1. Matthew Turk
  2. yt.milestones

Commits

Anthony Scopatz  committed db68368

Added r-slicing to cylindrical rays.

  • Participants
  • Parent commits 5790570
  • Branches default

Comments (0)

Files changed (1)

File cylindrical_rays.ipynb

View file
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [WARNING  ] 2012-08-23 15:36:29,721 integer runtime parameter checkpointfilenumber overwrites a simulation scalar of the same name\n"
+        "yt : [WARNING  ] 2012-08-23 16:51:37,725 integer runtime parameter checkpointfilenumber overwrites a simulation scalar of the same name\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [WARNING  ] 2012-08-23 15:36:29,722 integer runtime parameter forcedplotfilenumber overwrites a simulation scalar of the same name\n"
+        "yt : [WARNING  ] 2012-08-23 16:51:37,725 integer runtime parameter forcedplotfilenumber overwrites a simulation scalar of the same name\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [WARNING  ] 2012-08-23 15:36:29,722 integer runtime parameter nbegin overwrites a simulation scalar of the same name\n"
+        "yt : [WARNING  ] 2012-08-23 16:51:37,726 integer runtime parameter nbegin overwrites a simulation scalar of the same name\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [WARNING  ] 2012-08-23 15:36:29,723 integer runtime parameter plotfilenumber overwrites a simulation scalar of the same name\n"
+        "yt : [WARNING  ] 2012-08-23 16:51:37,726 integer runtime parameter plotfilenumber overwrites a simulation scalar of the same name\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [INFO     ] 2012-08-23 15:36:29,731 Parameters: current_time              = 8.00057343882e-10\n"
+        "yt : [INFO     ] 2012-08-23 16:51:37,737 Parameters: current_time              = 8.00057343882e-10\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [INFO     ] 2012-08-23 15:36:29,732 Parameters: domain_dimensions         = [48 96  1]\n"
+        "yt : [INFO     ] 2012-08-23 16:51:37,738 Parameters: domain_dimensions         = [48 96  1]\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [INFO     ] 2012-08-23 15:36:29,733 Parameters: domain_left_edge          = [ 0.     -1.2288  0.    ]\n"
+        "yt : [INFO     ] 2012-08-23 16:51:37,739 Parameters: domain_left_edge          = [ 0.     -1.2288  0.    ]\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [INFO     ] 2012-08-23 15:36:29,734 Parameters: domain_right_edge         = [ 1.2288      1.2288      6.28318531]\n"
+        "yt : [INFO     ] 2012-08-23 16:51:37,740 Parameters: domain_right_edge         = [ 1.2288      1.2288      6.28318531]\n"
        ]
       },
       {
        "output_type": "stream",
        "stream": "stderr",
        "text": [
-        "yt : [INFO     ] 2012-08-23 15:36:29,736 Parameters: cosmological_simulation   = 0.0\n"
+        "yt : [INFO     ] 2012-08-23 16:51:37,741 Parameters: cosmological_simulation   = 0.0\n"
        ]
       }
      ],
      "collapsed": false,
      "input": [
       "# Ray tracer\n",
-      "E = (1.0, 1.0, 0.0)\n",
-      "#F = (1.0, 1.0, na.pi)\n",
-      "F = (1.0, 1.0, 0.75 * na.pi)\n",
+      "E = (0.5, 0.0, 0.0)\n",
+      "F = (1.0, 0.0, 0.75*na.pi)\n",
       "\n",
-      "Ecart = na.array((2.0, 0.0, 1.0))\n",
-      "Fcart = na.array((-2.0, na.sin(0.75*na.pi), 1.0))\n",
+      "Ecart = na.array((E[0]*na.cos(E[2]), E[0]*na.sin(E[2]), E[1]))\n",
+      "Fcart = na.array((F[0]*na.cos(F[2]), F[0]*na.sin(F[2]), F[1]))\n",
       "\n",
       "D = Fcart - Ecart"
      ],
      "language": "python",
      "metadata": {},
      "outputs": [],
-     "prompt_number": 14
+     "prompt_number": 8
     },
     {
      "cell_type": "code",
      "collapsed": false,
      "input": [
-      "zmask = np.logical_or(pf.h.grid_left_edge[:,1] == 0.0, pf.h.grid_right_edge[:,1] == 0.0)\n",
+      "zmask = np.logical_or(pf.h.grid_left_edge[:,1] == E[1], pf.h.grid_right_edge[:,1] == E[1])\n",
       "r = pf.h.grid_right_edge[zmask,0]\n",
+      "\n",
       "a = (D**2)[:2].sum()\n",
       "b = (2*D*Ecart)[:2].sum()\n",
       "c = (Ecart**2)[:2].sum() - r**2\n",
       "\n",
-      "tp = (-b + na.sqrt(b**2 - 4*a*c)) / (2*a)  # tp1 = -b/(2*a) + na.sqrt(0.25*(b/a)**2 - c/a)\n",
       "tm = (-b - na.sqrt(b**2 - 4*a*c)) / (2*a)\n",
+      "tp = (-b + na.sqrt(b**2 - 4*a*c)) / (2*a)  \n",
       "\n",
-      "tpmask = ~np.isnan(tp)\n",
-      "tmmask = ~np.isnan(tm)\n",
+      "tmmask = np.logical_and(~np.isnan(tm), r <= E[0])\n",
+      "tpmask = np.logical_and(~np.isnan(tp), r <= F[0])\n",
       "t = np.concatenate([tm[tmmask][::-1], tp[tpmask]])\n",
       "ti = t.argsort()\n",
       "t = t[ti]\n",
      "language": "python",
      "metadata": {},
      "outputs": [],
-     "prompt_number": 15
+     "prompt_number": 9
     },
     {
      "cell_type": "code",
      "outputs": [
       {
        "output_type": "pyout",
-       "prompt_number": 18,
+       "prompt_number": 10,
        "text": [
-        "[<matplotlib.lines.Line2D at 0x909c550>]"
+        "[<matplotlib.lines.Line2D at 0x7fa1c8155bd0>]"
        ]
       },
       {
        "output_type": "display_data",
-       "png": "iVBORw0KGgoAAAANSUhEUgAAAXUAAAD9CAYAAABDaefJAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n"
+       "png": "iVBORw0KGgoAAAANSUhEUgAAAX8AAAD9CAYAAABUS3cAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/w+DzqEu27J6qAAAAABJRU5ErkJggg==\n"
       }
      ],
-     "prompt_number": 18
+     "prompt_number": 10
     },
     {
      "cell_type": "code",
      "input": [],
      "language": "python",
      "metadata": {},
-     "outputs": []
+     "outputs": [],
+     "prompt_number": 4
     }
    ],
    "metadata": {}