Contents:
Method for data scoring:
This assessment is provided as a (very rough) guide to the quality of the experimental data.
The lower a score, the better.The score is computed by:
Score = aveDev + sqrt(varDev) + (2004-pubYear)/5 + pubTypeFactor.
The pubTypeFactor is a cheezy measure of the amount of peer review a data set has received.
It is given by:
- Lab Report: 5
- Journal: 0
- Book: 2
- Conf. Proc.: 4
- Thesis: 2
- Prog. Report: 6
- Private Comm.: 10
The aveDev is computed by first creating an average evaluation from all the evaluations for the specific reaction.
With that, we loop over the data and compute the average deviation from the average evaluation and the data.
The uncertainty on the data are not taken into account (this is done in the evaluation scoring).
Similarly, the varDev is the variation in the deviation of the data and the average evaluation.
Together, these two quantities tell us whether the data has the right magnitude (low aveDev)
and the right shape (low varDev).
In addition to the raw score, some comments are generated.
They come in two flavors: a raw assessment of the shape of the data and an assessment of the score itself.
The assessment of the shape is straightforward and relies on aveDev and varDev:
- OK Data: if aveDev<5., agrees well with exiting evaluations
- Shape Data: if varDev<2. has same shape as existing evaluation so is likely shape data
- Possible Shape Data: if varDev<4. it has a different shape but still may be shape data
- Wrong Shape: if varDev<10. it has the wrong shape, but not so wrong that it possibly couldn't be salvaged
- Useless: aveDev and/or varDev are unacceptably high rendering this data probably useless
The assessment of the score itself is just a straightforward comment regarding the Score:
- bad score: if 20.<Score<40.
- really bad score: if Score>=40.
Method for fit quality assesment:
This assessment is provided as a guide to determine whether it is possible to perform a reasonable fit to the data.
If so, then we recommend either finding a recent evaluation in which the evaluators actually performed a fit
to the data or performing a fit yourself.
We now list the column labels and their meaning:
- nData: Number of data points in all sets within this energy region
- aveDeltaEn: The average spacing between adjacent data points' incident energy
- sqrt(varDeltaEn): The variation in the average spacing between adjacent data points' incident energy
- minEn: Minimum incident energy for all data points in this energy range
- maxEn: Maximum incident energy for all data points in this energy range
- Comment: Comment stating our opinion whether this data can be fit reliably:
- OK to fit:aveDeltaEn < Cutoffs[i][0] and sqrt(varDeltaEn) < Cutoffs[i][1]
- Points not dense enough:aveDeltaEn > Cutoffs[i][0]
- Erratic point spacing:sqrt(varDeltaEn) > Cutoffs[i][1]
- Useless: Both varDeltaEn and aveDeltaEn are inadequate to support fitting in this energy range
The Cutoff Table is given by:
| Region | Cutoff[i][0] (MeV) | Cutoff[i][1] (MeV) | Comment |
| Thermal (< 10-6 MeV) |
1.0 |
1.0 |
Only need one point to pin down 1/v rise |
| Resonance (10-6 - 10-3 MeV) |
1e-05 |
1e-05 |
Gets even 10 eV spacing |
| Unresolved Resonance (10-3 - 1 MeV) |
0.01 |
0.01 |
Gets even 10 keV spacing |
| High Energy (> 1 MeV) |
1.0 |
1.0 |
Gets even 1 MeV spacing |
Method for evaluation scoring:
To determine whether an evaluation is suitable, we first compute the evaluation's
:
Eventually we will compare each evaluation's
to a predetermined parameter
for each energy range and choose the evaluation with the lowest
from all of the suitable evaluations. These calculations come with the following caveats:
- We did not account for uncertainty in the incident energies
- We did not account for any kind of resolution effects or energy spread in the incident energy
- We comment that we did not compute covariance matrices for data
- We did not consider coupled (i.e. ratio) data
- We made no attempt to evaluate data itself. This implies that, for example, badly normalized data is not fixed.
Given the information in data quality assessment, we can in remove bad data and may implement this feature when
we believe the scoring is trustworthy.
Once we have computed the for each
of the evaluations in an energy region, we can loop through them and make a guess as to which evaluation is the "best" one.
Only evaluations that have a <
1010 are considered (This sounds like a lot, but one big dataset in the resonance region with small uncertainties
and bad normalization can do it!).
