Offline Data Analysis Software

1. Offline Data Analysis Software • Objectives + requirements • Field model + simulation • Pre-fit corrections • Fits + results • Post-fit corrections • Conclusions Steve Snow, Paul S Miyagawa (University of Manchester) John Hart (Rutherford Appleton Laboratory)

2. Objectives • A useful test of the Standard Model would be measurement of W mass with uncertainty of 25 MeV per lepton type per experiment. • W mass derived from the position of the falling edge of the transverse mass distribution. • Momentum scale will be dominant uncertainty in W mass measurement: • Need to keep uncertainty in momentum down to ~15 MeV. • Measure isolated muon tracks with pT ~ 40 GeV over large range of η: • Uncertainty in energy loss negligible. • Concentrate on alignment and B-field. • Momentum accuracy depends on ∫r(rmax - r)Bzdr : • Field at intermediate radii, as measured by the sagitta, is most important. • Typical sagitta will be ~1 mm and target accuracy would be 0.02%, implying a systematic error on sagitta of 0.02 μm: • Technically impossible! • Would need to use Z mass, which is nearby and very well known. • In reality, the limit on silicon alignment, even with infinite statistics and ideal algorithms, will be ~1 μm. • We target an accuracy of 0.05% on sagitta to ensure that B-field measurement is not the limiting factor on momentum accuracy. 2nd ATLAS Magnetic Field Workshop

3. Mapper Survey Requirements • At first workshop, survey requirements of mapper machine relative to Inner Detector were reported as ~1 mm and ~0.2 mrad. • Reinvestigated these requirements. • 1 mm survey error in x (or y) significant for high η tracks. • 1 mm survey error in z significant for endcap tracks. • 0.1 mrad rotation around x (or y) axis significant for high η tracks. • Conclusions remain unchanged. 2nd ATLAS Magnetic Field Workshop

4. Field Model • Basis of model is field due to coil of nominal dimensions. • Added in field due to magnetised iron (4% of total field). Distribution of iron magnetisation taken from simple FEA model using FlexPDE. • Model is symmetric in  and even in z. • Field can be displaced and rotated relative to mapping machine. 2nd ATLAS Magnetic Field Workshop

5. Field Mapping Simulation • Field model sampled on a grid of 90 z-positions × 8 -angles (defined by encoder values of machine). • 4 calibration points (2 near centre, 2 near one end) visited after every 25 measurements. • Used current design of machine to determine positions of 48 Hall + 1 NMR probes. • At each map point, wrote the following data to file: • Time stamp • Solenoid current • z and  encoder values • Field modulus measured by 5 NMR probes (1 moving + 4 fixed) • 3 field components measured by 48 Hall probes 2nd ATLAS Magnetic Field Workshop

6. Simulated Errors Six cumulative levels of error added to simulated data: • No errors. • Random errors in each measurement of: • Solenoid current, 1 A • NMR B-field modulus, 0.1 G • Each component of Hall probe B-field, 3 G • Drifts by random walk process in each measurement of: • Solenoid current, 1 A • Each component of Hall probe B-field, 0.1 G • Random calibration scale and alignment errors, which are constant for each run: • Each component of Hall probe B-field, 0.05% scale • Rotation about 3 axes of each Hall probe triplet, 0.1 mrad • Symmetry axis of field model displaced and rotated relative to the mapper machine axis. These misalignments are assumed to be measured perfectly by the surveyors. • Each Hall probe triplet has a systematic rotation of 1 mrad about the axis which mixes the Br and Bz components. 2nd ATLAS Magnetic Field Workshop

7. Correction for Current Drift Average B-field of 4 NMR probes used to calculate “actual” solenoid current. Scale all measurements to a reference current (7600 A). Effect of drift in current removed Calibration capable of coping with any sort of drift. 2nd ATLAS Magnetic Field Workshop

8. Correction for Hall Probe Drift • Mapping machine regularly returns to fixed calibration positions • Near coil centre to calibrate Bz • Near coil end for Br • No special calibration point for B • Each channel is calibrated to a reference time (beginning of run) • Offsets from calibration points used to determine offsets for measurements between calibrations 2nd ATLAS Magnetic Field Workshop

9. Geometrical Fit • Sum of simple fields known to obey Maxwell’s equations: • Long-thin coil (5 mm longer, 5 mm thinner than nominal) • Short-fat coil (5 mm shorter, 5 mm fatter) • Four terms of Fourier-Bessel series (for magnetisation) • Use Minuit to minimise a 2 fit between the fitting function and simulated data using 12 free parameters: • 1, mixing ratio of long-thin and short-fat solenoids • 2, scaling factors for length and field of the combined solenoids • 4, coefficients of Fourier-Bessel series • 3, offset of field coordinates from mapper coordinates • 2, angles between field coordinates and mapper coordinates • Fit dominated by large number of Hall probe points. • Could introduce one overall scale factor for all Hall probes as free parameter and take true scale from NMR probes 2nd ATLAS Magnetic Field Workshop

10. Fourier-Bessel Fit (1) • General fit able to describe any field obeying Maxwell’s equations. • Uses only the field measurements on the surface of a bounding cylinder, including the ends. • Parameterisation based on ALEPH 94-162 (Stephen Thorn) and proceeds in three stages: • Bz on the cylindrical surface is fitted as Fourier series, giving terms with φ variation of form cos(nφ+α), with radial variation In(κr) (modified Bessel function). • Bzmeas – Bz(1) on the cylinder ends is fitted as a series of Bessel functions, Jn(λjr) where the λj are chosen so the terms vanish for r = rcyl. The z-dependence is of form cosh(μz) or sinh(μz). • The multipole terms are calculated from the measurements of Bron the cylindrical surface, averaged over z, after subtraction of the contribution to Br from the terms above. (The only relevant terms in Bz are those that are odd in z.) 2nd ATLAS Magnetic Field Workshop

11. Fourier-Bessel Fit (2) Comments: • The fit is very quick since the coefficients are calculated directly as sums over the measured field except for a simple linear fit to the Jn coefficients. • A typical fit could have up to ~500 terms in the first category taking into account combinations of 25 terms in z, 5 terms in φ and their respective phases, but most of the φ-dependent terms are too small to be measurable and are set to zero. In the second category, up to ~120 terms could be calculated but again most φ-dependent terms are negligible. • The In and cosh or sinh terms decrease more or less exponentially with increasing distance from the outer cylinder or cylinder ends. This means that higher order terms contribute very little over most of the field volume and have little effect on the measured momentum. • A poor fit indicates measurement errors rather than an incorrect model. 2nd ATLAS Magnetic Field Workshop

12. Fit Quality with Ideal Performance • Quality of fit measured by comparing track sagitta in field model with track sagitta in fitted field: • Used 260 track directions equally spaced in η and . • Results shown below are for no errors added (error level 1): • Sagitta error not quite zero due to interpolation errors and finite number of terms. • Error is negligible compared to target level of 5×10-4. 2nd ATLAS Magnetic Field Workshop

13. Fit Quality with Expected Performance • Expected mapper performance corresponds to error level 5. • Both fits accurate within target level of 5×10-4. 2nd ATLAS Magnetic Field Workshop

14. Error level Geometrical fit Tables of Results • Tables give statistics of Bz error over 25000 points and sagitta error over 260 trajectories. • Correction procedure does not completely eliminate effects of drifts, so some runs (eg, 24) have normalisation errors of ~ 1×10-4. • Geometrical fit relatively uninfluenced until error level 6 (systematic probe tilts). Fourier-Bessel fit 2nd ATLAS Magnetic Field Workshop

15. Probe Normalisation and Alignment Correction Field should be very uniform near z=0, with Bz at maximum and Br=0. We take advantage of this information to perform two calibrations: • Inter-calibration of Bz scale between four Hall probes at common radius: • Find Bz value at z=0 using quadratic fit in range |z| < 0.25 m. • Normalise each probe to average scale of four at that radius. • Removal of z-r rotation: • Find Br value at z=0 using linear fit in range |z| < 0.25 m. • Assuming that any non-zero Br value is due to z-r rotation, correct the angle of the probe. Correlation between the generated probe misalignment and the value reconstructed by this method. CorrC26 data (systematic 1 mrad rotation). Still need to implement fully these corrections in the Fourier-Bessel fit: • Preliminary results indicate performance better than 1 mrad requirement. 2nd ATLAS Magnetic Field Workshop

16. Fourier-Bessel fit * Fit results including probe normalisation and alignment corrections are shown with an asterisk by the file name. Tables of Results Geometrical fit • Tables give statistics of Bz error over 25000 points and sagitta error over 260 trajectories. • Correction procedure does not completely eliminate effects of drifts, so some runs (eg, 24) have normalisation errors of ~ 1×10-4. • Geometrical fit relatively uninfluenced until error level 6 (systematic probe tilts). 2nd ATLAS Magnetic Field Workshop

17. Conclusions • With no simulated errors, both fits have excellent technical accuracy. • With realistic simulated errors, both fits give results within target of 5×10-4. • Fourier-Bessel fit more sensitive to random measurement errors: • Due to more free parameters. • F-B fit designed for any solenoid-like field, whereas geometrical fit is specifically for the ATLAS solenoid. • Probe normalisation + alignment correction helps correct for systematic tilts of Hall probes. • Require survey of mapper machine relative to Inner Detector to be accurate to ~1 mm and ~1 mrad. • With a few refinements, the code will be ready for release. 2nd ATLAS Magnetic Field Workshop