newherapdf_nov26

The New HERAPDF
Nov26 2008 HERA SFgroup
AM Cooper-Sarkar
Appears compatible with HERAPDF0.1 when doing fits at Q20=4.0 GeV2
But humpy gluon is Chisq favoured and there are two humpy mimima
Fitting with Q20=1.9 GeV2 is tricky
Straight gluon is favoured but shape of d-valence has changed a lot and is softer
than d-bar at high-x
This situation is improved if GMVFN rather than ZMVFN dynamical heavy quark
scheme is used
Moving from QCDNUM16 to QCDNUM17 gives the possibility of greater
accuracy at high-x
A combination of QCDNUM17 quadratic spline interpolation on the high-x grid,
and the use of GMVFN give results compatible with HERAPDF0.1 (and by
implication with CTEQ MRSTW etc)
Fit to new post-Oct20 2008 data set, using formalism of HERAPDF0.1 gives very
compatible results
19 extra points brings total to 592 data points: Chisq 573
100errors added in quadrature- no procedural as yet
Chisq/ndp for NCe+ data increases to 1.10
Let’s see the comparison of new to old in more detail for various flavours
Conclusion: compatible
Now look at alternative parametrizations: first ZEUS-JETS
optimization gives the same form of parametrization as in April 2008
Fit to new post-Oct20 2008 data set,
using ZEUS-JETS parametrization
gives total chisq 574 and very
compatible results:
both to old ZEUS-JETS
and to new fit using HERAPDF0.1
form of parametrization
Inbetween style
Zeus-jets style
H1-style
The PDF uncertainties for the inbetween parametrization (left) are more
conservative than those for the zeus-jets parametrization (middle) and than the
H1 parametrization (right) just as we found earlier.
Let’s see the comparison of new inbetween to new zeus-jets in more detail for
various flavours
Conclusion: broadly compatible, need model dependence to really judge
Now look at alternative parametrizations: humpy version of inbetween
Fit to new post-Oct20 2008 data set,
using humpy gluon parametrization
with ‘inbetween’
Humpy-2 Chisq 559 is now preferred
Humpy-2 is preferred by NC e+ data:
chisq/ndp 1.06 compared to 1.10
It is not preffered by CCe+:
chisq/ndp 0.82 compared to 0.76.
Other data set Chisq do not change
much
Let’s see the comparison of Humpy-2 to straight in more detail for various
flavours
Conclusion: more or less as we saw it when investigating HERAPDF0.1
Now look at alternative parametrizations: another humpy version of inbetween
But I have not been following strict
historical order
First I found another humpy solution with
an identical Chisq=559
Humpy-1 is preferred by NC e+ data:
chisq/ndp 1.05 compared to 1.10
It is not preffered by CCe+:
chisq/ndp 0.90 compared to 0.76.
This change in the CCe+ fit is what is
giving the big change in d-valence
(HERA-II CCe+ data could change this).
It’s not the humpiness it’s the
valence that worries me it’s
the valence shapes
The resulting d-valence is very softsofter than dbar at high-x
Let’s see the comparison of Humpy-1 to straight in more detail for various
flavours
Conclusion: not at all like we saw when investigating HERAPDF0.1
MUCH more model dependent
NOTE: Joel found only Humpy-1, Gang-li found Humpy-2, I can find both if my
start parameters are shifted
Now start looking at jobs with low Q20, so that we can move to dynamic heavy
quark treatment and to the NNLO QCDNUM programme
This is a fit with Q20=1.9 using
the ‘inbetween’ parametrization.
For Q20=1.9 the straight gluon
solution is preferred.
Now we see that the alternative
behaviour of the valence has
nothing to do with Humpyness of
the gluon.
It is hard to get the original
HERAPDF0.1 valence shapes
with low Q20.
Remember
HERAPDF NEW
means fitting with
Q20=4 exactly as for
HERAPDF0.1 AND it
is compatible with
HERAPDF0.1
In fact low Q20 fits are much
less stable
Again d-valence is softer than dbar at high-x
Let’s see the comparison of Q20=1.9 to Q20=4.0 straight gluon solutions for
fits to the new data in more detail for various flavours
Conclusion: very large model dependence- and some amount of
embarassment wrt what we have already said to the world- also if we use this
Q20=1.9 minimum for our central value we’ll be way out of line with
CTEQ/MSTW
Also if you look in detail you will see that dbar > dvalence at large x> 0.7
NOTE the problem is not
just the low start point
We used to get
compatible shapes from
q20=2.0 before the new
data were added
This shows the OLD
Q20=2.0 compared to
HERAPDF NEW with
Q20=4.0
Let’s see the comparison of Q20=2.0 to Q20=4.0 straight gluon solutions for
fits to the older data in more detail for various flavours
Conclusion: only modest model dependence for older data
The new data is now emphasing the preferences of the NCe+ sample such
that the ‘needs’ of the CCe+ sample are downgraded- we have thus relatively
less information on d-valence.
We have find a way to live with this - until we combine HERA-II (or possibly
assign larger weight CCe+ - CTEQ have done this sort of thing in the past).
Now move to a dynamically generated heavy quark variable flavour number scheme
Of course the newer type of solution could
be the correct one but since it has some
nasty features I have tried various ways to
recover the old one. Many unsuccessful
tries with penalty Chisq, setting limits on
parameters-even with fully hessian error
treatment
Some success with moving to Robert
Thorne’s heavy quark variable flavour
number scheme (2008 version more
compatible to ACOT)
Since the point of the move to low Q20
was to implement these schemes this is
relatively good news.
I remain somewhat uneasy about all this
and the d-valence remains somewhat
soft. The work comparing this to similar
jobs with Q20=4.0, alternative
parametrizations etc. has yet to be done.
Now move to NNLO QCDNUM:QCDNUM17-beta-02 BUT USE it at NLO
Early problems with beta-00, beta01 are now solved
NOTE this programme is NOT
BACKWARD COMPATIBLE you
have to completely re-write your
user routine.
Motivation to use it at NLO: Botje
pays more attention to accuracy at
high-x
Linear and quadratic spline
interpolation on the x grid is
available.
GOOD NEWS: linear is compatible
to QCDNUM16– I have recovered
a similar unpleasant soft d-valence
solution (see next page)
BAD NEWS: quadratic is
significantly different – it has
somewhat harder d-valence (but
still not harder than d-bar)
Compatibility of QCDNUM 16/17:
Well compatibility between QCDNUM versions is GOOD, though still getting the soft dvalence doesn’t thrill me
Compare the Q20=1.9 solution with
QCDNUM16– red line
To the Q20=1.9 solution with
QCDNUM17 linear spline
interpolation –black line and
yellow band
Comparison of linear and quadratic spline interpolation in more detail
Incompatibility of quadratic and linear is BAD since it means we’ve
probably never been sufficiently accurate at high-x
Finally move to a dynamically generated heavy quark variable flavour number scheme
But using QCDNUM17 linear and quadratic
Compare the Q20=1.9 solution with
QCDNUM16– red line
To the Q20=1.9 solution with
QCDNUM17 linear spline
interpolation –black line and also
note that the quadratic red-line dvalence is harder
Comparison of linear and quadratic spline interpolation for dynamical
heavy quark variable flavour number in more detail
Conclusions
Everything looks OK (definition: similar to what we saw before) if we stick to
Q20=4.0
(except that we have a double minimum in humpy solutions!)
We could stick to Q20=4 even for dynamic heavy quark generation since it is
POSSIBLE to evolve backwards
However we’ll have to face low Q20 for NNLO eventually. Low Q20 fits are tricky.
For Q20~2.0 we have a ‘new look’ for the d-valence (and a bit for u-valence)
which is not compatible to what we saw before and has dbar> dvalence at high-x
(Because the CCe+ data don’t have so much relative weight)
Moving from ZMVFN- GMVFN partly mitigates this –but much checking back to all
types of jobs must be done and H1/ZEUS fitters must agree on the results for this
type of job and its very slow
Moving from QCDNUM16 to QCDNUM17 quadratic interpolation also mitigates
this further (if combined with GMVFN) – but much checking of all types of jobs
must be done and H1/ZEUS fitters must agree on the results for this type of job
NOT QUICKLY DONE
EXTRAS
Use of RTVFN in HERAPDF0.1 was compatible
with the standard fit
But for HERAPDFNEW it seems to be pushing
us towards an alternative and nicer minimum
than the one found in ZMVFN for Q20=1.9