Ok, got it. 

Well I think the single hit make sense.

Since you are doing low level studies, why don’t you shoot the particle straight into the sensor to start with and have all else being ideal I..e B-field aligned with sensor. That should remove the particle angle out of this to avoid confusion.

I probably missed this, but what is the problem with our “track resolution” (resolution of what?)? 

I haven’t seen pulls using the covariance matrix from GBL either before or after vertexing (i.e pulls of the vertex distributions) in the ideal MC simulation. Are they not right and would we expect these small effects to be the main cause in that case?


On Nov 10, 2015, at 10:47 AM, Graf, Norman A. <[log in to unmask]> wrote:

Hello Pelle,
 These are not residuals from track fits. These are even more basic than that. These are comparisons between where the MC particle crossed the sensor (SimTrackerHit position) and where the sensor response simulation placed the hit (either one or two-strip cluster u position).
 I started looking into this to better understand our track resolution, and more importantly, the track covariance matrix, both of which are necessary for the vertex fits. In the final track fits many things contribute to the covariance: the intrinsic hit measurement uncertainty, the material effects, and the field propagation. I started with what I thought would be the easiest to understand. When I didn't understand the distributions I thought I'd throw it out there to see if anyone else had ever looked at this basic quantity.
 Tim looked at these years ago, but has not tracked down those low-level validation studies. If no one else has done these studies I'll continue my investigations. 

From: Hansson Adrian, Per Ola
Sent: Tuesday, November 10, 2015 9:34 AM
To: Graf, Norman A.
Cc: madphox-tracking; hps-software; [log in to unmask]
Subject: Re: SVT MC hit resolution
Hi Norman,

just to understand a little better. There is no material at all here? This is SeedTracker fits?

The green looks narrower than a simple box with our readout pitch but we also have intermediate strips that couple some of the charge to the nearby readout strip. I think if you select single strip hit clusters you have that intermediate strip that “biases”  you to be closer to the correct position because that strips tries to pull other strips above threshold. This should lead to better resolution, I think?

For the magenta distribution, I’m not sure I understand exactly what you meant with the simtrackerhit position and strip hit position?  Generally, if the track has a tanLambda of, say, 0.02 then you can see effects of roughly  150um*0.02=3um in the vertical direction for axial by propagating to the wrong place on the sensor. That track angle effect could skew things. 

Also, it would be really interesting to see how the fit chi2 (per df and chi2 probability) look and also the residuals across all the planes look like? Are the pulls ok in this idealistic case? If the fit is not ideal I guess you would expect to have some artifacts on the different hit residuals as well.


On Nov 10, 2015, at 7:46 AM, Graf, Norman A. <[log in to unmask]> wrote:

​Hello All,
 I'm trying to better understand our tracking resolution and have decided to go back to the basics. I'm starting with our simulated hit resolution and right away have found something which puzzles me. I generated single, full energy electrons and reconstructed them with the simple digitization. I'm plotting what I think is the u position of the MC SimTrackerHit compared to the u measurement resulting from our simulation of the response of the sensor to the energy deposition of that SimTrackerHit. I've attached a plot of the difference for single-strip (green) and two-strip (magenta) clusters. 


The green distribution is roughly what I would have expected: a box distribution for the single-hit measurements. But the width is narrower than what I would have naively expected. This seems to indicate that our charge spreading is larger than I would have naively expected.
The magenta distribution is narrower, which is what I would have expected, but other than that I don't understand it. 
I should note that I did not ensure that the SimTrackerHit was propagated back to the plane at which the strip hits are defined. This could have some effect in skewing these distributions, but I would expect that to be less for the axial planes (which is what is plotted here for layer 1).
Has anyone looked at our MC hit distributions at this level?


Use REPLY-ALL to reply to list

To unsubscribe from the HPS-SOFTWARE list, click the following link: