CMS 3D CMS Logo

MuonSegFit.cc
Go to the documentation of this file.
1 // ------------------------- //
2 // MuonSegFit.cc
3 // Created: 11.05.2015
4 // Based on CSCSegFit.cc
5 // ------------------------- //
6 
9 
12 
13 bool MuonSegFit::fit(void) {
14  if (fitdone())
15  return fitdone_; // don't redo fit unnecessarily
16  short n = nhits();
17  if (n < 2) {
18  edm::LogVerbatim("MuonSegFit") << "[MuonSegFit::fit] - cannot fit just 1 hit!!";
19  } else if (n == 2) {
20  fit2();
21  } else if (2 * n <= MaxHits2) {
22  fitlsq();
23  } else {
24  edm::LogVerbatim("MuonSegFit") << "[MuonSegFit::fit] - cannot fit more than " << MaxHits2 / 2 << " hits!!";
25  }
26  return fitdone_;
27 }
28 
29 void MuonSegFit::fit2(void) {
30  // Just join the two points
31  // Equation of straight line between (x1, y1) and (x2, y2) in xy-plane is
32  // y = mx + c
33  // with m = (y2-y1)/(x2-x1)
34  // and c = (y1*x2-x2*y1)/(x2-x1)
35  //
36  // Now we will make two straight lines
37  // one in xz-plane, another in yz-plane
38  // x = uz + c1
39  // y = vz + c2
40  // 1) Check whether hits are on the same layer
41  // should be done before, so removed
42  // -------------------------------------------
43 
44  MuonRecHitContainer::const_iterator ih = hits_.begin();
45  // 2) Global Positions of hit 1 and 2 and
46  // Local Positions of hit 1 and 2 w.r.t. reference GEM Eta Partition
47  // ---------------------------------------------------------------------
48  LocalPoint h1pos = (*ih)->localPosition();
49  ++ih;
50  LocalPoint h2pos = (*ih)->localPosition();
51 
52  // 3) Now make straight line between the two points in local coords
53  // ----------------------------------------------------------------
54  float dz = h2pos.z() - h1pos.z();
55  if (dz != 0.0) {
56  uslope_ = (h2pos.x() - h1pos.x()) / dz;
57  vslope_ = (h2pos.y() - h1pos.y()) / dz;
58  }
59 
60  float uintercept = (h1pos.x() * h2pos.z() - h2pos.x() * h1pos.z()) / dz;
61  float vintercept = (h1pos.y() * h2pos.z() - h2pos.y() * h1pos.z()) / dz;
62 
63  // calculate local position (variable: intercept_)
64  intercept_ = LocalPoint(uintercept, vintercept, 0.);
65 
66  // calculate the local direction (variable: localdir_)
67  setOutFromIP();
68 
69  //@@ NOT SURE WHAT IS SENSIBLE FOR THESE...
70  chi2_ = 0.;
71  ndof_ = 0;
72 
73  fitdone_ = true;
74 }
75 
76 void MuonSegFit::fitlsq(void) {
77  // Linear least-squares fit to up to 6 GEM rechits, one per layer in a GEM chamber.
78  // Comments adapted from Tim Cox' comments in the original GEMSegAlgoSK algorithm.
79 
80  // Fit to the local x, y rechit coordinates in z projection
81  // The strip measurement controls the precision of x
82  // The wire measurement controls the precision of y.
83  // Typical precision: u (strip, sigma~200um), v (wire, sigma~1cm)
84 
85  // Set up the normal equations for the least-squares fit as a matrix equation
86 
87  // We have a vector of measurements m, which is a 2n x 1 dim matrix
88  // The transpose mT is (u1, v1, u2, v2, ..., un, vn) where
89  // ui is the strip-associated measurement and
90  // vi is the wire-associated measurement
91  // for a given rechit i.
92 
93  // The fit is to
94  // u = u0 + uz * z
95  // v = v0 + vz * z
96  // where u0, uz, v0, vz are the parameters to be obtained from the fit.
97 
98  // These are contained in a vector p which is a 4x1 dim matrix, and
99  // its transpose pT is (u0, v0, uz, vz). Note the ordering!
100 
101  // The covariance matrix for each pair of measurements is 2 x 2 and
102  // the inverse of this is the error matrix E.
103  // The error matrix for the whole set of n measurements is a diagonal
104  // matrix with diagonal elements the individual 2 x 2 error matrices
105  // (because the inverse of a diagonal matrix is a diagonal matrix
106  // with each element the inverse of the original.)
107 
108  // In function 'weightMatrix()', the variable 'matrix' is filled with this
109  // block-diagonal overall covariance matrix. Then 'matrix' is inverted to the
110  // block-diagonal error matrix, and returned.
111 
112  // Define the matrix A as
113  // 1 0 z1 0
114  // 0 1 0 z1
115  // 1 0 z2 0
116  // 0 1 0 z2
117  // .. .. .. ..
118  // 1 0 zn 0
119  // 0 1 0 zn
120 
121  // This matrix A is set up and returned by function 'derivativeMatrix()'.
122 
123  // Then the normal equations are described by the matrix equation
124  //
125  // (AT E A)p = (AT E)m
126  //
127  // where AT is the transpose of A.
128 
129  // Call the combined matrix on the LHS, M, and that on the RHS, B:
130  // M p = B
131 
132  // We solve this for the parameter vector, p.
133  // The elements of M and B then involve sums over the hits
134 
135  // The covariance matrix of the parameters is obtained by
136  // (AT E A)^-1 calculated in 'covarianceMatrix()'.
137 
138  SMatrix4 M; // 4x4, init to 0
139  SVector4 B; // 4x1, init to 0;
140 
141  MuonRecHitContainer::const_iterator ih = hits_.begin();
142 
143  // Loop over the GEMRecHits and make small (2x2) matrices used to fill the blockdiagonal covariance matrix E^-1
144  for (ih = hits_.begin(); ih != hits_.end(); ++ih) {
145  LocalPoint lp = (*ih)->localPosition();
146  // Local position of hit w.r.t. chamber
147  double u = lp.x();
148  double v = lp.y();
149  double z = lp.z();
150 
151  // Covariance matrix of local errors
152  SMatrixSym2 IC; // 2x2, init to 0
153 
154  IC(0, 0) = (*ih)->localPositionError().xx();
155  IC(1, 1) = (*ih)->localPositionError().yy();
156  //@@ NOT SURE WHICH OFF-DIAGONAL ELEMENT MUST BE DEFINED BUT (1,0) WORKS
157  //@@ (and SMatrix enforces symmetry)
158  IC(1, 0) = (*ih)->localPositionError().xy();
159  // IC(0,1) = IC(1,0);
160 
161 #ifdef EDM_ML_DEBUG
162  edm::LogVerbatim("MuonSegFitMatrixDetails")
163  << "[MuonSegFit::fit] 2x2 covariance matrix for this GEMRecHit :: [[" << IC(0, 0) << ", " << IC(0, 1) << "]["
164  << IC(1, 0) << "," << IC(1, 1) << "]]";
165 #endif
166 
167  // Invert covariance matrix (and trap if it fails!)
168  bool ok = IC.Invert();
169  if (!ok) {
170  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::fit] Failed to invert covariance matrix: \n" << IC;
171  // return ok; //@@ SHOULD PASS THIS BACK TO CALLER?
172  }
173 
174  M(0, 0) += IC(0, 0);
175  M(0, 1) += IC(0, 1);
176  M(0, 2) += IC(0, 0) * z;
177  M(0, 3) += IC(0, 1) * z;
178  B(0) += u * IC(0, 0) + v * IC(0, 1);
179 
180  M(1, 0) += IC(1, 0);
181  M(1, 1) += IC(1, 1);
182  M(1, 2) += IC(1, 0) * z;
183  M(1, 3) += IC(1, 1) * z;
184  B(1) += u * IC(1, 0) + v * IC(1, 1);
185 
186  M(2, 0) += IC(0, 0) * z;
187  M(2, 1) += IC(0, 1) * z;
188  M(2, 2) += IC(0, 0) * z * z;
189  M(2, 3) += IC(0, 1) * z * z;
190  B(2) += (u * IC(0, 0) + v * IC(0, 1)) * z;
191 
192  M(3, 0) += IC(1, 0) * z;
193  M(3, 1) += IC(1, 1) * z;
194  M(3, 2) += IC(1, 0) * z * z;
195  M(3, 3) += IC(1, 1) * z * z;
196  B(3) += (u * IC(1, 0) + v * IC(1, 1)) * z;
197  }
198 
199  SVector4 p;
200  bool ok = M.Invert();
201  if (!ok) {
202  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::fit] Failed to invert matrix: \n" << M;
203  // return ok; //@@ SHOULD PASS THIS BACK TO CALLER?
204  } else {
205  p = M * B;
206  }
207 
208 #ifdef EDM_ML_DEBUG
209  LogTrace("MuonSegFitMatrixDetails") << "[MuonSegFit::fit] p = " << p(0) << ", " << p(1) << ", " << p(2) << ", "
210  << p(3);
211 #endif
212 
213  // fill member variables (note origin has local z = 0)
214  // intercept_
215  intercept_ = LocalPoint(p(0), p(1), 0.);
216 
217  // localdir_ - set so segment points outwards from IP
218  uslope_ = p(2);
219  vslope_ = p(3);
220  setOutFromIP();
221 
222  // calculate chi2 of fit
223  setChi2();
224 
225  // flag fit has been done
226  fitdone_ = true;
227 }
228 
230  double chsq = 0.;
231 
232  MuonRecHitContainer::const_iterator ih;
233 
234  for (ih = hits_.begin(); ih != hits_.end(); ++ih) {
235  LocalPoint lp = (*ih)->localPosition();
236 
237  double u = lp.x();
238  double v = lp.y();
239  double z = lp.z();
240 
241  double du = intercept_.x() + uslope_ * z - u;
242  double dv = intercept_.y() + vslope_ * z - v;
243 
244  SMatrixSym2 IC; // 2x2, init to 0
245 
246  IC(0, 0) = (*ih)->localPositionError().xx();
247  // IC(0,1) = (*ih)->localPositionError().xy();
248  IC(1, 0) = (*ih)->localPositionError().xy();
249  IC(1, 1) = (*ih)->localPositionError().yy();
250  // IC(1,0) = IC(0,1);
251 
252 #ifdef EDM_ML_DEBUG
253  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::setChi2] IC before = \n" << IC;
254 #endif
255 
256  // Invert covariance matrix
257  bool ok = IC.Invert();
258  if (!ok) {
259  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::setChi2] Failed to invert covariance matrix: \n"
260  << IC;
261  // return ok;
262  }
263  chsq += du * du * IC(0, 0) + 2. * du * dv * IC(0, 1) + dv * dv * IC(1, 1);
264 
265 #ifdef EDM_ML_DEBUG
266  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::setChi2] IC after = \n" << IC;
267  edm::LogVerbatim("MuonSegFit")
268  << "[GEM RecHit ] Contribution to Chi^2 of this hit :: du^2*Cov(0,0) + 2*du*dv*Cov(0,1) + dv^2*IC(1,1) = "
269  << du * du << "*" << IC(0, 0) << " + 2.*" << du << "*" << dv << "*" << IC(0, 1) << " + " << dv * dv << "*"
270  << IC(1, 1) << " = " << chsq;
271 #endif
272  }
273 
274  // fill member variables
275  chi2_ = chsq;
276  ndof_ = 2. * hits_.size() - 4;
277 
278 #ifdef EDM_ML_DEBUG
279  edm::LogVerbatim("MuonSegFit") << "[MuonSegFit::setChi2] chi2/ndof = " << chi2_ << "/" << ndof_;
280  edm::LogVerbatim("MuonSegFit") << "-----------------------------------------------------";
281 
282  // check fit quality ... maybe write a separate function later on
283  // that is there only for debugging issues
284 
285  edm::LogVerbatim("MuonSegFit") << "[GEM Segment with Local Direction = " << localdir_ << " and Local Position "
286  << intercept_ << "] can be written as:";
287  edm::LogVerbatim("MuonSegFit") << "[ x ] = " << localdir_.x() << " * t + " << intercept_.x();
288  edm::LogVerbatim("MuonSegFit") << "[ y ] = " << localdir_.y() << " * t + " << intercept_.y();
289  edm::LogVerbatim("MuonSegFit") << "[ z ] = " << localdir_.z() << " * t + " << intercept_.z();
290  edm::LogVerbatim("MuonSegFit")
291  << "Now extrapolate to each of the GEMRecHits XY plane (so constant z = RH LP.z()) to obtain [x1,y1]";
292 #endif
293 }
294 
296  bool ok = true;
297 
298  SMatrixSym12 matrix = ROOT::Math::SMatrixIdentity(); // 12x12, init to 1's on diag
299 
300  int row = 0;
301 
302  for (MuonRecHitContainer::const_iterator it = hits_.begin(); it != hits_.end(); ++it) {
303  // Note scaleXError allows rescaling the x error if necessary
304  matrix(row, row) = scaleXError() * (*it)->localPositionError().xx();
305  matrix(row, row + 1) = (*it)->localPositionError().xy();
306  ++row;
307  matrix(row, row - 1) = (*it)->localPositionError().xy();
308  matrix(row, row) = (*it)->localPositionError().yy();
309  ++row;
310  }
311 
312  ok = matrix.Invert(); // invert in place
313  if (!ok) {
314  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::weightMatrix] Failed to invert matrix: \n" << matrix;
315  // return ok; //@@ SHOULD PASS THIS BACK TO CALLER?
316  }
317  return matrix;
318 }
319 
321  SMatrix12by4 matrix; // 12x4, init to 0
322  int row = 0;
323 
324  for (MuonRecHitContainer::const_iterator it = hits_.begin(); it != hits_.end(); ++it) {
325  LocalPoint lp = (*it)->localPosition();
326 
327  float z = lp.z();
328 
329  matrix(row, 0) = 1.;
330  matrix(row, 2) = z;
331  ++row;
332  matrix(row, 1) = 1.;
333  matrix(row, 3) = z;
334  ++row;
335  }
336  return matrix;
337 }
338 
340  // Set direction of segment to point from IP outwards
341  // (Incorrect for particles not coming from IP, of course.)
342 
343  double dxdz = uslope_;
344  double dydz = vslope_;
345  double dz = 1. / sqrt(1. + dxdz * dxdz + dydz * dydz);
346  double dx = dz * dxdz;
347  double dy = dz * dydz;
348  LocalVector localDir(dx, dy, dz);
349 
350  localdir_ = (localDir).unit();
351 #ifdef EDM_ML_DEBUG
352  edm::LogVerbatim("MuonSegFit") << "[MuonSegFit::setOutFromIP] :: dxdz = uslope_ = " << std::setw(9) << uslope_
353  << " dydz = vslope_ = " << std::setw(9) << vslope_ << " local dir = " << localDir;
354  edm::LogVerbatim("MuonSegFit") << "[MuonSegFit::setOutFromIP] :: ==> local dir = " << localdir_
355  << " localdir.phi = " << localdir_.phi();
356 #endif
357 }
358 
362 #ifdef EDM_ML_DEBUG
363  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::covarianceMatrix] weights matrix W: \n" << weights;
364  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::covarianceMatrix] derivatives matrix A: \n" << A;
365 #endif
366 
367  // (AT W A)^-1
368  // e.g. See http://www.phys.ufl.edu/~avery/fitting.html, part I
369 
370  bool ok;
371  SMatrixSym4 result = ROOT::Math::SimilarityT(A, weights);
372 #ifdef EDM_ML_DEBUG
373  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::covarianceMatrix] (AT W A): \n" << result;
374 #endif
375 
376  ok = result.Invert(); // inverts in place
377  if (!ok) {
378  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::calculateError] Failed to invert matrix: \n" << result;
379  // return ok; //@@ SHOULD PASS THIS BACK TO CALLER?
380  }
381 #ifdef EDM_ML_DEBUG
382  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::covarianceMatrix] (AT W A)^-1: \n" << result;
383 #endif
384 
385  // reorder components to match TrackingRecHit interface (GEMSegment isa TrackingRecHit)
386  // i.e. slopes first, then positions
388 
389  return flipped;
390 }
391 
393  // The GEMSegment needs the error matrix re-arranged to match
394  // parameters in order (uz, vz, u0, v0)
395  // where uz, vz = slopes, u0, v0 = intercepts
396 
397 #ifdef EDM_ML_DEBUG
398  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::flipErrors] input: \n" << a;
399 #endif
400 
401  AlgebraicSymMatrix hold(4, 0.);
402 
403  for (short j = 0; j != 4; ++j) {
404  for (short i = 0; i != 4; ++i) {
405  hold(i + 1, j + 1) = a(i, j); // SMatrix counts from 0, AlgebraicMatrix from 1
406  }
407  }
408 
409 #ifdef EDM_ML_DEBUG
410  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::flipErrors] after copy:";
411  edm::LogVerbatim("MuonSegFitMatrixDetails")
412  << "(" << hold(1, 1) << " " << hold(1, 2) << " " << hold(1, 3) << " " << hold(1, 4);
413  edm::LogVerbatim("MuonSegFitMatrixDetails")
414  << " " << hold(2, 1) << " " << hold(2, 2) << " " << hold(2, 3) << " " << hold(2, 4);
415  edm::LogVerbatim("MuonSegFitMatrixDetails")
416  << " " << hold(3, 1) << " " << hold(3, 2) << " " << hold(3, 3) << " " << hold(3, 4);
417  edm::LogVerbatim("MuonSegFitMatrixDetails")
418  << " " << hold(4, 1) << " " << hold(4, 2) << " " << hold(4, 3) << " " << hold(4, 4) << ")";
419 #endif
420 
421  // errors on slopes into upper left
422  hold(1, 1) = a(2, 2);
423  hold(1, 2) = a(2, 3);
424  hold(2, 1) = a(3, 2);
425  hold(2, 2) = a(3, 3);
426 
427  // errors on positions into lower right
428  hold(3, 3) = a(0, 0);
429  hold(3, 4) = a(0, 1);
430  hold(4, 3) = a(1, 0);
431  hold(4, 4) = a(1, 1);
432 
433  // must also interchange off-diagonal elements of off-diagonal 2x2 submatrices
434  hold(4, 1) = a(1, 2);
435  hold(3, 2) = a(0, 3);
436  hold(2, 3) = a(3, 0); // = a(0,3)
437  hold(1, 4) = a(2, 1); // = a(1,2)
438 
439 #ifdef EDM_ML_DEBUG
440  edm::LogVerbatim("MuonSegFitMatrixDetails") << "[MuonSegFit::flipErrors] after flip:";
441  edm::LogVerbatim("MuonSegFitMatrixDetails")
442  << "(" << hold(1, 1) << " " << hold(1, 2) << " " << hold(1, 3) << " " << hold(1, 4);
443  edm::LogVerbatim("MuonSegFitMatrixDetails")
444  << " " << hold(2, 1) << " " << hold(2, 2) << " " << hold(2, 3) << " " << hold(2, 4);
445  edm::LogVerbatim("MuonSegFitMatrixDetails")
446  << " " << hold(3, 1) << " " << hold(3, 2) << " " << hold(3, 3) << " " << hold(3, 4);
447  edm::LogVerbatim("MuonSegFitMatrixDetails")
448  << " " << hold(4, 1) << " " << hold(4, 2) << " " << hold(4, 3) << " " << hold(4, 4) << ")";
449 #endif
450 
451  return hold;
452 }
453 
454 float MuonSegFit::xfit(float z) const {
455  //@@ ADD THIS TO EACH ACCESSOR OF FIT RESULTS?
456  // if ( !fitdone() ) fit();
457  return intercept_.x() + uslope_ * z;
458 }
459 
460 float MuonSegFit::yfit(float z) const { return intercept_.y() + vslope_ * z; }
461 
462 float MuonSegFit::xdev(float x, float z) const { return intercept_.x() + uslope_ * z - x; }
463 
464 float MuonSegFit::ydev(float y, float z) const { return intercept_.y() + vslope_ * z - y; }
465 
466 float MuonSegFit::Rdev(float x, float y, float z) const {
467  return sqrt(xdev(x, z) * xdev(x, z) + ydev(y, z) * ydev(y, z));
468 }
Vector3DBase< float, LocalTag >
DDAxes::y
mps_fire.i
i
Definition: mps_fire.py:428
MuonSegFit::setChi2
void setChi2(void)
Definition: MuonSegFit.cc:229
MuonSegFit::hits_
MuonRecHitContainer hits_
Definition: MuonSegFit.h:114
MessageLogger.h
dqmiodumpmetadata.n
n
Definition: dqmiodumpmetadata.py:28
MuonSegFit::ndof_
int ndof_
Definition: MuonSegFit.h:120
makeMuonMisalignmentScenario.matrix
list matrix
Definition: makeMuonMisalignmentScenario.py:141
detailsBasic3DVector::z
float float float z
Definition: extBasic3DVector.h:14
PV3DBase::x
T x() const
Definition: PV3DBase.h:59
MuonSegFit::Rdev
float Rdev(float x, float y, float z) const
Definition: MuonSegFit.cc:466
MuonSegFit::ydev
float ydev(float y, float z) const
Definition: MuonSegFit.cc:464
MuonSegFit::fitdone_
bool fitdone_
Definition: MuonSegFit.h:122
MuonSegFit::covarianceMatrix
AlgebraicSymMatrix covarianceMatrix(void)
Definition: MuonSegFit.cc:359
DDAxes::x
align::LocalPoint
Point3DBase< Scalar, LocalTag > LocalPoint
Definition: Definitions.h:30
findQualityFiles.v
v
Definition: findQualityFiles.py:179
convertSQLiteXML.ok
bool ok
Definition: convertSQLiteXML.py:98
MuonSegFit::fit2
void fit2(void)
Definition: MuonSegFit.cc:29
PV3DBase::z
T z() const
Definition: PV3DBase.h:61
MuonSegFit::uslope_
float uslope_
Definition: MuonSegFit.h:115
MuonSegFit::SMatrixSym12
ROOT::Math::SMatrix< double, MaxHits2, MaxHits2, ROOT::Math::MatRepSym< double, MaxHits2 > > SMatrixSym12
Definition: MuonSegFit.h:43
HLT_FULL_cff.weights
weights
Definition: HLT_FULL_cff.py:99207
mathSSE::sqrt
T sqrt(T t)
Definition: SSEVec.h:19
DDAxes::z
BeamSpotPI::dxdz
Definition: BeamSpotPayloadInspectorHelper.h:37
MuonSegFit::MaxHits2
static const int MaxHits2
Definition: MuonSegFit.h:41
MuonSegFit::fitdone
bool fitdone() const
Definition: MuonSegFit.h:93
Point3DBase< float, LocalTag >
MuonSegFit::fitlsq
void fitlsq(void)
Definition: MuonSegFit.cc:76
MuonSegFit::SMatrix12by4
ROOT::Math::SMatrix< double, MaxHits2, 4 > SMatrix12by4
Definition: MuonSegFit.h:46
MuonSegFit::nhits
size_t nhits(void) const
Definition: MuonSegFit.h:88
MuonSegFit::localdir_
LocalVector localdir_
Definition: MuonSegFit.h:118
MuonSegFit::flipErrors
AlgebraicSymMatrix flipErrors(const SMatrixSym4 &)
Definition: MuonSegFit.cc:392
a
double a
Definition: hdecay.h:119
AlCaHLTBitMon_ParallelJobs.p
def p
Definition: AlCaHLTBitMon_ParallelJobs.py:153
MuonSegFit::xfit
float xfit(float z) const
Definition: MuonSegFit.cc:454
A
BeamSpotPI::dydz
Definition: BeamSpotPayloadInspectorHelper.h:38
PV3DBase::y
T y() const
Definition: PV3DBase.h:60
MuonSegFit::intercept_
LocalPoint intercept_
Definition: MuonSegFit.h:117
MuonSegFit::SMatrix4
ROOT::Math::SMatrix< double, 4 > SMatrix4
Definition: MuonSegFit.h:49
MuonSegFit::yfit
float yfit(float z) const
Definition: MuonSegFit.cc:460
AlgebraicSymMatrix
CLHEP::HepSymMatrix AlgebraicSymMatrix
Definition: AlgebraicObjects.h:15
MuonSegFit::derivativeMatrix
SMatrix12by4 derivativeMatrix(void)
Definition: MuonSegFit.cc:320
PVValHelper::dy
Definition: PVValidationHelpers.h:50
MuonSegFit::weightMatrix
SMatrixSym12 weightMatrix(void)
Definition: MuonSegFit.cc:295
MuonSegFit::vslope_
float vslope_
Definition: MuonSegFit.h:116
MuonSegFit::scaleXError
double scaleXError(void) const
Definition: MuonSegFit.h:87
unit
Basic3DVector unit() const
Definition: Basic3DVectorLD.h:162
TtFullHadDaughter::B
static const std::string B
Definition: TtFullHadronicEvent.h:9
MaterialEffects_cfi.A
A
Definition: MaterialEffects_cfi.py:11
MuonSegFit::SVector4
ROOT::Math::SVector< double, 4 > SVector4
Definition: MuonSegFit.h:56
PVValHelper::dz
Definition: PVValidationHelpers.h:51
edm::LogVerbatim
Log< level::Info, true > LogVerbatim
Definition: MessageLogger.h:128
genVertex_cff.x
x
Definition: genVertex_cff.py:13
detailsBasic3DVector::y
float float y
Definition: extBasic3DVector.h:14
MuonSegFit::fit
bool fit(void)
Definition: MuonSegFit.cc:13
MuonSegFit::setOutFromIP
void setOutFromIP(void)
Definition: MuonSegFit.cc:339
mps_fire.result
result
Definition: mps_fire.py:311
LogTrace
#define LogTrace(id)
Definition: MessageLogger.h:234
MuonSegFit::SMatrixSym4
ROOT::Math::SMatrix< double, 4, 4, ROOT::Math::MatRepSym< double, 4 > > SMatrixSym4
Definition: MuonSegFit.h:50
ParameterSet.h
dqmiolumiharvest.j
j
Definition: dqmiolumiharvest.py:66
MuonSegFit::chi2_
double chi2_
Definition: MuonSegFit.h:119
MuonSegFit::SMatrixSym2
ROOT::Math::SMatrix< double, 2, 2, ROOT::Math::MatRepSym< double, 2 > > SMatrixSym2
Definition: MuonSegFit.h:53
GlobalPoint.h
MuonSegFit.h
PVValHelper::dx
Definition: PVValidationHelpers.h:49
PV3DBase::phi
Geom::Phi< T > phi() const
Definition: PV3DBase.h:66
MuonSegFit::xdev
float xdev(float x, float z) const
Definition: MuonSegFit.cc:462