Belle II Software  release-08-01-10
KalmanCalculator.cc
1 /**************************************************************************
2  * basf2 (Belle II Analysis Software Framework) *
3  * Author: The Belle II Collaboration *
4  * External Contributor: Wouter Hulsbergen *
5  * *
6  * See git log for contributors and copyright holders. *
7  * This file is licensed under LGPL-3.0, see LICENSE.md. *
8  **************************************************************************/
9 
10 #include <analysis/VertexFitting/TreeFitter/KalmanCalculator.h>
11 
12 namespace TreeFitter {
13 
15  int sizeRes,
16  int sizeState
17  ) :
18  m_constrDim(sizeRes),
19  m_chisq(1e10),
20  m_res(sizeRes),
21  m_G(sizeRes, sizeState),
22  m_R(sizeRes, sizeRes),
23  m_Rinverse(sizeRes, sizeRes),
24  m_K(sizeState, sizeRes),
25  m_CGt(sizeState, sizeRes)
26  {
27  m_R = Eigen::Matrix < double, -1, -1, 0, 7, 7 >::Zero(m_constrDim, m_constrDim);
28  }
29 
30 
32  const Eigen::Matrix < double, -1, 1, 0, 7, 1 > & residuals,
33  const Eigen::Matrix < double, -1, -1, 0, 7, MAX_MATRIX_SIZE > & G,
34  const FitParams& fitparams,
35  const Eigen::Matrix < double, -1, -1, 0, 7, 7 > * V,
36  double weight)
37  {
38  m_res = residuals;
39  m_G = G;
40 
41  Eigen::Matrix < double, -1, -1, 0, MAX_MATRIX_SIZE, MAX_MATRIX_SIZE > C = fitparams.getCovariance().triangularView<Eigen::Lower>();
42 
43  m_CGt = C.selfadjointView<Eigen::Lower>() * G.transpose();
44  Eigen::Matrix < double, -1, -1, 0, 7, 7 > Rtemp = G * m_CGt;
45  if (V && (weight) && ((*V).diagonal().array() != 0).all()) {
46 
47  const Eigen::Matrix < double, -1, -1, 0, 7, 7 > weightedV =
48  weight * (*V).selfadjointView<Eigen::Lower>();
49 
50  m_R = Rtemp + weightedV;
51 
52  } else {
53  m_R = Rtemp.triangularView<Eigen::Lower>();
54  }
55 
56  Eigen::Matrix < double, -1, -1, 0, 7, 7 > RInvtemp;
57  RInvtemp = m_R.selfadjointView<Eigen::Lower>();
58  m_Rinverse = RInvtemp.inverse();
59  if (!m_Rinverse.allFinite()) { return ErrCode(ErrCode::Status::inversionerror); }
60 
61  m_K = m_CGt * m_Rinverse.selfadjointView<Eigen::Lower>();
62  return ErrCode(ErrCode::Status::success);
63  }
64 
66  {
67  double eps = Eigen::NumTraits<double>::epsilon();
68  fitparams.getStateVector() = (fitparams.getStateVector().array() - (m_K * m_res).array()).matrix()
69  + eps * Eigen::MatrixXd::Identity(fitparams.getStateVector().rows(), fitparams.getStateVector().cols());
70  m_chisq = m_res.transpose() * m_Rinverse.selfadjointView<Eigen::Lower>() * m_res;
71  }
72 
74  {
75  double eps = Eigen::NumTraits<double>::epsilon();
76  Eigen::Matrix < double, -1, 1, 0, 7, 1 > res_prime =
77  m_res + m_G * ((oldState.getStateVector().array() - fitparams.getStateVector().array()).matrix() +
78  + eps * Eigen::MatrixXd::Identity(fitparams.getStateVector().rows(), fitparams.getStateVector().cols()));
79  fitparams.getStateVector() = (oldState.getStateVector().array() - (m_K * res_prime).array()).matrix()
80  + eps * Eigen::MatrixXd::Identity(oldState.getStateVector().rows(), oldState.getStateVector().cols());
81 
82  m_chisq = res_prime.transpose() * m_Rinverse.selfadjointView<Eigen::Lower>() * res_prime;
83  }
84 
85  TREEFITTER_NO_STACK_WARNING
86 
88  {
89  Eigen::Matrix < double, -1, -1, 0, MAX_MATRIX_SIZE, MAX_MATRIX_SIZE > fitCov =
90  fitparams.getCovariance().triangularView<Eigen::Lower>();
91 
92  Eigen::Matrix < double, -1, -1, 0, MAX_MATRIX_SIZE, MAX_MATRIX_SIZE > GRinvGt =
93  m_G.transpose() * m_Rinverse.selfadjointView<Eigen::Lower>() * m_G;
94 
95  //fitcov is sym so no transpose needed (not that it would have runtime cost)
96  Eigen::Matrix < double, -1, -1, 0, MAX_MATRIX_SIZE, MAX_MATRIX_SIZE > deltaCov =
97  fitCov.selfadjointView<Eigen::Lower>() * GRinvGt * fitCov.selfadjointView<Eigen::Lower>();
98 
99  double eps = Eigen::NumTraits<double>::epsilon();
100  Eigen::Matrix < double, -1, -1, 0, MAX_MATRIX_SIZE, MAX_MATRIX_SIZE > delta =
101  (fitCov.array() - deltaCov.array()).matrix() + eps * Eigen::MatrixXd::Identity(fitCov.rows(), fitCov.cols());
102 
103  fitparams.getCovariance().triangularView<Eigen::Lower>() = delta.triangularView<Eigen::Lower>();
104 
105  }//end function
106 
107  TREEFITTER_RESTORE_WARNINGS
108 
109 }// end namespace
abstract errorocode be aware that the default is success
Definition: ErrCode.h:14
Class to store and manage fitparams (statevector)
Definition: FitParams.h:20
Eigen::Matrix< double, -1, -1, 0, MAX_MATRIX_SIZE, MAX_MATRIX_SIZE > & getCovariance()
getter for the states covariance
Definition: FitParams.h:53
Eigen::Matrix< double, -1, 1, 0, MAX_MATRIX_SIZE, 1 > & getStateVector()
getter for the fit parameters/statevector
Definition: FitParams.h:65
void updateState(FitParams &fitparams)
update statevector
ErrCode calculateGainMatrix(const Eigen::Matrix< double, -1, 1, 0, 7, 1 > &residuals, const Eigen::Matrix< double, -1, -1, 0, 7, MAX_MATRIX_SIZE > &G, const FitParams &fitparams, const Eigen::Matrix< double, -1, -1, 0, 7, 7 > *V=0, double weight=1)
init the kalman machienery
Eigen::Matrix< double, -1, -1, 0, 7, 7 > m_Rinverse
R inverse.
KalmanCalculator(int sizeRes, int sizeState)
constructor
Eigen::Matrix< double, -1, -1, 0, 7, 7 > m_R
R residual covariance.
Eigen::Matrix< double, -1, -1, 0, 7, MAX_MATRIX_SIZE > m_G
G former H, transforms covraince of {residuals}<->{x,p,E}.
Eigen::Matrix< double, -1, 1, 0, 7, 1 > m_res
we know the max sizes of the matrices we assume the tree is smaller than MAX_MATRIX_SIZE parameters a...
void updateCovariance(FitParams &fitparams)
update the statevectors covariance
Eigen::Matrix< double, -1, -1, 0, MAX_MATRIX_SIZE, 7 > m_K
K kalman gain matrix.
int m_constrDim
dimension of the constraint
Eigen::Matrix< double, -1, -1, 0, MAX_MATRIX_SIZE, 7 > m_CGt
C times G^t