CMS 3D CMS Logo

List of all members | Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | Private Types | Private Member Functions | Private Attributes
HcalPulseShapes Class Reference

#include <HcalPulseShapes.h>

Public Types

typedef HcalPulseShape Shape
 

Public Member Functions

void beginRun (edm::EventSetup const &es)
 
void beginRun (const HcalDbService *conditions)
 
const ShapedefaultShape (const HcalDetId &detId) const
 in case of conditions problems More...
 
std::map< int, Shape const * > const & get_all_shapes () const
 
const ShapegetShape (int shapeType) const
 
const ShapehbShape () const
 
 HcalPulseShapes ()
 
 HcalPulseShapes (edm::ConsumesCollector iC)
 
const ShapeheShape () const
 
const ShapehfShape () const
 
const ShapehoShape (bool sipm=false) const
 
const Shapeshape (const HcalDetId &detId) const
 automatically figures out which shape to return More...
 
const ShapeshapeForReco (const HcalDetId &detId) const
 
 ~HcalPulseShapes ()
 

Static Public Member Functions

static double analyticPulseShapeSiPMHE (double t)
 
static double analyticPulseShapeSiPMHO (double t)
 
template<class F1 , class F2 >
static std::vector< double > convolve (unsigned nbin, F1 f1, F2 f2)
 
static double generatePhotonTime (CLHEP::HepRandomEngine *engine, unsigned int signalShape)
 
static double generatePhotonTime203 (CLHEP::HepRandomEngine *engine)
 
static double generatePhotonTime206 (CLHEP::HepRandomEngine *engine)
 
static std::vector< double > normalize (std::vector< double > nt, unsigned nbin)
 
static std::vector< double > normalizeShift (std::vector< double > nt, unsigned nbin, int shift)
 
static double Y11203 (double t)
 
static double Y11206 (double t)
 

Static Public Attributes

static constexpr float deltaTSiPM_ = 0.5
 
static constexpr float invDeltaTSiPM_ = 2.0
 
static const int nBinsSiPM_ = 250
 
static constexpr float Y11MAX203_ = 0.04
 
static constexpr float Y11MAX206_ = 0.08
 
static constexpr float Y11RANGE_ = nBinsSiPM_
 

Private Types

typedef std::map< int, const Shape * > ShapeMap
 

Private Member Functions

void computeHFShape ()
 
void computeHPDShape (float, float, float, float, float, float, float, float, Shape &)
 
void computeSiPMShapeData2017 ()
 
void computeSiPMShapeData2018 ()
 
const HcalPulseShapecomputeSiPMShapeHE203 ()
 
const HcalPulseShapecomputeSiPMShapeHE206 ()
 
void computeSiPMShapeHO ()
 
void computeSiPMShapeMCRecoRun3 ()
 

Private Attributes

Shape hfShape_
 
Shape hpdBV30Shape_v2
 
Shape hpdBV30ShapeMC_v2
 
Shape hpdShape_
 
Shape hpdShape_v2
 
Shape hpdShape_v3
 
Shape hpdShapeMC_v2
 
Shape hpdShapeMC_v3
 
Shape siPMShapeData2017_
 
Shape siPMShapeData2018_
 
Shape siPMShapeHO_
 
Shape siPMShapeMCRecoRun3_
 
const HcalDbServicetheDbService
 
edm::ESGetToken< HcalDbService, HcalDbRecordtheDbServiceToken
 
ShapeMap theShapes
 

Detailed Description

Author
J. Mans - Minnesota

Definition at line 23 of file HcalPulseShapes.h.

Member Typedef Documentation

◆ Shape

Definition at line 25 of file HcalPulseShapes.h.

◆ ShapeMap

typedef std::map<int, const Shape*> HcalPulseShapes::ShapeMap
private

Definition at line 121 of file HcalPulseShapes.h.

Constructor & Destructor Documentation

◆ HcalPulseShapes() [1/2]

HcalPulseShapes::HcalPulseShapes ( )

Definition at line 14 of file HcalPulseShapes.cc.

References computeHFShape(), computeHPDShape(), computeSiPMShapeData2017(), computeSiPMShapeData2018(), computeSiPMShapeHE203(), computeSiPMShapeHE206(), computeSiPMShapeHO(), computeSiPMShapeMCRecoRun3(), hfShape_, hpdBV30Shape_v2, hpdBV30ShapeMC_v2, hpdShape_, hpdShape_v2, hpdShape_v3, hpdShapeMC_v2, hpdShapeMC_v3, siPMShapeData2017_, siPMShapeData2018_, siPMShapeHO_, siPMShapeMCRecoRun3_, and theShapes.

14  : theDbService(nullptr), theShapes() {
15  /*
16 
17 Reco MC
18 -------------------------------------------------------------------------------------------
19 000 not used (reserved)
20 101 101 hpdShape_ HPD (original version)
21 102 102 =101 HPD BV 30 volts in HBP iphi54
22 103 123 hpdShape_v2,hpdShapeMC_v2 HPD (2011. oct version)
23 104 124 hpdBV30Shape_v2,hpdBV30ShapeMC_v2 HPD bv30 in HBP iph54
24 105 125 hpdShape_v2,hpdShapeMC_v2 HPD (2011.11.12 version)
25 201 201 siPMShapeHO_ SiPMs Zecotec shape (HO)
26 202 202 =201, SiPMs Hamamatsu shape (HO)
27 205 203 siPMShapeData2017_,siPMShapeMC2017_ SiPMs from Data, Hamamatsu shape (HE 2017)
28 207 206 siPMShapeData2018_,siPMShapeMC2018_ SiPMs from Data, Hamamatsu shape (HE 2018)
29 207 208 siPMShapeData2018_,siPMShapeMCRecoRun3_ SiPMs from Data, 2021 MC phase scan
30 301 301 hfShape_ regular HF PMT shape
31 401 401 regular ZDC shape
32 -------------------------------------------------------------------------------------------
33 
34 */
35 
36  float ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3;
37 
38  // HPD Shape Version 1 (used before CMSSW5, until Oct 2011)
39  ts1 = 8.;
40  ts2 = 10.;
41  ts3 = 29.3;
42  thpd = 4.0;
43  tpre = 9.0;
44  wd1 = 2.0;
45  wd2 = 0.7;
46  wd3 = 1.0;
47  computeHPDShape(ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3, hpdShape_);
48  theShapes[101] = &hpdShape_;
49  theShapes[102] = theShapes[101];
50 
51  // HPD Shape Version 2 for CMSSW 5. Nov 2011 (RECO and MC separately)
52  ts1 = 8.;
53  ts2 = 10.;
54  ts3 = 25.0;
55  thpd = 4.0;
56  tpre = 9.0;
57  wd1 = 2.0;
58  wd2 = 0.7;
59  wd3 = 1.0;
60  computeHPDShape(ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3, hpdShape_v2);
61  theShapes[103] = &hpdShape_v2;
62 
63  ts1 = 8.;
64  ts2 = 10.;
65  ts3 = 29.3;
66  thpd = 4.0;
67  tpre = 7.0;
68  wd1 = 2.0;
69  wd2 = 0.7;
70  wd3 = 1.0;
71  computeHPDShape(ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3, hpdShapeMC_v2);
72  theShapes[123] = &hpdShapeMC_v2;
73 
74  // HPD Shape Version 3 for CMSSW 5. Nov 2011 (RECO and MC separately)
75  ts1 = 8.;
76  ts2 = 19.;
77  ts3 = 29.3;
78  thpd = 4.0;
79  tpre = 9.0;
80  wd1 = 2.0;
81  wd2 = 0.7;
82  wd3 = 0.32;
83  computeHPDShape(ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3, hpdShape_v3);
84  theShapes[105] = &hpdShape_v3;
85 
86  ts1 = 8.;
87  ts2 = 10.;
88  ts3 = 22.3;
89  thpd = 4.0;
90  tpre = 7.0;
91  wd1 = 2.0;
92  wd2 = 0.7;
93  wd3 = 1.0;
94  computeHPDShape(ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3, hpdShapeMC_v3);
95  theShapes[125] = &hpdShapeMC_v3;
96 
97  // HPD with Bias Voltage 30 volts, wider pulse. (HBPlus iphi54)
98 
99  ts1 = 8.;
100  ts2 = 12.;
101  ts3 = 31.7;
102  thpd = 9.0;
103  tpre = 9.0;
104  wd1 = 2.0;
105  wd2 = 0.7;
106  wd3 = 1.0;
107  computeHPDShape(ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3, hpdBV30Shape_v2);
108  theShapes[104] = &hpdBV30Shape_v2;
109 
110  ts1 = 8.;
111  ts2 = 12.;
112  ts3 = 31.7;
113  thpd = 9.0;
114  tpre = 9.0;
115  wd1 = 2.0;
116  wd2 = 0.7;
117  wd3 = 1.0;
118  computeHPDShape(ts1, ts2, ts3, thpd, tpre, wd1, wd2, wd3, hpdBV30ShapeMC_v2);
120 
121  // HF and SiPM
122 
123  computeHFShape();
128 
129  theShapes[201] = &siPMShapeHO_;
130  theShapes[202] = theShapes[201];
131  theShapes[203] = &(computeSiPMShapeHE203());
133  theShapes[206] = &(computeSiPMShapeHE206());
136  theShapes[301] = &hfShape_;
137  //theShapes[401] = new CaloCachedShapeIntegrator(&theZDCShape);
138 }
void computeSiPMShapeMCRecoRun3()
const HcalPulseShape & computeSiPMShapeHE206()
void computeSiPMShapeData2017()
const HcalPulseShape & computeSiPMShapeHE203()
void computeSiPMShapeData2018()
const HcalDbService * theDbService
void computeHPDShape(float, float, float, float, float, float, float, float, Shape &)

◆ HcalPulseShapes() [2/2]

HcalPulseShapes::HcalPulseShapes ( edm::ConsumesCollector  iC)
explicit

◆ ~HcalPulseShapes()

HcalPulseShapes::~HcalPulseShapes ( )

Definition at line 144 of file HcalPulseShapes.cc.

144 {}

Member Function Documentation

◆ analyticPulseShapeSiPMHE()

double HcalPulseShapes::analyticPulseShapeSiPMHE ( double  t)
static

Definition at line 575 of file HcalPulseShapes.cc.

References onePulse(), and submitPVValidationJobs::t.

Referenced by HcalSiPMShape::computeShape(), computeSiPMShapeHE203(), and computeSiPMShapeHE206().

575  {
576  // taken from fit to laser measurement taken by Iouri M. in Spring 2016.
577  double A1(5.204 / 6.94419), sigma1_shape(0.5387), theta1_loc(-0.3976), m1_scale(4.428);
578  double A2(1.855 / 6.94419), sigma2_shape(0.8132), theta2_loc(7.025), m2_scale(12.29);
579  return onePulse(t, A1, sigma1_shape, theta1_loc, m1_scale) + onePulse(t, A2, sigma2_shape, theta2_loc, m2_scale);
580 }
double onePulse(double t, double A, double sigma, double theta, double m)

◆ analyticPulseShapeSiPMHO()

double HcalPulseShapes::analyticPulseShapeSiPMHO ( double  t)
static

Definition at line 568 of file HcalPulseShapes.cc.

References alignmentValidation::c1, gexp(), and submitPVValidationJobs::t.

Referenced by HcalSiPMShape::computeShape().

568  {
569  // HO SiPM pulse shape fit from Jake Anderson ca. 2013
570  double A1(0.08757), c1(-0.5257), t01(2.4013), s1(0.6721);
571  double A2(0.007598), c2(-0.1501), t02(6.9412), s2(0.8710);
572  return gexp(t, A1, c1, t01, s1) + gexp(t, A2, c2, t02, s2);
573 }
double gexp(double t, double A, double c, double t0, double s)

◆ beginRun() [1/2]

void HcalPulseShapes::beginRun ( edm::EventSetup const &  es)

Definition at line 146 of file HcalPulseShapes.cc.

References edm::EventSetup::getData(), theDbService, and theDbServiceToken.

Referenced by HcalPulseContainmentManager::beginRun().

146 { theDbService = &es.getData(theDbServiceToken); }
edm::ESGetToken< HcalDbService, HcalDbRecord > theDbServiceToken
const HcalDbService * theDbService

◆ beginRun() [2/2]

void HcalPulseShapes::beginRun ( const HcalDbService conditions)

Definition at line 148 of file HcalPulseShapes.cc.

References submitPVValidationJobs::conditions, and theDbService.

◆ computeHFShape()

void HcalPulseShapes::computeHFShape ( )
private

Definition at line 247 of file HcalPulseShapes.cc.

References JetChargeProducer_cfi::exp, hfShape_, dqmiolumiharvest::j, reco::ParticleMasses::k0, LaserDQM_cfg::p1, SiStripOfflineCRack_cfg::p2, HcalPulseShape::setNBin(), and HcalPulseShape::setShapeBin().

Referenced by HcalPulseShapes().

247  {
248  // first create pulse shape over a range of time 0 ns to 255 ns in 1 ns steps
249  unsigned int nbin = 256;
250  hfShape_.setNBin(nbin);
251  std::vector<float> ntmp(nbin, 0.0); //
252 
253  const float k0 = 0.7956; // shape parameters
254  const float p2 = 1.355;
255  const float p4 = 2.327;
256  const float p1 = 4.3; // position parameter
257 
258  float norm = 0.0;
259 
260  for (unsigned int j = 0; j < 25 && j < nbin; ++j) {
261  float r0 = j - p1;
262  float sigma0 = (r0 < 0) ? p2 : p2 * p4;
263  r0 /= sigma0;
264  if (r0 < k0)
265  ntmp[j] = exp(-0.5 * r0 * r0);
266  else
267  ntmp[j] = exp(0.5 * k0 * k0 - k0 * r0);
268  norm += ntmp[j];
269  }
270  // normalize pulse area to 1.0
271  for (unsigned int j = 0; j < 25 && j < nbin; ++j) {
272  ntmp[j] /= norm;
273  hfShape_.setShapeBin(j, ntmp[j]);
274  }
275 }
void setNBin(int n)
void setShapeBin(int i, float f)

◆ computeHPDShape()

void HcalPulseShapes::computeHPDShape ( float  ts1,
float  ts2,
float  ts3,
float  thpd,
float  tpre,
float  wd1,
float  wd2,
float  wd3,
Shape tmphpdShape_ 
)
private

Definition at line 151 of file HcalPulseShapes.cc.

References JetChargeProducer_cfi::exp, nano_mu_digi_cff::float, mps_fire::i, createfilelist::int, dqmiolumiharvest::j, dqmdumpme::k, HcalPulseShape::setNBin(), HcalPulseShape::setShapeBin(), RandomServiceHelper::t1, RandomServiceHelper::t2, RandomServiceHelper::t3, and tmax.

Referenced by HcalPulseShapes().

152  {
153  // pulse shape time constants in ns
154  /*
155  const float ts1 = 8.; // scintillation time constants : 1,2,3
156  const float ts2 = 10.;
157  const float ts3 = 29.3;
158  const float thpd = 4.; // HPD current collection drift time
159  const float tpre = 9.; // preamp time constant (refit on TB04 data)
160 
161  const float wd1 = 2.; // relative weights of decay exponents
162  const float wd2 = 0.7;
163  const float wd3 = 1.;
164 */
165  // pulse shape components over a range of time 0 ns to 255 ns in 1 ns steps
166  unsigned int nbin = 256;
167  tmphpdShape_.setNBin(nbin);
168  std::vector<float> ntmp(nbin, 0.0); // zeroing output pulse shape
169  std::vector<float> nth(nbin, 0.0); // zeroing HPD drift shape
170  std::vector<float> ntp(nbin, 0.0); // zeroing Binkley preamp shape
171  std::vector<float> ntd(nbin, 0.0); // zeroing Scintillator decay shape
172 
173  unsigned int i, j, k;
174  float norm;
175 
176  // HPD starts at I and rises to 2I in thpd of time
177  norm = 0.0;
178  for (j = 0; j < thpd && j < nbin; j++) {
179  nth[j] = 1.0 + ((float)j) / thpd;
180  norm += nth[j];
181  }
182  // normalize integrated current to 1.0
183  for (j = 0; j < thpd && j < nbin; j++) {
184  nth[j] /= norm;
185  }
186 
187  // Binkley shape over 6 time constants
188  norm = 0.0;
189  for (j = 0; j < 6 * tpre && j < nbin; j++) {
190  ntp[j] = ((float)j) * exp(-((float)(j * j)) / (tpre * tpre));
191  norm += ntp[j];
192  }
193  // normalize pulse area to 1.0
194  for (j = 0; j < 6 * tpre && j < nbin; j++) {
195  ntp[j] /= norm;
196  }
197 
198  // ignore stochastic variation of photoelectron emission
199  // <...>
200 
201  // effective tile plus wave-length shifter decay time over 4 time constants
202  unsigned int tmax = 6 * (int)ts3;
203 
204  norm = 0.0;
205  for (j = 0; j < tmax && j < nbin; j++) {
206  ntd[j] = wd1 * exp(-((float)j) / ts1) + wd2 * exp(-((float)j) / ts2) + wd3 * exp(-((float)j) / ts3);
207  norm += ntd[j];
208  }
209  // normalize pulse area to 1.0
210  for (j = 0; j < tmax && j < nbin; j++) {
211  ntd[j] /= norm;
212  }
213 
214  unsigned int t1, t2, t3, t4;
215  for (i = 0; i < tmax && i < nbin; i++) {
216  t1 = i;
217  // t2 = t1 + top*rand;
218  // ignoring jitter from optical path length
219  t2 = t1;
220  for (j = 0; j < thpd && j < nbin; j++) {
221  t3 = t2 + j;
222  for (k = 0; k < 4 * tpre && k < nbin; k++) { // here "4" is set deliberately,
223  t4 = t3 + k; // as in test fortran toy MC ...
224  if (t4 < nbin) {
225  unsigned int ntb = t4;
226  ntmp[ntb] += ntd[i] * nth[j] * ntp[k];
227  }
228  }
229  }
230  }
231 
232  // normalize for 1 GeV pulse height
233  norm = 0.;
234  for (i = 0; i < nbin; i++) {
235  norm += ntmp[i];
236  }
237 
238  for (i = 0; i < nbin; i++) {
239  ntmp[i] /= norm;
240  }
241 
242  for (i = 0; i < nbin; i++) {
243  tmphpdShape_.setShapeBin(i, ntmp[i]);
244  }
245 }
static const double tmax[3]

◆ computeSiPMShapeData2017()

void HcalPulseShapes::computeSiPMShapeData2017 ( )
private

Definition at line 380 of file HcalPulseShapes.cc.

References dqmiolumiharvest::j, nt, HcalPulseShape::setNBin(), HcalPulseShape::setShapeBin(), and siPMShapeData2017_.

Referenced by HcalPulseShapes().

380  {
381  //From Jay Lawhorn: derived from data Edward Laird phase scan may2017
382  //https://indico.cern.ch/event/641978/contributions/2604491/attachments/1468666/2271582/17-05-31-hcal-hep17-pulse-shape.pdf
383  //Run numbers are 294736-294740 and 294929-294950
384 
385  unsigned int nbin = 250;
386 
387  std::array<float, 250> nt{
388  {3.97958e-29, 1.11634e-22, 9.96106e-18, 6.25334e-14, 5.08863e-11, 8.59141e-09, 4.32285e-07, 8.56617e-06,
389  8.28549e-05, 0.000461447, 0.00168052, 0.00441395, 0.00901637, 0.0151806, 0.0220314, 0.028528,
390  0.0338471, 0.0375578, 0.0395985, 0.0401567, 0.0395398, 0.0380776, 0.0360669, 0.0337474,
391  0.0312984, 0.0288457, 0.0264721, 0.0242276, 0.0221393, 0.0202181, 0.0184647, 0.0168731,
392  0.0154335, 0.0141346, 0.0129639, 0.0119094, 0.0109594, 0.0101031, 0.0093305, 0.00863267,
393  0.0080015, 0.00742977, 0.00691107, 0.00643969, 0.00601059, 0.00561931, 0.00526188, 0.00493483,
394  0.00463505, 0.00435981, 0.00410667, 0.00387348, 0.00365832, 0.00345949, 0.00327547, 0.0031049,
395  0.00294656, 0.00279938, 0.00266237, 0.00253467, 0.00241548, 0.0023041, 0.00219989, 0.00210227,
396  0.00201072, 0.00192476, 0.00184397, 0.00176795, 0.00169634, 0.00162884, 0.00156512, 0.00150494,
397  0.00144803, 0.00139418, 0.00134317, 0.00129481, 0.00124894, 0.00120537, 0.00116398, 0.00112461,
398  0.00108715, 0.00105147, 0.00101747, 0.000985042, 0.000954096, 0.000924545, 0.000896308, 0.000869311,
399  0.000843482, 0.000818758, 0.000795077, 0.000772383, 0.000750623, 0.000729747, 0.00070971, 0.000690466,
400  0.000671977, 0.000654204, 0.00063711, 0.000620663, 0.000604831, 0.000589584, 0.000574894, 0.000560735,
401  0.000547081, 0.00053391, 0.0005212, 0.000508929, 0.000497078, 0.000485628, 0.000474561, 0.000463862,
402  0.000453514, 0.000443501, 0.000433811, 0.000424429, 0.000415343, 0.00040654, 0.00039801, 0.000389741,
403  0.000381722, 0.000373944, 0.000366398, 0.000359074, 0.000351964, 0.00034506, 0.000338353, 0.000331838,
404  0.000325505, 0.00031935, 0.000313365, 0.000307544, 0.000301881, 0.000296371, 0.000291009, 0.000285788,
405  0.000280705, 0.000275755, 0.000270932, 0.000266233, 0.000261653, 0.00025719, 0.000252837, 0.000248593,
406  0.000244454, 0.000240416, 0.000236475, 0.00023263, 0.000228876, 0.000225212, 0.000221633, 0.000218138,
407  0.000214724, 0.000211389, 0.00020813, 0.000204945, 0.000201831, 0.000198787, 0.000195811, 0.0001929,
408  0.000190053, 0.000187268, 0.000184543, 0.000181876, 0.000179266, 0.000176711, 0.00017421, 0.000171761,
409  0.000169363, 0.000167014, 0.000164713, 0.000162459, 0.00016025, 0.000158086, 0.000155964, 0.000153885,
410  0.000151847, 0.000149848, 0.000147888, 0.000145966, 0.000144081, 0.000142232, 0.000140418, 0.000138638,
411  0.000136891, 0.000135177, 0.000133494, 0.000131843, 0.000130221, 0.00012863, 0.000127066, 0.000125531,
412  0.000124023, 0.000122543, 0.000121088, 0.000119658, 0.000118254, 0.000116874, 0.000115518, 0.000114185,
413  0.000112875, 0.000111587, 0.000110321, 0.000109076, 0.000107851, 0.000106648, 0.000105464, 0.000104299,
414  0.000103154, 0.000102027, 0.000100918, 9.98271e-05, 9.87537e-05, 9.76974e-05, 9.66578e-05, 9.56346e-05,
415  9.46274e-05, 9.3636e-05, 9.26599e-05, 9.16989e-05, 9.07526e-05, 8.98208e-05, 8.89032e-05, 8.79995e-05,
416  8.71093e-05, 8.62325e-05, 8.53688e-05, 8.45179e-05, 8.36796e-05, 8.28536e-05, 8.20397e-05, 8.12376e-05,
417  8.04471e-05, 7.96681e-05, 7.89002e-05, 7.81433e-05, 7.73972e-05, 7.66616e-05, 7.59364e-05, 7.52213e-05,
418  7.45163e-05, 7.3821e-05, 7.31354e-05, 7.24592e-05, 7.17923e-05, 7.11345e-05, 7.04856e-05, 6.98455e-05,
419  6.9214e-05, 6.8591e-05}};
420 
422 
423  double norm = 0.;
424  for (unsigned int j = 0; j < nbin; ++j) {
425  norm += (nt[j] > 0) ? nt[j] : 0.;
426  }
427 
428  for (unsigned int j = 0; j < nbin; ++j) {
429  nt[j] /= norm;
431  }
432 }
void setNBin(int n)
void setShapeBin(int i, float f)
int nt
Definition: AMPTWrapper.h:42

◆ computeSiPMShapeData2018()

void HcalPulseShapes::computeSiPMShapeData2018 ( )
private

Definition at line 327 of file HcalPulseShapes.cc.

References dqmiolumiharvest::j, nt, HcalPulseShape::setNBin(), HcalPulseShape::setShapeBin(), and siPMShapeData2018_.

Referenced by HcalPulseShapes().

327  {
328  //Combination of all phase scan data (May,Jul,Oct2017)
329  //runs: 294736-294740, 294929-294950, 298594-298598 and 305744-305758
330 
331  unsigned int nbin = 250;
332 
333  std::array<float, 250> nt{
334  {5.22174e-12, 7.04852e-10, 3.49584e-08, 7.78029e-07, 9.11847e-06, 6.39666e-05, 0.000297587, 0.000996661,
335  0.00256618, 0.00535396, 0.00944073, 0.0145521, 0.020145, 0.0255936, 0.0303632, 0.0341078,
336  0.0366849, 0.0381183, 0.0385392, 0.0381327, 0.0370956, 0.0356113, 0.0338366, 0.0318978,
337  0.029891, 0.0278866, 0.0259336, 0.0240643, 0.0222981, 0.0206453, 0.0191097, 0.0176902,
338  0.0163832, 0.0151829, 0.0140826, 0.0130752, 0.0121533, 0.01131, 0.0105382, 0.00983178,
339  0.00918467, 0.00859143, 0.00804709, 0.0075471, 0.00708733, 0.00666406, 0.00627393, 0.00591389,
340  0.00558122, 0.00527344, 0.00498834, 0.00472392, 0.00447837, 0.00425007, 0.00403754, 0.00383947,
341  0.00365465, 0.00348199, 0.00332052, 0.00316934, 0.00302764, 0.0028947, 0.00276983, 0.00265242,
342  0.00254193, 0.00243785, 0.00233971, 0.00224709, 0.0021596, 0.00207687, 0.0019986, 0.00192447,
343  0.00185421, 0.00178756, 0.0017243, 0.00166419, 0.00160705, 0.00155268, 0.00150093, 0.00145162,
344  0.00140461, 0.00135976, 0.00131696, 0.00127607, 0.00123699, 0.00119962, 0.00116386, 0.00112963,
345  0.00109683, 0.0010654, 0.00103526, 0.00100634, 0.000978578, 0.000951917, 0.000926299, 0.000901672,
346  0.000877987, 0.000855198, 0.00083326, 0.000812133, 0.000791778, 0.000772159, 0.000753242, 0.000734994,
347  0.000717384, 0.000700385, 0.000683967, 0.000668107, 0.000652779, 0.00063796, 0.000623629, 0.000609764,
348  0.000596346, 0.000583356, 0.000570777, 0.000558592, 0.000546785, 0.00053534, 0.000524243, 0.000513481,
349  0.00050304, 0.000492907, 0.000483072, 0.000473523, 0.000464248, 0.000455238, 0.000446483, 0.000437974,
350  0.0004297, 0.000421655, 0.00041383, 0.000406216, 0.000398807, 0.000391595, 0.000384574, 0.000377736,
351  0.000371076, 0.000364588, 0.000358266, 0.000352104, 0.000346097, 0.00034024, 0.000334528, 0.000328956,
352  0.00032352, 0.000318216, 0.000313039, 0.000307986, 0.000303052, 0.000298234, 0.000293528, 0.000288931,
353  0.000284439, 0.00028005, 0.000275761, 0.000271567, 0.000267468, 0.000263459, 0.000259538, 0.000255703,
354  0.000251951, 0.00024828, 0.000244688, 0.000241172, 0.00023773, 0.000234361, 0.000231061, 0.00022783,
355  0.000224666, 0.000221566, 0.000218528, 0.000215553, 0.000212636, 0.000209778, 0.000206977, 0.00020423,
356  0.000201537, 0.000198896, 0.000196307, 0.000193767, 0.000191275, 0.000188831, 0.000186432, 0.000184079,
357  0.000181769, 0.000179502, 0.000177277, 0.000175092, 0.000172947, 0.000170841, 0.000168772, 0.000166741,
358  0.000164745, 0.000162785, 0.000160859, 0.000158967, 0.000157108, 0.00015528, 0.000153484, 0.000151719,
359  0.000149984, 0.000148278, 0.000146601, 0.000144951, 0.000143329, 0.000141734, 0.000140165, 0.000138622,
360  0.000137104, 0.00013561, 0.000134141, 0.000132695, 0.000131272, 0.000129871, 0.000128493, 0.000127136,
361  0.000125801, 0.000124486, 0.000123191, 0.000121917, 0.000120662, 0.000119426, 0.000118209, 0.00011701,
362  0.000115829, 0.000114665, 0.000113519, 0.00011239, 0.000111278, 0.000110182, 0.000109102, 0.000108037,
363  0.000106988, 0.000105954, 0.000104935, 0.00010393, 0.000102939, 0.000101963, 0.000101, 0.000100051,
364  9.91146e-05, 9.81915e-05, 9.7281e-05, 9.63831e-05, 9.54975e-05, 9.46239e-05, 9.37621e-05, 9.2912e-05,
365  9.20733e-05, 9.12458e-05}};
366 
368 
369  double norm = 0.;
370  for (unsigned int j = 0; j < nbin; ++j) {
371  norm += (nt[j] > 0) ? nt[j] : 0.;
372  }
373 
374  for (unsigned int j = 0; j < nbin; ++j) {
375  nt[j] /= norm;
377  }
378 }
void setNBin(int n)
void setShapeBin(int i, float f)
int nt
Definition: AMPTWrapper.h:42

◆ computeSiPMShapeHE203()

const HcalPulseShape & HcalPulseShapes::computeSiPMShapeHE203 ( )
private

Definition at line 485 of file HcalPulseShapes.cc.

References analyticPulseShapeSiPMHE(), convolve(), nBinsSiPM_, normalize(), and Y11203().

Referenced by HcalPulseShapes().

485  {
486  //numerical convolution of SiPM pulse + WLS fiber shape
487  static const HcalPulseShape siPMShapeMC2017(
489  return siPMShapeMC2017;
490 }
static const int nBinsSiPM_
static double analyticPulseShapeSiPMHE(double t)
static std::vector< double > normalize(std::vector< double > nt, unsigned nbin)
static std::vector< double > convolve(unsigned nbin, F1 f1, F2 f2)
static double Y11203(double t)

◆ computeSiPMShapeHE206()

const HcalPulseShape & HcalPulseShapes::computeSiPMShapeHE206 ( )
private

Definition at line 492 of file HcalPulseShapes.cc.

References analyticPulseShapeSiPMHE(), convolve(), nBinsSiPM_, normalizeShift(), and Y11206().

Referenced by HcalPulseShapes().

492  {
493  //numerical convolution of SiPM pulse + WLS fiber shape
494  //shift: aligning 206 phase closer to 205 in order to have good reco agreement
495  static const HcalPulseShape siPMShapeMC2018(
497  return siPMShapeMC2018;
498 }
static std::vector< double > normalizeShift(std::vector< double > nt, unsigned nbin, int shift)
static const int nBinsSiPM_
static double analyticPulseShapeSiPMHE(double t)
static std::vector< double > convolve(unsigned nbin, F1 f1, F2 f2)
static double Y11206(double t)

◆ computeSiPMShapeHO()

void HcalPulseShapes::computeSiPMShapeHO ( )
private

Definition at line 434 of file HcalPulseShapes.cc.

References dqmiolumiharvest::j, nt, HcalPulseShape::setNBin(), HcalPulseShape::setShapeBin(), and siPMShapeHO_.

Referenced by HcalPulseShapes().

434  {
435  unsigned int nbin = 128;
436 
437  //From Jake Anderson: toy MC convolution of SiPM pulse + WLS fiber shape + SiPM nonlinear response
438  std::array<float, 128> nt{
439  {2.782980485851731e-6, 4.518134885954626e-5, 2.7689305197392056e-4, 9.18328418900969e-4,
440  .002110072599166349, .003867856860331454, .006120046224897771, .008754774090536956,
441  0.0116469503358586, .01467007449455966, .01770489955229477, .02064621450689512,
442  .02340678093764222, .02591874610854916, .02813325527435303, 0.0300189241965647,
443  .03155968107671164, .03275234052577155, .03360415306318798, .03413048377960748,
444  .03435270899678218, .03429637464659661, .03398962975487166, .03346192884394954,
445  .03274298516247742, .03186195009136525, .03084679116113031, 0.0297238406141036,
446  .02851748748929785, .02724998816332392, .02594137274487424, .02460942736731527,
447  .02326973510736116, .02193576080366117, 0.0206189674254987, .01932895378564653,
448  0.0180736052958666, .01685925112650875, 0.0156908225633535, .01457200857138456,
449  .01350540559602467, .01249265947824805, .01153459805300423, .01063135355597282,
450  .009782474412011936, .008987026319784546, 0.00824368281357106, .007550805679909604,
451  .006906515742762193, .006308754629755056, .005755338185695127, .005244002229973356,
452  .004772441359900532, .004338341490928299, .003939406800854143, 0.00357338171220501,
453  0.0032380685079891, .002931341133259233, .002651155690306086, .002395558090237333,
454  .002162689279320922, .001950788415487319, .001758194329648101, .001583345567913682,
455  .001424779275191974, .001281129147671334, 0.00115112265163774, .001033577678808199,
456  9.273987838127585e-4, 8.315731274976846e-4, 7.451662302008696e-4, 6.673176219006913e-4,
457  5.972364609644049e-4, 5.341971801529036e-4, 4.775352065178378e-4, 4.266427928961177e-4,
458  3.8096498904225923e-4, 3.3999577417327287e-4, 3.032743659102713e-4, 2.703817158798329e-4,
459  2.4093719775272793e-4, 2.145954900503894e-4, 1.9104365317752797e-4, 1.6999839784346724e-4,
460  1.5120354022478893e-4, 1.3442763782650755e-4, 1.1946179895521507e-4, 1.0611765796993575e-4,
461  9.422550797617687e-5, 8.363258233342666e-5, 7.420147621931836e-5, 6.580869950304933e-5,
462  5.834335229919868e-5, 5.17059147771959e-5, 4.5807143072062634e-5, 4.0567063461299446e-5,
463  3.591405732740723e-5, 3.178402980354131e-5, 2.811965539165646e-5, 2.4869694240316126e-5,
464  2.1988373166730962e-5, 1.9434825899529382e-5, 1.717258740121378e-5, 1.5169137499243157e-5,
465  1.339548941011129e-5, 1.1825819079078403e-5, 1.0437131581057595e-5, 9.208961130078894e-6,
466  8.12310153137994e-6, 7.163364176588591e-6, 6.315360932244386e-6, 5.566309502463164e-6,
467  4.904859063429651e-6, 4.320934164082596e-6, 3.8055950719111903e-6, 3.350912911083174e-6,
468  2.9498580949517117e-6, 2.596200697612328e-6, 2.2844215378879293e-6, 2.0096328693141094e-6,
469  1.7675076766686654e-6, 1.5542166787225756e-6, 1.366372225473431e-6, 1.200978365778838e-6,
470  1.0553864128982371e-6, 9.272554464808518e-7, 8.145171945902259e-7, 7.153448381918271e-7}};
471 
472  siPMShapeHO_.setNBin(nbin);
473 
474  double norm = 0.;
475  for (unsigned int j = 0; j < nbin; ++j) {
476  norm += (nt[j] > 0) ? nt[j] : 0.;
477  }
478 
479  for (unsigned int j = 0; j < nbin; ++j) {
480  nt[j] /= norm;
482  }
483 }
void setNBin(int n)
void setShapeBin(int i, float f)
int nt
Definition: AMPTWrapper.h:42

◆ computeSiPMShapeMCRecoRun3()

void HcalPulseShapes::computeSiPMShapeMCRecoRun3 ( )
private

Definition at line 276 of file HcalPulseShapes.cc.

References dqmiolumiharvest::j, nt, HcalPulseShape::setNBin(), HcalPulseShape::setShapeBin(), and siPMShapeMCRecoRun3_.

Referenced by HcalPulseShapes().

276  {
277  //modified shape 206
278  //7.2 ns shift in 206
279  unsigned int nbin = 250;
280  std::array<float, 250> nt{
281  {0, 0, 0, 0, 0, 0, 0, 0.000117468,
282  0.0031549, 0.0117368, 0.0219974, 0.0305776, 0.0365429, 0.0400524, 0.0415915, 0.0416765,
283  0.0408111, 0.0394627, 0.0379353, 0.0363688, 0.0348152, 0.0332891, 0.0317923, 0.0303237,
284  0.028883, 0.0274714, 0.0260914, 0.0247462, 0.0234392, 0.0221738, 0.0209531, 0.0197793,
285  0.0186544, 0.0175796, 0.0165556, 0.0155823, 0.0146596, 0.0137866, 0.0129623, 0.0121853,
286  0.0114539, 0.0107665, 0.0101213, 0.0095162, 0.00894934, 0.00841873, 0.0079224, 0.00745841,
287  0.00702487, 0.00661995, 0.00624189, 0.00588898, 0.00555961, 0.00525223, 0.00496539, 0.0046977,
288  0.00444786, 0.00421464, 0.00399689, 0.00379353, 0.00360355, 0.00342602, 0.00326004, 0.0031048,
289  0.00295954, 0.00282355, 0.00269616, 0.00257676, 0.00246479, 0.00235972, 0.00226106, 0.00216834,
290  0.00208117, 0.00199914, 0.00192189, 0.0018491, 0.00178044, 0.00171565, 0.00165445, 0.00159659,
291  0.00154186, 0.00149003, 0.00144092, 0.00139435, 0.00135015, 0.00130816, 0.00126825, 0.00123027,
292  0.00119412, 0.00115966, 0.0011268, 0.00109544, 0.00106548, 0.00103685, 0.00100946, 0.000983242,
293  0.000958125, 0.000934047, 0.000910949, 0.000888775, 0.000867475, 0.000847, 0.000827306, 0.000808352,
294  0.000790097, 0.000772506, 0.000755545, 0.000739182, 0.000723387, 0.000708132, 0.00069339, 0.000679138,
295  0.000665352, 0.00065201, 0.000639091, 0.000626577, 0.00061445, 0.000602692, 0.000591287, 0.00058022,
296  0.000569477, 0.000559044, 0.000548908, 0.000539058, 0.000529481, 0.000520167, 0.000511106, 0.000502288,
297  0.000493704, 0.000485344, 0.000477201, 0.000469266, 0.000459912, 0.000448544, 0.000437961, 0.000428079,
298  0.000418825, 0.000410133, 0.000401945, 0.00039421, 0.000386883, 0.000379924, 0.000373298, 0.000366973,
299  0.000360922, 0.00035512, 0.000349545, 0.000344179, 0.000339003, 0.000334002, 0.000329163, 0.000324475,
300  0.000319925, 0.000315504, 0.000311204, 0.000307017, 0.000302935, 0.000298954, 0.000295066, 0.000291267,
301  0.000287553, 0.000283919, 0.000280361, 0.000276877, 0.000273462, 0.000270114, 0.000266831, 0.000263609,
302  0.000260447, 0.000257343, 0.000254295, 0.0002513, 0.000248358, 0.000245467, 0.000242625, 0.000239831,
303  0.000237083, 0.000234381, 0.000231723, 0.000229109, 0.000226536, 0.000224005, 0.000221514, 0.000219062,
304  0.000216648, 0.000214272, 0.000211933, 0.00020963, 0.000207362, 0.000205129, 0.000202929, 0.000200763,
305  0.000198629, 0.000196526, 0.000194455, 0.000192415, 0.000190405, 0.000188424, 0.000186472, 0.000184548,
306  0.000182653, 0.000180784, 0.000178943, 0.000177127, 0.000175338, 0.000173574, 0.000171835, 0.00017012,
307  0.000168429, 0.000166762, 0.000165119, 0.000163498, 0.000161899, 0.000160322, 0.000158767, 0.000157233,
308  0.000155721, 0.000154228, 0.000152756, 0.000151304, 0.000149871, 0.000148457, 0.000147062, 0.000145686,
309  0.000144327, 0.000142987, 0.000141664, 0.000140359, 0.000139071, 0.000137799, 0.000136544, 0.000135305,
310  0.000134082, 0.000132874, 0.000131682, 0.000130505, 0.000129344, 0.000128196, 0.000127064, 0.000125945,
311  0.00012484, 0.00012375, 0.000122672, 0.000121608, 0.000120558, 0.00011952, 0.000118495, 0.000117482,
312  0.000116482, 0.000115493}};
313 
315 
316  double norm = 0.;
317  for (unsigned int j = 0; j < nbin; ++j) {
318  norm += (nt[j] > 0) ? nt[j] : 0.;
319  }
320 
321  for (unsigned int j = 0; j < nbin; ++j) {
322  nt[j] /= norm;
324  }
325 }
void setNBin(int n)
void setShapeBin(int i, float f)
int nt
Definition: AMPTWrapper.h:42

◆ convolve()

template<class F1 , class F2 >
static std::vector<double> HcalPulseShapes::convolve ( unsigned  nbin,
F1  f1,
F2  f2 
)
inlinestatic

Definition at line 61 of file HcalPulseShapes.h.

References DeadROC_duringRun::f1, DeadROC_duringRun::f2, mps_fire::i, edm::isNotFinite(), dqmiolumiharvest::j, SiStripPI::min, mps_fire::result, and createJobs::tmp.

Referenced by computeSiPMShapeHE203(), and computeSiPMShapeHE206().

61  {
62  std::vector<double> result(2 * nbin - 1, 0.);
63  for (unsigned i = 0; i < 2 * nbin - 1; ++i) {
64  for (unsigned j = 0; j < std::min(i + 1, nbin); ++j) {
65  double tmp = f1(j) * f2(i - j);
66  if (edm::isNotFinite(tmp))
67  continue;
68  result[i] += tmp;
69  }
70  }
71  return result;
72  }
constexpr bool isNotFinite(T x)
Definition: isFinite.h:9
tmp
align.sh
Definition: createJobs.py:716

◆ defaultShape()

const HcalPulseShapes::Shape & HcalPulseShapes::defaultShape ( const HcalDetId detId) const

in case of conditions problems

Definition at line 538 of file HcalPulseShapes.cc.

References hcalRecHitTable_cff::detId, Exception, hbShape(), HcalBarrel, HcalEndcap, HcalForward, HcalOuter, heShape(), hfShape(), and hoShape().

Referenced by shape(), and shapeForReco().

538  {
539  edm::LogWarning("HcalPulseShapes") << "Cannot find HCAL MC Params ";
540  HcalSubdetector subdet = detId.subdet();
541  switch (subdet) {
542  case HcalBarrel:
543  return hbShape();
544  case HcalEndcap:
545  return heShape();
546  case HcalForward:
547  return hfShape();
548  case HcalOuter:
549  //FIXME doesn't look for SiPMs
550  return hoShape(false);
551  default:
552  throw cms::Exception("HcalPulseShapes") << "unknown detId";
553  break;
554  }
555 }
HcalSubdetector
Definition: HcalAssistant.h:31
const Shape & hoShape(bool sipm=false) const
const Shape & hbShape() const
const Shape & heShape() const
const Shape & hfShape() const
Log< level::Warning, false > LogWarning

◆ generatePhotonTime()

double HcalPulseShapes::generatePhotonTime ( CLHEP::HepRandomEngine *  engine,
unsigned int  signalShape 
)
static

Definition at line 582 of file HcalPulseShapes.cc.

References generatePhotonTime203(), and generatePhotonTime206().

Referenced by HcalSiPMHitResponse::add().

582  {
583  if (signalShape == 206)
584  return generatePhotonTime206(engine);
585  else
586  return generatePhotonTime203(engine);
587 }
static double generatePhotonTime206(CLHEP::HepRandomEngine *engine)
static double generatePhotonTime203(CLHEP::HepRandomEngine *engine)

◆ generatePhotonTime203()

double HcalPulseShapes::generatePhotonTime203 ( CLHEP::HepRandomEngine *  engine)
static

Definition at line 589 of file HcalPulseShapes.cc.

References mps_fire::result, Y11203(), Y11MAX203_, and Y11RANGE_.

Referenced by generatePhotonTime().

589  {
590  double result(0.);
591  while (true) {
592  result = CLHEP::RandFlat::shoot(engine, HcalPulseShapes::Y11RANGE_);
593  if (CLHEP::RandFlat::shoot(engine, HcalPulseShapes::Y11MAX203_) < HcalPulseShapes::Y11203(result))
594  return result;
595  }
596 }
static constexpr float Y11MAX203_
static double Y11203(double t)
static constexpr float Y11RANGE_

◆ generatePhotonTime206()

double HcalPulseShapes::generatePhotonTime206 ( CLHEP::HepRandomEngine *  engine)
static

Definition at line 598 of file HcalPulseShapes.cc.

References mps_fire::result, Y11206(), Y11MAX206_, and Y11RANGE_.

Referenced by generatePhotonTime().

598  {
599  double result(0.);
600  while (true) {
601  result = CLHEP::RandFlat::shoot(engine, HcalPulseShapes::Y11RANGE_);
602  if (CLHEP::RandFlat::shoot(engine, HcalPulseShapes::Y11MAX206_) < HcalPulseShapes::Y11206(result))
603  return result;
604  }
605 }
static constexpr float Y11MAX206_
static double Y11206(double t)
static constexpr float Y11RANGE_

◆ get_all_shapes()

std::map<int, Shape const*> const& HcalPulseShapes::get_all_shapes ( ) const
inline

Definition at line 103 of file HcalPulseShapes.h.

References theShapes.

103 { return theShapes; }

◆ getShape()

const HcalPulseShapes::Shape & HcalPulseShapes::getShape ( int  shapeType) const

Definition at line 500 of file HcalPulseShapes.cc.

References Exception, hpdShape_, and theShapes.

Referenced by ALPAKA_ACCELERATOR_NAMESPACE::HcalRecoParamWithPulseShapeESProducer::produce(), SimpleHBHEPhase1Algo::reconstruct(), MahiFit::resetPulseShapeTemplate(), and HcalShape::setShape().

500  {
501  ShapeMap::const_iterator shapeMapItr = theShapes.find(shapeType);
502  if (shapeMapItr == theShapes.end()) {
503  throw cms::Exception("HcalPulseShapes") << "unknown shapeType";
504  return hpdShape_; // should not return this, but...
505  } else {
506  return *(shapeMapItr->second);
507  }
508 }

◆ hbShape()

const Shape& HcalPulseShapes::hbShape ( ) const
inline

Definition at line 34 of file HcalPulseShapes.h.

References hpdShape_.

Referenced by defaultShape(), and HBHEPulseShapeFlagSetter::Initialize().

34 { return hpdShape_; }

◆ heShape()

const Shape& HcalPulseShapes::heShape ( ) const
inline

Definition at line 35 of file HcalPulseShapes.h.

References hpdShape_.

Referenced by defaultShape().

35 { return hpdShape_; }

◆ hfShape()

const Shape& HcalPulseShapes::hfShape ( ) const
inline

Definition at line 36 of file HcalPulseShapes.h.

References hfShape_.

Referenced by defaultShape().

36 { return hfShape_; }

◆ hoShape()

const Shape& HcalPulseShapes::hoShape ( bool  sipm = false) const
inline

Definition at line 37 of file HcalPulseShapes.h.

References hpdShape_, and siPMShapeHO_.

Referenced by defaultShape().

37 { return sipm ? siPMShapeHO_ : hpdShape_; }

◆ normalize()

static std::vector<double> HcalPulseShapes::normalize ( std::vector< double >  nt,
unsigned  nbin 
)
inlinestatic

Definition at line 73 of file HcalPulseShapes.h.

References dqmiolumiharvest::j, and nt.

Referenced by computeSiPMShapeHE203().

73  {
74  //skip first bin, always 0
75  double norm = 0.;
76  for (unsigned int j = 1; j <= nbin; ++j) {
77  norm += (nt[j] > 0) ? nt[j] : 0.;
78  }
79 
80  double normInv = 1. / norm;
81  for (unsigned int j = 1; j <= nbin; ++j) {
82  nt[j] *= normInv;
83  }
84 
85  return nt;
86  }
int nt
Definition: AMPTWrapper.h:42

◆ normalizeShift()

static std::vector<double> HcalPulseShapes::normalizeShift ( std::vector< double >  nt,
unsigned  nbin,
int  shift 
)
inlinestatic

Definition at line 87 of file HcalPulseShapes.h.

References createfilelist::int, dqmiolumiharvest::j, SiStripPI::max, nt, and edm::shift.

Referenced by computeSiPMShapeHE206().

87  {
88  //skip first bin, always 0
89  double norm = 0.;
90  for (unsigned int j = std::max(1, -1 * shift); j <= nbin; j++) {
91  norm += std::max(0., nt[j - shift]);
92  }
93  double normInv = 1. / norm;
94  std::vector<double> nt2(nt.size(), 0);
95  for (int j = 1; j <= (int)nbin; j++) {
96  if (j - shift >= 0) {
97  nt2[j] = nt[j - shift] * normInv;
98  }
99  }
100  return nt2;
101  }
int nt
Definition: AMPTWrapper.h:42
static unsigned int const shift

◆ shape()

const HcalPulseShapes::Shape & HcalPulseShapes::shape ( const HcalDetId detId) const

automatically figures out which shape to return

Definition at line 510 of file HcalPulseShapes.cc.

References defaultShape(), hcalRecHitTable_cff::detId, HcalDbService::getHcalMCParam(), HcalMCParam::signalShape(), theDbService, and theShapes.

510  {
511  if (!theDbService) {
512  return defaultShape(detId);
513  }
514  int shapeType = theDbService->getHcalMCParam(detId)->signalShape();
515 
516  ShapeMap::const_iterator shapeMapItr = theShapes.find(shapeType);
517  if (shapeMapItr == theShapes.end()) {
518  return defaultShape(detId);
519  } else {
520  return *(shapeMapItr->second);
521  }
522 }
const Shape & defaultShape(const HcalDetId &detId) const
in case of conditions problems
const HcalMCParam * getHcalMCParam(const HcalGenericDetId &fId) const
unsigned int signalShape() const
Definition: HcalMCParam.h:38
const HcalDbService * theDbService

◆ shapeForReco()

const HcalPulseShapes::Shape & HcalPulseShapes::shapeForReco ( const HcalDetId detId) const

Definition at line 524 of file HcalPulseShapes.cc.

References defaultShape(), hcalRecHitTable_cff::detId, HcalDbService::getHcalRecoParam(), theDbService, and theShapes.

Referenced by HcalPulseContainmentManager::get().

524  {
525  if (!theDbService) {
526  return defaultShape(detId);
527  }
528  int shapeType = theDbService->getHcalRecoParam(detId.rawId())->pulseShapeID();
529 
530  ShapeMap::const_iterator shapeMapItr = theShapes.find(shapeType);
531  if (shapeMapItr == theShapes.end()) {
532  return defaultShape(detId);
533  } else {
534  return *(shapeMapItr->second);
535  }
536 }
const Shape & defaultShape(const HcalDetId &detId) const
in case of conditions problems
const HcalDbService * theDbService
const HcalRecoParam * getHcalRecoParam(const HcalGenericDetId &fId) const

◆ Y11203()

double HcalPulseShapes::Y11203 ( double  t)
static

◆ Y11206()

double HcalPulseShapes::Y11206 ( double  t)
static

Definition at line 611 of file HcalPulseShapes.cc.

References A, JetChargeProducer_cfi::exp, l1tPhase2CaloJetEmulator_cfi::fit, DivergingColor::frac, dqmiodumpmetadata::n, edm::shift, submitPVValidationJobs::t, FrontierCondition_GT_autoExpress_cfi::t0, and heppy_batch::val.

Referenced by computeSiPMShapeHE206(), and generatePhotonTime206().

611  {
612  //Shifting phase to have better comparison of digi shape with data
613  //If necessary, further digi phase adjustment can be done here:
614  //SimCalorimetry/HcalSimProducers/python/hcalSimParameters_cfi.py
615  //by changing "timePhase"
616  double shift = 7.2;
617 
618  //Fit From Deconvolved Data
619  double A, n, t0, fit;
620  A = 0.104204;
621  n = 0.44064;
622  t0 = 10.0186;
623  if (t > shift)
624  fit = A * (1 - exp(-(t - shift) / n)) * exp(-(t - shift) / t0);
625  else
626  fit = 0.0;
627 
628  //Correction Term
629  double norm, mpv, sigma, corTerm;
630  norm = 0.0809882;
631  mpv = 0;
632  sigma = 20;
633  if (t > shift)
634  corTerm = norm * TMath::Landau((t - shift), mpv, sigma);
635  else
636  corTerm = 0.0;
637 
638  //Overall Y11
639  double frac = 0.11;
640  double val = (1 - frac) * fit + frac * corTerm;
641 
642  if (val >= 0)
643  return val;
644  else
645  return 0.0;
646 }
static unsigned int const shift
Definition: APVGainStruct.h:7

Member Data Documentation

◆ deltaTSiPM_

constexpr float HcalPulseShapes::deltaTSiPM_ = 0.5
static

Definition at line 47 of file HcalPulseShapes.h.

Referenced by HcalSiPMShape::computeShape().

◆ hfShape_

Shape HcalPulseShapes::hfShape_
private

Definition at line 114 of file HcalPulseShapes.h.

Referenced by computeHFShape(), HcalPulseShapes(), and hfShape().

◆ hpdBV30Shape_v2

Shape HcalPulseShapes::hpdBV30Shape_v2
private

Definition at line 118 of file HcalPulseShapes.h.

Referenced by HcalPulseShapes().

◆ hpdBV30ShapeMC_v2

Shape HcalPulseShapes::hpdBV30ShapeMC_v2
private

Definition at line 118 of file HcalPulseShapes.h.

Referenced by HcalPulseShapes().

◆ hpdShape_

Shape HcalPulseShapes::hpdShape_
private

Definition at line 114 of file HcalPulseShapes.h.

Referenced by getShape(), hbShape(), HcalPulseShapes(), heShape(), and hoShape().

◆ hpdShape_v2

Shape HcalPulseShapes::hpdShape_v2
private

Definition at line 116 of file HcalPulseShapes.h.

Referenced by HcalPulseShapes().

◆ hpdShape_v3

Shape HcalPulseShapes::hpdShape_v3
private

Definition at line 117 of file HcalPulseShapes.h.

Referenced by HcalPulseShapes().

◆ hpdShapeMC_v2

Shape HcalPulseShapes::hpdShapeMC_v2
private

Definition at line 116 of file HcalPulseShapes.h.

Referenced by HcalPulseShapes().

◆ hpdShapeMC_v3

Shape HcalPulseShapes::hpdShapeMC_v3
private

Definition at line 117 of file HcalPulseShapes.h.

Referenced by HcalPulseShapes().

◆ invDeltaTSiPM_

constexpr float HcalPulseShapes::invDeltaTSiPM_ = 2.0
static

◆ nBinsSiPM_

const int HcalPulseShapes::nBinsSiPM_ = 250
static

Definition at line 46 of file HcalPulseShapes.h.

Referenced by computeSiPMShapeHE203(), and computeSiPMShapeHE206().

◆ siPMShapeData2017_

Shape HcalPulseShapes::siPMShapeData2017_
private

Definition at line 115 of file HcalPulseShapes.h.

Referenced by computeSiPMShapeData2017(), and HcalPulseShapes().

◆ siPMShapeData2018_

Shape HcalPulseShapes::siPMShapeData2018_
private

Definition at line 115 of file HcalPulseShapes.h.

Referenced by computeSiPMShapeData2018(), and HcalPulseShapes().

◆ siPMShapeHO_

Shape HcalPulseShapes::siPMShapeHO_
private

Definition at line 114 of file HcalPulseShapes.h.

Referenced by computeSiPMShapeHO(), HcalPulseShapes(), and hoShape().

◆ siPMShapeMCRecoRun3_

Shape HcalPulseShapes::siPMShapeMCRecoRun3_
private

Definition at line 115 of file HcalPulseShapes.h.

Referenced by computeSiPMShapeMCRecoRun3(), and HcalPulseShapes().

◆ theDbService

const HcalDbService* HcalPulseShapes::theDbService
private

Definition at line 120 of file HcalPulseShapes.h.

Referenced by beginRun(), shape(), and shapeForReco().

◆ theDbServiceToken

edm::ESGetToken<HcalDbService, HcalDbRecord> HcalPulseShapes::theDbServiceToken
private

Definition at line 119 of file HcalPulseShapes.h.

Referenced by beginRun(), and HcalPulseShapes().

◆ theShapes

ShapeMap HcalPulseShapes::theShapes
private

Definition at line 122 of file HcalPulseShapes.h.

Referenced by get_all_shapes(), getShape(), HcalPulseShapes(), shape(), and shapeForReco().

◆ Y11MAX203_

constexpr float HcalPulseShapes::Y11MAX203_ = 0.04
static

Definition at line 52 of file HcalPulseShapes.h.

Referenced by generatePhotonTime203().

◆ Y11MAX206_

constexpr float HcalPulseShapes::Y11MAX206_ = 0.08
static

Definition at line 53 of file HcalPulseShapes.h.

Referenced by generatePhotonTime206().

◆ Y11RANGE_

constexpr float HcalPulseShapes::Y11RANGE_ = nBinsSiPM_
static

Definition at line 51 of file HcalPulseShapes.h.

Referenced by generatePhotonTime203(), and generatePhotonTime206().