreover, we make no assumption on the existence of a crepant resolution. If we allow ODPs and hope for a smoothing result, it is known that a global condition is necessary, as Namikawa [ 26 ] provided an example of a Calabi-Yau threefold with one ordinary double point that is not smoothable. We have the following Corollary of Theorem 1.5 : Corollary 1.18 . Suppose Y is a singular Calabi-Yau threefold with rational isolated lci singularities. Assume moreover that we have the following equality of Hodge-Du Bois numbers \underline{h}^{2,2}(Y)=\underline{h}^{1,1}(Y). Then Y is smoothable. For threefolds with rational singularities, Park and Popa [ 30 , Theorem E] show the equality \underline{h}^{2,2}=\underline{h}^{1,1} if and only if Y is \mathbb{Q} -factorial (see also [ 25 , Theorem 3.2.] which assumes rational isolated hypersurface singularities). Hence, Corollary 1.18 recovers Theorem 1.9 of Namikawa-Steenbrink and, moreover, generalizes it to the case of lci singularities. In higher dimensions, one should think of our Theorem 1.5 as a generalization of Theorem 1.9 , where the condition \underline{h}^{n-1,2}=\underline{h}^{1,n-2} is the natural replacement for \mathbb{Q} -factoriality. The equality \underline{h}^{n-1,2}=\underline{h}^{1,n-2} is related to Poincaré duality in the singular case. For a more detailed discussion of this, see Remark 4.10 . We now discuss how our results relate to prior work in the case when Y is a Calabi-Yau of higher dimensions. In the general case of lci singularities, to our knowledge, no prior results were known in dimension \geq 4. Back to the case of hypersurface singularities, we can replace the requirement in ( 1 ) by the hypothesis on the Hodge-Du Bois numbers as in Theorem 1.5 . One can think of our stronger condition as implying the vanishing of an appropriate morphism, rather than the vanishing of a particular element as in ( 1 ). We hope this condition is more natural as it directly relates to Poincaré duality. We now move to the case of singular Fano varieties. We have the following Corollary of Theorem 3.8 : Corollary 1.19 . Suppose Y is a singular Fano threefold with rational isolated lci singularities. Then Y is smoothable. Note that this recovers and generalizes to the lci case the previous work for Fano threefolds of Namikawa and Friedman-Laza. In higher dimensions, and in the general setting of lci singularities, no prior results appear to be known. Back to the case of hypersurface singularities, note that our Theorem 3.8 recovers (i) of Theorem 1.15 of Friedman-Laza and shows that none of the cohomological conditions in (ii) or (iii) are needed. Finally, let us briefly comment on our Theorem 3.1 , which shows \mathbb{T}_{Y}^{2}=0 (and, in particular, unobstructedness) for Fano varieties with isolated lci singularities. We note that this is new when Y has Du Bois but not rational singularities (see Remark 3.2 ). Unobstructedness for Fano varieties was previously known in the lci 1-Du Bois (not necessarily isolated) case by Friedman-Laza [ 10 , Theorem 4.5] with a simple proof as in our work, which uses Nakano-type vanishing. Moreover, Friedman [ 12 , Theorem 3.17] recently proved unobstructedness for Fano varieties with rational isolated lci singularities by a different approach involving T^{1} -lifting (first proving an unobstructedness result in the log Calabi-Yau case). 1.4. Strategy of proof The common theme in the smoothing results is the comparison between global and local deformation spaces. Globally, first-order deformations of Y are parametrized by \mathbb{T}^{1}_{Y}=\Ext^{1}(\Omega^{1}_{Y},\mathcal{O}_{Y}) , while locally the deformation of the germ (Y,x) at a singular point x is governed by T^{1}_{Y,x} , where T^{1}_{Y} is the sheaf \mathcal{E}xt^{1}(\Omega^{1}_{Y},\mathcal{O}_{Y}), supported on the singular points. Since the singularities are isolated lci, local deformations are unobstructed and each point has local smoothing directions. Global deformations of Y are unobstructed as well, so classes in \mathbb{T}^{1}_{Y} integrate to actual deformations. If \eta were surjective we could lift local smoothing directions directly, but this is rarely the case; instead, we must identify suitable global classes whose images provide the required local smoothings. In the case of hypersurface singularities, one has a good handle on which one-parameter global deformations give smoothings of a singularity x in terms of its Kodaira-Spencer class u\in T^{1}_{Y,x}. (for more details see Proposition 2.7 ). In the case of lci singularities there is no longer such a criterion. Instead, we use a result of Gross (see Lemma 2.9 ) which, in our situation, roughly says that if \operatorname{im}(\mathbb{T}^{1}_{Y}\to T^{1}_{Y,x})\not\subset m_{x}T^{1}_{Y,x} then the general fiber of the Kuranishi family over \text{Def}(Y) smooths the singular point x. Note that we can no longer control which specific first-deformation gives a smoothing, but, as long as the whole image of \mathbb{T}^{1}_{Y} doesn’t land in m_{x}T^{1}_{Y,x} we can obtain some smoothing. Let us briefly explain how one lifts appropriate local directions to global ones in the case of a Calabi-Yau variety with 1-liminal singularities. Let \pi:\widehat{Y}\to Y be a strong log resolution with exceptional divisor E. If x is a singularity, let (X_{x},x) denote its germ, where X_{x} is a contractible Stein representative. By restriction we have a log resolution \pi:\widehat{X}_{x}\to X_{x} with exceptional divisor E_{x}=\pi^{-1}(x). Consider the commutative diagram below with exact row (coming from the local cohomology long exact sequence): {\lx@inpgf@ignorespaces{\mathbb{T}^{1}_{Y}}} {\lx@inpgf@ignorespaces{\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x}))}} {\lx@inpgf@ignorespaces{H^{2}(\Omega_{\widehat{Y}}^{n-1}(\log E)(-E))}} {\lx@inpgf@ignorespaces{\bigoplus\limits_{x\in\mathrm{Sing}(Y)}T^{1}_{Y,x}}} \scriptstyle{\lx@inpgf@ignorespaces h} \scriptstyle{\lx@inpgf@ignorespaces\eta} \scriptstyle{\lx@inpgf@ignorespaces h^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces\oplus s_{x}} Putting together Proposition 4.4 with the condition on the Hodge-Du Bois numbers, one can show h^{\prime} is zero, so h is surjective. This is the key result that allows us to lift local first-order deformations to global ones. One concludes by showing that the image of \eta cannot be contained in m_{x}T^{1}_{Y,x}. Unlike in the hypersurface case, for the more general case of lci singularities the vanishing of the morphism h^{\prime} is crucial. Otherwise, it seems difficult to say which elements not in m_{x}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})) are mapped to zero by h^{\prime}. This was not a concern in the hypersurface case where m_{x}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x}))=0 so any nontrivial element will come from a smoothing direction in which case one can just impose a condition along the lines of ( 1 ). Acknowledgements. I am grateful to Mihnea Popa for insightful conversations throughout the project and for constant support and encouragement during the preparation of this paper. I am also thankful to Wanchun Shen and Anh Duc Vo for many useful discussions, and to Robert Friedman and Sung Gi Park for helpful comments and suggestions that have sharpened some of the statements of this paper. 2. Preliminaries 2.1. Hodge Theory of singularities Let Y be an algebraic complex variety. To extend the usual de Rham complex from the smooth case to singular varieties, fix a hyperresolution \varepsilon_{\bullet}:Y_{\bullet}\to Y . Following ideas of Deligne, Du Bois defined \underline{\Omega}_{Y}^{\bullet}:=\mathbf{R}\varepsilon_{\bullet*}\Omega_{Y_{\bullet}}^{\bullet} , an object in the derived category of filtered differential complexes on Y , and proved that it is independent of the chosen hyperresolution [ 4 ] . One can associate a filtration F^{p}\underline{\Omega}_{Y}^{\bullet}:=\mathbf{R}\varepsilon_{\bullet*}\Omega_{Y_{\bullet}}^{\geq p} by recalling that \Omega_{Y_{i}}^{\bullet} is filtered by \Omega_{Y_{i}}^{\geq p}. The p -th Du Bois complex of Y is then \underline{\Omega}_{Y}^{p}:=\operatorname{gr}_{F}^{p}\underline{\Omega}_{Y}^{\bullet}[p] . For further background on Du Bois complexes, see [ 4 ] , [ 17 , Chapter V] , and [ 31 , §7.3] . We collect some properties of Du Bois complexes (1) For every p\geq 0 there is a natural morphism \Omega_{Y}^{p}\to\underline{\Omega}_{Y}^{p} , which is an isomorphism when Y is smooth. Here \Omega_{Y}^{p} denotes the sheaf of Kähler differentials on Y ; see [ 4 ] or [ 31 , p. 175] . In particular, for all i>0 , the cohomology sheaves \mathcal{H}^{i}(\underline{\Omega}_{Y}^{p}) are supported on the singular locus of Y . (2) We have \underline{\Omega}^{p}_{Y}=0 for p<0 and p>n , where n=\dim Y . Moreover \mathcal{H}^{q}\underline{\Omega}^{p}_{Y}=0 if p+q>n (see [ 37 , (4.1)] ). (3) There is a Hodge-to-de Rham spectral sequence E_{1}^{p,q}=\mathbb{H}^{q}(Y,\underline{\Omega}_{Y}^{\,p})\Rightarrow H^{p+q}(Y,\mathbb{C}) , which degenerates at E_{1} when Y is projective (see [ 4 , Thm. 4.5(iii)] or [ 31 , Prop. 7.24] ). (4) Let Z\subset Y be a closed reduced subscheme, and let \pi:\widehat{Y}\to Y be a resolution of singularities that is an isomorphism over Y\setminus Z . Denote by E:=(\pi^{-1}(Z))_{\mathrm{red}} the reduced exceptional divisor, which is a simple normal crossing divisor. Then, by [ 31 , Ex. 7.25] , for every p there is an exact triangle \mathbf{R}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E)\longrightarrow\underline{\Omega}_{Y}^{\,p}\longrightarrow\underline{\Omega}_{Z}^{\,p}\xrightarrow{+1}. In particular, if Y has isolated singularities \mathbf{R}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E)\simeq\underline{\Omega}_{Y}^{p} for p>0. Definition 2.1 . The Hodge-Du Bois numbers of Y are defined by \underline{h}^{p,q}(Y)=\dim_{\mathbb{C}}\mathbb{H}^{q}(Y,\underline{\Omega}_{Y}^{p}). If Y is smooth, then \underline{\Omega}_{Y}^{\bullet}=\Omega_{Y}^{\bullet} and \underline{h}^{p,q}(Y)=h^{p,q}(Y). Let \pi:\widehat{Y}\to Y be a strong log resolution with exceptional divisor E. If x is a singularity, let (X_{x},x) denote its germ, where X_{x} is a contractible Stein representative. By restriction we have a log resolution \pi:\widehat{X}_{x}\to X_{x} with exceptional divisor E_{x}=\pi^{-1}(x). Let U_{x}:=X_{x}\setminus\{x\}. Let us now restrict our attention to the case when Y has isolated lci singularities . As mentioned above for p>0 , \underline{\Omega}_{Y}^{p}\simeq\mathbf{R}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E) Then \mathcal{H}^{q}(\underline{\Omega}_{Y}^{p})=R^{q}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E), which for q>0 is supported on the singular points. Then H^{0}(R^{q}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E))=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{q}(\Omega^{p}_{\widehat{X}_{x}}(\log E_{x})(-E_{x})), since X_{x} is a small Stein neighborhood around each singular point. The dimensions of these spaces are the so-called Du Bois invariants of the isolated singularity. Definition 2.2 . [ 38 ] For q>0, define the Du Bois invariants of an isolated singularity (X,x) to be b^{p,q}:=\dim H^{q}(\widehat{X},\Omega^{p}_{\widehat{X}}(\log E)(-E)). Theorem 2.3 ( [ 38 , Theorem 5] ) . If (X,x) is an isolated lci singularity, then b^{p,q}=0 for all q>0 unless p+q\in\{n-1,n\} . Thus, by the above theorem we see that in the case of isolated lci singularities if q>0, we have \mathcal{H}^{q}(\underline{\Omega}_{Y}^{p})=0 except for p+q\in\{n-1,n\}. Steenbrink also defines the link invariants: \ell^{p,q}:=\dim H^{q}(E,\Omega_{\widehat{X}}^{p}(\log E)|_{E})=\dim\text{Gr}_{F}^{p}H^{p+q}(L). The link L of an n -dimensional isolated complete intersection singularity has H^{i}(L)=0 except for i\in\{0,n-1,n,2n-1\} . Hence, this immediately implies \ell^{p,q}=0 unless p+q\in\{0,n-1,n,2n-1\} . Remark 2.4 . Notice that using the restriction short exact sequence and the results on vanishing of b^{p,q} and the vanishing of the cohomology of the link one immediately also has that if (X,x) is an isolated lci singularity then H^{q}(\widehat{X},\Omega^{p}_{\widehat{X}}(\log E))=0 for all q>0 unless p+q\in\{n-1,n,2n-1\}. 2.2. Introduction to higher singularities Because of their close connection to the minimal model program, rational and Du Bois singularities have been studied extensively. In particular, Kawamata log terminal singularities are rational, while log canonical singularities are Du Bois. Advances in Hodge theory have prompted intensive study of the higher versions: k -Du Bois and k -rational singularities (see [ 21 ] , [ 18 ] , [ 8 ] , [ 7 ] , [ 24 ] , [ 2 ] , [ 36 ] ). When k=0 one recovers the classical notions (rational and Du Bois), and as k increases, the singularities get progressively milder, until they become smooth. We begin by recalling the definitions of k -rational and k -Du Bois, and then explain how this definition becomes more manageable in our situation i.e., in the case of rational isolated lci singularities. We focus specifically on the 1-rational and 1-Du Bois conditions, which will be useful notions for us in the context of local deformations. Let Y be a complex algebraic variety with lci singularities. For each integer p\geq 0 there is a natural composition \Omega_{Y}^{p}\xlongrightarrow{\varphi_{p}}\underline{\Omega}_{Y}^{p}\xlongrightarrow{\phi_{p}}\mathbb{D}_{Y}(\underline{\Omega}_{Y}^{n-p}). We say that Y has k -Du Bois if \varphi_{p}:\Omega_{Y}^{p}\to\underline{\Omega}_{Y}^{p} is an isomorphism in D^{b}_{\mathrm{Coh}}(Y) for all 0\leq p\leq k. On the other hand, Y is said to have k -rational singularities if the canonical morphisms \phi_{p}\circ\varphi_{p}:\Omega_{Y}^{p}\to\mathbb{D}_{Y}(\underline{\Omega}_{Y}^{n-p}) are isomorphisms for all 0\leq p\leq k , where \mathbb{D}_{Y}(-):={\bf R}\mathcal{H}om(-,\omega_{Y}^{\bullet})[-n] . If we moreover assume isolated singularities then Y has k -Du Bois singularities if and only if \Omega^{p}_{Y}\cong R^{0}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E) and R^{q}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E)=0\quad\text{for all }p\leq k\text{ and all }q>0. Even more, if n\geq 2k+1 the former condition is automatically satisfied for lci singularities (see [ 7 , Lemma 3.19] ). A similar criterion holds for k -rational singularities. To summarize, we have: Definition 2.5 . Let Y have isolated lci singularities and \dim Y=n\geq 2k+1. Then Y has k -Du Bois if and only if R^{q}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)(-E)=0\quad\text{for all }p\leq k\text{ and all }q>0. Also, Y has k -rational singularities if and only if R^{q}\pi_{*}\Omega^{p}_{\widehat{Y}}(\log E)=0\quad\text{for all }p\leq k\text{ and all }q>0. Finally, we say a singularity is called k -liminal if it is k -Du Bois but not k -rational. We now explain the 1-rational and 1-Du Bois conditions under the additional assumption that the singularities are already rational. Let (X,x) be a rational isolated lci singularity. Then by Theorem 2.3 to check that (X,x) is a 1-Du Bois singularity is equivalent to having H^{n-1}(\Omega^{1}_{\widehat{X}}(\log E)(-E))=0 and H^{n-2}(\Omega^{1}_{\widehat{X}}(\log E)(-E))=0. But notice that the first equality follows from extra vanishing [ 37 , page 1369] . Similarly, by Remark 2.4 to check that (X,x) is 1-rational is equivalent to having H^{n-1}(\Omega^{1}_{\widehat{X}}(\log E))=0 and H^{n-2}(\Omega^{1}_{\widehat{X}}(\log E))=0. The first equality is proved for isolated rational singularities by Mustaţă-Olano-Popa [ 22 ] . Hence, we have the following definition which will be used throughout the paper: Definition 2.6 . [ 10 , Definition 2.6.] Let (X,x) be an isolated rational lci singularity. (i) We say (X,x) is 1-rational if and only if H^{n-2}(\widehat{X};\Omega^{1}_{\widehat{X}}(\log E))=0 . By duality this is also equivalent to H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E))=0. We also say a singularity which is not 1-rational is 1-irrational. (ii) (X,x) is 1-Du Bois if and only if H^{n-2}(\widehat{X};\Omega^{1}_{\widehat{X}}(\log E)(-E))=0. By duality this is also equivalent to H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E))=0. 2.3. Local deformations and smoothings Globally, one has a deformation space \text{Def}(Y) whose tangent space is a finite-dimensional vector space \mathbb{T}^{1}_{Y}=\Ext^{1}(\Omega^{1}_{Y},\mathcal{O}_{Y}) which classifies first-order deformations of Y. Locally, for every singular point x of Y one can study the local deformation \text{Def}(Y,x) of the germ (Y,x). This has a tangent space given by the finite-dimensional vector space T^{1}_{Y,x}, where T^{1}_{Y} is the sheaf \mathcal{E}xt^{1}(\Omega^{1}_{Y},\mathcal{O}_{Y}), supported on the singular points. As before, T^{1}_{Y,x} parameterizes first-order local deformations of the complex germ. There is a natural map \text{Def}(Y)\to\prod_{x\in Y_{\text{sing}}}\text{Def}{(Y,x)} given by restricting to the germs. On Zariski tangent spaces, this gives a homomorphism \mathbb{T}^{1}_{Y}\xrightarrow{\eta}H^{0}(T^{1}_{Y})=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}T^{1}_{Y,x}. In this section we focus on understanding the local deformations as well as criteria for global smoothing. We begin by recalling some of the results on deforming complex space germs. For more details see for example [ 14 , Chapter 2.1] Let (X,x) be the germ of an isolated lci singularity. The infinitesimal deformation theory of (X,x) is controlled by the \mathcal{O}_{X,x} -modules T^{i}_{(X,x)}\coloneqq\operatorname{Ext}^{i}_{\mathcal{O}_{X,x}}(\Omega^{1}_{X,x},\mathcal{O}_{X,x})\qquad(i=1,2), where T^{1}_{(X,x)} is the Zariski tangent space to the deformation functor and T^{2}_{(X,x)} is the obstruction space. Since (X,x) is lci, T^{2}_{(X,x)}=0 and, in particular, the deformation functor is unobstructed. Say (X,x) has embedding dimension e , defined by f_{1}=\cdots=f_{k}=0,\qquad f_{i}\in\mathbb{C}\{x_{1},\ldots,x_{e}\}. Hence \mathcal{O}_{X,x}=\mathbb{C}\{x_{1},\ldots,x_{e}\}/(f_{1},\ldots,f_{k}). The module of first-order deformations is T^{1}_{(X,x)}\cong\mathcal{O}_{X,x}^{\oplus k}/J, where J denotes the submodule of \mathcal{O}_{X,x}^{\oplus k} generated by the columns \left(\frac{\partial f_{1}}{\partial x_{i}},\ldots,\frac{\partial f_{k}}{\partial x_{i}}\right),\qquad i=1,\ldots,e. There exists an explicit miniversal deformation with smooth base space \phi:(\mathcal{X},x)\longrightarrow(S,0),\qquad\dim_{\mathbb{C}}S=\dim_{\mathbb{C}}T^{1}_{(X,x)} and the Kodaira-Spencer map gives the identification T^{1}_{(X,x)}\cong T_{S,0}. Note that (X,x) is smoothable. In fact, its discriminant locus is a reduced and irreducible hypersurface germ in (S,0) (see [ 39 , page 645] ). We now discuss criteria for determining when global deformations map to such smoothing directions. Smoothing hypersurface singularities . In the case when (X,x) is the germ of an isolated hypersurface singularity, one has a nice criterion for identifying when a one parameter deformation is a smoothing, formulated in terms of its Kodaira-Spencer class. In this setting, the infinitesimal deformations of (X,x) are parametrized by the module T^{1}_{(X,x)}\cong\mathcal{O}_{X,x}/J, where J=\left(\frac{\partial f}{\partial x_{1}},\ldots,\frac{\partial f}{\partial x_{e}}\right) is the Jacobian ideal associated to the defining equation f\in\mathbb{C}\{x_{1},\ldots,x_{e}\} of X . In particular, T^{1}_{X,x} is a cyclic \mathcal{O}_{X,x} -module. The following tells us that a deformation has smooth total space if the Kodaira-Spencer class is a generator of T^{1}_{X,x} as an \mathcal{O}_{X,x} -module. Proposition 2.7 . Let \mathcal{X}\to\Delta be a one-parameter deformation of (X,x). Then \mathcal{X} is smooth in a neighborhood of x if and only if the Kodaira-Spencer class u\in T^{1}_{X,x} is not in m_{x}T^{1}_{X,x}, if and only if u is a generator of T^{1}_{X,x} as an \mathcal{O}_{X,x} -module. Proof. See, for example, [ 10 , Lemma 1.9] . ∎ Remark 2.8 . Note that, in particular, this gives a criterion for when a deformation is a smoothing of (X,x). Smoothing lci singularities . In the case of isolated lci singularities, we no longer have a criterion that tells us when a one-parameter deformation is a smoothing in terms of its Kodaira-Spencer class. Instead, we will use the following variant of a lemma of Gross [ 15 , Lemma 3.7] : Lemma 2.9 . Let (X,0) be the germ of an isolated lci singularity and F^{\prime}:(X^{\prime},0)\to(S^{\prime},0) be a flat deformation of (X,0) , where S^{\prime} is non-singular at 0\in S^{\prime} with tangent space T=T_{S^{\prime},0} . Since T^{1} is the tangent space to the base of the miniversal deformation of (X,0) we obtain a unique differential T\to T^{1}. If \operatorname{im}(T\to T^{1})\not\subset mT^{1}, then a general fiber of X^{\prime}\to S^{\prime} is non-singular. When Y is a singular Calabi-Yau or Fano variety, the idea will be to apply this to S^{\prime}=\text{Def}(Y) which is smooth by the unobstructedness result. To obtain a smoothing, one would like to find an element in \mathbb{T}^{1}_{Y} whose image in T^{1}_{Y,x} is not in m_{x}T^{1}_{Y,x} for each singular point x. Exact sequences in local deformations. Let (X,x) be the germ of an isolated rational lci singularity. In order to lift local first-order deformations to global ones it is often useful to look at appropriate quotients of T^{1}_{X,x} which fit into certain global long exact sequences. To do this, we consider the identification H^{0}(X,T^{1}_{X})\cong H^{2}_{x}(X,T^{0}_{X})\cong H^{1}(T_{U}) due to Schlessinger [ 35 , Theorem 2] . Since X has canonical singularities (see [ 20 , Theorem 11.1] ), there exists an effective divisor D with \operatorname{Supp}(D)\subseteq E such that K_{\widehat{X}}\cong\mathcal{O}_{\widehat{X}}(D) and so T_{\widehat{X}}\cong\Omega^{n-1}_{\widehat{X}}(-D). Hence, if we consider the inclusions \Omega^{n-1}_{\widehat{X}}(\log E)(-E)\subseteq\Omega^{n-1}_{\widehat{X}}\subseteq\Omega^{n-1}_{\widehat{X}}(\log E) notice that they all restrict to T_{U} on the smooth locus U, which we have identified with \widehat{X}\setminus E . Therefore, one can identify the first-order deformations with H^{1}(U;\Omega^{n-1}_{\widehat{X}}|_{U}),\;H^{1}(U;\Omega^{n-1}_{\widehat{X}}(\log E)|_{U}),\;\text{or}\;H^{1}(U;\Omega^{n-1}_{\widehat{X}}(\log E)(-E)|_{U}) which all fit into local cohomology long exact sequences. Following Friedman-Laza, we recall without proof some of the results we will need for the local deformation theory. For more details one should consult [ 10 , Section 2] . See also [ 25 , Section 1] . There exists a short exact sequence 0\to H^{1}(\Omega^{n-1}_{\widehat{X}}(\log E))\to H^{1}(\Omega^{n-1}_{U})\to H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E))\to 0. Hence, in the case when H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E))\neq 0 , i.e. when the singularities are not 1-Du Bois (see Definition 2.6 ), to construct global smoothings the idea will be to consider appropriate elements in H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E)) that come from global first-order deformations. For the more general case of 1-irrational singularities (see Definition 2.6 ), one looks at the image in H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E)) as we explain below. Then there exists an exact sequence coming from local cohomology: 0\to H^{1}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E))\to H^{1}(\Omega^{n-1}_{U})\to H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E))\to 0. and, moreover, H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E)) sits in the following exact sequence: 0\to\text{Gr}_{F}^{n-1}H^{n}(L)\to H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E))\to H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E))\to 0, where L is the link of the singularity. The image of H^{1}(\Omega^{n-1}_{\widehat{X}})\to H^{1}(\Omega^{n-1}_{U}) can be identified with the image of H^{1}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E))\to H^{1}(\Omega^{n-1}_{U}). Finally, we have a commutative diagram as follows: {\lx@inpgf@ignorespaces{H^{1}_{E}(\Omega^{n-1}_{\widehat{X}})}} {\lx@inpgf@ignorespaces{H^{1}(\Omega^{n-1}_{\widehat{X}})}} {\lx@inpgf@ignorespaces{H^{1}(\Omega^{n-1}_{U})}} {\lx@inpgf@ignorespaces{H^{2}_{E}(\Omega^{n-1}_{\widehat{X}})}} {\lx@inpgf@ignorespaces 0} {\lx@inpgf@ignorespaces{H^{1}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E))}} {\lx@inpgf@ignorespaces{H^{1}(\Omega^{n-1}_{U})}} {\lx@inpgf@ignorespaces{H^{2}_{E}(\Omega^{n-1}_{\widehat{X}}(\log E)(-E))}} {\lx@inpgf@ignorespaces 0} which identifies the image of {H^{1}(\Omega^{n-1}_{U})}\to{H^{2}_{E}(\Omega^{n-1}_{\widehat{X}})} with H^{2}_{E}(\Omega^{n-1}_{\widehat{X}_{x}}(\log E)(-E)). 3. Smoothing Fano varieties We begin by proving unobstructedness results for Fano varieties (cf. [ 10 , Theorem 4.5] and [ 12 , Theorem 3.17] ). Theorem 3.1 . Let Y be a Fano variety with isolated lci singularities. Then \mathbb{T}^{i}_{Y}=0 for i\geq 2 and, in particular, deformations of Y are unobstructed. Proof. We have \mathbb{T}^{i}_{Y}=\Ext^{i}(\Omega_{Y}^{1},\mathcal{O}_{Y}) and so by Serre duality it suffices to show H^{n-i}(\Omega_{Y}^{1}\otimes\omega_{Y})=0. Consider the following hypercohomology spectral sequence E_{2}^{p,q}=H^{p}(\mathcal{H}^{q}\underline{\Omega}_{Y}^{1}\otimes\omega_{Y})\Rightarrow\mathbb{H}^{p+q}(\underline{\Omega}_{Y}^{1}\otimes\omega_{Y}). Note that \mathcal{H}^{0}(\underline{\Omega}_{Y}^{1})\cong\Omega_{Y}^{1} since \Omega_{Y}^{1} is reflexive and \mathcal{H}^{0}(\underline{\Omega}_{Y}^{1}) is torsion-free. Therefore, H^{n-i}(\Omega_{Y}^{1}\otimes\omega_{Y})=E^{n-i,0}_{2} and by Theorem 2.3 we have E^{0,n-i-1}_{2}=0. Hence, H^{n-i}(\Omega_{Y}^{1}\otimes\omega_{Y})=E^{n-i,0}_{\infty}\hookrightarrow\mathbb{H}^{n-i}(\underline{\Omega}_{Y}^{1}\otimes\omega_{Y}). But since n-i+10. Then, as in the previous theorem, we wish to show H^{n-i}(\Omega_{Y}^{1}\otimes\omega_{Y})=0. Note since n-i\geq 2 we have H^{n-i}(\Omega_{Y}^{1}\otimes\omega_{Y})=H^{n-i}(\Omega_{Y}^{[1]}\otimes\omega_{Y})=H^{n-i}(\mathcal{H}^{0}(\underline{\Omega}_{Y}^{1})\otimes\omega_{Y}). We would like to make use of a Nakano-type vanishing for singular varieties. Since \text{lcdef}(Y)=0 we have H^{n-i}(\underline{\Omega}_{Y}^{1}\otimes\omega_{Y})=0 by the dual Nakano-type vanishing of Popa-Shen [ 33 , Theorem C] . Finally, to prove that H^{n-i}(\Omega_{Y}^{1}\otimes\omega_{Y})\hookrightarrow H^{n-i}(\underline{\Omega}_{Y}^{1}\otimes\omega_{Y}) it would suffice to have \mathcal{H}^{n-i-1}(\underline{\Omega}^{1}_{Y})=0. This follows from a result of Popa-Shen-Vo [ 32 , Proposition F] which, in our case, says \mathcal{H}^{k}(\underline{\Omega}^{1}_{Y})=0 for 00 , the sheaves R^{q}\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)(-E) are supported on finitely many points, hence H^{p}(R^{q}\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)(-E))=0 for p>0 . Therefore E_{2}^{p,q}=0 whenever both p,q>0. Note, R^{n-2}\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)(-E) is a skyscraper sheaf with stalk at x equal to H^{n-2}(\Omega^{1}_{\widehat{X}}(\log E_{x})(-E_{x})). Hence, H^{0}(R^{n-2}\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)(-E))\cong\bigoplus_{x\in\mathrm{Sing}(Y)}H^{n-2}(\Omega^{1}_{\widehat{X}_{x}}(\log E_{x})(-E_{x})). We would like to show \displaystyle H^{n-2}(\Omega_{\widehat{Y}}^{1}(\log E)(-E))=E_{\infty}^{n-2}\rightarrow E_{2}^{0,n-2} \displaystyle=H^{0}(R^{n-2}\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)(-E)) \displaystyle=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{n-2}(\Omega^{1}_{\widehat{X}_{x}}(\log E_{x})(-E_{x})) is zero. In other words, we want to show E_{\infty}^{0,n-2}=0. Note E_{2}^{n-2,0}=E_{\infty}^{n-2,0} which is immediate for n=3 and for n>3 follows from E_{2}^{0,n-3}=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{n-3}(\Omega^{1}_{\widehat{X}_{x}}(\log E_{x})(-E_{x}))=0 by [ 38 , Theorem 5] , as the singularities are isolated lci. Hence, we obtain an injection E_{2}^{n-2,0}=E_{\infty}^{n-2,0}\hookrightarrow E_{\infty}^{n-2}=\mathbb{H}^{n-2}(\underline{\Omega}_{Y}^{1}). However, by [ 32 , Proposition 5.2] , E_{\infty}^{n-2,0}=E_{2}^{n-2,0}=H^{n-2}(\mathcal{H}^{0}(\underline{\Omega}_{Y}^{1}))\rightarrow E_{\infty}^{n-2}=\mathbb{H}^{n-2}(\underline{\Omega}_{Y}^{1}) is also surjective, and so, it is an isomorphism (cf. also [ 12 , Remark 2.5] ). Note [ 32 , Proposition 5.2] makes use of the E_{1} -degeneration of the Hodge-to-de Rham spectral sequence. We conclude E_{\infty}^{0,n-2}=0 and so the morphism we were interested in is zero. ∎ The first result concerns smoothing a projective Calabi-Yau variety whose singularities are not 1-Du Bois. In this case, no global condition is needed to show the variety is smoothable. Theorem 4.5 . Suppose Y is a singular Calabi-Yau variety with rational, isolated lci singularities. Then Y can be deformed to a Calabi-Yau variety whose singularities are 1-Du Bois. Moreover, if one assumes in addition that all the singularities of Y are not 1-Du Bois, then Y is smoothable, and every small smoothing is Calabi-Yau. Proof. As in the Proof of Theorem 3.8 , it suffices to prove that one can find a deformation which smooths out the points that are not 1-Du Bois. Recall one can identify T^{1}_{Y,x} with H^{2}_{x}(X_{x},T^{0}_{X_{x}})\cong H^{1}(U_{x},\Omega^{n-1}_{U_{x}}) and, moreover, a global version of this says \mathbb{T}^{1}_{Y}\cong H^{1}(V,\Omega_{V}^{n-1}) (see [ 10 , Lemma 4.1] ). Consider the following commutative diagram: {\lx@inpgf@ignorespaces{\mathbb{T}^{1}=H^{1}(V,\Omega_{V}^{n-1})}} {\lx@inpgf@ignorespaces{H^{2}_{E}(\Omega_{\widehat{Y}}^{n-1}(\log E))}} {\lx@inpgf@ignorespaces{H^{2}(\Omega_{\widehat{Y}}^{n-1}(\log E))}} {\lx@inpgf@ignorespaces{\mathbb{T}^{1}=H^{1}(V,\Omega_{V}^{n-1})}} {\lx@inpgf@ignorespaces{H^{2}_{E}(\Omega_{\widehat{Y}}^{n-1})}} {\lx@inpgf@ignorespaces{H^{2}(\Omega_{\widehat{Y}}^{n-1})}} {\lx@inpgf@ignorespaces{H^{0}(T^{1}_{Y})\cong H^{2}_{Z}(T^{0}_{Y})\cong\bigoplus H^{1}(U_{x},\Omega^{n-1}_{U_{x}})}} \scriptstyle{\lx@inpgf@ignorespaces f} \scriptstyle{\lx@inpgf@ignorespaces f^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces g} \scriptstyle{\lx@inpgf@ignorespaces\eta} \scriptstyle{\lx@inpgf@ignorespaces\psi} \scriptstyle{\lx@inpgf@ignorespaces g^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces\varphi} \scriptstyle{\lx@inpgf@ignorespaces\phi} As in Theorem 3.8 the diagram comes from the fact that T^{0}_{Y}\cong\Omega_{Y}^{[n-1]}\cong\pi_{*}(\Omega^{n-1}_{\widehat{Y}})\cong\pi_{*}(\Omega^{n-1}_{\widehat{Y}}(\log E)) (see [ 19 , Theorems 1.4 and 1.5] ). First, by excision H^{2}_{E}(\Omega_{\widehat{Y}}^{n-1}(\log E))\cong\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})). Hence, the morphism H^{0}(T^{1}_{Y})\rightarrow H^{2}_{E}(\Omega_{\widehat{Y}}^{n-1}(\log E)) is given by \bigoplus\limits_{x\in\mathrm{Sing}(Y)}\varphi_{x} where \varphi_{x} is the composition H^{1}(U_{x},\Omega^{n-1}_{U_{x}})\xrightarrow{\phi_{x}}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1})\xrightarrow{\psi_{x}}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})). By Section 2.3 we have the following short exact sequence 0\rightarrow H^{1}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x}))\rightarrow H^{1}(U_{x},\Omega^{n-1}_{U_{x}})\xrightarrow{\varphi_{x}}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x}))\rightarrow 0. For each singularity that is not 1-Du Bois we obtain that each H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})) is nonzero. The idea is to apply Lemma 2.9 to S^{\prime}=\text{Def}(Y) , which is smooth by Friedman’s unobstructedness Theorem 1.14 . Consider the composition T=\mathbb{T}^{1}_{Y}=H^{1}(\Omega_{V}^{n-1})\to T^{1}_{Y,x}\twoheadrightarrow H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})). From Lemma 2.9 if \operatorname{im}(H^{1}(\Omega_{V}^{n-1})\to T^{1}_{Y,x})\not\subset m_{x}T^{1}_{Y,x} then we can find a deformation that smooths the singular point x. Therefore, following the composition, we find that if \operatorname{im}(H^{1}(\Omega_{V}^{n-1})\to H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})))\not\subset m_{x}H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})) then we can find a deformation that smooths the singular point x , which was not 1-Du Bois. Consider \beta^{\prime}_{x}\in H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})) but such that \beta^{\prime}_{x}\notin m_{x}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})). Now, by Proposition 4.4 the map H^{2}_{E}(\Omega_{\widehat{Y}}^{n-1}(\log E))\rightarrow H^{2}(\Omega_{\widehat{Y}}^{n-1}(\log E)) is zero. Hence, the element \beta^{\prime}=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}\beta^{\prime}_{x} comes from a global first-order deformation and, therefore, \operatorname{im}(H^{1}(\Omega_{V}^{n-1})\to H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})))\not\subset m_{x}H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})). ∎ Remark 4.6 . As mentioned in Remark 1.7 , to obtain smoothings if one assumes further that Y has hypersurface singularities, by using Proposition 2.7 we can show the stronger result. Indeed, one shows there exists a flat analytic deformation over a disk \mathcal{Y}\to\Delta of Y with smooth total space \mathcal{Y} . Let us briefly explain the argument. Consider \alpha_{x}\in H^{1}(U_{x},\Omega^{n-1}_{U_{x}}) such that it is a first-order smoothing in the component T^{1}_{Y,x} i.e. it is a generator of the cyclic \mathcal{O}_{Y,x} -module T^{1}_{Y,x}. Let \beta_{x}=\varphi_{x}(\alpha_{x}) be its image in H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})). We claim that \beta_{x}\neq 0 for each x. Otherwise, \alpha_{x} would be in the kernel of H^{1}(U_{x},\Omega^{n-1}_{U_{x}})\rightarrow H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})) which is a contradiction as that would force the kernel to be the whole H^{1}(U_{x},\Omega^{n-1}_{U_{x}}) since \alpha_{x} is a generator of the cyclic module. Now, as in the proof above, by Proposition 4.4 the map H^{2}_{E}(\Omega_{\widehat{Y}}^{n-1}(\log E))\xrightarrow{f^{\prime}}H^{2}(\Omega_{\widehat{Y}}^{n-1}(\log E)) is zero. Hence, the element \beta=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}\beta_{x} comes from a global first-order deformation y\in\mathbb{T}^{1} i.e. f(y)=\beta . By Friedman’s unobstructedness Theorem 1.14 , we find a global deformation whose Kodaira-Spencer class is y. Let \eta(y)=\oplus\gamma_{x}. By the commutativity of the diagram \alpha_{x} and \gamma_{x} both map to \beta_{x}. Hence they differ by an element of m_{x}T^{1}_{Y,x} and so \gamma_{x} is a first-order smoothing as well. Hence, we have found a global 1-parameter deformation with smooth total space that smooths out every singular point. Let us now explain how the results above can be used to prove Proposition 1.12 . Proof of Proposition 1.12 . First, one handles the smoothing of 1-liminal points as in the proof of [ 10 , Theorem 5.8] and obtains \theta_{2}\in\mathbb{T}^{1}_{Y} which maps to a generator in each component T^{1}_{Y,x} where x is 1-liminal. Now for the points that are not 1-Du Bois. For each such point x pick an element in \beta_{x}\in H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x}))\neq 0 which comes from a generator in T^{1}_{Y,x}. Then, since f^{\prime}=0 , we find an element \theta_{1}\in\mathbb{T}^{1}_{Y} which maps to a generator of T^{1}_{Y,x} for each x that is not 1 Du Bois. Finally, pick a general linear combination of \theta_{1} and \theta_{2} to obtain a first-order smoothing of Y. ∎ In the smoothing statement, we would now like to relax the not 1-Du Bois condition to not 1-rational. In other words, we would like to be able to smooth out 1-liminal points. As explained in Section 1.3 , one should not expect a smoothing in the absence of additional global conditions. In the case of threefolds, Namikawa-Steenbrink show \mathbb{Q} -factoriality is a global condition which, together with appropriate local conditions on the singularities, guarantees smoothability. In higher dimensions, we find that the appropriate global condition on Y is the equality of Hodge-Du Bois numbers \underline{h}^{n-1,2}=\underline{h}^{1,n-2}, which for threefolds with rational singularities is equivalent to the familiar \mathbb{Q} -factoriality condition (see [ 30 , Theorem E] and [ 25 , Theorem 3.2] ). Theorem 4.7 . Suppose Y is a singular Calabi-Yau variety of dimension n with rational, isolated lci singularities. Assume moreover that we have the following equality of Hodge-Du Bois numbers \underline{h}^{n-1,2}(Y)=\underline{h}^{1,n-2}(Y). Then Y can be deformed to a Calabi-Yau variety whose singularities are 1-rational. Moreover, if one assumes in addition that all the singularities of Y are not 1-rational, then Y is smoothable, and every small smoothing is Calabi-Yau. Proof. As in the Proof of Theorem 3.8 , it suffices to prove that one can find a deformation which smooths out the points that are not 1-rational. Consider the following commutative diagram with exact row coming from the local cohomology sequence: {\lx@inpgf@ignorespaces{\mathbb{T}^{1}=H^{1}(V,\Omega^{n-1}_{V})}} {\lx@inpgf@ignorespaces{\oplus H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1})}} {\lx@inpgf@ignorespaces{H^{2}(\Omega_{\widehat{Y}}^{n-1})}} {\lx@inpgf@ignorespaces{H^{0}(T^{1}_{Y})\cong H^{2}_{Z}(T^{0}_{Y})\cong\bigoplus H^{1}(U_{x},\Omega^{n-1}_{U_{x}})}} \scriptstyle{\lx@inpgf@ignorespaces g} \scriptstyle{\lx@inpgf@ignorespaces\eta} \scriptstyle{\lx@inpgf@ignorespaces g^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces\oplus\phi_{x}} Recall by the result recalled in Section 2.3 we can factor \varphi_{x} via H^{1}(U_{x},\Omega^{n-1}_{U_{x}})\twoheadrightarrow H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x}))\hookrightarrow H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}). Moreover, H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})) is nonzero by the assumption that we are looking at the 1-irrational singular points. Consider the following commutative diagram with exact rows coming from the local cohomology sequence: {\lx@inpgf@ignorespaces{\mathbb{T}^{1}=H^{1}(V,\Omega^{n-1}_{V})}} {\lx@inpgf@ignorespaces{\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x}))}} {\lx@inpgf@ignorespaces{H^{2}(\Omega_{\widehat{Y}}^{n-1}(\log E))}} {\lx@inpgf@ignorespaces{\mathbb{T}^{1}=H^{1}(V,\Omega^{n-1}_{V})}} {\lx@inpgf@ignorespaces{\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1})}} {\lx@inpgf@ignorespaces{H^{2}(\Omega_{\widehat{Y}}^{n-1})}} {\lx@inpgf@ignorespaces{\mathbb{T}^{1}=H^{1}(V,\Omega^{n-1}_{V})}} {\lx@inpgf@ignorespaces{\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x}))}} {\lx@inpgf@ignorespaces{H^{2}(\Omega_{\widehat{Y}}^{n-1}(\log E)(-E))}} {\lx@inpgf@ignorespaces{H^{2}_{Z}(T^{0}_{Y})\cong\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{1}(U_{x},\Omega^{n-1}_{U_{x}})}} \scriptstyle{\lx@inpgf@ignorespaces f} \scriptstyle{\lx@inpgf@ignorespaces f^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces g} \scriptstyle{\lx@inpgf@ignorespaces\psi_{x}} \scriptstyle{\lx@inpgf@ignorespaces g^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces h} \scriptstyle{\lx@inpgf@ignorespaces\eta} \scriptstyle{\lx@inpgf@ignorespaces\iota_{x}} \scriptstyle{\lx@inpgf@ignorespaces h^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces s_{x}} As in Theorem 3.8 the diagram comes from the fact that since we have rational singularities T^{0}_{Y}\cong\Omega_{Y}^{[n-1]}\cong\pi_{*}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E))\cong\pi_{*}(\Omega^{n-1}_{\widehat{Y}})\cong\pi_{*}(\Omega^{n-1}_{\widehat{Y}}(\log E)) (see [ 19 , Theorems 1.4–1.6] ). We now show that the composition H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x}))\xhookrightarrow{\iota_{x}}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1})\xrightarrow{g^{\prime}}H^{2}(\Omega_{\widehat{Y}}^{n-1}) is zero, which then allows us to lift an appropriate element in each component to obtain a global smoothing. From Proposition 4.4 the morphism \bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega^{n-1}_{\widehat{X}_{x}}(\log E_{x}))\xrightarrow{f^{\prime}}H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)) is zero and, therefore, so is the composition \bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega^{n-1}_{\widehat{X}_{x}}(\log E_{x})(-E_{x}))\to\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega^{n-1}_{\widehat{X}_{x}}(\log E_{x}))\rightarrow H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)). Claim. The equality \underline{h}^{n-1,2}=\underline{h}^{1,n-2} implies H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E))(-E)\to H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)) is an isomorphism. Assuming the claim for the moment, we obtain that the morphism \bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}_{E_{x}}(\Omega^{n-1}_{\widehat{X}_{x}}(\log E_{x})(-E_{x}))\xrightarrow{h^{\prime}}H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E)) is zero. As before we apply Lemma 2.9 to S^{\prime}=\text{Def}(Y) , which is smooth by Friedman’s unobstructedness Theorem 1.14 . Consider the composition T=\mathbb{T}^{1}_{Y}=H^{1}(\Omega_{V}^{n-1})\to T^{1}_{Y,x}\twoheadrightarrow H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})). Therefore, following the composition, we find that if \operatorname{im}(H^{1}(\Omega_{V}^{n-1})\to H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})))\not\subset m_{x}H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})) then we can find a deformation that smooths the singular point x. Consider \beta^{\prime}_{x}\in H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})) but such that \beta^{\prime}_{x}\notin m_{x}H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})). Now, h^{\prime} is zero. Hence, the element \beta^{\prime}=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}\beta^{\prime}_{x} comes from a global first-order deformation and, therefore, \operatorname{im}(H^{1}(\Omega_{V}^{n-1})\to H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})))\not\subset m_{x}H_{E_{x}}^{2}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x})). Hence, we can smooth out the 1-irrational points. To prove the claim, let us consider the long exact sequence H^{1}(\Omega^{n-1}_{\widehat{Y}}(\log E)|_{E})\to H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E))\to H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E))\to H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)|_{E}) and note that the last term H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)|_{E})\cong\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{2}(\Omega^{n-1}_{\widehat{X}_{x}}(\log E_{x})|_{E_{x}})=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}\Gr_{F}^{n-1}H^{n+1}(L_{x}). But since the singularities are lci, H^{n+1}(L_{x})=0 and so H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)|_{E})=0. Therefore, we have a surjection H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E))\twoheadrightarrow H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)). Now notice that \dim H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E))=\dim\mathbb{H}^{2}(\underline{\Omega}^{n-1}_{Y})=\underline{h}^{n-1,2} and similarly by Serre duality \dim H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E))=\dim H^{n-2}(\Omega^{1}_{\widehat{Y}}(\log E)(-E))=\underline{h}^{1,n-2}. By our assumption on the equality of the Hodge-Du Bois numbers, H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E))\twoheadrightarrow H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)) must then be an isomorphism. ∎ Remark 4.8 . In the previous theorem, one can check that the condition \underline{h}^{n-1,2}=\underline{h}^{1,n-2} is in fact equivalent to the morphism H^{2}_{E}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E))\to H^{2}(\Omega^{n-1}_{\widehat{Y}}(\log E)(-E)) being zero. Indeed, assuming the morphism is zero we have that dually the morphism H^{n-2}(\Omega_{\widehat{Y}}^{1}(\log E))\rightarrow\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{n-2}(\Omega^{1}_{\widehat{X}_{x}}(\log E_{x})) is zero. Consider the following hypercohomology spectral sequence E_{2}^{p,q}=H^{p}(R^{q}\pi_{*}\Omega_{\widehat{Y}}^{1}(\log E))\Rightarrow H^{p+q}(\Omega_{\widehat{Y}}^{1}(\log E)). Therefore E_{2}^{p,q}=0 whenever both p,q>0. We know \displaystyle H^{n-2}(\Omega_{\widehat{Y}}^{1}(\log E))=E_{\infty}^{n-2}\rightarrow E_{2}^{0,n-2} \displaystyle=H^{0}(R^{n-2}\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)) \displaystyle=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{n-2}(\Omega^{1}_{\widehat{X}_{x}}(\log E_{x})) is zero. In other words, we have E_{\infty}^{0,n-2}=0. Note E_{2}^{n-2,0}=E_{\infty}^{n-2,0} which is immediate for n=3 and for n>3 follows from E_{2}^{0,n-3}=\bigoplus\limits_{x\in\mathrm{Sing}(Y)}H^{n-3}(\Omega^{1}_{\widehat{X}_{x}}(\log E_{x}))=0 by [ 38 , Theorem 5] , as the singularities are isolated lci. Hence, we obtain an isomorphism E_{\infty}^{n-2,0}=H^{n-2}(\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E))\cong H^{n-2}(\Omega_{\widehat{Y}}^{1}(\log E)). But \pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)\cong\pi_{*}\Omega^{1}_{\widehat{Y}}(\log E)(-E) (and, in fact, the same as reflexive differentials on the base by Kebekus-Schnell [ 19 ] , so just \Omega^{1}_{Y} since in our situation this is already reflexive) so we conclude H^{n-2}(\Omega^{1}_{\widehat{Y}}(\log E)(-E))\cong H^{n-2}(\Omega_{\widehat{Y}}^{1}(\log E)) which gives the equality \underline{h}^{n-1,2}=\underline{h}^{1,n-2}. Remark 4.9 . In higher dimensions, the condition \underline{h}^{n-1,2}=\underline{h}^{1,n-2} would be immediate if all the singularities were 1-rational ( [ 7 , Corollary 3.22, Theorem 3.24] and [ 29 , Theorem A] ), in which case \underline{h}^{p,q}(Y)=\underline{h}^{q,p}(Y)=\underline{h}^{\,n-p,\,n-q}(Y)=\underline{h}^{\,n-q,\,n-p}(Y),\qquad\text{for all }0\leq p\leq 1\text{ and }0\leq q\leq n. However, this is too strong an assumption for smoothability, as our arguments rely on the fact that the singularities are not 1-rational i.e., that H^{2}_{E_{x}}(\Omega_{\widehat{X}_{x}}^{n-1}(\log E_{x})(-E_{x}))\neq 0. Remark 4.10 . We conclude with a discussion of the equality \underline{h}^{n-1,2}=\underline{h}^{1,n-2} and its relation to the Poincaré duality of the variety. For threefolds, the assumption \underline{h}^{2,2}=\underline{h}^{1,1} is also implied by the Poincaré duality of the threefold. In fact, it is equivalent to the vertical symmetry of the Hodge-Du Bois diamond. To see this, since the singularities are rational, the border of the diamond is symmetric ( [ 7 , Corollary 3.22, Theorem 3.24] and [ 29 , Theorem A] ). More precisely, we have \underline{h}^{3,3}=\underline{h}^{0,0} and \underline{h}^{3,1}=\underline{h}^{1,3}=\underline{h}^{2,0}=\underline{h}^{0,2} \underline{h}^{3,2}=\underline{h}^{2,3}=\underline{h}^{1,0}=\underline{h}^{0,1}. Hence, one has full vertical symmetry of the Hodge-Du Bois diamond of Y if and only if \underline{h}^{2,2}=\underline{h}^{1,1} if and only if Y is \mathbb{Q} -factorial. For fourfolds, the condition \underline{h}^{3,2}=\underline{h}^{1,2} is still equivalent to the full vertical symmetry of the Hodge-Du Bois diamond. As before, rational singularities imply that the border of the Hodge-Du Bois diamond is symmetric. Moreover, the top part of the Hodge-Du Bois diamond ( p+q\geq 4 ), coincides with intersection cohomology since the singularities are isolated. Since the singularities are lci, we have \text{lcdef}(Y)=0, and so, by [ 29 , Corollary 8.9] the Hodge structure on H^{3}(Y,\mathbb{Q}) is pure and we have H^{i}(Y,\mathbb{Q})\simeq IH^{i}(Y,\mathbb{Q})\quad\text{for }i\in\{1,2\}. Therefore, except for the middle row, one has horizontal symmetry. Moreover, vertical symmetry is equivalent to b_{3}(Y)=b_{5}(Y) , which, by the horizontal symmetry, is equivalent to our condition \underline{h}^{3,2}=\underline{h}^{1,2}. Of course, in dimension n\geq 4 the condition \underline{h}^{n-1,2}=\underline{h}^{1,n-2} is still implied by (but weaker than) the vertical symmetry of the Hodge-Du Bois diamond (i.e. by b_{n-1}(Y)=b_{n+1}(Y) ). Department of Mathematics, Harvard University, 1 Oxford Street, Cambridge, MA 02138, USA E-mail address : astenie@math.harvard.edu