{"id":4440,"date":"2020-03-07T02:19:21","date_gmt":"2020-03-07T02:19:21","guid":{"rendered":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/?p=4440"},"modified":"2020-03-09T01:15:01","modified_gmt":"2020-03-09T01:15:01","slug":"purity-part-1","status":"publish","type":"post","link":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/?p=4440","title":{"rendered":"Purity (part 1)"},"content":{"rendered":"<p>Let f : X &rarr; Y be a dominant, finite type morphism of integral Noetherian schemes. We assume X is normal and Y regular. In this and the next blog post we define Sing(f) to be the closed set of points of X where f isn&#8217;t smooth and we define Disc(f) to be the image of Sing(f) in Y.<\/p>\n<p>In this setting a purity result is one that gives an upper bound on the codimension of Sing(f) in X or a lower bound on the codimension of Disc(f) in Y.<\/p>\n<p>When the relative dimension of f is 0, then the Stacks project has <a href=\"https:\/\/stacks.math.columbia.edu\/tag\/0BJE\">Zariski-Nagata purity of branch locus<\/a>. If f is generically finite, then the Stacks project has <a href=\"https:\/\/stacks.math.columbia.edu\/tag\/0EA1\">purity of ramification locus<\/a> and over fields we even have that the complement of Sing(f) is affine in X (<a href=\"https:\/\/stacks.math.columbia.edu\/tag\/0ECA\">Tag 0ECA<\/a>).<\/p>\n<p>But what if the relative dimension is &gt; 0?<\/p>\n<p>For the rest of this blog post, let&#8217;s say f has relative dimension m &ge; 0 in the sense that every irreducible component of every fibre has dimension m.<\/p>\n<p>The simplest case is when Y and X are smooth over a field k with Y of dimension n and X of dimension n + m. In fact, let&#8217;s assume Y and X are both affine spaces over k. Picking coordinates x_1, &#8230;, x_{n + m} on X and coordinates y_1, &#8230;, y_n on Y and say f = (f_1, &#8230;, f_n) is given in coordinates by y_i = f_i(x_1, &#8230;, x_{n + m}) for some polynomials f_i. Then of course the locus Sing(f) is the set of points of X where the n &times; (n + m) matrix of partial derivatives (d f_i \/ d x_j) does not have maximal rank. We conclude immediately that codim Sing(f) &le; m + 1. Taking the image in Y we find codim Disc(f) &leq; m + 1.<\/p>\n<p>In fact, this is all you can say! Here are some examples to show this is true. (We assume the characteristic of k is not 2.)<\/p>\n<p>Relative dimension m = 0.<\/p>\n<p>(A) For n &ge; 1 we have the example (f_1, &#8230;, f_n) = (x_1, &#8230;, x_{n &#8211; 1}, x_n^2). Then Sing(f) = V(x_n) and Disc(f) = V(y_n).<\/p>\n<p>Relative dimension m = 1.<\/p>\n<p>(A) For n &ge; 1 we have the example (f_1, &#8230;, f_n) = (x_1, &#8230;, x_{n &#8211; 1}, x_nx_{n + 1}). Then Sing(f) = V(x_n, x_{n + 1}) and Disc(f) = V(y_n). Here Disc(f) is a divisor in Y and the map Sing(f) &rarr; Disc(f) is an isomorphism. This is the usual example of a family of nodal curves.<\/p>\n<p>(B) For n &ge; 2 we have the example (f_1, &#8230;, f_n) = (x_1, &#8230;, x_{n &#8211; 1}, x_1x_n + x_{n + 1}^2). Then Sing(f) = V(x_1, x_{n + 1}) and Disc(f) = V(y_1, y_n). Here Disc(f) has codimension 2 in Y, the fibres over points of Disc(f) are nonreduced, and the map Sing(f) &rarr; Disc(f) has fibres of dimension 1. Thanks to Will Sawin for finding this example.<\/p>\n<p>Relative dimension m = 2.<\/p>\n<p>(A) For n &ge; 1 we have the example (f_1, &#8230;, f_n) = (x_1, &#8230;, x_{n &#8211; 1}, x_n^2 + x_{n + 1}^2 + x_{n + 2}^2). Here Sing(f) = V(x_n, x_{n + 1}, x_{n + 2}) and Disc(f) = V(y_n). Here Disc(f) is a divisor in Y and the map Sing(f) &rarr; Disc(f) is an isomorphism. This is the usual example of a family of nodal surfaces.<\/p>\n<p>(B) For n &ge; 2 we have the example (f_1, &#8230;, f_n) = (x_1, &#8230;, x_{n &#8211; 1}, x_1 x_n + x_{n + 1}^2 + x_{n + 2}^2). Here Sing(f) = V(x_1, x_{n + 1}, x_{n + 2}) and Disc(f) = V(y_1, y_n). Here Disc(f) has codimension 2 in Y, the fibres of f over points of Disc(f) are planes meeting in lines, and the nap Sing(f) &rarr; Disc(f) has fibres of dimension 1.<\/p>\n<p>(C) For n &ge; 3 we have the example (f_1, &#8230;, f_n) = (x_1, &#8230;, x_{n &#8211; 1}, x_1 x_n + x_2 x_{n + 1} + x_{n + 2}^2). Here Sing(f) = V(x_1, x_2, x_{n + 2}) and Disc(f) = V(y_1, y_2, y_n). Here Disc(f) has codimension 3 in Y, the fibres of f over points of Disc(f) are nonreduced, and the map Sing(f) &rarr; Disc(f) has fibres of dimension 2.<\/p>\n<p>It seems pretty clear that we can keep going in this manner for arbitrary m. Enjoy!<\/p>\n<p><b>Edit 3\/8\/2020<\/b> Will Sawin gives this general example: the map from A^{2n+m} with coordinates (x_1,&#8230;,x_n, y_1,&#8230;,y_n,z_1,&#8230;.z_m) to A^{n+1} given by (x_1,&#8230;,x_n , x_1y_1+ &#8230;. + x_ny_n + z_1^2 + &#8230;. + z_m^2) is a map with (n+m-1)-dimensional fibers which has a codimension n+1 locus on the base where the codimension of the singularities of the fiber is exactly m-1 (the locus where the x_i and z_i all vanish), showing that all such pairs appear sharply.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Let f : X &rarr; Y be a dominant, finite type morphism of integral Noetherian schemes. We assume X is normal and Y regular. In this and the next blog post we define Sing(f) to be the closed set of &hellip; <a href=\"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/?p=4440\">Continue reading <span class=\"meta-nav\">&rarr;<\/span><\/a><\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-4440","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4440","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4440"}],"version-history":[{"count":16,"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4440\/revisions"}],"predecessor-version":[{"id":4462,"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4440\/revisions\/4462"}],"wp:attachment":[{"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4440"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4440"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.math.columbia.edu\/~dejong\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4440"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}