Beam setup changes destroy integration / event generation

[Christian Speckner]

At the moment, the automatic integrations performed when unstable is executed use the beam setup which is in place at the moment of execution. This can lead to unexpected results (e.g. when the beam has been polarized) or even complete failure (e.g. if the beams are set up for collisions). My solution to this would be to locally deactivate beams during the execution of unstable --- what do the other WHIZARDs think?

[dwiesler]

I confirm the beam setup mismatch error using the example casc_dec.sin file, when the initial integration of the decays is commented out.

CS' solution sounds good to me, although I really don't know this part of the code very well.

[Juergen Reuter]

Sounds like something that could be really solved quickly. Therefore I give it high priority.

[kilian]

Solution in r2521: ignore any previous beam statement. If there is a previous integration result for the same decay process, use it, but only if it had no beam specified. Otherwise, do a new integration.

[dwiesler]

Seems as if this is not yet finished.

The example casc_dec.sin ist not working anymore, since after the first request of simulation, WHIZARD claims no beam setup is allowed and starts to integrate again, thereby, and I'm not sure if these issues are connected, letting the efficiencies run into NaN's.

| Integrating process 'full':
|=============================================================================|
| It      Calls  Integral[fb]  Error[fb]   Err[%]    Acc  Eff[%]   Chi2 N[It] |
|=============================================================================|
   1      10000  5.7412271E-01  9.34E-02   16.27   16.27*   1.46
   2      10000  3.4179670E-01  3.19E-02    9.34    9.34*   0.71
   3      10000  3.3950505E-01  1.05E-02    3.10    3.10*   2.28
   4      10000  3.4469844E-01  9.49E-03    2.75    2.75*   2.35
   5      10000  3.5140005E-01  5.60E-03    1.59    1.59*   5.70
   6      10000  3.6864745E-01  1.30E-02    3.52    3.52    1.83
   7      10000  3.6175569E-01  8.02E-03    2.22    2.22*   2.19
   8      10000  3.5707403E-01  7.57E-03    2.12    2.12*   2.30
   9      10000  4.5893769E-01  9.05E-02   19.71   19.71    0.22
  10      10000  4.2119482E-01  4.78E-02   11.36   11.36*   0.64
|-----------------------------------------------------------------------------|
  10     100000  3.5400073E-01  3.29E-03    0.93    2.93    0.64    1.61  10
|-----------------------------------------------------------------------------|
  11      20000  3.5824875E-01  4.58E-03    1.28    1.81*   1.90
  12      20000  3.6609526E-01  3.40E-03    0.93    1.31*   1.11
  13      20000  3.5271111E-01  2.36E-03    0.67    0.95*    NaN
  14      20000  3.4684695E-01  1.89E-03    0.55    0.77*    NaN
  15      20000  3.5238471E-01  1.69E-03    0.48    0.68*   0.18
|-----------------------------------------------------------------------------|
  15     100000  3.5235818E-01  1.03E-03    0.29    0.92    0.18    6.62   5
|-----------------------------------------------------------------------------|
|=============================================================================|
  15     100000  3.5235818E-01  1.03E-03    0.29    0.92    0.18    6.62   5
|=============================================================================|
| Process 'full':
|    time estimate for generating 10000 unweighted events: 27m:35s
|-----------------------------------------------------------------------------|
n_events =        10000
$title = "Full process"
$description = "$p + p \to u+ \bar u \to \bar{\tilde u}_1 + u + \tilde e_{12}^+ + e^-$"
$xlabel = "$M_{\rm inv}(ue^-)$"
$title = "Factorized process with complete spin correlations"
$description = "$p + p \to u\bar u \to \bar{\tilde u}_1 + 
  (\tilde u_1 \to u + (\tilde\chi_2^0 \to \tilde e_{12}^+ + e^-))$"
$title = "Factorized process with classical spin correlations"
$title = "Factorized process with isotropic decay"
$sample = "casc_dec_full"
| Initializing simulation for processes full:
| Simulation mode = unweighted, event_normalization = '1', unpolarized_events
Warning: Discarding previous result for process 'full': no beam setup allowed
| Initializating integration for process full:
| Read phase-space configuration from file 'full.phs'...
| ... found 49 phase space channels, collected in 25 groves.
| Phase space: found 65 equivalences between channels.
| iterations = 10:10000, 5:20000
Warning: No cuts have been defined.
| Reading integration grids and results from file 'full.vg':
| Beam setup has changed, discarding old grid file
| Creating VAMP integration grids:
| Using phase-space channel equivalences.
| 10000 calls, 49 channels, 8 dimensions, 20 bins, stratified = T
| Integrating process 'full':
|=============================================================================|
| It      Calls  Integral[fb]  Error[fb]   Err[%]    Acc  Eff[%]   Chi2 N[It] |
|=============================================================================|
   1      10000  1.4981199E+00  1.08E+00   71.81   71.81*   0.65
   2      10000  1.7247537E+00  1.03E+00   59.73   59.73*   0.44
   3      10000  1.3102155E+00  4.01E-01   30.62   30.62*   0.10
   4      10000  9.9336372E-01  5.13E-02    5.17    5.17*   0.83
   5      10000  1.0751755E+00  4.98E-02    4.63    4.63*   0.87
   6      10000  1.4915800E+00  3.10E-01   20.78   20.78    0.18
   7      10000  1.1730722E+00  3.88E-02    3.31    3.31*   1.18
   8      10000  1.2105528E+00  3.10E-02    2.56    2.56*   2.32
   9      10000  9.7380981E-01  1.63E-02    1.67    1.67*    NaN
  10      10000  1.4885763E+00  1.03E-01    6.89    6.89    1.28
|-----------------------------------------------------------------------------|
  10     100000  1.0497769E+00  1.25E-02    1.19    3.77    1.28    9.05  10
|-----------------------------------------------------------------------------|
  11      20000  1.0470912E+00  2.30E-02    2.19    3.10*    NaN
  12      20000  1.0465725E+00  1.44E-02    1.37    1.94*    NaN
  13      20000  1.0583460E+00  9.75E-03    0.92    1.30*    NaN
  14      20000  1.0689840E+00  7.93E-03    0.74    1.05*    NaN
  15      20000  1.0774136E+00  7.45E-03    0.69    0.98*    NaN
|-----------------------------------------------------------------------------|
  15     100000  1.0668394E+00  4.42E-03    0.41    1.31     NaN    1.39   5
|-----------------------------------------------------------------------------|
|=============================================================================|
  15     100000  1.0668394E+00  4.42E-03    0.41    1.31     NaN    1.39   5
|=============================================================================|
| Process 'full':
|    time estimate for generating 10000 unweighted events: -24855d:-3h:-14m:-8s
|-----------------------------------------------------------------------------|

The time estimates obviously gets affected here.

Afterwards, eventgeneration breaks w/ the following error on OSX:

| Process 'full':
|    time estimate for generating 10000 unweighted events: -24855d:-3h:-14m:-8s
|-----------------------------------------------------------------------------|
******************************************************************************
******************************************************************************
*** FATAL ERROR: Process '	' is not available for simulation.
******************************************************************************
******************************************************************************
| There were no errors and    6 warning(s).
WHIZARD run aborted.

or runs forever on debian i386...

[Juergen Reuter]

Good that this has been found. Maybe we should add a test where the examples are also run to see whether they are still working. This has to be resolved before we can release.

[dwiesler]

A remark: the NaN's in the efficiency were introduced (at least for me) w/ the improved phase space setup of r2508 and are probably not linked with the OSX fatal error described above.

[kilian]

There were two unrelated bugs, neither of them connected to phase space.

1. The check for missing integrals in 'simulate' was called with wrong args; I should remember to identify optional args always by name, not by position.
2. In casc_decay.sin, the process lacked a cut on Pt of the outgoing quark, there was a t-channel singularity. This was the cause of the nan-s.

Fixed in r2529.