better compatibility through greater accuracy

prog/python/qqgg/parton_density_function_stuff.org

@@ -489,14 +489,14 @@ draw_histogram(ax, part_hist)
 Now, it would be interesting to know the total cross section.
 So let's define the increments for VEGAS.
 #+begin_src jupyter-python :exports both :results raw drawer
-  increments = np.array([3, 100, 100])
+  increments = np.array([4, 100, 100])
   tex_value(
       np.prod(increments), prefix=r"K=", prec=0, save=("results/pdf/", "num_increments.tex")
   )
 #+end_src
 
 #+RESULTS:
-: \(K=30000\)
+: \(K=40000\)
 
 And calculate the XS.
 #+begin_src jupyter-python :exports both :results raw drawer
@@ -511,14 +511,14 @@ And calculate the XS.
 
   xs_int_res = monte_carlo.integrate_vegas_nd(
       dist_η_vec,
-      [interval_η, [pdf.xMin, 1], [pdf.xMin, 1]],
-      epsilon=1e-11,
+      [interval_η, [pdf.xMin, pdf.xMax], [pdf.xMin, pdf.xMax]],
+      epsilon=1e-11/2,
       proc=1,
       increment_epsilon=.02,
       alpha=1.8,
       num_increments=increments,
       num_points_per_cube=10,
-      cache="cache/pdf/total_xs_2_5_20_take20",
+      cache="cache/pdf/total_xs_2_5_20_take22",
   )
 
   total_xs = gev_to_pb(np.array(xs_int_res.combined_result)) * 2 * np.pi
@@ -528,7 +528,7 @@ And calculate the XS.
 #+RESULTS:
 :RESULTS:
 : Loading Cache:  integrate_vegas_nd
-: array([3.86911687e+01, 1.96651183e-02])
+: array([3.86891167e+01, 9.39243388e-03])
 :END:
 
 #+begin_src jupyter-python :exports both :results raw drawer
@@ -547,7 +547,7 @@ there are two identical protons.
 #+end_src
 
 #+RESULTS:
-: -0.0012964079498659457
+: 0.006924228292849407
 
 The efficiency will be around:
 #+begin_src jupyter-python :exports both :results raw drawer
@@ -555,7 +555,7 @@ The efficiency will be around:
 #+end_src
 
 #+RESULTS:
-: 32.89204347314658
+: 36.68699853098365
 
 Let's export those results for TeX:
 #+begin_src jupyter-python :exports both :results raw drawer
@@ -691,7 +691,7 @@ Lets plot how the pdf looks.
 
 Overestimating the upper bounds helps with bias.
 #+begin_src jupyter-python :exports both :results raw drawer
-  overestimate = 1.1
+  overestimate = 1.0
   tex_value(
       (overestimate - 1) * 100,
       unit=r"\percent",
@@ -702,7 +702,7 @@ Overestimating the upper bounds helps with bias.
 
 #+RESULTS:
 :RESULTS:
-: \(\SI{10}{\percent}\)
+: \(\SI{0}{\percent}\)
 [[file:./.ob-jupyter/542b03d025920448ba653b470ec6492cbdd1e4a7.png]]
 [[file:./.ob-jupyter/d47db0dde9ae59979f271a7cba8dfc46be3f1dd3.png]]
 [[file:./.ob-jupyter/7fe9d3bd60427cf20af835649efbcbaafefbb3e0.png]]
@@ -718,7 +718,7 @@ figure out the cpu mapping.
       cubes=xs_int_res.cubes,
       proc="auto",
       report_efficiency=True,
-      cache="cache/pdf/total_xs_10000_000_2_5_take8",
+      cache="cache/pdf/total_xs_10000_000_2_5_take11",
       status_path="/tmp/status1",
       overestimate_factor=overestimate,
   )
@@ -728,7 +728,7 @@ figure out the cpu mapping.
 #+RESULTS:
 :RESULTS:
 : Loading Cache:  sample_unweighted_array
-: 0.29610040880251154
+: 0.3615754237122427
 :END:
 
 That does look pretty good eh? So lets save it along with the sample size.
@@ -744,7 +744,7 @@ That does look pretty good eh? So lets save it along with the sample size.
 #+end_src
 
 #+RESULTS:
-: \(\mathfrak{e}=\SI{30}{\percent}\)
+: \(\mathfrak{e}=\SI{36}{\percent}\)
 
 ** Observables
 Let's look at a histogramm of eta samples.
@@ -757,7 +757,7 @@ Let's look at a histogramm of eta samples.
 #+RESULTS:
 :RESULTS:
 : 10000000
-[[file:./.ob-jupyter/0b1b4f39201dac86ebfbfb8953561cfe81a6c70f.png]]
+[[file:./.ob-jupyter/764bd95aedbef68e7709a84780df593399e347a4.png]]
 :END:
 
 Let's use a uniform histogram image size.
@@ -786,7 +786,7 @@ And now we compare all the observables with sherpa.
 #+end_src
 
 #+RESULTS:
-[[file:./.ob-jupyter/a32ee488f4357426e3acecb1a5baaeddc367ee9b.png]]
+[[file:./.ob-jupyter/800bffd38987eae852b74eb2483698169c08c4de.png]]
 
 Hah! there we have it!
 
@@ -817,11 +817,10 @@ both equal.
 #+end_src
 
 #+RESULTS:
-[[file:./.ob-jupyter/deeb122e09b948ea416db671bfe1838aedeae84a.png]]
+[[file:./.ob-jupyter/45cc8fbfa523c8faf704505d716ebc299a1a44fd.png]]
 
 The invariant mass is not constant anymore.
 #+begin_src jupyter-python :exports both :results raw drawer
-  bins = np.logspace(*np.log10([2 * min_pT, 2 * e_proton]), 51)
   yoda_hist_inv_m = yoda_to_numpy(yoda_file["/MC_DIPHOTON_PROTON/inv_m"])
 
   fig, (ax, ax_ratio) = draw_ratio_plot(
@@ -842,7 +841,7 @@ The invariant mass is not constant anymore.
 #+end_src
 
 #+RESULTS:
-[[file:./.ob-jupyter/a7708226adeb9782b68ffb10003c280d8dd19ef2.png]]
+[[file:./.ob-jupyter/9ad48ff715bb6445bf5a0c515e4a4f41593146fe.png]]
 
 The cosθ distribution looks more like the paronic one.
 #+begin_src jupyter-python :exports both :results raw drawer
@@ -861,7 +860,7 @@ The cosθ distribution looks more like the paronic one.
 #+end_src
 
 #+RESULTS:
-[[file:./.ob-jupyter/5bf3849e8196878ba4585a09e496e852b9969866.png]]
+[[file:./.ob-jupyter/d834978baba81dba0e2f052f1965b62ae736c151.png]]
 
 
 #+begin_src jupyter-python :exports both :results raw drawer
@@ -883,7 +882,7 @@ The cosθ distribution looks more like the paronic one.
 #+end_src
 
 #+RESULTS:
-[[file:./.ob-jupyter/cd50dc4eb341d959871bf4288ff3620556a53970.png]]
+[[file:./.ob-jupyter/22100d74fc1e07aa9ac75e22dd38ce49eaad89d4.png]]
 
 In this case the opening angles are the same because the CS frame is
 the same as the ordinary rest frame. The z-axis is the beam axis
@@ -907,4 +906,4 @@ because pT=0!
 #+end_src
 
 #+RESULTS:
-[[file:./.ob-jupyter/9ae6d76460a0d99d6e64cd85c8d0b712984b93d6.png]]
+[[file:./.ob-jupyter/28abf9fcf1840bc31a5819d380a8eff0c2ef3e43.png]]

prog/python/qqgg/results/pdf/my_sigma.tex

@@ -1 +1 @@
-\(\sigma = \SI{38.691\pm 0.020}{\pico\barn}\)
+\(\sigma = \SI{38.689\pm 0.009}{\pico\barn}\)

prog/python/qqgg/results/pdf/num_increments.tex

@@ -1 +1 @@
-\(K=30000\)
+\(K=40000\)

prog/python/qqgg/results/pdf/overesimate.tex

@@ -1 +1 @@
-\(\SI{10}{\percent}\)
+\(\SI{0}{\percent}\)

prog/python/qqgg/results/pdf/samp_eff.tex

@@ -1 +1 @@
-\(\mathfrak{e}=\SI{30}{\percent}\)
+\(\mathfrak{e}=\SI{36}{\percent}\)