#197 closed defect (invalid)
integration error rises when including PDFs
| Reported by: | sschmidt | Owned by: | kilian |
|---|---|---|---|
| Priority: | P1 | Milestone: | |
| Component: | core | Version: | 2.0.0rc2 |
| Severity: | critical | Keywords: | |
| Cc: |
Description
When replacing
sqrts = 500 GeV beams = U,u
from test/jets.sin by
sqrts = 2000 GeV beams = p,p -> lhapdf
and integrating the processes given in jets.sin, the integration error is about a factor of 10 higher.
E.g. the jj cross section if for U,u beams:
|=============================================================================| | It Calls Integral[fb] Error[fb] Err[%] Acc Eff[%] Chi2 N[It] | |=============================================================================| 1 20000 1.4024531E+06 9.18E+03 0.65 0.93* 5.12 2 20000 1.4163212E+06 9.16E+03 0.65 0.91* 5.20 3 20000 1.3745491E+06 9.59E+03 0.70 0.99 5.04 |=============================================================================| 3 60000 1.3984734E+06 5.37E+03 0.38 0.94 5.04 5.10 3 |=============================================================================|
and for the p,p-beams
|=============================================================================| | It Calls Integral[fb] Error[fb] Err[%] Acc Eff[%] Chi2 N[It] | |=============================================================================| 1 20000 8.1056636E+04 5.40E+03 6.66 9.42* 0.37 2 20000 8.5659810E+04 5.59E+03 6.52 9.23* 0.41 3 20000 8.5594942E+04 5.14E+03 6.01 8.50* 0.44 |=============================================================================| 3 60000 8.4118602E+04 3.10E+03 3.68 9.02 0.44 0.24 3 |=============================================================================|
Thanks to Daniel for reporting this.
Can the inclusion of PDFs enlarge the integration errors by a factor of 10?
Change History (7)
comment:1 Changed 15 years ago by
comment:2 Changed 15 years ago by
| Resolution: | → wontfix |
|---|---|
| Status: | new → closed |
comment:3 Changed 15 years ago by
| Resolution: | wontfix |
|---|---|
| Status: | closed → reopened |
Here is Daniel's comment:
Das is der W1 output:
! WHIZARD run for process jetstest:
!=============================================================================
! It Calls Integral[fb] Error[fb] Err[%] Acc Eff[%] Chi2 N[It]
!-----------------------------------------------------------------------------
! Reading cut configuration data from file whizard.cut1
! Replacing default cuts by user-defined cuts.
cut pT of 1 within 1.00000E+02 1.00000E+04
cut pT of 2 within 1.00000E+02 1.00000E+04
cut pT of 4 within 1.00000E+02 1.00000E+04
cut delta-r of 1 2 within 1.00000E+00 1.00000E+01
cut delta-r of 1 4 within 1.00000E+00 1.00000E+01
cut delta-r of 2 4 within 1.00000E+00 1.00000E+01
! Preparing (fixed weights): 1 sample of 10000 calls ...
1 10000 6.8893107E+00 2.80E+00 40.61 40.61* 0.15
0.00 1
!-----------------------------------------------------------------------------
! Adapting (variable wgts.): 5 samples of 10000 calls ...
2 10000 4.6720992E+00 1.38E+00 29.49 29.49* 0.18
3 10000 7.4050624E+00 9.71E-01 13.12 13.12* 0.32
4 10000 8.2889482E+00 4.18E-01 5.04 5.04* 0.89
5 10000 8.0754186E+00 2.83E-01 3.50 3.50* 1.84
6 10000 7.3433128E+00 2.11E-01 2.88 2.88* 2.57
!-----------------------------------------------------------------------------
! Integrating (fixed wgts.): 3 samples of 10000 calls ...
7 30000 7.6536289E+00 1.51E-01 1.97 3.42 0.97
0.19 3
!-----------------------------------------------------------------------------
!
! Time estimate for generating 10000 unweighted events: 0h 00m 42s
!=============================================================================
! Summary (all processes):
!-----------------------------------------------------------------------------
! Process ID Integral[fb] Error[fb] Err[%] Frac[%]
!-----------------------------------------------------------------------------
jetstest 7.6536289E+00 1.51E-01 1.97 100.00
!-----------------------------------------------------------------------------
sum 7.6536289E+00 1.51E-01 1.97 100.00
!=============================================================================
und das der vergleichbare W2:
| Creating grids
| 20000 calls, 25 channels, 7 dimensions, 20 bins, stratified = T
|=============================================================================|
| It Calls Integral[fb] Error[fb] Err[%] Acc Eff[%]
| Chi2 N[It] |
|=============================================================================|
1 20000 7.9530373E+00 2.32E+00 29.14 41.22* 0.18
2 20000 1.3998675E+01 4.39E+00 31.34 44.32 0.12
3 20000 5.3754824E+00 1.14E+00 21.18 29.95* 0.13
4 20000 4.9941640E+00 1.12E+00 22.39 31.66 0.11
5 20000 7.1907576E+00 2.00E+00 27.82 39.34 0.09
6 20000 7.8801646E+00 1.54E+00 19.58 27.69* 0.10
7 20000 8.5053854E+00 1.75E+00 20.55 29.07 0.12
8 20000 5.0550115E+00 1.00E+00 19.87 28.11* 0.10
9 20000 6.9491414E+00 1.31E+00 18.82 26.61* 0.10
10 20000 6.4434584E+00 1.35E+00 20.98 29.66 0.09
|=============================================================================|
10 200000 6.2559147E+00 4.50E-01 7.20 32.20 0.09
| 1.14 10
|=============================================================================|
beides mal fuer u,U -> d,D,g mit lhapdf aktiviert. Da is definitiv
noch wat faul bezgl des Phasenraums mit aktivierten Strukturfunktionen
fuer W2!
hier der Inhalt des sindarin files:
alias j = u:d:s:U:D:S:g
process jets = u, U -> d, D, g
compile
cuts = all Pt > 100 GeV [j]
and
all Dist > 1 [j,j]
sqrts = 2000
iterations = 10:20000
beams = p, p -> lhapdf
integrate (jets)
n_events = 1000
$file_lhef = "jets.lhef"
simulate (jets)
comment:5 Changed 15 years ago by
| Priority: | P3 → P1 |
|---|---|
| Severity: | normal → critical |
I'll do this next, it's essential for LHC...
comment:6 Changed 15 years ago by
| Resolution: | → invalid |
|---|---|
| Status: | reopened → closed |
... arrggh!!
It took me several hours to find out that in the sample run, the grids are not adapted at all. The problem is here:
iterations = 1:10000
If there is only one pass, the grids are kept fixed. There is NO problem with the structure functions.
If I set (as in the W1 run above)
iterations = 6:10000, 3:10000
the output is
|=============================================================================| | It Calls Integral[fb] Error[fb] Err[%] Acc Eff[%] Chi2 N[It] | |=============================================================================| 1 10000 9.6935749E+00 3.02E+00 31.16 31.16* 0.39 2 10000 6.7269146E+00 5.97E-01 8.88 8.88* 0.26 3 10000 6.9282520E+00 3.66E-01 5.28 5.28* 1.73 4 10000 7.0721804E+00 4.28E-01 6.06 6.06 1.30 5 10000 7.4295492E+00 3.32E-01 4.47 4.47* 1.67 6 10000 7.5572470E+00 3.21E-01 4.24 4.24* 1.75 |-----------------------------------------------------------------------------| 6 60000 7.2511865E+00 1.70E-01 2.34 5.74 1.75 0.72 6 |-----------------------------------------------------------------------------| 7 10000 8.1755060E+00 3.61E-01 4.42 4.42 1.58 8 10000 7.9508841E+00 4.25E-01 5.34 5.34 1.54 9 10000 7.9376105E+00 3.45E-01 4.35 4.35* 1.75 |=============================================================================| 9 30000 8.0253353E+00 2.15E-01 2.68 4.64 1.75 0.13 3 |=============================================================================|
which is ok.
I close this ticket and open another one on the confusing behavior of integration passes.
Note: See
TracTickets for help on using
tickets.
That's unavoidable. Having sqrts non-constant makes a lot of a difference. Ideas for improvement welcome, of course. (Usually, spending extra iterations helps, but you will never reach the same level of accuracy)
You should compare W1/W2, if W1 is much better here, there is a point, and you should reopen the ticket.