From: David Wells Date: Thu, 18 Jan 2018 18:37:43 +0000 (-0500) Subject: Implement a 2D-specific chart point estimate. X-Git-Tag: v9.0.0-rc1~544^2~4 X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=7a2c1d182e9d4e4dc20237acb96d5449cbecc0ce;p=dealii.git Implement a 2D-specific chart point estimate. The current algorithm calculates a reasonable guess for the chart point in 3D. Using this 3D algorithm in 2D results in relatively bad guesses: for i = 5: affine approx chart point: 0.5 3.64977e-14 pulled back chart point: 1 0.5 for i = 6: affine approx chart point: 0.5 0.5 pulled back chart point: 0.5 8.678e-17 for i = 7: affine approx chart point: 0.5 1 pulled back chart point: 0.5 1 This patch adds a 2D-specific algorithm that simply averages the two adjacent vertices (e.g., chart point 5 is assumed to be an average of chart points 1 and 3). This issue was reported by Juan Carlos Araujo Cabarcas, who found a test case where the affine guess was sufficiently inaccurate that TransfiniteInterpolationManifold::pull_back failed to find the right point. This change fixes that test case. --- diff --git a/source/grid/manifold_lib.cc b/source/grid/manifold_lib.cc index 48f2878e12..43989a311a 100644 --- a/source/grid/manifold_lib.cc +++ b/source/grid/manifold_lib.cc @@ -1825,12 +1825,68 @@ TransfiniteInterpolationManifold chart_points[i] = pull_back(cell, surrounding_points[i], p3); } // 8 points usually form either a cube or a rectangle with vertices - // and line mid points. assume a cube here which gives us some new - // initial guesses - else if (surrounding_points.size() == 8 && i > 4) + // and line mid points. Get the initial guess with line segment + // midpoints in 2D and assuming a cube for 3D. + else if (surrounding_points.size() == 8 && + ((dim == 3 && i > 4) || (dim == 2 && i > 3))) { - const Point guess = chart_points[i-4] + - (chart_points[4] - chart_points[0]); + Point guess; + switch (dim) + { + case 2: + // inline the standard numbering + // + // 2 - 7 - 3 + // | | + // 4 5 + // | | + // 0 - 6 - 1 + // + // to calculate guesses based on averaging already computed + // chart points. + switch (i) + { + case 4: + guess = 0.5*(chart_points[0] + chart_points[2]); + break; + case 5: + guess = 0.5*(chart_points[1] + chart_points[3]); + break; + case 6: + guess = 0.5*(chart_points[0] + chart_points[1]); + break; + case 7: + guess = 0.5*(chart_points[2] + chart_points[3]); + break; + default: + Assert(false, ExcInternalError()); + } + break; + case 3: + // Assuming that we are in 3D and have the points around a + // cube numbered as + // + // 6-------7 + // /| /| + // / / | + // / | / | + // 4-------5 | + // | 2- -|- -3 + // | / | / + // | | / + // |/ |/ + // 0-------1 + // + // (where vertex 2 is the back left vertex) we can estimate + // where chart points 5 - 7 are by computing the height (in + // chart coordinates) as c4 - c0 and then adding that onto the + // appropriate bottom vertex. + guess = chart_points[i - 4] + (chart_points[4] - chart_points[0]); + break; + default: + Assert(false, ExcInternalError()); + } + chart_points[i] = pull_back(cell, surrounding_points[i], guess); } else