Closed scarlehoff closed 2 years ago
The numbers are not absolutely crazy for the scale variations (which means that if I missed something at least it does not multiply a divergence).
In order to test the scale variations I can also generate a grid for r_muR=2, r_muF=0.5 for instance. @cschwan how do I get the results for that? What would --scales 3
or --scales 7
produce?
@scarlehoff I suppose you can leave the scale variation set to its default (7-point scale variation), but give convolute
the extra switch -a
(short for --absolute
), which will explicitly show you all seven scale-varied results.
In order to test the scale variations I can also generate a grid for r_muR=2, r_muF=0.5 for instance.
However, for this particular scale variation you'll need the 9-point one.
Ok -s 9 -a
is what I need. Thanks!
One last comment:
b etal dsig/detal (1,1) (2,2) (0.5,0.5) (2,1) (1,2) (0.5,1) (1,0.5)
[] [pb] [pb] [pb] [pb] [pb] [pb] [pb] [pb]
--+-------------------+-------------------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------
0 0 0.21000000000000002 5.8160276e2 5.8160276e2 5.9586544e2 5.7033788e2 5.8173849e2 5.9153986e2 5.8143363e2 5.6518854e2
1 0.21000000000000002 0.42000000000000004 5.8305367e2 5.8305367e2 5.9740188e2 5.7166953e2 5.8318310e2 5.9309082e2 5.8289238e2 5.6652444e2
2 0.42000000000000004 0.63 5.8507036e2 5.8507036e2 5.9952938e2 5.7351192e2 5.8522952e2 5.9520110e2 5.8487204e2 5.6842278e2
3 0.63 0.8400000000000001 5.8978094e2 5.8978094e2 6.0445348e2 5.7789617e2 5.8985141e2 6.0024712e2 5.8969313e2 5.7274408e2
4 0.8400000000000001 1.05 5.9404180e2 5.9404180e2 6.0883674e2 5.8190653e2 5.9407032e2 6.0470514e2 5.9400625e2 5.7675040e2
5 1.05 1.37 6.0055591e2 6.0055591e2 6.1572014e2 5.8782319e2 6.0043869e2 6.1183196e2 6.0070196e2 5.8261226e2
6 1.37 1.52 6.0807392e2 6.0807392e2 6.2360037e2 5.9472526e2 6.0764516e2 6.2016255e2 6.0860817e2 5.8930040e2
7 1.52 1.74 6.0948805e2 6.0948805e2 6.2532209e2 5.9560835e2 6.0896514e2 6.2213068e2 6.1013961e2 5.9024710e2
8 1.74 1.95 6.1029012e2 6.1029012e2 6.2631099e2 5.9590317e2 6.0925596e2 6.2391246e2 6.1157873e2 5.9024184e2
9 1.95 2.18 6.0109307e2 6.0109307e2 6.1710921e2 5.8635416e2 5.9964338e2 6.1550267e2 6.0289943e2 5.8061802e2
10 2.18 2.5 5.7361389e2 5.7361389e2 5.8869057e2 5.5929736e2 5.7108844e2 5.8881951e2 5.7676068e2 5.5301373e2
will show you the results as tuples of the form (xi_R, xi_F)
.
Great, it works :)
Results computed "by hand":
(1,1) --- (0.5, 0.5) --- (2,1) --- (1,2) --- (2,2) --- (0.5, 2) --- (2, 0.5)
440.5 --- 471.8 --- 421.0 --- 435.6 --- 411.4 --- 468.2 --- 441.4
results produced by pineappl
Pineappl results:
bin x0 x1 diff (1,1) (2,2) (0.5,0.5) (2,1) (1,2) (0.5,1) (1,0.5) (2,0.5) (0.5,2)
---+-------+-------+----+----+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------+-----------
0 7.10428 7.10428 -0.2 -0.2 4.4126323e2 4.4126323e2 4.1193325e2 4.7313431e2 4.2144816e2 4.3644362e2 4.6793163e2 4.5582043e2 4.4179080e2 4.7050294e2
Results computed "by hand":
Do you have these results with more digits? For some of them the accuracy is a bit worse maybe - are they computed using separate MC runs? If so, what are the MC uncertainties for them?
I added one extra digit (the mc error is around 0.1 for everyone).
The only one that worries me is (0.5, 2.0) but since (0.5, 2.0) is ok I'm less worried.
This is what I find (differences are given in per mille):
scale diff
1 (1,1) 1.7326447
2 (2,2) 1.2961838
3 (0.5,0.5) 2.8281263
4 (2,1) 1.0645131
5 (1,2) 1.9366850
6 (2,0.5) 0.8853647
7 (0.5,2) 4.9187100
and the differences in multiples of the MC uncertainty (0.1):
scale sigma
1 (1,1) 7.6323
2 (2,2) 5.3325
3 (0.5,0.5) 13.3431
4 (2,1) 4.4816
5 (1,2) 8.4362
6 (2,0.5) 3.9080
7 (0.5,2) 23.0294
yeah, as I said, 0.5-2 is the only one that worries me. I wonder whether vrap implements its own alpha_s since the other bigger difference is 0.5,0.5.
At NLO we have the same kind of differences.
If you want we can do a proper check etc to know where the differences are coming from and all that, but I think most of it is due to the interpolation itself (already almost a 2 per-mille difference for (1,1)) which makes sense since we are subtracting numerical divergences inside the pineappl grid. Then these differences increase a bit more when the values grow (as alpha_s grow).
I would consider these numbers as ok x)
Addresses #13
For the central scale seems to work (pineappl and vrap produce the same result). It would be good to have a third code to properly check the results of course... (beyond the fact that they haven't changed since the initial version).