(UNITEXT For Physics) Michele Livan, Richard Wigmans - Calorimetry For Collider Physics, An Introduction-Springer International Publishing (2019)
(UNITEXT For Physics) Michele Livan, Richard Wigmans - Calorimetry For Collider Physics, An Introduction-Springer International Publishing (2019)
(UNITEXT For Physics) Michele Livan, Richard Wigmans - Calorimetry For Collider Physics, An Introduction-Springer International Publishing (2019)
Michele Livan
Richard Wigmans
Calorimetry
for Collider
Physics, an
Introduction
UNITEXT for Physics
Series Editors
Michele Cini, University of Rome Tor Vergata, Roma, Italy
Attilio Ferrari, University of Turin, Turin, Italy
Stefano Forte, University of Milan, Milan, Italy
Guido Montagna, University of Pavia, Pavia, Italy
Oreste Nicrosini, University of Pavia, Pavia, Italy
Luca Peliti, University of Napoli, Naples, Italy
Alberto Rotondi, Pavia, Italy
Paolo Biscari, Politecnico di Milano, Milan, Italy
Nicola Manini, University of Milan, Milan, Italy
Morten Hjorth-Jensen, University of Oslo, Oslo, Norway
UNITEXT for Physics series, formerly UNITEXT Collana di Fisica e Astronomia,
publishes textbooks and monographs in Physics and Astronomy, mainly in English
language, characterized of a didactic style and comprehensiveness. The books
published in UNITEXT for Physics series are addressed to graduate and advanced
graduate students, but also to scientists and researchers as important resources for
their education, knowledge and teaching.
123
Michele Livan Richard Wigmans
Dipartimento di Fisica Physics and Astronomy Department
Università di Pavia Texas Tech University
Pavia, Italy Lubbock, TX, USA
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
“Dual-readout calorimetry”
For Nadia and Nazzi
with love
Preface
In the past 40 years, calorimetry has become an extremely important technique for
the study of the subatomic structure of matter. Especially experiments that make use
of the increasingly powerful accelerators and storage rings rely heavily on
calorimetry. Calorimeters fulfill a number of crucial tasks, ranging from event
selection and triggering to precision measurements of the four-vectors of individual
particles and jets and of the energy flow in the events (missing energy, etc.). This
development has benefited in no small part from the improved understanding of the
working of these, in many respects somewhat mysterious, instruments.
Much has been learned about calorimetry, primarily thanks to dedicated R&D
projects. This information is contained in a very large number of papers, scattered
in the scientific literature. Over the years, many review articles have been pub-
lished, in which the state-of-the-art is summarized. However, these review papers
usually concentrate on specific aspects and their educational value is typically rather
limited.
We decided to write this book as a compact educational introduction for those
students (and others) who are confronted for the first time with calorimetric particle
detection. It is based on our experience teaching courses on instrumentation at our
universities and at summer schools. We assume a basic understanding of physics,
including quantum mechanics and (sub)atomic physics as taught in a typical
undergraduate curriculum.
The book is subdivided into five parts. After a general introductory chapter,
Part I concentrates on the physics that is relevant for the interactions of particles
with matter, for shower development and the signals produced by calorimeters.
Part II deals with the aspects of shower development that affect the calorimeter
performance in practice. Part III is exclusively dedicated to the specific problems
encountered in hadron calorimetry, and the methods that have been developed to
mitigate the effects of these problems. Part IV deals with the challenges encountered
when operating a calorimeter system in an experiment. This part concerns mainly
calibration issues, which are in practice frequently misunderstood and mishandled,
even by experienced experimentalists. In Part V, we describe the state of the art in
calorimeter performance. We also dedicate in this part a chapter to Particle Flow
ix
x Preface
xi
xii Contents
The term calorimetry means “measurement of heat”. It was first used in thermody-
namics, the branch of physics that relates energy, heat, temperature and work. The
calorie, a unit of energy, was introduced by Nicolas Clément in 1824. It is defined as
the energy needed to increase the temperature of one gram of water by one degree
Celsius at a pressure of one atmosphere. Even though the unit is based on the metric
system, it has become somewhat obsolete after the introduction of the SI system of
units, where it has been replaced by the Joule: 1 cal = 4.184 J.
Figure 1.1 shows examples of calorimeters that can be used for measurements in
this context. The isolated container in Fig. 1.1a can, for example, be used to measure
the latent heat of ice. When adding 10 g of 0◦ ice to 100 g of 20◦ water, the temperature
of the mixture is observed to drop to 11 ◦ C. From this, one can conclude that the latent
heat of ice (E L ) is
Next, the container may be placed in the Sun, and after one hour the temperature
has increased from 11◦ to 38◦ . The container has thus received 27 × 110 = 2970
calories, i.e., 0.825 cal/s or 3.45 J/s. Since the exposed surface area is 30 cm2 , this
corresponds to 1.15 kW/m2 . Of course, such a measurement of the solar constant
requires the container to be perfectly isolated, so that all the heat received from the
Sun is indeed used to increase the temperature of the water, and for nothing else.
In practice, instruments of this type are therefore a bit more sophisticated
(Fig. 1.1b). They are used, for example, to measure the energy released in certain
chemical reactions, or the specific activity of radioactive samples. In the latter appli-
cation, it offers the possibility for a non-destructive assay of fissile materials, such
as plutonium samples. For example, the specific power of the isotopes 238 Pu and
239
Pu amounts to 567 and 1.9 mW/g, respectively. A calorimetric measurement may
therefore provide information about the relative content of the short-lived (88 yr)
238
Pu isotope and hence on the possible production mechanism of the sample.
In nuclear and particle physics, and more recently in astrophysics experiments,
calorimeters are used to measure the energy of particles produced in the reactions
that are being studied. There are similarities and differences with the instruments
mentioned above. They are similar in the sense that a good measurement requires
that the entire energy of the particle is indeed deposited in the sensitive volume
of the instrument, any leakage implies a mis-measurement of that energy. The main
difference derives from the fact that the energies one wants to measure in these exper-
iments are many orders of magnitude smaller. The most energetic manmade particles
are nowadays produced at the Large Hadron Collider at CERN (Genève, Switzer-
land). The protons accelerated in this machine travel at a speed of 99.999999% of
the speed of light and carry a kinetic energy that is about 9000 times larger than
the energy contained in their mass (E = mc2 ). Yet, this kinetic energy is equivalent
to only 0.0000003 cal. Even a simultaneous dump of a million such particles in a
water container will thus have a negligible effect on the temperature of that water.
For this reason, these calorimeters are based on other mechanisms than a tempera-
ture increase of the instrument. These mechanisms may include the production of
ionization charge, light or sound.
Fig. 1.2 Spectrum of γ rays from the decay of 60 Co, measured with a scintillating CsI(Na) crystal
a reasonable precision in this way. Before that, A crucial aspect of the new detector
technique was the use of photomultiplier tubes (PMTs), which made it possible to
convert the scintillation light produced by the crystals into electric signals.
Figure 1.2 shows an example of a γ ray spectrum from the decay of 60 Co, measured
with a CsI crystal doped with sodium. The two peaks represent events in which the
total energy carried by the 1.17 and 1.33 MeV γ s was deposited inside the crystal.
The background “ridge” separating the “backscatter peak” and the “Compton edge”
is the result of events in which some energy leakage out of the crystal occurred.
Details about the various processes that play a role in this are given in the next
chapter.
One of the interesting aspects of Fig. 1.2 is the fact that the two γ s are clearly
recognized as such. In spite of the fact that their energies differ by only 13%, the two
peaks are almost completely resolved by the detector, which is thus said to have a
good energy resolution. As we will see later, there are γ ray detectors with an even
much better energy resolution, but this feature of the scintillating crystals generated
the interest of scientists working on experiments in which particles with much higher
energies than carried by these nuclear γ s were produced.
An example of an experiment where crystals were applied was E-70 [1], one of
the first experiments carried out with the 200 GeV proton accelerator (a world record
at that time) at Fermilab in the early 1970s. Like many experiments in those days, the
experimental setup consisted of a magnetic spectrometer with two arms (Fig. 1.3).
These arms consisted each of a series of wire chambers, which were used to track the
charged particles produced in the target. The momentum of these particles could be
selected with powerful magnets. Neutral particles, and in particular the γ s produced
in the decay of the abundantly produced π 0 s, were of course not affected by the
4 1 Calorimetry—From Thermodynamics to Particle Detection
Fig. 1.3 The magnetic spectrometer of the Columbia/FNAL experiment E-70. From: Appel, J.A.
et al. (1975). Nucl. Instr. and Meth. 127, 495
magnets and did not produce any signals in the wire chambers. They were detected
in arrays of crystals installed at the end of both spectrometer arms.1
The crystals used in this experiment were not based on the production of scintilla-
tion light. Instead, the signals they produced were the result of the Čerenkov effect,
which leads to the production of light by particles that travel faster than the speed of
light in the medium they traverse, lead glass in this case. Details of this mechanism
are described in the next subsection. The main advantage of these lead glass crystals
was the cost, which is only a small fraction of the cost of scintillating crystals.
Whereas these crystals turned out to be excellent detectors for electrons and γ rays
in an energy domain that was several orders of magnitude beyond that for which they
were initially intended, attempts to detect also other particles, such as protons and
charged pions, led to disappointing results. After the first successful use of NaI(Tl)
crystals for detecting GeV-type electrons and γ s by Hofstadter and his colleagues at
SLAC in the late 1960s [3], it was believed that hadrons could be detected with the
same type of precision, provided the detectors were made sufficiently large. To test
this idea, the researchers collected as many NaI(Tl) crystals as they could get their
hands on and assembled these to a calorimeter with a total mass of about 450 kg (0.12
m3 ). They exposed this device to particle beams (predominantly pions, with small
admixtures of electrons and muons of the same energy) of different energies, ranging
from 4–16 GeV. Figure 1.4 shows the signal distribution for 8 GeV π − measured
with this detector [4].
1 The primary experimental goal of this experiment was the detection of high-energy electrons and
positrons produced in the target. These studies led in 1977 to the discovery of the b-quark [2].
1.2 Short History 5
They interpreted the observation that the signals from the pions were, on average,
only half as large as those from the electrons as evidence that about half of the energy
leaked out of this detector. The fact that Monte Carlo simulations indicated that this
leakage was actually much smaller was blamed on a flaw in the simulations. The
results obtained at other energies (4, 12 and 16 GeV) were very similar to those shown
in the figure: pion signals that were typically only half as large as those of electrons of
the same energy, an asymmetric hadronic response function, and a hadronic energy
resolution (σ/E) that was approximately independent of the energy. Their conclusion
from these observations was that the calorimeter was too small to perform well
for hadron detection. However, tests of a huge, 60-ton homogeneous calorimeter
consisting of liquid scintillator built a few years later at Fermilab revealed that shower
leakage was not the dominating problem for the poor hadronic performance. Tests of
that detector showed that pion signals were also substantially smaller than electron
ones for the same energy, and that the hadronic energy resolution essentially did not
improve with increasing energy [5].
We now know that the basic reason for the different hadronic and em calorimeter
responses lies in the fact that in the absorption of hadronic showers, a significant
fraction of the energy carried by the showering particle is invisible, i.e., it does
not contribute to the calorimeter signal. The main source of invisible energy is the
energy used to release nucleons from nuclei, including the nuclear recoil energy.
Additional, smaller contributions come from neutrinos and muons (mainly from π
6 1 Calorimetry—From Thermodynamics to Particle Detection
Fig. 1.5 The Spaghetti Calorimeter, which was instrumental in unraveling the secrets of excellent
hadronic calorimeter performance (a). Photograph courtesy CERN. Example of the experimental
performance that can be obtained with an excellent hadronic calorimeter (b). From: Young, G.R.
et al. (1989). Nucl. Instr. and Meth. A279, 503
and K decay in flight) escaping the detector, and from signal saturation for densely
ionizing particles. These effects are discussed in detail in Chap. 2.
Once the reasons for the poor performance of hadron calorimeters were under-
stood, dedicated efforts could be undertaken to improve that performance. These
efforts, which are described in detail in Chap. 8, were very successful. An example
of a calorimeter developed in the context of this R&D work is depicted in Fig. 1.5a.
This so-called Spaghetti Calorimeter was for 25 years holder of the best hadronic
performance characteristics. A special feature of this calorimeter was that it was
compensating, i.e., it produced signals for electrons and hadrons of the same energy
that were, on average, equally large (unlike the crystals from Fig. 1.4). An example
of results that were obtained with a compensating calorimeter is shown in Fig. 1.5b.
This is a signal distribution from the early days of heavy-ion acceleration with the
CERN SPS. The beam was obtained by accelerating 32 S ions to an energy of 200
GeV per nucleon. However, it turned out that the beam contained “satellites” con-
sisting of ions with the same charge/mass ratio, such as 4 He, 12 C, 16 O, etc. [6]. This
was discovered by dumping the beam of the accelerated ions into a (very good)
calorimeter. The satellites showed up as peaks with intensities at the level of 10−3 in
the spectrum. Note that the resolution in this plot is very similar to that of the γ ray
spectrum shown in Fig. 1.2.
In recent years, even better results than those obtained with compensating
calorimeters have become possible thanks to the use of the so-called dual-readout
technique. This technique and the possibilities it offers are the topic of Sect. 8.3.
1.2 Short History 7
Calorimeters such as the one shown in Fig. 1.5a differ from the crystals in the
sense that they are sampling devices. The functions of particle absorption and signal
generation are in these detectors performed by different materials, e.g., lead and
plastic scintillator in this example. The ratio between the passive material (lead)
and the active one (scintillator) determines the sampling fraction, i.e., the fraction
of the energy of the absorbed particle that is deposited in the active material and
thus contributes to the calorimeter signal. In this calorimeter, that fraction is 2.4%.
An important reason why almost all calorimeters used nowadays in particle physics
experiments are of the sampling type is cost, since the signal generating materials are
typically much cheaper than the absorber material. Also the fact that the density of
the absorber material can be much larger is an important consideration, especially in
4π experiments, since it makes it possible to construct much more compact detectors.
Yet, the small sampling fraction does not necessarily imply that the energy resolu-
tion of sampling calorimeters is worse than for homogeneous ones, i.e., calorimeters
such as crystals where the functions of particle absorption and signal generation are
performed by the same material. As a matter of fact, sampling calorimeters such as
the one shown in Fig. 1.5a measured hadrons with an energy resolution that was a
factor of three better than for the best homogeneous calorimeter [5, 7].
However, the good energy resolution that can be obtained for high-energy particles
was not the only, and not even the main, reason why calorimeters have become the
most important components of most modern experiments in particle physics, and
definitely in experiments carried out at high-energy particle colliders such as the
Large Hadron Collider at CERN. It turned out that calorimeters are ideal instruments
for providing information that is crucial in the context of the physics to be studied
in such experiments and, especially, that they can provide this information very fast.
Figure 1.6 shows a picture of one of the first large calorimeters that operated in an
accelerator based experiment. The experiment, WA1, was intended to study interac-
tions induced by neutrinos produced by the Super Proton Synchrotron at CERN. This
detector had an instrumented mass of about 1 kiloton. It consisted of large slabs of
iron, interleaved with active material that generated the signals. In the upstream part
of the detector (the first 6 m), this active material consisted of scintillating plastic.
There were in total 105 such sampling layers. This part of the detector served as the
neutrino target. In the rear part of the detector, the iron slabs were thicker (15 cm
instead of 5 cm), and they were interleaved with wire chambers, which allowed the
localization of passing charged particles with good precision.
By comparing to total signal registered in all scintillator plates combined with
preset thresholds, the occurrence of neutrino interactions in the target section could
be determined in real time. The rear part of the detector made it possible to determine
the trajectory of muons in case these were produced in the interactions. Since the
iron was also magnetized, the momenta of such muons could be measured as well.
The detection of muons was a crucial aspect of this experiment, since these particles
distinguished charged current neutrino reactions from neutral current ones. More-
over, dimuon events were a signature of the production of charmed particles. These
were hot topics in the 1970s, and this experiment made major contributions in this
8 1 Calorimetry—From Thermodynamics to Particle Detection
Fig. 1.6 The WA1 neutrino detector combination that operated at CERN from 1976 to 1984. Large
slabs of absorber material (iron) are interleaved with layers of a plastic scintillator. The rear part of
the detector, located at the left-hand side of the picture, is instrumented with wire chambers intended
for tracking muons generated in charged current interactions and/or charmed particle production.
Photograph courtesy CERN
domain, as well as to the understanding of details of the nucleon structure, for which
(anti-)neutrinos are very important tools.
The role of calorimeters became even more important with the advent of colliding-
beam accelerators. A crucial aspect of experiments at such colliders is hermetic cov-
erage, i.e., the detectors have to cover as large a fraction as possible of the 4π stera-
dians surrounding the interaction point of the colliding particles. If that is achieved,
the detectors can provide crucial information on the energy flow in the events. For
example, a comparison between the total signals measured in any two hemispheres
gives information about the missing transverse energy. This may indicate the pro-
duction of an energetic neutrino, which does not leave a trace in the calorimeter. The
calorimeter data can also be used to identify other types of particles produced in the
interactions. For example, the energy deposit profile and/or the time structure of the
signals are important tools to recognize electrons, and the fact that a charged particle
penetrates the calorimeter indicates that it is a muon.
The total measured calorimeter signal is a measure for the centrality of the collision
(deep inelastic scattering), and the concentration of energy in certain regions of the
calorimeter (“jets”) is a signature for a fragmenting quark of gluon produced in
the interactions. Since the rate of the particle interactions in such colliders may be
extremely high (e.g., O(GHz) in the Large Hadron Collider), the calorimeter is also
crucial for determining which of these events should be retained for further, offline,
1.2 Short History 9
Fig. 1.7 The UA2 calorimeter that operated at CERN’s proton–anti-proton collider (S p p̄S) from
1980 to 1986 and played a crucial role in the discovery of the W and Z bosons [8]. Photograph
courtesy CERN
inspection. This triggering capability is possibly the most important reason for the
dominant role of calorimeters in modern collider experiments.
In 1982, the UA1 and UA2 (see Fig. 1.7) experiments discovered the W intermedi-
ate vector boson [9, 10] on the basis of the signatures mentioned above: an energetic
charged lepton (electron, muon), in combination with missing transverse energy
(caused by the neutrino accompanying the charged lepton in the decay W → lν.
This discovery contributed in no small measure to the way in which experiments are
designed nowadays.
A relatively recent development is that of imaging calorimeters. These are based on
liquid-argon technology, instrumented as Time Projection Chambers, and produce
event images that rival those from bubble chambers (Fig. 1.8). These devices are
typically used in neutrino oscillation experiments, located at different distances from
the source [11]. For example, neutrinos produced at Fermilab are (scheduled to be)
detected at distances varying from a few km (μBOONE) to 1,300 km (DUNE) [12].
Whereas measuring the energy carried by these particles is an important goal of such
experiments, they typically emphasize complete imaging of the detected events.
10 1 Calorimetry—From Thermodynamics to Particle Detection
Fig. 1.8 Examples of cosmic ray events in a large liquid argon TPC. Diagram a shows an interaction
by a charged particle, in which three charged particles are produced, plus possibly some neutral
objects. Several δ-rays are also visible along the track of the particle that re-interacts after about 2
m. Diagram b shows a textbook example of a developing em shower in argon. From: Ereditato, A.
et al. (2013). J. Instrumentation 8, P07002
Fig. 1.9 A sector of the CMS ECAL, which consists of scintillating PbWO4 crystals. Photograph
courtesy CERN
[22]. Liquid argon is used on a large scale for imaging calorimeters (Fig. 1.8), but
in this case the ionization charge produced in the particle absorption process is the
source of the signals.
By far the largest homogeneous calorimeters in use are based on the production of
Čerenkov light in water. SuperKamiokande (Fig. 1.10) operates a detector containing
55,000 ton of ultrapure water in a Japanese mine [23]. Even larger instruments
use the natural environment for particle detection. For example, IceCube [24] has
instrumented 1 km3 of Antarctic ice, and several other efforts are under way to do
the same in the Mediterranean [25]. The Auger experiment [26] uses (scintillation)
light production in a very large volume of the Earth’s atmosphere to measure the
highest energy cosmic rays that reach us from outer space.
The calorimeter systems used in experiments at the highest energy particle col-
liders are typically of the sampling type, mainly because of cost and compactness
considerations, although crystals are still popular at e+ e− rings, especially those
operating at the J/Ψ [16] and Υ [27] resonances. The active material in sampling
calorimeters generates signals in the form of (scintillation or Čerenkov) light or
electric charge, by direct ionization or the production of electron-hole pairs (semi-
conductors). The absorber material is typically lead for the detection of em particles,
and iron or copper for hadrons. In some cases, depleted uranium has been used, based
on the (incorrect) assumption that this would provide unique advantages.
12 1 Calorimetry—From Thermodynamics to Particle Detection
Fig. 1.10 Picture taken inside the Super-Kamiokande detector, showing part of the 11,200 PMTs
that detect the Čerenkov light when the vessel is filled with 55 kilotons of ultrapure water. Note the
three technicians standing on the left. Photograph courtesy Super-Kamiokande
The structure of these sampling calorimeters has evolved over time. Initially, it
was assumed that the active and passive layers had to be arranged in the from of a
“sandwich” structure, with the layers oriented perpendicular to the travel direction
of the envisaged particles (Fig. 1.11a). However, in the 1980s it was discovered
that there was absolutely no fundamental reason that made this necessary, and that
other arrangements could offer major practical advantages, while barely affecting
the calorimeter performance. In many modern calorimeters, the active layers are
therefore distributed in very different ways inside the absorber structure. For example,
in the ATLAS hadron calorimeter, tiles of plastic scintillator are oriented along the
direction of the incoming particles (Fig. 1.11b). And in the CMS HF calorimeter,
large numbers of thin quartz fibers that run in the same direction as the incoming
particles collect Čerenkov light produced in the absorption process and transport it
to the rear end of the calorimeter (Fig. 1.11c). In the ATLAS em calorimeters, the
active medium (liquid argon in this case) is contained in a sort of “accordion-like”
structure of containers inside the lead absorber structure (Fig. 1.12).
The reason for these particular choices lies in the fact that the mentioned calorime-
ters have to operate in a 4π environment. The chosen structure offers some clear
advantages compared with the “sandwich” one. For example,
1.3 Types of Calorimeters Used as Particle Detectors 13
Fig. 1.11 Different sampling structures used in calorimeters. The “sandwich” structure, used in
the ZEUS calorimeters (a). The “tile” structure, used in the ATLAS hadron calorimeter (b). Image
courtesy CERN. The “spaghetti” structure, used in the very forward region of CMS (c). Photograph
courtesy CERN
Fig. 1.12 The “accordion” structure used in the ATLAS em calorimeter. From: Aad, G. et al. (2010).
Eur. Phys. J. C70, 755
14 1 Calorimetry—From Thermodynamics to Particle Detection
• The so-called “dead” volume, i.e., the part of the space surrounding the interaction
point of the beams that is not covered by the calorimeter, is limited. This dead
volume is the result of the need to support the multi-ton detectors and to export
the signals they produce to the outside world.
• It makes it easier to realize a projective detector geometry
• In the case of liquid argon (LAr), It limits the distance between the locations
where the ionization charge is produced and the front-end electronics. This limits
the capacitance and thus the electronic noise of the system.
In calorimeters that are not part of a 4π detector system, there is still a strong
preference for the much simpler sandwich structure. As an example, we mention the
MINOS detector, which detects neutrinos produced at Fermilab, 800 km away [28].
This calorimeter consists of 2.54 cm thick iron absorber plates interleaved with 1 cm
plastic scintillator (cf. Fig. 1.6).
In this section, the various mechanisms through which calorimeter signals are gen-
erated are described. The discussion is limited to the general aspects that make these
mechanisms useful for application as active media for calorimetric particle detection.
Specific features, such as signal quenching effects for densely ionizing particles or
non-linearities caused by active media operating in the Geiger mode, are elaborated
upon in the relevant later chapters.
1.4.1 Scintillation
When charged particles traverse matter, they lose energy through the electromagnetic
interaction with the Coulomb fields of the electrons. This energy may be used to ionize
the atoms or molecules of which the traversed medium is composed, or to bring these
atoms/molecules into an excited state. Scintillation is a phenomenon associated with
the latter process.
The excited atomic or molecular states are unstable. Usually, the excited atom or
molecule quickly returns to the ground state. In this process, the excitation energy
is released in the form of one or more photons. The timescale of this process is
determined by the excitation energy, by the number of available return paths, and
by the quantum numbers of the states involved (wavefunction overlap). When the
energy differences are such that the emitted photons are in the visible domain, this
process is called fluorescence or scintillation. Typical timescales range in that case
from 10−12 to 10−6 s, although exceptions in either direction may occur. In general,
the timescales get shorter as the molecules get more complex. This can be simply
understood from the fact that the density of excited states, and therefore the number of
1.4 Techniques Used for Signal Generation in Calorimetry 15
different ways in which an excited molecule can get rid of its excess energy increases
sharply with the complexity of the molecule.
Relatively simple scintillating crystals, such as NaI(Tl) and BGO, have decay
times of several hundred ns, orders of magnitude longer than the decay times of
complex organic scintillators, such as those based on anthracene or polystyrene.
Historically, scintillation was the first physics process to be used for the generation
of calorimetric signals. And until this day, a large number of calorimeters in a wide
variety of particle physics experiments rely upon scintillation light as the prime
source of information. Two inventions have played a crucial role in the development
of scintillator-based particle detectors in general, and calorimeters in particular:
• The photomultiplier tube. Almost 70 years old, the PMT which allows the con-
version of single photons into electric signals, is still playing a crucial role in
many experiments. The sensitivity to single photons and the essentially noise-free
amplification offered by these devices are very attractive features [29]. Although
some new devices based on semiconductor applications (such as the Hybrid Photo
Detector [30], the Avalanche Photo Diode [31] and especially the silicon photomul-
tiplier [32, 33]) have more or less successfully addressed some PMT weaknesses,
such as the sensitivity to external magnetic fields, the essentially noise-free signal
amplification offered by PMTs is still an extremely attractive feature.
• Wavelength shifters. These devices absorb the scintillation light and re-emit it at
a lower energy (longer wavelength). This development made it possible to apply
scintillator-based calorimeters in experiments requiring hermetic coverage, such
as the 4π experiments in a colliding-beam setup. The light produced in scintillator
plates oriented perpendicular to the direction of the incoming particles can be
wavelength-shifted and at the same time redirected towards the rear end of the
calorimeters, where it can be converted into electric signals. Figure 1.13 shows
schematically the readout of scintillator calorimeters with and without wavelength-
shifting plates. The price to be paid for these advantages is a loss of light, because
of inefficiencies in the process and a longer signal duration, since the wavelength
shifters are usually somewhat slower than the scintillators whose light they shift.
The development of plastic optical fibers has also greatly influenced the design of
scintillator calorimeters, especially after it became clear that there is no reason why
active material has to be oriented in a particular direction. Scintillating fibers usually
consist of a polystyrene core (index of refraction n = 1.59), surrounded by one or
several layers of cladding with (gradually) lower values of n.
Unlike the optical fibers used for telecommunication purposes, which are designed
to transport light injected along the fiber axis, the scintillating fibers used in particle
physics experiments are both the source of the light (generated isotropically) and the
medium through which this light is transported to a place where it can be converted
into an electric signal.
The fraction of the light that is trapped is proportional to the
numerical aperture n 2core − n 2clad , and most of the light is traveling near the critical
angle, defined as θcr = arcsin (n clad /n core ) [35].
16 1 Calorimetry—From Thermodynamics to Particle Detection
Fig. 1.13 Schematic of the readout systems of scintillator calorimeters without (a) and with (b)
wavelength-shifting plates [34]
Apart from these chemically doped optical fibers, undoped plastic fibers are also
being applied in particle detectors. For example, clear plastic fibers based on a PMMA
core (n = 1.49) surrounded by lower-index (fluorinated plastic) cladding material
is used to detect Čerenkov light produced in the particle absorption process (see
Sect. 1.4.2).
Optical fibers are being used in many calorimeters, either as the active medium
sampling the showers, or as wavelength shifters, converting the scintillation light,
e.g., blue light from scintillator plates, to a longer wavelength (e.g., green) and
transporting it to light detectors located in a convenient position.
Among the advantages offered by such fibers, we mention
• The perfectly hermetic calorimeter structure that can be achieved,
• The very high sampling frequency (good energy and position resolution!) that can
be obtained using fibers as the active medium,
• The high signal speed that can be obtained,
• The arbitrary granularity allowed by the fiber structure,
• The high light yield that can be achieved, and
• The excellent cost/performance ratio
We elaborate on the relevance of these factors, and on the specific advantages of
plastic as active material for hadron calorimeters in the next chapters.
When a charged particle travels faster than the speed of light in a certain medium
(v > c/n, or β = v/c > 1/n, with n the medium’s index of refraction), it loses energy
by emitting Čerenkov radiation. This radiation is emitted at a characteristic angle,
the Čerenkov angle θC = arccos (nβ)−1 , with the direction of the particle. Therefore,
1.4 Techniques Used for Signal Generation in Calorimetry 17
Fig. 1.14 The principle of Čerenkov light emission by a superluminal particle. In a time t, the
particle travels a distance vt, while the light it emits travels a distance ct/n. The wavefronts of the
light emitted by such a particle form a cone with half-opening angle θC
this radiation forms a cone with half-opening angle θC (Fig. 1.14). The amount of
energy is proportional to sin2 θC [29, 36].
The spectrum of this Čerenkov radiation exhibits a characteristic 1/λ2 dependence
and, therefore, the visible part of the Čerenkov spectrum is experienced as blue
light. This blue light can be abundantly observed in highly radioactive environments,
e.g., the moderating liquids in nuclear reactors. It is also a source of light deep in the
oceans, where it is created by penetrating cosmic rays.
The emission of Čerenkov light is only a very minor source contributing to the
energy loss of the particles. For example, in water, a charged particle with β 1
loses about 400 eV/cm in the form of visible Čerenkov photons. That is some four
orders of magnitude less than its energy loss through other processes, in particular
ionization (2 MeV/cm).
Since the Čerenkov mechanism is sensitive to the velocity of particles, it can be
used to determine the mass of particles of which the momentum has been determined
by means of deflection in a magnetic field. A variety of devices (threshold Čerenkov
counters, differential Čerenkov counters, ring imaging Čerenkov detectors) have
been developed to separate electrons, pions, kaons, protons and deuterons from each
other, exploiting this effect.
As we will see in later chapters, calorimeters based on the detection of Čerenkov
light exhibit some interesting properties, which may be ideal for certain very specific
applications, e.g., jet detection very close to the beam pipe in LHC experiments, or
dual-readout calorimetry (Sect. 8.3).
A very important aspect of Čerenkov light is its instantaneous character. There are
no delaying factors, such as the lifetime of a metastable excited state, which affect
the time characteristics of detectors based on scintillation light. Therefore, Čerenkov
detectors, including calorimeters, are the instruments of choice for experiments in
which ultimate signal speed is required.
18 1 Calorimetry—From Thermodynamics to Particle Detection
Fig. 1.15 Schematic drawing of the ATLAS LAr calorimeter system. In the middle, the barrel
cryostat hosts two electromagnetic wheels, at each end the end cap cryostats host two concentric
electromagnetic wheels, two hadronic wheels and three forward calorimeter wheels. Image courtesy
CERN
1.4.3 Ionization
When charged particles traverse matter, they may ionize the atoms of which this
matter consists. One or several electrons are released from their Coulomb field in
this process, leaving behind an ionized atom. Collection of these liberated electrons
is applied as the signal-producing technique in a wide variety of particle detectors.
The electrons produced along the trajectory of the ionizing particle may or may not
be amplified in this process.
In ionization chambers based on liquid media, no amplification takes place. A
potential difference applied over the gap containing the liquid separates the electrons
from the ions. The electrons are collected at the anode, the ions at the cathode. In
order for this method to work properly, the mean free path of the electrons in the liquid
should be long, considerably longer than the size of the gap between the electrodes.
Therefore, noble liquids such as argon, krypton and xenon, which have no desire
to capture loose electrons wandering around since all the electronic shells of their
atoms are filled, are the media of choice in these detectors. The fact that the use
of such liquids requires cryogenic operating conditions typically represents a major
experimental challenge. Especially, hermetic coverage in a 4π environment is very
hard to achieve (Fig. 1.15).
To ensure a sufficiently long mean free drift path for the electrons, very stringent
purity standards have to be met in noble-liquid ionization chambers. In particular,
1.4 Techniques Used for Signal Generation in Calorimetry 19
since the energy acquired by the electrons is proportional to the electric field strength,
this multiplication process works best in the immediate vicinity of the anode (small
values of r ). For this reason, the anode is often made of very thin (30 µm) wires.
Wire chambers may operate in a variety of modes, depending on the type of gas
mixture and on the voltage difference between anode and cathode: the proportional
mode, the streamer mode, the Geiger mode, etc. The time needed for the charge
to arrive at the anode may provide information about the spatial coordinates of the
particle that caused the signals. This principle is applied in drift chambers and in
Time Projection Chambers [40, 41]. A large number of calorimeter systems rely on
wire chambers or tubes of some sort to provide the experimental signals. Especially
when very large surface areas have to be covered, this is often the most cost effective
solution available.
Thin wires make such detectors very fragile. A short circuit caused by a broken
wire may wipe out a large section of the detector system, as has been experienced
by several experiments. In the past 25 years, we have witnessed an enormous devel-
opment of detectors in which the strong electric fields needed for the multiplication
of the ionization charge are created in some other way than by means of thin wires.
The general term for such devices is “micropattern gas detectors.” Thin gaps, tiny
holes or some micromesh structure provide the field-shaping geometry, and new
developments in photolithography and microelectronics have led to a new class of
extraordinary detectors [41]. Not surprisingly, these devices have become increas-
ingly popular with designers of tracking systems for new experiments.
Finally, there are also solid state devices that are being used as detectors of charges
produced by passing ionizing particles. Silicon, germanium and gallium arsenide
have all been applied for particle detection since more than 50 years. The outer-shell
atomic levels of these semiconductor crystals exhibit a band structure, consisting of
a valence band and a conduction band, separated by a “forbidden” energy gap, in
which no energy levels are available. An ionizing particle passing through such a
semiconductor excites electrons from the valence band into the conduction band. For
every electron that jumps into the conduction band, a hole remains in the valence
band. This hole is positive relative to the sea of negative electrons in the valence
band. Therefore, it acts as a positive charge carrier, and its movement through the
semiconductor crystal constitutes an electric current, just as does the movement of
the electrons in the conduction band. The electron–hole pairs created by ionizing
particles may be collected by means of an electric field.
This technique has several advantages. The energy gap between the valence and
conduction bands is very narrow, typically of the order of 1 eV and, therefore, very
little energy is required for the production of one electron–hole pair. For example,
in silicon every 3.6 eV of deposited radiation energy yields one electron–hole pair.
This is typically one order of magnitude less than the energy needed to produce
one electron–ion pair in gases and two orders of magnitude less than the energy
required for the production of one photoelectron in scintillation counters. Therefore,
semiconductor crystals offer the potential of excellent energy resolution in detectors
in which fluctuations in the number of primary charge carriers are the limiting factor
for resolution.
1.4 Techniques Used for Signal Generation in Calorimetry 21
Fig. 1.16 Example of a silicon photomultiplier, and the spectra it produces when exposed to two
different sources of visible light [34]
(a) Some elementary excitations require very little energy. For example, Cooper
pairs in superconductors have binding energies in the μeV–meV range and may
be broken by phonon absorption.
(b) The specific heat for dielectric crystals and for superconductors decreases to
very small values at these low temperatures.
(c) Thermal noise in the detectors and the associated electronics becomes very small.
(d) Some materials exhibit specific behavior (e.g., change in magnetization, latent
heat release) that may provide detector signals.
The devices that have been proposed in this context are typically still in the early
phases of the R&D process. In many cases, this R&D involves fundamental research
in solid-state physics and materials science. However, some devices have reached
the stage where practical applications have been successfully demonstrated. Among
these, we mention
• Bolometers, which are based on principle (b). These are calorimeters in the true
sense of the word, since the energy deposit of particles (in an insulating crystal at
very low temperature) is measured with a resistive thermometer.
• Superconducting Tunnel Junctions, in which the quasi-particles and -holes (Cooper
pairs) excited by incident radiation tunnel through a thin layer separating two
superconducting materials.
• Superheated Superconducting Granules, which are based on the fact that certain
type I superconductors can exhibit metastable states, in which the material remains
superconducting in external magnetic fields exceeding the critical field. These
detectors are usually prepared as a colloid of small (diameter 1–100 µm) metallic
granules suspended in a dielectric matrix (e.g., paraffin). Heat deposited by an
interacting particle may drive one of several granules from the superconducting
to the normal state. The resulting change in magnetic flux (disappearance of the
Meissner effect) may be recorded by a pickup coil.
Fig. 1.17 The principle on which a cryogenic calorimeter is based (a). Spectrum of X-rays of
titanium nitrate, measured with a cryogenic calorimeter, and with a standard Si(Li) semiconductor
detector (b). From: Pretzl, K. (2000). Nucl. Instr. and Meth. A454 114
At the other end of the energy scale covered by particle calorimeters, i.e., in the Joule
domain (>1018 eV), attempts are being made to use acoustic signals, especially for
the detection of cosmic neutrinos with ultra-high energies. The pressure signals are
produced by the particle showers that evolve when neutrinos interact with nuclei in
water. The resulting energy deposition in a cylindrical volume of a few centimeters
24 1 Calorimetry—From Thermodynamics to Particle Detection
in radius and several meters in length leads to a local heating of the medium which
is instantaneous on the hydrodynamic time scale. This temperature change induces
an expansion or contraction of the medium depending on its volume expansion
coefficient. After propagating several hundreds of meters in sea water, the pressure
pulse created in this way has a characteristic frequency spectrum that is expected to
peak around 10 kHz [48] and could provide detailed information about the particle
that created it.
The study of acoustic particle detection is motivated by two potentially major
advantages over an optical neutrino telescope:
1. The attenuation length in sea water is about 5 km (1 km) for 10 kHz (20 kHz)
signals. This is one to two orders of magnitude larger than for Čerenkov light
detected by the telescope.
2. The sensors can be more compact and the readout electronics simpler for acoustic
measurements.
In principle, this offers the promise of a much simpler telescope covering a much
larger fiducial volume. Potential disadvantages include:
1. The speed of sound is small compared to that of light. Therefore, coincidence
windows between two spatially separated sensors have to be relatively large.
2. There is substantial acoustic background in the sea (whales!).
These disadvantages limit the applicability of this technique to the very high end
of the investigated energy spectrum of the neutrinos. To particle physicists, >1018
eV is an extremely high energy. However, we are dealing here with a macroscopic
phenomenon, “mini explosions” in which a total energy of ∼1 Joule is released.
In the Mediterranean, the ANTARES telescope infrastructure has been used to
study acoustic particle detection [49]. The so-called AMADEUS system consists of
six “acoustic clusters,” each comprising six acoustic sensors (hydrophones) that are
arranged at distances of ≈1 m from each other. Figure 1.18a shows one such cluster.
The hydrophones use piezo-electric elements for the broad-band recording of signals
with frequencies up to 125 kHz (Fig. 1.18b).
At first sight, the South Pole ice cap seems to be a more favorable environment than
the Mediterranean Sea for detecting high-energy neutrinos using acoustic signals.
The noise level is expected to be much lower, because of the complete absence of
sound-emitting life forms. The speed of sound is also much higher in ice than in
water, which means that coincidence windows can be much shorter, thus reducing
the probability for fake trigger signals. Theoretical estimates put the attenuation
length for sound waves with a typical frequency of 20 kHz at 8 km [50], dominated
by absorption rather than scattering.
Given these favorable conditions, the option of installing an array of acoustic
detectors, to be operated in conjunction with the light detecting IceCube instruments,
seemed very attractive, especially because the fiducial volume could be substantially
increased, perhaps by several orders of magnitude. For this reason, a subgroup of
IceCube researchers set out to measure the relevant parameters in the IceCube envi-
ronment. This project became known as SPATS, the South Pole Acoustic Test Setup.
1.4 Techniques Used for Signal Generation in Calorimetry 25
Fig. 1.18 Drawing of a standard acoustic cluster with hydrophones (a), and a schematic drawing
of one of the hydrophones (b) tested in the ANTARES setup. From: Aguilar, J.A. et al. (2005).
Nucl. Instr. and Meth. A626–627, 128
An array of acoustic transmitters and sensors was set up in some of the IceCube
holes, at depths varying from 80 to 500 m below the surface and spaced horizontally
up to 690 m [51]. Some results of the tests include [52]:
1. The speed of sound quickly increases with depth and reached a maximum value
of ∼3,900 m/s at a depth of 200 m.
2. The noise level decreases slightly with increasing depth, it is stable in time and
exhibits no correlation with surface conditions (wind, temperature). There is some
correlation with human activity, especially if that involves heavy machinery.
3. The relationship between the measured noise floor and the minimum detectable
neutrino energy is not clear.
4. The attenuation length was consistently measured, using a variety of methods,
sensors, distances and positions, to be ∼300 m.
The last result was a great surprise. It is not understood where the calculations went
so wrong. But in any case, it has important consequences for the design of a hybrid
detector system, which would have to be scaled down considerably in size from the
originally envisaged 100 km3 to have any sensitivity at all.
In the meantime, more than 35 years after the idea of acoustic neutrino detection
was first proposed, not a single neutrino has ever been observed producing both
optical and acoustic signals, neither at the South Pole, nor in the Mediterranean Sea,
nor in Lake Baikal (where this technique was also tried).
26 1 Calorimetry—From Thermodynamics to Particle Detection
References
49. Aguilar, J.A., et al.: Nucl. Instrum. Methods A626–627, 128 (2005)
50. Price, B.J.: J. Geophys. Res. 111, B02201 (2006)
51. Abbasi, R., et al.: Astropart. Phys. 34, 382 (2011)
52. Laihem, K.: Nucl. Instrum. Methods A692, 192 (2012)
Part I
The Basics of Calorimetry
Chapter 2
Interactions of Particles with Matter
2.1 Introduction
In the absorption process, the energy carried by the incoming particle is eventually
distributed in a typically large number of intermediate steps to electrons and (con-
stituents of) nuclei that are part of the absorber structure. Eventually, all the available
energy of the incoming particle is shared among a very large number of these so
called shower particles, each of which carries so little kinetic energy that all that is
left for them to do is to ionize or excite the atoms/molecules of the medium they
traverse or (in the case of neutrons) scatter off atomic nuclei before being captured
by one. The ionization charge and/or the light produced in these processes forms the
basis of the calorimeter signals.
In this part, we present the basic facts that determine the performance of calorime-
ters. In this chapter, the various processes that play a role in the absorption of energetic
particles in matter are described. In Chap. 3, the entire absorption process is being
considered. And finally, in Chap. 4 some important experimental phenomena that
affect the calorimeter signals are discussed.
The best known energy-loss mechanism contributing to the absorption process is the
electromagnetic interaction experienced by charged particles traversing matter. The
particles ionize the medium, if their energy is at least sufficient to release the atomic
electrons from the Coulomb fields generated by the atomic nuclei. This process also
forms the principle on which many particle detectors are based, since the liberated
electrons may be collected by means of an electric field and yield an electric signal.
Already at energies above 100 MeV, and in many materials even at energies consid-
erably lower than that, by far the principal source of energy loss by electrons and
positrons is bremsstrahlung. In their passage through matter, electrons and positrons
radiate photons as a result of the Coulomb interaction with the electric fields gener-
ated by the atomic nuclei. The energy spectrum of these photons falls off as 1/E. It
extends, in principle, all the way to the energy of the radiating particle, but in general
each emitted photon carries only a small fraction of this energy.
In this process, the electron (or positron) itself undergoes a (usually small) change
in direction. This deviation depends on the angle and the energy of the emitted
photon, which in turn depend on the strength of the Coulomb field, i.e., on the Z of
the absorber material.
These radiative processes, which dominate the absorption of high-energy electrons
and positrons, play a role for any charged particle traversing matter. However, for
heavier charged particles the competition with ionization as the main source of energy
loss only starts to play a role at much higher energies. The critical energy, εc , which
may be defined as the energy at which the average energy losses from radiation
processes equal those from ionization, is higher by a factor (m/m e )2 , where m and
m e are the particle and the electron mass, respectively. The critical energy of the
next-lightest charged particle, the muon (m μ ≈ 207m e ), is thus about 40,000 times
larger than that of the electron. For hadrons such as pions (m π ≈ 273m e ), kaons
(m K ≈ 966m e ) and protons (m p ≈ 1836m e ), the critical energy is correspondingly
larger, but is in practice irrelevant for calorimetry because of the nuclear interactions
to which these particles are subject.
The energy loss mechanisms for electrons and positrons are governed by the
laws of Quantum Electrodynamics (QED) and can be calculated with a high degree
of accuracy. The relative importance of ionization and radiation losses at a given
energy depends primarily on the electron density of the medium in which the shower
develops. This density is roughly proportional to the (average) Z of the medium,
since the number of atoms per unit volume is, within a factor of about two, the same
for all materials in the solid state.
2.2 The Electromagnetic Interaction 33
Fig. 2.1 Cross sections for the processes through which the particles composing electromagnetic
showers lose their energy, in various absorber materials. To the left are shown the cross sections for
pair production, Compton scattering and photoelectric effect in carbon (a), iron (b) and uranium
(c). To the right, the fractional energy losses by radiation and ionization are given as a function of
the electron energy in carbon (d), iron (e) and uranium (f) [1]
Results of calculations on the energy loss mechanisms for electrons are shown as
a function of energy in Fig. 2.1, for three different absorber materials: carbon (Z = 6,
Fig. 2.1d), iron (Z = 26, Fig. 2.1e) and uranium (Z = 92, Fig. 2.1f) [2]. The energy
at which the energy losses from ionization equal those from radiation decreases from
about 95 MeV for carbon, to 28 MeV for iron, to 9 MeV for uranium.
The Particle Data Group [3] prefers a slightly different definition of the critical
energy (at least for electrons), originally formulated by Rossi [4]. In this definition,
εc is the energy at which the ionization loss per radiation length (X 0 ) equals the elec-
tron energy. The radiation length is defined as the distance over which high-energy
electrons and positrons lose, on average, 63.2% (1 − 1/e) of their energy through ra-
diation. This variable was introduced to describe the development of electromagnetic
showers in a material independent way (see Sect. 3.3.1 for details).
dE
(ΔE)ion = X0 = E (2.1)
dx ion
This definition would thus be equivalent to the first one if the energy loss to
bremsstrahlung were given by
34 2 Interactions of Particles with Matter
dE E
= (2.2)
dx brems X0
which is true at very high energies, where ionization losses are negligible, but which
is only an approximation in the energy regime near εc . The difference between the
two definitions is illustrated in Fig. 2.2, for electrons in copper. In this figure, Eq. 2.2
is represented by the dashed line. Using this alternative definition, the PDG has fitted
the d E/d x data tabulated by Pages [2] and gives the following expressions for the
critical energy:
610 MeV
εc = (2.3)
Z + 1.24
710 MeV
εc = (2.4)
Z + 0.92
for gases. These formulae fit the data from the mentioned d E/d x tables to within
∼4%, with the largest deviations occurring at the highest Z values. For example, for
uranium, Eq. 2.3 gives εc = 6.54 MeV, while the data tabulated in [2] fulfill Rossi’s
condition at an energy of 6.75 MeV.
The εc values found in this way are systematically smaller than the ones following
from the other definition, where εc is the energy at which ionization losses equal
radiation losses. The differences range from ∼15% for carbon to ∼35% for uranium
(see also Fig. 2.2).
In the following, we will use the definition of Rossi and the PDG.
2.2 The Electromagnetic Interaction 35
The other particles subjected to only the em interaction, the muons, behave, at
the same energies, in a very different way than electrons and positrons. Whereas
radiative processes dominate the energy loss of the latter particles, muons traversing
dense material lose their energy up to very high energies (100 GeV or higher) pri-
marily through ionization and δ-rays. These mechanisms account for energy losses
of typically only 1–2 MeV g−1 cm2 and, therefore, it takes very substantial amounts
of material to absorb high-energy muons.
For this reason, experiments in which cosmic muons constitute a major source of
undesirable background have to be located in deep mines or under high mountains,
since these muons may sometimes penetrate several kilometers of the Earth’s crust.
For the same reason, the CERN high-energy neutrino beam that was used for many
experiments in the West Area (1963–1998) was equipped with a 300 m long iron
shield. The neutrinos were produced from pion and kaon decay (π, K → νμ μ) and
the muons had to be absorbed in the space between the production target and the
neutrino detectors. In iron, muons lose energy at a rate of about 1.1 GeV/m (about a
factor of three higher than in the soil of which the CERN site is composed).
Higher-order QED processes, such as bremsstrahlung and direct e+ e− pair pro-
duction, do also occur in muon absorption. However, compared with electrons, they
are suppressed by a scale factor of (m μ /m e )2 ≈ 40, 000. Therefore, the critical en-
ergy at which muons lose, on average, equal amounts of energy through radiation
and ionization is at least 200 GeV. Just as for electrons, the contribution of these
higher-order QED processes to a muon’s energy loss is strongly Z dependent [5].
For example, the average energy loss of 500 GeV muons in lead increases (with
respect to ionization losses) by a factor of 5.8 because of these effects. In iron, this
factor amounts to 2.5 and in aluminium 1.8.
At energies below 100 GeV, the average energy loss is primarily determined by
ionization, in all absorber materials. The mean energy loss per unit path length,
d E/d x, is given by the Bethe–Bloch formula [4]:
Z 1 1 2m e c2 β 2 γ 2 Tmax δ
− d E/d x = K z 2 ln − β 2
− (2.5)
A β2 2 I2 2
in which Tmax represents the maximum kinetic energy that can be imparted to an
electron in a single collision, I is the mean excitation energy of the absorber material,
δ a correction term describing the density effect, and the proportionality constant K
equals 4π NA re2 m e c2 .
The quantity d E/d x, which is often referred to as the specific ionization or the
ionization density, has a characteristic energy dependence, which is governed by
the product of the velocity (β) and the Lorentz factor (γ ) of the particles (Fig. 2.3).
For relativistic muons, d E/d x falls rapidly with increasing β, reaches a minimum
value near β = 0.96, and then exhibits what is called the relativistic rise, to level off
at values of 1–2 MeV g−1 cm2 in most materials. Muons, or other particles with unity
36 2 Interactions of Particles with Matter
Fig. 2.3 The average energy loss per unit path length (−d E/d x) for positive muons in copper,
given as a function of the product of the Lorentz variables βγ . For muon momenta in the range from
∼5 MeV/c to ∼50 GeV/c, this energy loss is well described by the Bethe-Bloch formula (Eq. 2.5.)
From: Particle Data Group, Tanabashi. M. et al. (2018), Phys. Rev. D98, 030001
charge such as pions, with an energy corresponding to that at which the d E/d x
curve reaches its minimum, are called minimum ionizing particles, or mips,
In relatively thin amounts of material, such as those represented by a typical
calorimeter, the total energy loss ΔE/Δx may differ substantially from the value
calculated on the basis of d E/d x. This is because of the relatively small number
of collisions with atomic electrons, and the very large fluctuations in energy transfer
that may occur in such collisions. Therefore, the energy loss distributions measured
with (thin) calorimeters reach their maximum (i.e., most probable) value in general
below the value calculated on the basis of d E/d x and have a long tail toward
large energy losses, the so-called Landau tail [6]. Only for very substantial amounts
of matter, e.g., equivalent to 100 m of water, the energy loss distribution becomes
approximately Gaussian.
It should be emphasized that the Bethe-Bloch formula does not only apply to
muons, but to the ionization losses of all charged particles. In that sense, it is an
extremely important formula for calorimetry in general, since the signals produced
by calorimeters are the result of processes in which the atoms of the absorber material
are excited. The β −2 dependence of the energy loss for non-relativistic particles turns
out to have crucial consequences for the signals from hadronic showers.
2.2 The Electromagnetic Interaction 37
The quantum of the em interaction, the photon (γ ) is mainly affected by four dif-
ferent processes: the photoelectric effect, coherent (Rayleigh) scattering, incoherent
(Compton) scattering and electron–positron pair production.
At low energies, this is the most likely process to occur. In this process, an atom
absorbs the photon and emits an electron. The atom, which is left in an excited
state, returns to the ground state by the emission of Auger electrons or X-rays.
The photoelectric cross section is extremely dependent on the available number of
electrons, and thus on the Z value of the absorber material. It scales with Z n , with
the power n between 4 and 5. The photoelectric cross section varies with the photon
energy as E −3 , so that this process rapidly loses its importance as the energy increases.
In uranium, the highest-Z material that can be used for calorimeter construction, the
cross section for photoelectric effect is dominating for energies below 700 keV, for
iron inelastic scattering already starts to dominate above 100 keV (see Fig. 2.5).
This (coherent) process is also important at low energies. In this process, the photon
is deflected by the atomic electrons. However, the photon does not lose energy. There-
fore, Rayleigh scattering affects the spatial distribution of the energy deposition, but
it does not contribute to the energy deposition process itself.
∼5 MeV (see Fig. 2.5). As we shall see in Sect. 3.4, typically at least half of the total
energy is deposited by such γ s in the absorption process of multi-GeV electrons,
positrons or photons. Compton scattering is therefore a very important process for
understanding the fine details of calorimetry.
The angular distribution of the Compton recoil electrons exhibits a preference
for the direction of the incoming photons (cos φ = 1), but there is also a substantial
isotropic component in the forward hemisphere (the requirements of momentum
and energy conservation prevent the electrons from being scattered in the backward
hemisphere).
Since the photoelectric effect, in which the photon is absorbed and thus disappears,
only plays a role at low energies, many γ s in the MeV energy range are absorbed in
a sequence of Compton scattering processes, in which the photon energy is reduced
in a number of steps down to the point where the final absorption in a photoelectric
process occurs. In each step, an amount of energy equal to
ζ (1 − cos θ )
T = Eγ (2.7)
1 + ζ (1 − cos θ )
is transferred to the struck electron. In this process, the angular preference still visible
for the first scattering in this sequence quickly disappears. Most of the Compton-
and photoelectrons produced in this sequential absorption process are isotropically
distributed with respect to the direction of the initial γ .
The cross section for Compton scattering is much less dependent on the Z value
of the absorber material than the cross section for photoelectric effect. The Compton
cross section is almost proportional to Z , i.e., proportional to the number of target
electrons in the nuclei.
As for the photoelectric effect, the cross section for Compton scattering decreases
with increasing photon energy, albeit much less steeply: σ ∼ 1/E. Therefore, above a
certain threshold energy, Compton scattering becomes more likely than photoelectric
2.2 The Electromagnetic Interaction 39
absorption. This threshold ranges from 20 keV for carbon (Z = 6) to 700 keV for
uranium (Z = 92). The values for other elements can be derived from Fig. 2.5.
At energies larger than twice the electron rest mass, a photon may create, in the field of
a charged particle, an electron–positron pair. These particles produce bremsstrahlung
radiation as well as ionization along their paths. The electron is eventually absorbed
by an ion, while the positron annihilates with an electron. In the latter process, two
new photons are produced, each with an energy of 511 keV, the electron restmass, if
the annihilation takes place when the positron has come to rest.
Typically, more than 99% of the γ → e+ e− conversions are caused by nuclear
electromagnetic fields. For low-Z elements and at high energies, e+ e− creation in the
fields of the atomic electrons also contributes significantly to the total pair production
cross section.
The cross section for pair production rises with energy and reaches an asymptotic
value at very high energies (>1 GeV). This cross section is related to the radiation
length of the absorber material (see Sect. 3.3.1).
The relative importance of the processes through which photons are absorbed
depends strongly on the photon energy and on the electron density (∼Z ) of the
medium. This is illustrated in Fig. 2.1, which shows the cross sections for these
three processes as a function of energy in carbon (Z = 6, Fig. 2.1a), iron (Z = 26,
Fig. 2.1b) and uranium (Z = 92, Fig. 2.1c).
Since the cross sections for the photoelectric effect and for Compton scattering
decrease with energy, and the cross section for pair production increases, pair pro-
duction is the most likely process to occur at high energies.
40 2 Interactions of Particles with Matter
At energies in the range 5–20 MeV, a modest role may be played by photonuclear
reactions, e.g., γ n, γ p or photo-induced nuclear fission. The cross sections for such
reactions reach a maximum value at the so-called giant dipole resonance, when the
photon energy is approximately equal to the marginal binding energy of the proton or
neutron, i.e., the difference in nuclear binding energy between the target nucleus and
the nuclei with one nucleon less. The cross sections for these processes usually do
not exceed 1% of the total cross section for the processes mentioned in the previous
subsections [7].
Table 2.1 The specific ionization energy loss of minimum ionizing particles in various absorber
materials, and the average energy lost by minimum ionizing protons over a distance of one nuclear
interaction length. Data from [3]
Absorber Z d E/d x (mip) λint ΔE/λint
(MeV g−1 cm2 ) (g cm−2 ) (MeV)
Carbon 6 1.742 85.8 149
Aluminum 13 1.615 107.2 173
Iron 26 1.451 132.1 192
Copper 29 1.403 137.3 193
Tin 50 1.263 166.7 211
Tungsten 74 1.145 191.9 220
Lead 82 1.122 199.6 224
Uranium 92 1.081 209.0 226
When an incoming high-energy hadron strikes an atomic nucleus, the most likely
process to occur is spallation. Spallation is usually described as a two-stage process:
a fast intranuclear cascade, followed by a slower evaporation stage. The incoming
hadron makes quasi-free collisions with nucleons inside the struck nucleus. The
affected nucleons start traveling themselves through the nucleus and collide with
other nucleons. In this way, a cascade of fast nucleons develops. At this stage, pions
and other unstable hadrons may also be created if the transferred energy is sufficiently
high. Some of the particles taking part in this cascade reach the nuclear boundary and
escape. Others get caught and distribute their kinetic energy among the remaining
nucleons in the nucleus.
The second step of the spallation reaction consists of a de-excitation of the
intermediate nucleus. This is achieved by evaporating a certain number of parti-
cles, predominantly free nucleons, but sometimes also αs or even heavier nucleon
aggregates, until the excitation energy is less than the binding energy of one nucleon.
The remaining energy, typically a few MeV, is released in the form of γ -rays. In very
heavy nuclei, e.g., uranium, the intermediate nucleus may also fission.
Much experimental information on spallation reactions has been accumulated
during the past 70 years. Rudstam [8] has given a useful empirical formula, valid
within broad limits either of energies (>50 MeV) or of atomic mass (A > 20), which
gives a satisfactory description of spallation cross sections. When a particle of energy
E hits a target with atomic mass AT , the relative cross sections σ for the production
of spallation products (Z f , A f ) are given by the relation
σ (Z f , A f ) ∼ exp −P(AT − A f ) × exp −R|Z f − S A f + T A2f |3/2 (2.8)
in which E is expressed in MeV and the parameters P, R, S and T have the fol-
lowing values: P = 20E −0.77 for E < 2100 MeV, P = 0.056 for E > 2100 MeV,
R = 11.8A−0.45
f , S = 0.486, T = 0.00038.
Figure 2.6 shows the cross sections for nuclides that can be produced from 238 U
spallation induced by a 2 GeV hadron, computed with this formula. Hundreds of
different reactions occur with comparable probability. The largest cross section for
an exclusive reaction amounts to only ∼2% of the total spallation cross section, and
there are about 300 different reactions that contribute more than 0.1% to the total
spallation cross section! This example illustrates the enormous diversity of processes
that may occur in the nuclear sector of the hadronic interactions.
44 2 Interactions of Particles with Matter
Fig. 2.6 Cross sections for nuclides produced by spallation of 238 U, induced by a 2 GeV hadron.
The final-state nuclide is defined by the number of protons (ΔZ ) and neutrons (ΔN ) released from
the target nucleus [1]
Fig. 2.7 A proton-nucleus interaction in a nuclear emulsion stack. Photograph courtesy CERN
In the spallation reactions with absorber nuclei that take place in the absorption
of energetic hadrons, large numbers of nucleons and nucleon aggregates such as
α particles are produced. Equation 2.8 provides quantitative information about this
process. For example, in the reactions of 1 GeV hadrons and 208
82 Pb nuclei, on average,
2.7 protons and 12.8 neutrons are produced. The large discrepancy between the
numbers of protons and neutrons released in these reactions is even more striking
at lower incident energies. Figure 2.8a shows the average numbers of protons and
46 2 Interactions of Particles with Matter
Fig. 2.8 The average numbers of protons and neutrons produced in spallation reactions on 208 82 Pb
(a) or 56
26 Fe (b), as a function of the energy of the incoming hadron. The neutrons are split up in an
evaporation and a cascade component [1]
neutrons as a function of the incident kinetic energy E. For energies smaller than 200
MeV, the probability that at least one proton is emitted drops below 50%. However,
on average, 7 neutrons still come off at 200 MeV.
This indicates that the protons that are produced in the spallation processes on
lead are almost exclusively produced in the fast cascade step. In the evaporation stage
of the reaction, almost all emitted nucleons are neutrons. This is not surprising, since
the Coulomb barrier for protons in a lead nucleus is ∼12 MeV. Therefore, in the
evaporation stage, where fragments are released with a kinetic energy of typically a
few MeV (some fraction of the binding energy per nucleon, which amounts to ∼7.9
MeV in lead), not many charged particles are expected to emerge from the nucleus.
In the fast cascade step, protons and neutrons are emitted in a ratio that, on average,
reflects the numerical presence of these nucleons in the target nuclei. In 208
82 Pb, one
may thus expect for every cascade proton about 1.5 cascade neutrons (126/82).
These considerations make it possible to split the total nucleon production in the
spallation reactions induced by our 1 GeV hadrons (∼16 nucleons) into a cascade
component and an evaporation component: 9 evaporation neutrons and 7 cascade
nucleons (2.8 protons and 4.2 neutrons). The cascade nucleons, in particular the
cascade neutrons, are likely to induce themselves new spallation reactions, further
increasing the numbers of evaporation neutrons.
The distributions of the numbers of protons and neutrons released from their
nuclear environment in spallation reactions are markedly different when the lead
target is replaced by an iron one. The dependence of the average numbers of such
2.3 The Strong Interaction 47
protons and neutrons on the energy of the incident hadron is given in Fig. 2.8b. Among
the most characteristic differences, we mention:
• The strong asymmetry between protons and neutrons found in the case of lead is
almost absent for reactions with iron nuclei.
• The total number of nucleons released in collisions with iron nuclei is considerably
smaller than that for collisions with lead nuclei at the same energy.
The proton/neutron asymmetry in lead is a consequence of the Coulomb barrier
(∼12 MeV), which prevents protons from being emitted by an excited nucleus in the
evaporation stage. In iron, this barrier is considerably lower (∼5 MeV). Therefore,
the probabilities for an excited Fe nucleus to emit a proton or a neutron are not very
different from each other.
The particles emitted in the evaporation stage of the reactions are emitted isotrop-
ically, but the cascade particles have a dominating momentum component along the
direction of the incoming particle. Therefore, the residual target nucleus undergoes a
net recoil, in which it acquires a kinetic energy of the order of m/M, where m and M
denote the total mass of the cascade nucleons and the residual nucleus, respectively.
This recoil energy will, in general, not result in a measurable calorimeter signal and
therefore has to be considered part of the invisible component of the shower energy.
Let us now return to the earlier example of 1 GeV hadrons striking 208 82 Pb nuclei
and examine what happens to the energy in the “average” spallation reaction. In
order to release the 16 nucleons from the lead nucleus, 16 × 7.9 = 126 MeV of
nuclear binding energy has to be provided. The 9 evaporation neutrons carry a total of
∼27 MeV of kinetic energy (for T = 2 MeV, see Fig. 2.9). The remaining 1000–153
= 847 MeV is shared among the target nucleus (30 MeV recoil energy) and 7 cascade
nucleons, about 117 MeV each. The range of 117 MeV protons in lead amounts to
∼2 cm [9], considerably less than the nuclear interaction length (17 cm). Therefore,
these cascade protons are most likely to lose their energy by ionizing lead atoms.
The (on average, 4.2) cascade neutrons, on the other hand, will induce new spal-
lation reactions. Especially in high-Z materials such as lead, they typically initiate
nuclear reactions of the type ZA X (n, yn) ZA X, with A = A − y + 1, in which the
total number of evaporation neutrons produced in the shower absorption process
thus increases by (y − 1). In practice, almost all neutrons that are present in the
absorber structure a few nanoseconds after the start of the shower process are thus
of the evaporation type.
A significant fraction of the hadronic shower energy is thus carried by large numbers
of soft neutrons. The absorption of these neutrons in dense material proceeds very
differently from that of the other types of shower particles encountered so far. Elec-
trons, photons, charged mesons and protons are all subject to the electromagnetic
interaction. Neutrons depend entirely on the strong (and sometimes the weak) inter-
action in order to be absorbed in matter. This has very important consequences for
48 2 Interactions of Particles with Matter
Fig. 2.9 Kinetic energy spectrum of evaporation neutrons, produced according to a Maxwell dis-
tribution with a temperature of 2 MeV. For comparison, the spectrum for a temperature of 3 MeV
is given as well [1]
calorimetry. These consequences may range from very beneficial to very detrimental
[10] and are extensively discussed in Chaps. 3 and 7. In this chapter we concentrate
on the neutron spectra and on the absorption mechanisms.
The kinetic energy spectrum of the evaporation neutrons is usually described by
a Boltzmann–Maxwell distribution
dN √
= E exp(−E/T ) (2.9)
dE
with a temperature T of about 2 MeV, so that the average kinetic energy of these
neutrons amounts to about 3 MeV at production (see Fig. 2.9).
This means that these neutrons have obtained a kinetic energy that is typically of
the order of one-third to one-half of the binding energy that confined them to their
parent nuclei before the shower development occurred.
Experimental measurements have revealed that the numbers of neutrons produced
in the hadronic absorption process are large. Leroy et al. [11] measured the produc-
tion rates of thermalized neutrons in high-energy hadron showers, by analyzing
the induced radioactivity resulting from neutron-capture reactions in the absorber
material. They found rates of ∼20 neutrons per GeV of energy in lead and up to 60
neutrons per GeV in 238 U, where nuclear fission causes a significant multiplication
of the neutron production rates.
Based on knowledge of the spectrum of the neutrons, the average total kinetic
energy carried by these neutrons, and the fluctuations in this total kinetic energy may
be calculated. Some results of these calculations, for the energy spectrum given in
Fig. 2.9, are shown in Fig. 2.10.
Because of the Central Limit Theorem and because of the large numbers of neu-
trons involved, the fluctuations in this total kinetic energy are relatively small. For
2.3 The Strong Interaction 49
Fig. 2.10 Distribution of the total kinetic energy carried by 100 and 1,000 evaporation neutrons
[1]
In this subsection we discuss the various mechanisms through which the evaporation
neutrons lose their kinetic energy and are eventually absorbed in dense matter.
values of f amount thus to 50, 3.4 and 0.96% for collisions with hydrogen, iron
and lead, respectively. It is of course no surprise that, in terms of energy loss, the
elastic scattering process is most efficient in hydrogen. A large cross section and a
considerable fraction of energy lost in each collision make hydrogen-rich compounds
the material of choice for neutron shielding purposes, e.g., in nuclear reactors.
As a result, neutrons in the mentioned energy bracket are sampled very differently
from charged particles in calorimeter structures containing hydrogen in the active
components. This is further discussed in Sect. 8.2.
When the neutrons generated in hadronic shower development have lost (almost) all
of their kinetic energy in collisions with the target material, one of two things may
happen: they decay, or they get captured by an atomic nucleus. Since the timescale
for the first process is very long (average lifetime ∼15 min) and the cross section
for the second process usually large, capture is much more likely to occur. When a
neutron is captured by an atomic nucleus, the nuclear binding energy that had to be
supplied to the nucleus when the neutron was released (invisible energy) is gained
back. The excited “compound” nucleus usually gets rid of this excess energy by
emitting γ -rays. In some light nuclei, such as 6 Li and 10 B, the capture of a neutron
may be followed by the emission of an α particle.
The neutron capture process is distinctly different from the processes through
which charged particles, such as electrons and protons, get absorbed in the shower
development process. After losing their kinetic energy through ionization of the
calorimeter materials, these charged particles just become part of the absorbing
structure, while the neutrons transform an absorber nucleus into another type of
nucleus.
Another process which illustrates that neutrons may be sampled very differently from
charged particles takes place at energies between 3 and 20 MeV. At these energies,
neutrons frequently release α particles from the nuclei with which they interact, for
example through (n, α) reactions. This is particularly true for 12 C, a key ingredient
of organic materials such as plastics or gases used in wire chambers. Neutrons above
10 MeV may split this nucleus into three α particles, the reverse process of the one
that starts the CNO cycle in aging stars. This process alone accounts for ∼60% of
the inelastic n 12 C cross section. In other light gases such as oxygen and fluorine, α
production is also quite abundant, while in higher-Z materials like iron or copper, α
production takes place in only ∼3% of the inelastic reactions.
When produced in wire chambers, such αs may give rise to signals that are orders
of magnitude larger than the ones caused by minimum ionizing particles and, there-
2.3 The Strong Interaction 51
fore, it may be an important process in calorimeters that use such wire chambers for
generating the signals (Sect. 8.2).
The fact that the energy loss process of neutrons in a given material is extremely
dependent on subtleties of that material’s nuclear structure, becomes also clear when
we examine the role of inelastic scattering.
In this process, part of the neutron’s kinetic energy is used to bring a nucleus in
an excited state. The nucleus releases this excitation energy in the form of one or
several γ s, whose (combined) energy corresponds to the energy loss of the neutron.
The contribution of this process to the energy loss of the neutrons produced in the
calorimeter depends completely on details of the nuclear level structure.
In some materials, e.g., lead, it becomes insignificant below energies as high
as 2.6 MeV (because it takes that much energy to bring the most abundant lead
isotope, 208 Pb, from its ground state into the lowest excited state), in other materials
it continues to play a role down to energies well below 1 MeV. For example, the
first excited state of the most abundant isotope of iron, 56 Fe, is located 0.85 MeV
above the ground state. The cross section for inelastic scattering processes of the
type (n, n γ ), in which neutrons in the energy range of 1–6 MeV lose 0.85 MeV, is
more than one barn [12]. This is one of the reasons why steel-reinforced concrete is
a good shielding material for MeV-type neutrons.
References
Fig. 3.1 The energy deposit as a function of depth, for 1, 10, 100 and 1,000 GeV electron showers
developing in a block of copper. In order to compare the energy deposit profiles, the integrals of
these curves have been normalized to the same value. The vertical scale gives the energy deposit per
cm of copper, as a percentage of the energy of the showering particle. Results of EGS4 calculations
[1]
5 cm for 1 GeV electrons. For every order of magnitude in energy, this maximum
shifts by ∼3.5 cm deeper into the detector, to reach a depth of about 16 cm for electron
showers of 1 TeV. The copper thickness needed to absorb 99% of the shower energy
rises from 23 cm at 1 GeV, via 28 cm at 10 GeV and 33 cm at 100 GeV, to 39 cm at
1 TeV.
Electron and γ induced showers are different in the way they start developing
(Fig. 3.2). Upon entering an absorber medium, electrons start to radiate immediately.
In the first radiation length, they lose, on average, 63.2% of their kinetic energy to
bremsstrahlung. On the other hand, high-energy γ s travel, on average, 9/7 radiation
lengths in the absorbing medium before their first interaction.
This difference between the interaction mechanisms has two types of conse-
quences for the em showers initiated by photons and by electrons/positrons:
1. Photon-induced showers deposit their energy, on average, deeper inside the
absorbing structure than do em showers induced by charged particles of the same
energy.
2. The fluctuations in the amount of energy deposited in a given slab of material are
larger for showers induced by photons than for showers induced by e+ or e− .
The first effect results from the fact that the photons travel a certain distance in the
absorbing structure before they start losing energy, while electrons and positrons
3.1 Electromagnetic Showers 55
Fig. 3.2 The different starts of em showers initiated by high-energy electrons (left) and γ s (right)
entering an absorbing medium [1]
start losing energy immediately upon their entry. Moreover, the starting point of the
photon-induced showers fluctuates from event to event, which leads to the second
effect.
These effects are illustrated in Fig. 3.3, which shows the distribution of the energy
deposited by 10 GeV electrons and 10 GeV photons in a 5X 0 (2.8 cm) thick slab
of lead [2]. On average, the electrons deposit more energy in this material than the
photons (2.10 GeV vs. 1.48 GeV). However, the fluctuations in the energy deposited
by the photons are clearly larger than those in the energy deposited by the electrons
(0.86 GeV vs. 0.64 GeV). The distribution for the photon showers exhibits an excess
near zero, which is the result of photons penetrating (almost) the entire slab without
interacting. The “punch-thru” probability for a high-energy γ is in this example
exp (−35/9) ≈ 2%.
The different effects of dead material installed in front of the calorimeter on elec-
trons/positrons and γ s is particularly relevant for the ATLAS experiment, where the
electromagnetic calorimeter is “hidden” in a cryostat, although the effects are less
dramatic than suggested in Fig. 3.3 because this cryostat is made of aluminium. How-
ever, another consequence of the differences between electron and γ induced show-
ers for ATLAS is that the very complicated calibration scheme that was developed
for electrons showering in the three longitudinal segments of the electromagnetic
calorimeter is not necessarily the optimal solution for γ detection in this calorimeter.
More on this aspect in Chap. 9.
56 3 Shower Development
Fig. 3.4 The energy deposit as a function of depth, for pion showers of different energies developing
in the ZEUS 238 U/plastic-scintillator calorimeter. These profiles are measured with respect to the
starting point of the showers. From: Catanesi, M.G. et al. (1987). Nucl. Instr. and Meth. A260, 43
Fig. 3.5 Schematic depiction of a hadron shower. The energy carried by the hadron is typically
deposited in the form of an electromagnetic and a non-electromagnetic component. The em compo-
nent is the result of π 0 s and ηs produced in the nuclear reactions. The non-em component consists of
charged hadrons, and nuclear fragments. Some fraction of the energy transferred to this component
(the “invisible” energy needed to break apart nuclei excited in this process) does not contribute to
the calorimeter signals [1]
ZEUS experiment [4]. In stark contrast with the profiles for em showers (Fig. 3.1), the
depth at which the hadronic profiles reach their maximum value barely changes with
the energy of the incoming particles. This maximum is invariably located around
1λint beyond the starting point of the showers. However, the downward slope of the
absorption profile clearly becomes less steep as the hadron energy increases.
The reason for the fact that the hadronic shower maximum is (almost) indepen-
dent of the hadron energy is illustrated by Fig. 3.5, which schematically depicts the
start of a typical high-energy hadron shower. In the first interaction, some fraction of
the energy is used for the production of one or several π 0 s. These decay into γ s and
develop electromagnetic showers. This em shower component is a typical feature of
hadron showers. On average, one third of the energy of the incoming high-energy
hadron is carried away by π 0 s. In 238 U, the absorber material for which the results
shown in Fig. 3.4 were obtained, the radiation length is 3.2 mm. One nuclear interac-
tion length (100 mm) is thus equivalent to about 30X 0 . As illustrated by Fig. 3.1, the
58 3 Shower Development
Fig. 3.6 Longitudinal profiles for four different showers induced by 270 GeV pions in a
lead/iron/plastic-scintillator calorimeter. Data from [5]
em showers resulting from the production of π 0 s in the first nuclear interaction have
thus completely been absorbed after the hadron shower has penetrated the absorber
material over a distance of 1λint beyond its starting point, regardless of the energy of
these π 0 s and thus of the energy of the showering hadron. Since the relevant scale
for the development of the em shower component is so much shorter than that of the
non-em component, the energy deposit profile of the absorbed hadron is in its first
stage completely determined by the em shower component. Therefore, the depth at
which the energy deposited in a slice of given thickness reaches its maximum value
is independent of the hadron energy.
It is crucial to note that the absorption profiles shown in Fig. 3.4 are averaged over
large numbers of showers. The energy deposit profiles of individual hadron showers
may deviate substantially from these averages. In individual showers, the production
of energetic π 0 s may occur at later stages of the shower development, i.e., initiated
by secondary or tertiary shower particles. The em shower components created by
such π 0 s develop in that case in completely different regions of the absorbing vol-
ume, and this will lead to energy deposit profiles that differ considerably from the
average. Figure 3.6 shows a few examples of non-average energy deposit profiles,
for pion showers developing in a lead/iron/plastic-scintillator sandwich calorimeter
[5]. The absorber structure consisted of 40 lead plates (3.1 mm thick), followed by
26 iron plates (2.5 cm thick). The energy deposited in this structure was measured
in every individual scintillator plate, so that detailed event-by-event information on
the longitudinal shower development was obtained.
Figure 3.6a shows a shower with an energy deposit profile that corresponds
roughly with the profile averaged over a large number of showers. However, individ-
ual beam particles may penetrate deep into the detector before initiating a nuclear
reaction (Fig. 3.6b, c). Figure 3.6b depicts an event in which a large fraction of the
energy was transferred to one or several π 0 s in this first nuclear interaction. Also in
3.2 Hadron Showers 59
the event shown in Fig. 3.6c, energetic π 0 s were produced in the first nuclear interac-
tion. However, in addition, an energetic charged hadron was produced at that point.
This particle traveled about one interaction length deeper into the absorber and then
transferred almost all its energy to π 0 s in a second-generation interaction, leading
to a two-peak structure in the longitudinal energy deposit profile. In Fig. 3.6d, even
three generations of π 0 production can be distinguished.
Such “stochastic” energy deposit profiles are by no means exceptional in hadronic
shower development. They are a consequence of the fact that the π 0 s produced in
the shower development deposit the energy they carry in a much smaller absorber
volume than other shower particles. Therefore, the hadronic energy deposit profiles
directly reflect the large event-by-event fluctuations that may occur in both the energy
carried by these π 0 s and the position in the absorber where they are generated.
The large difference between the values of the radiation length and the nuclear
interaction length, which is responsible for the phenomena discussed in Sect. 3.2
(Figs. 3.4 and 3.6), is a consequence of the difference between the cross sections for
the reactions mediated by the em and the strong interactions. These depend on the A
and Z values of the absorbing medium. In the following subsections, this dependence
is described.
7 A
σ (E → ∞) = (3.1)
9 NA X 0
in which X 0 is expressed in g cm−2 and the ratio of Avogadro’s number (NA ) and the
atomic weight (A) denotes the number of atoms per gram of material. This implies
that the mean free path of very-high-energy photons equals 97 X 0 . For approximate
calculations, which are accurate to within 3%, the Particle Data Group [6] recom-
mends the following expression:
716.4 A
X0 = √ g cm−2 (3.2)
Z (Z + 1) ln (287/ Z )
60 3 Shower Development
in which Vi and X i are the fraction by volume and the radiation length (expressed in
mm) of the ith component of the mixture. Equation 3.3 may, for example, be used
to calculate the effective radiation length of a calorimeter consisting of a variety of
different materials. Let us, as an example, consider a lead/liquid-argon calorimeter
consisting of 5 mm thick lead plates, separated by 3 mm wide LAr-filled gaps.
The radiation lengths of lead and LAr are 5.6 mm and 140 mm, respectively, and
the fractional volume occupied by these elements is 62.5% for lead and 37.5% for
argon. Therefore, we find for the effective radiation length: X eff = [0.625/5.6 +
0.375/140]−1 = 8.75 mm, only slightly less than the value one would obtain if the
argon were replaced by vacuum (X eff = 5.6/0.625 = 8.96 mm).
If the argon were contained in 0.8 mm thick stainless steel containers, separated
from the lead plates by 0.2 mm of air, the volume ratio of the different materials
in the calorimeter structure would be as follows: lead/argon/iron/air = 5/3/1.6/0.4
= 50%/30%/16%/4%. With the radiation lengths for iron and air being 17.6 mm
and 300 m, respectively, the effective radiation length of this structure becomes:
X eff = [0.5/5.6 + 0.3/140 + 0.16/17.6 + 0.04/30, 000]−1 = 9.95 mm.
The radiation length of a compound can be calculated in a similar way, using the
equation
1
= mi / Xi (3.4)
X0 i
in which m i and X i are the fraction (by mass) and the radiation length (expressed
in g cm−2 ) of the ith component of the compound. Let us, for example, calculate
the radiation length of lead-tungstate crystals (PbWO4 ). The mass ratio of the ele-
ments of which these crystals are composed is as follows: lead/tungsten/oxygen =
207.19/183.85/64.0 = 45.5%/40.4%/14.1%. The radiation lengths of these elements
are 6.37, 6.76 and 34.24 g cm−2 , respectively. Therefore, we find for the radiation
length of lead-tungstate: X 0 = [0.455/6.37 + 0.404/6.76 + 0.141/34.24]−1 =
7.39 g cm−2 . Since the density of these crystals amounts to 8.30 g cm−3 , their radi-
ation length equals 8.9 mm.
Fig. 3.7 Average energy fraction contained in an infinitely long cylinder of absorber material, as a
function of the radius of this cylinder. Results of EGS4 calculations for various absorber materials
and different energies [1]
X0
ρM = E s (3.5)
εc
√
in which the scale energy E s , defined as m e c2 4π/α, equals 21.2 MeV. Typically,
∼85–90% of the shower energy is deposited in a cylinder with radius ρM around the
shower axis (Fig. 3.7).
The Molière radii for mixtures or compounds of different elements may be cal-
culated in the same way as the radiation length for such mixtures or compounds
was obtained, replacing X i in Eqs. 3.3 and 3.4 by ρi . Let us, for example, calculate
the Molière radius of BGO crystals, which have the following chemical composi-
tion: Bi3 Ge4 O12 . The mass ratio of the elements of which these crystals are com-
posed is as follows: bismuth/germanium/oxygen = (209.0×3)/(72.6×4)/(16.0×12)
= 56.5%/26.2%/17.3%. The radiation lengths of these elements are 6.32, 12.25
and 34.24 g cm−2 , respectively. For the critical energies, we use Eq. 2.3, which
gives values of 7.24 MeV for bismuth, 18.4 MeV for germanium and 66 MeV for
oxygen. This leads to Molière radii for the different crystal components of 18.5
g cm−2 (Bi), 14.4 g cm−2 (Ge) and 11.0 g cm−2 (O), respectively. When com-
bining all these data, we find for the Molière radius of bismuth-germanium oxide:
ρM = [0.565/18.5 + 0.262/14.4 + 0.173/11.0]−1 = 15.5 g cm−2 . Since the den-
sity of these crystals amounts to 7.13 g cm−3 , their Molière radius thus equals about
22 mm.
The Molière radius is much less Z dependent than the radiation length. This can
be seen from Eqs. 2.3, 3.2 and 3.5. The radiation length scales in first approximation
with A/Z 2 (Eq. 3.2). If we assume that A is proportional to Z , which is roughly true,
the radiation length (expressed in g cm−2 ) decreases with increasing Z like 1/Z .
The same is approximately true for the critical energy (Eq. 2.3). Since the Molière
62 3 Shower Development
radius is defined as the ratio of the radiation length and the critical energy, the Z
dependence cancels in first approximation.
This difference in Z dependence may be illustrated by comparing two materials
that are frequently used as absorbers in calorimeters, copper (Z = 29) and lead
(Z = 82). The densities are not very different: 8.96 g cm−3 for copper versus
11.35 g cm−3 for lead. The radiation lengths for these materials reflect the large
difference in Z : 14.3 mm for copper versus 5.6 mm for lead, almost a factor of three
difference. However, the values for the Molière radii show a completely different
pattern: 15.2 mm for copper versus 16.0 mm for lead.
As a consequence, the development of em showers in these two absorber materials
has very different characteristics. In the longitudinal direction, it takes about three
times as much copper as lead (in cm) to contain these showers. However, laterally,
the showers in copper are even narrower than those in lead.
This definition is thus equivalent to the one for the mean free path of high-energy
photons, which was found to be equal to 9/7 of a radiation length (Eq. 3.1). And just
as the mean free path of photons is inversely proportional to the total cross section
for photon-induced reactions, λint is inversely proportional to the total cross section
for nuclear interactions:
A
σtot = (3.7)
NA λint
This cross section is determined by the size of the projectiles and the size of the
target nuclei. The cross section of the target nuclei is determined by their radius
squared. And since the volume of these nuclei (and thus r 3 ) scales with the atomic
weight A, the cross section scales with A2/3 . From Eq. 3.7, it then follows that λint
scales with A1/3 , when expressed in units of g cm−2 (which eliminates differences
in material density).
The smallest values for λint , around 10 cm, are found for high-density, high-Z
materials such as tungsten, gold, platinum and uranium. For frequently used absorber
materials such as iron and copper, the interaction length is less than twice as long
(a 60–70% increase compared with uranium). This is quite different from the situation
encountered earlier for the radiation length, which increases by about a factor of five
3.3 Material Dependence 63
going from uranium to iron. We will come back to these differences when discussing
calorimetric particle identification (Sect. 11.3).
The nuclear interaction lengths for mixtures of different elements or for a com-
pound can be determined in the same way as discussed for the radiation length and
the Molière radius (Sects. 3.3.1 and 3.3.2).
In Sect. 3.2, we saw that, in uranium, the nuclear interaction is more than 30 times
larger than the radiation length. Given the Z and A dependence of these scaling
variables, the ratio λint / X 0 depends on these material parameters as Z 2 A−2/3 , and
thus increases (almost proportionally) with the Z value of the absorber material.
Figure 3.8 shows the actual value of this ratio as a function of the Z of some
absorber materials that may be used in calorimeters. It increases from 4.4 for alu-
minium (Z = 13) and 9.5 for iron (Z = 26) to 27 for tungsten (Z = 74 and 33 for
uranium (Z = 92). The large difference between X 0 and λint makes high-Z absorber
materials ideally suited for particle identification. When a beam consisting of a mix-
ture of high-energy electrons and pions is sent through a 5 mm thick plate of lead,
almost all (96%) of the pions will traverse this plate without strongly interacting. On
the other hand, the electrons lose a considerable fraction of their energy by radiating
large numbers of bremsstrahlung photons. This feature is the founding principle of
some preshower detectors (Sect. 5.4).
Eventually, all the available energy of the incoming particle is shared among a very
large number of shower particles, each of which carries very little kinetic energy,
and that energy is transferred to the atoms and molecules of the absorber medium.
And even though calorimeters are often intended to measure energy deposits at the
level of 109 eV and up, their performance is in practice determined by what happens
at the MeV, keV and sometimes eV levels, simply because the particles that carry
these low energies are so abundantly produced in the absorption process.
To illustrate the importance of the last stages of the shower development, Fig. 3.9
shows what happens to the energy carried by a 10 GeV electron that is absorbed
in a block of material, as a function of the Z -value of that material. If we take as
an example lead, which is often used as absorber material in calorimeters intended
for detecting such electrons, it turns out that almost 70% of the entire energy is
deposited by shower electrons that carry less than 4 MeV kinetic energy. Electrons
of less than 1 MeV even carry 40% of the total energy, which means that there must
thus be at least 4,000 such electrons, and probably many more. Figure 2.5 shows that
these soft electrons are almost exclusively the result of Compton scattering and the
photoelectric effect, since these processes are the most likely ones to occur when γ s
with energies below 4 MeV interact with the absorber material.
To understand the peculiarities of electromagnetic calorimeters, one thus needs to
understand the relevant characteristics of Compton scattering and the photoelectric
effect. We mention three examples.
1. The radiation length (X 0 ) was introduced as a parameter to describe electromag-
netic shower development in a material independent way. Longitudinal shower
profiles should thus be material independent when expressed in terms of X 0 .
Figure 3.10 shows that this is by no means the case.
Figure 5.1a also illustrates the inadequacy of this parameter for determining the
calorimeter depth needed to contain em showers at the level of 99%. For 10
GeV electrons, this depth ranges from 16X 0 to 22X 0 , depending on the chosen
absorber material. These discrepancies are a result of the fact that X 0 is defined
based on the properties of the high-energy component of the showers, where elec-
trons predominantly lose energy by radiation (Bremsstrahlung) and γ s interact
by producing e+ e− pairs. These processes do not play an important role in the
last stages of the shower development during which, as shown above, a major
fraction of the total energy is deposited.
2. There is no preferential direction for the electrons produced in Compton scat-
tering and by the photoelectric effect. Because of the dominant role of these
processes in em shower development, a large fraction of the shower particles thus
travel in random directions with respect to the particle that initiated the shower
and whose properties are being measured with the calorimeter (Fig. 3.11). This
means that the orientation of the active layers of a sampling calorimeter can be
chosen as desired, without serious implications for the calorimetric performance
of the detector.
The first generation of sampling calorimeters used in particle physics experiments
consisted almost exclusively of instruments of the “sandwich type,” i.e., detectors
composed of alternating layers of absorber and active material, oriented perpen-
dicular to the direction of the particles to be detected. Although this, from an
intuitive point of view, may seem to be the only correct choice, the R&D with
66 3 Shower Development
Fig. 3.11 Angular distribution of the shower particles (electrons and positrons) through which the
energy of a 1 GeV electron is absorbed in a lead-based fiber calorimeter. The angle between the
direction of the shower particles and the fiber axis (θz ) was chosen to be 3◦ in these EGS4 Monte
Carlo simulations. From: Acosta, D. et al. (1990). Nucl. Instr. and Meth. A294, 193
fiber calorimeters has proven that there is no need for such a geometry [7, 8]. The
notion that the active calorimeter layers do not necessarily have to be oriented
perpendicular to the direction of the incoming particles has had a considerable
impact on the design of detectors for new experiments (Fig. 1.11). Other orienta-
tions may offer considerable advantages in terms of detector hermeticity, readout,
granularity, etc. Apart from the “spaghetti” type of calorimeters built for a number
of experiments, this development is also illustrated by the liquid-argon calorime-
ters with an “accordion” geometry and the tile/fiber hadron calorimeter of the
ATLAS experiment at the LHC [9].
3. The cross section for the photoelectric effect is extremely Z -dependent (∝ Z 5 ).
This has very important consequences for sampling calorimeters that consist of
high-Z absorber material and low-Z active layers. Low-energy γ s produced in the
shower development will, for all practical purposes, only interact in the absorber
material, and the photoelectrons produced in this process will only contribute
to the signals if they are produced very close to a boundary layer. In practice,
they are much less efficiently sampled than the high-energy electron/positron
pairs produced in the early stages of the shower development. As a result, the
sampling fraction in such calorimeters decreases as the shower develops in depth
(see Fig. 3.13). Also, the fact that e/mi p = 1 in such calorimeters1 is the result of
this phenomenon. Some consequences for calorimetry are discussed in Sect. 7.3.
Fig. 3.12 Average range of electrons (a) and protons (b) in various absorber materials, as a function
of energy. Data from [10]
The fact that a large fraction of the em shower energy is deposited by very
soft electrons also has implications for the energy resolution of em sampling
calorimeters. This may be concluded from Fig. 3.12a, which shows the range of
electrons with energies from 20 keV to 10 MeV in several materials. For example,
1 mm of aluminium is enough to stop electrons with a kinetic energy up to 590 keV.
For silicon, a material that is increasingly being considered as active material for
sampling calorimeters, electrons with energies up to 115 keV stop in 100 µm thick
layers, while 330 keV electrons have a range of 500 µm [10]. If one replaced in
the same absorber structure 500 µm thick Si sensors by 100 µm thick ones, shower
electrons that escaped from the absorber layers carrying a kinetic energy between
115 keV and 330 keV would thus always produce the same signals in the thicker
sensors, while the signals in the thinner ones would depend on the angle of incidence.
Because of this additional source of fluctuations, the energy resolution would dete-
riorate as the sensors were made thinner and thinner. The fact that this does indeed
happen (Sect. 4.2) is an indication of the important contributions of shower electrons
from this very-low-energy bracket.
with liquid argon. Minimum ionizing particles lose, on average, 2.13 MeV cm−1 in
LAr and 20.5 MeV cm−1 in 238 U (see Table B.1), or 1.06 MeV and 6.15 MeV in one
active and one passive layer, respectively. Therefore, the sampling fraction of this
calorimeter amounts to
1.06
= 0.147, or 14.7%
1.06 + 6.15
While this definition of the sampling fraction is extremely simple and straightforward,
its connection with experimental data obtained with the calorimeter is not. This is
because the response to mips cannot be measured directly. It should be emphasized
that a mip is a hypothetical particle. As soon as a charged particle with an energy
for which d E/d x reaches its minimum value starts traveling through matter, it loses
energy and therefore ceases to be a mip.
For all practical purposes, muons are the closest thing nature provides us with
in terms of mips. However, even muons with an energy as low as 5 GeV are by no
means minimum ionizing particles. Since they are extremely relativistic (γ ∼ 50),
the energy loss per unit length is significantly larger than the minimum ionizing value
[12]. The increased specific ionization of muons with energies larger than the min-
imum ionizing value is due to phenomena such as δ-ray emission (relativistic rise),
bremsstrahlung, e+ e− pair production and, at very high energies, nuclear reactions.
The contribution of these effects to the total energy loss is strongly dependent on the
muon energy and on (the Z value of) the traversed material.
In practice, the experimental calorimeter response to mips is determined from
the signal distributions measured for muons of different energies, by estimating the
consequences of the above effects, thereby unfolding the mip part of the calorimeter
signals [13–16].
The sampling fraction of a calorimeter, while defined for minimum ionizing particles,
may be very different for other types of particles. This may have very large and
important consequences for the characteristics of sampling calorimeters, such as
the response function for different types of high-energy particles, and the energy
resolution with which these can be measured. In this context, we discuss the two
most important effects in that context.
1. Soft γ s.
In Sect. 2.2.3, we saw that soft γ s produced in em shower development pre-
dominantly interact with the absorber material through the photoelectric effect
(Fig. 2.5). The cross section for this process depends very sensitively on the Z
value. As pointed out in Sect. 3.4, this has important consequences for sampling
calorimeters that consist of high-Z absorber material and low-Z active layers.
Low-energy γ s are sampled much less efficiently than mips in such calorimeters.
3.5 The Sampling Fraction 69
Fig. 3.13 The sampling fraction changes in a developing shower. The local sampling fraction,
normalized to that for a minimum ionizing particle, is shown as a function of depth for 10 GeV
electrons in a Pb/scintillating-plastic calorimeter. Results of EGS4 calculations [1]
Fig. 3.14 The longitudinal distributions of 239 Np nuclei (from neutron capture in 238 U) produced
by hadron showers in various calorimeter configurations. From: Leroy, C., Sirois, Y. and Wigmans,
R. (1986). Nucl. Instr. and Meth. A252, 4
Table 3.1 Sampling fractions for neutrons and mips in various calorimeter structures. See text for
details
Structure ↓ 1 keV n 10 keV n (%) 100 keV n (%) 1 MeV n (%) mip (%)
Particle →
Fe/LH2 (1/1 vol.) 93.6 95.9 95.6 92.6 2.4
Pb/LH2 (1/1 vol.) 99.2 99.2 98.8 98.3 2.2
Fe/LAr (1/1 vol.) 2.0 2.8 11.4 21.5 15.7
(by ∼20%) in the number of nuclear fissions in this structure, compared with
the pure 238 U case. Nuclear fission of 238 U requires a neutron with a kinetic
energy of at least 1.5 MeV, and because of the moderating effect of the plastic,
the number of neutrons capable of inducing fission was smaller in this experi-
ment.
These experimental results show that neutrons lose a disproportionally large frac-
tion of their kinetic energy in collisions with hydrogen when the calorimeter struc-
ture contains this material. In addition, the recoil protons produced in this process
may also substantially contribute to the calorimeter signals. As a result, neutrons
in the energy bracket between 1 keV and a few MeV are sampled very differently
from charged particles in calorimeter structures that contain hydrogenous active
components. This has spectacular calorimetric consequences.
Table 3.1 shows the sampling fractions for neutrons of different energies for three
different structures: iron/liquid-hydrogen, lead/liquid-hydrogen and iron/liquid-
argon, each with equal volumes of metal and liquid. These sampling fractions
were calculated on the basis of the cross sections for elastic scattering and the
assumption that the average fraction of the neutron’s kinetic energy transferred
in elastic collisions with a nucleus containing A nucleons, f el , is given by
2A
f el = (3.8)
(A + 1)2
which implies f el values of 50% and 5% for hydrogen and argon, respectively.
Neutrons at energies between 1 keV and 1 MeV lose almost 100% of their kinetic
energy in the hydrogen in the Fe/H2 and Pb/H2 structures, while mips deposit only
a few percent of their energy in the hydrogen. Therefore, if hydrogen were the
active component of these structures, the sampling fraction for neutrons would
be larger than that for mips by a factor of 40–50 (i.e., n/mi p = 40–50). In the
liquid-argon case, mips are typically sampled more efficiently than soft neutrons
(i.e., n/mi p < 1).
The n/mi p ratio depends sensitively on the fraction of hydrogen contained in the
calorimeter structure. This is illustrated in Table 3.2 for the case of lead/hydrogen.
Whereas the sampling fraction for mips is almost proportional to the fraction of
hydrogen in the structure, the sampling fraction for neutrons changes by less than
a factor of three when the fraction of hydrogen is changed by two orders of magni-
72 3 Shower Development
Table 3.2 Sampling fractions for 1 MeV neutrons and mips in Pb structures containing different
fractions of liquid hydrogen, assumed to be the active calorimeter medium in these configurations
H2 fraction (vol. %) 1 MeV neutrons (%) mips (%) n/mi p ratio
1% 36.9 0.0227 1630
5% 75.3 0.118 640
10% 86.6 0.249 350
20% 93.5 0.558 170
50% (Pb/H2 = 1/1) 98.3 2.20 45
90% 99.8 16.8 5.9
tude. As a result, the n/mi p ratio spans a wide domain, from 5.9 in hydrogen-rich
structures (90 volume%) to 1630 in hydrogen-poor ones (1 volume %).
The neutrons are thus much more efficiently sampled than mips in these calorime-
ter structures, more so if the fraction of hydrogen is reduced. At first sight, it might
seem counter-intuitive that it takes a reduction in the fraction of neutron-sensitive
material to increase the relative calorimeter response to these particles (the n/mi p
ratio). However, this is easily understood from the fact that a change in the
hydrogen fraction affects the response to mips much stronger than it affects the
response to neutrons. The mips share their energy among the active and passive
materials according to the relative abundance of these materials in the structure.
If there is a small fraction of active material, only a small fraction of the mip’s
energy is deposited in this material.
The neutrons, on the other hand, have in practice almost no alternative for deposit-
ing their kinetic energy in the active material, even if this active material represents
only a small fraction of the total mass. This is because elastic scattering in the
lead absorber is an extremely inefficient process for losing energy. On average,
the neutrons lose only 0.96% of their kinetic energy in such collisions, versus
50% in hydrogen.
Apart from the effects discussed above, there are also differences between the
sampling fractions for mips and non-mip charged shower particles. For example,
the effects of the relativistic rise in Fig. 2.3 depends on the Z value of the traversed
material, and are thus in principle different for the active and passive layers of a
sampling calorimeter. However, these effects are typically only at the level of a few
percent. The differences may be considerably larger for extremely non-relativistic
shower particles. However, the range of these particles is very short, so that they
typically are not sampled, but rather range out in the material in which they are
produced. More information about this issue is given in [1].
References 73
References
4.1 Introduction
Calorimeter signals consist typically of electric pulses produced either directly by col-
lecting electric charge (e.g., in liquid-argon calorimeters or resistive plate chambers),
or by photons that produce such pulses as a result of the photoelectric effect (e.g., in
photomultiplier tubes, silicon photomultipliers or photodiodes). Since calorimeters
are intended for measuring energy, the total signal should be a good measure of
that energy. In practice, there are a number of instrumental effects that may spoil
the relationship. These effects should be clearly distinguished from the physics ef-
fects described in the previous subsections (invisible energy, inefficient sampling of
certain types of shower particles, or in late stages of the shower development, etc.).
Fig. 4.1 Average signal as a function of electron energy for the W/Si ECAL built by CALICE (a)
[1]. Residual signals from this detector, before and after taking out a 360 MeV offset (b). From:
Wigmans, R. (2018), J. Progr. Part. Nucl. Phys. 103, 109
and plot
as a function of the beam energy. The result is represented by the (black) squares in
Fig. 4.1b. They conclude that “the calorimeter is linear to within approximately 1%.”
This is highly misleading. When the calorimeter signals they actually measured are
used to check the linearity, i.e., when
is plotted as a function of the beam energy, the results, represented by the (red) full
circles in Fig. 4.1b, look quite different. We conclude from these results that the
authors measured a signal non-linearity of 5% over one decade in energy.
Signal linearity is a very important property, since it is a crucial ingredient for
the precision with which the energy of an unknown object that produces signals
in the calorimeter can be measured. The assumption that it is sufficient to know
the relationship between the signals and the deposited energy, even in the case of
non-proportionality, constitutes another frequent misconception.
Let us take a look at Eq. 4.2, which describes the relationship between the
measured signal (E mean ) and the corresponding energy (E)meas for this non-linear
calorimeter. A γ with an energy of 10 GeV would thus produce an average sig-
nal E mean of 10 − 0.36 = 9.64 GeV. For a γ of 5 GeV, the average signal would
be 5 − 0.36 = 4.64 GeV. Now consider what would happen to a 10 GeV π 0 that
4.2 Signal Linearity and Non-linearity 77
Fig. 4.2 Signal distributions for γ s and various hadrons decaying into all-γ final states. All particles
have the same nominal energy and the detector, which consists of two longitudinal segments and
has an intrinsic resolution of 0.5% for em showers of this energy, was calibrated with electrons such
as to optimize the energy resolution for these particles. From: Wigmans, R. and Zeyrek, M. (2002).
Nucl. Instr. and Meth. A485, 385
would be absorbed in this calorimeter. This particle would decay into two 5 GeV
γ s, and would be assigned an energy value of 2 × (4.64 + 0.36) = 10 GeV if the
two γ s would be recognized as two separate showers by the calorimeter. How-
ever, if the two showers were so close that the calorimeter would not recognize the
event as being caused by two separate particles, it would be assigned an energy of
2 × 4.64 + 0.36 = 9.64 GeV, i.e., 3.4% too low. In a non-linear calorimeter, the re-
constructed energy thus depends on whether the event in question is caused by one
particle or by several particles of which the showers overlap to the point that they
are not recognized as such.
Figure 4.2 illustrates this situation for a practical case. Electromagnetic showers
caused by different particles of the same total energy give signal distributions centered
around different values. This example is useful since it shows that the precision with
which the energy of an unknown object can be measured is only determined by the
energy resolution (i.e., the width of the curves) if the central value of the signal
distributions corresponds to the correct energy (50 GeV), which is not the case here.
Signal non-linearity is one of the most worrisome instrumental effects. There are
many possible causes. Miscalibration of the signals from the different sections of
a longitudinally segmented calorimeter, which is the origin of the effect shown in
Fig. 4.2, is a very common cause, discussed in detail in Chap. 9. Hadron calorimeters
are usually intrinsically non-linear, because the invisible energy (which does not
contribute to the calorimeter signals) represents an energy dependent fraction of the
total. The effects of this are discussed in Chap. 7.
In this chapter, instrumental effects that may lead to signal non-linearity are ad-
dressed. We distinguish two types of non-linearity, which have quite different con-
sequences for the calorimeter performance.
78 4 The Calorimeter Signals
The first type of non-linearity is caused by quenching effects in the signal producing
medium. It affects the signals from densely ionizing shower particles. Detector media
that are susceptible to this effect include scintillators and noble liquids. The latter are
based on ionization charge drifting over a rather long distance to an electrode that
converts the collected charge into a measurable signal. If the ionization density in
the liquid is large, then the probability for recombination of electrons and ions into
atoms along the track increases and the charge collected at the electrode decreases
as a result. A similar phenomenon occurs in scintillators, where the effects are well
described by Birks’ law:
dL d E/d x
= S (4.3)
dx 1 + kB · d E/d x
Fig. 4.4 The stopping power, normalized to the mip value (a) and the range (b) of densely ionizing
protons in silicon. Data from [3]
gases used in proportional wire chambers is very similar. Such a (recoil) proton may
be produced in elastic n − p scattering in that gas and deposit its entire 1 MeV energy
inside the wire chamber. On the other hand, a mip traversing the same wire chamber
typically deposits at least three orders of magnitude less energy in it [3]. This leads
to a phenomenon known as the Texas tower effect, in which one individual shower
particle can mimick a local deposit of an enormous amount of energy. It occurs in
calorimeters with a very small sampling fraction and a non-quenching active medium.
In Sect. 4.3, several practical examples of this effect are discussed. These examples
include calorimeters in which thin silicon sensors are used to produce the signals.
Figure 4.4a shows that the stopping power of silicon is larger than that for mips by
a factor ranging from 20 to 100 in the relevant energy range (10–1 MeV) of protons
produced in hadron shower development. The amplification factor of a proton signal
in such calorimeters is also determined by the range of these particles in the sensors
(Fig. 4.4b). The thinner the silicon, the more prone the detector will be to local spikes
in the measured energy deposit pattern.
The Texas tower effect does not play a role in calorimeters that use plastic scin-
tillator or liquid-argon as active material, because of the combined effects of signal
quenching and the larger thickness of the active layers. And since densely ioniz-
ing shower particles represent a more or less energy independent fraction of the
(non-π 0 ) shower component, the quenching effects do not adversely affect the per-
formance characteristics of such calorimeters. That is different for the second type
of non-linearity, which is the result of signal saturation.
80 4 The Calorimeter Signals
Fig. 4.5 Saturation effects in one of the towers of the SPACAL calorimeter (a). Shown are the
average signal (b) and the energy resolution (c) as a function of energy, measured when a beam of
electrons was sent into this tower [4]
Fig. 4.6 Event display for a 120 GeV π − showering in the CALICE digital hadron calorimeter.
From: Sefkow, F. et al. (2016). Rev. Mod. Phys. 88, 015003
• Non-linearity of the calorimeter response, i.e., the total calorimeter signal is not
proportional to the energy of the detected particle (Fig. 4.5b).
• Overestimated energy resolution. The energy resolution is determined by the com-
bined effects of all fluctuations that may occur in the shower development. Signal
saturation leads to the suppression of a certain source of fluctuations, and the actual
resolution is thus worse than measured (Fig. 4.5c).
A calorimeter of this type was built by the CALICE Collaboration [5]. The active
elements of this detector are resistive plate chambers (RPCs) with in total 500,000
small readout pads (a cross section of 1 × 1 cm2 each). These devices, which operate
in the saturated avalanche mode, produce a signal when they are traversed by a
charged particle. The calorimeter in question produces pretty energy deposit patterns
(Fig. 4.6), but is otherwise not a very good calorimeter. This is because the RPC does
not make a difference between 1, 3, 17 or 53 charged particles. It is a digital device,
with two options: yes or no.
It is clear that this type of calorimeter exhibits the effects outlined above: response
non-linearity (Fig. 4.7) and overestimated energy resolution. These effects are also
worse for em showers compared to hadronic ones because of the larger spatial density
of shower particles. The described effects could of course be mitigated by reducing
the size of the RPC readout pads, but the number of readout channels that would
have to be handled in that case might exceed the limit of what is reasonably possible
in that respect.
Similar, albeit probably easier to solve, problems are faced when silicon photo-
multipliers (SiPM) are used to detect the light signals produced by calorimeters. A
SiPM consists of a large number of very tiny pixels, each of which is a photon detec-
tor operating in the Geiger mode [6, 7]. At the moment of this writing, SiPMs with
up to 40,000 pixels per mm2 are the state of the art [8]. Since these pixels are also
digital detectors (yes/no), saturation effects will occur when two or more photons hit
the same pixel within the time used to collect the light signals, with the consequences
described above. In this case, these consequences might be mitigated by a further
reduction of the pixel size and/or a decrease of the intensity of the light signals to
which the SiPM is exposed.
4.3 The Texas Tower Effect 83
Fig. 4.8 Photograph of two CMS APDs (active area 5 × 5 mm2 ) mounted in a capsule (a). CMS
event display of a pp collision event, showing an isolated ECAL spike (top-right) simulating a 690
GeV transverse energy deposit (b). From: Petyt, D.A. (2012). Nucl. Instr. and Meth. A695, 293. A
nuclear reaction induced by a proton with a kinetic energy of 160 MeV in a photographic emulsion
(c). Photograph courtesy CERN
used, the hadronic section was exposed to beams of all available types of particles
(electrons, pions, kaons, protons, muons). The performance of both sections was
documented in detail, and found to be in agreement with expectations [10, 11].
Yet, when the entire calorimeter system was assembled and exposed to high-
energy hadrons, an unexpected surprise occurred [12]. In some fraction of the events,
anomalously large signals were observed. An example of this phenomenon is shown
in Fig. 4.8b. What was going on?
The APDs that convert the light produced in the crystals are also extremely sen-
sitive to ionizing particles. In fact, a mip traversing such an APD may create a
signal equivalent to several thousand light quanta [13]. Measurements performed
with muons traversing the CMS PbWO4 crystals revealed that a muon that passed
through the active layer of an APD generated a signal that was, on average, equivalent
to the signal from scintillation photons created by an energy deposit of 160 MeV
inside the crystal [14]. The signals produced by densely ionizing charged particles
are correspondingly larger.
In Fig. 4.8c, an example is given of interactions that are typical when a high-
energy hadron strikes an atomic nucleus. Several densely ionizing nuclear fragments
are visible in this picture, with d E/d x values that are up to 100 times larger than that
of a mip (see Fig. 4.4a). If such a nuclear interaction would take place in the vicinity
of an APD, the nuclear fragments traversing the active detector surface area could
produce a very large signal. Given the relationship between the signals from mips
and from scintillation photons, such an event could well mimick an energy deposit
of 100 GeV or more.
This phenomenon was only discovered when the em and hadronic calorimeter
sections were assembled together and exposed to high-energy hadrons. Since the em
section corresponds to about one nuclear interaction length, a substantial fraction
of hadrons entering the calorimeter start the shower development process in the
em section, and therefore the process that generates the “spikes” described above
becomes a realistic possibility.
4.3 The Texas Tower Effect 85
Fig. 4.9 Distinguishing characteristics of CMS “spike” events. Shown are the average pulse shape
(a) and the distribution of the so-called Swiss Cross topological variable for the highest energy
deposit in each event (b). See text for details. From: Petyt, D.A. (2012). Nucl. Instr. and Meth. A695,
293
As stated above, the nuclear reactions have to take place “close to” the sensor
surface of the APD in order to produce this effect. The scale on Fig. 4.8c clarifies
what “close” means in this context, i.e., within 100 µm or so. This also provides the
answer to the question how this problem could have been avoided. Since the two
APDs that read out each crystal are separated by several mm, a nuclear interaction
of the type discussed here would never affect both APDs, but only one of them.
Therefore, if the two APDs had been read out separately, instead of being treated
as one detector unit in the data acquisition system, “spike” events would be easily
recognized since the signal in only one of the APDs would be anomalously large,
while the signal from the other one would still provide the useful information one
would like to obtain from the event in question.
However, since the two groups that were responsible for the two calorimeter
sections worked in their own individual, separate universes, this problem was only
discovered when it was too late to make the corrections needed to avoid it. We include
this example here because it illustrates that it is important to realize that a calorimeter
built for a given experiment is not a stand-alone device, but is part of an integrated
system of detectors. The different components of this detector system may affect the
performance of each other and it is important to realize and test this in the earliest
possible stage of the experiment.
This phenomenon is very similar, at least in its consequences, to the Texas tower
effects in CDF, discussed above. Also here, one low-energy shower particle may cause
an event in which an anomalously large amount of energy seems to be deposited in
the detector (see Fig. 4.8). Contrary to CDF, CMS has made an effort to deal with the
real-life consequences of this phenomenon, and has succeeded in recognizing, and
eliminating, affected events to a reasonable extent. This is illustrated in Fig. 4.9, which
shows two characteristics that distinguish these “spike” events from the “normal”
86 4 The Calorimeter Signals
ones. One effective method is a cut on the so-called Swiss Cross variable, in which
the signal in each tower of the em calorimeter is compared to the sum of the signals
in the four neighboring towers. Monte Carlo simulations showed that a cut of events
in which E4/E1 < 0.05 is an effective tool for eliminating the “spike” events [12].
Also the time structure of the spike events is a differentiating characteristic. Since
the signals in this case are caused by shower particles traversing the APDs, they are
faster than the signals based on detection of the scintillation light generated by the
shower particles in the crystals. The latter are delayed because the molecules excited
in the scintillation process take some time to decay (∼10 ns). Figure 4.10 shows the
time distribution of the hits in the ECAL barrel with a reconstructed energy above
1 GeV, for an 80 ns time slice of the LHC operations. The 25 ns bunch structure is
clearly visible. The events caused by the scintillation light are characterized by the
peaks at −25 ns, 0 ns and +25 ns in this plot. These peaks are preceded by smaller
peaks that occur about 10 ns earlier. These are the spike events, and the figure shows
that these represent about 3% of the total. It should be emphasized that this plot
concerns the time characteristics of the events that survived the Swiss Cross cuts.
Based on the understanding of the underlying cause of this phenomenon, it is
expected that the frequency of these events will increase both with the luminosity
and with the center-of-mass energy of the pp collisions in the LHC. The reason
why this example is included in this context is that it is an inherent feature of the
CMS calorimeter system. Even though attempts to deal with the problem may seem
successful, it is good to keep in mind that any data selection based on the calorimeter
information alone may lead to a biased event sample. Also, it is inevitable that some
fraction of the events in which the process that causes a “spike” occurs will not be
eliminated by the cuts devised to deal with the problem.
4.3 The Texas Tower Effect 87
Given the origin of the problems described in this subsection, other examples of
catastrophic effects caused by a single shower particle may be expected in the High-
Granularity calorimeter that is scheduled to replace the radiation damaged endcaps
of the CMS calorimeter [15] in 2025. For example, in the hadronic section of this
instrument, 5 cm thick brass plates will be interleaved with 200 µm thick silicon
sensors that act as active material. The sampling fraction for mips is thus 6 · 10−4
in this (FH) section, and even less for showers. A nuclear interaction such as the
one shown in Fig. 4.8c may easily deposit 30 MeV in the silicon sensor if the event
takes place in the vicinity of the boundary between the active and passive material.
Since the silicon signals do not saturate for densely ionizing particles, this event will
be interpreted as a 50 GeV highly localized energy deposit. The nuclear reactions
responsible for this phenomenon are typically initiated by spallation neutrons with
kinetic energies of a few hundred MeV. Such neutrons may travel tens of centimeters
away from the shower axis before initiating the reaction that gives rise to the large
signal. They will also be prolifically produced in the absorption of soft hadrons that
constitute the “pileup” component of the events CMS is looking for, and one should
thus expect very large signals at (multiple) random locations in the calorimeter, for
essentially every bunch crossing. Just like the Texas Tower effect, the described
phenomena are a typical consequence of the development of hadron showers in a
calorimeter with a small sampling fraction and a non-saturating active medium, and
therefore play no role in calorimeters based on plastic-scintillator or liquid-argon
readout.
Fig. 4.11 Three modules of the CMS forward calorimeter, which consists of 2 × 12 such wedges.
The quartz fibers that serve as the active material in this sampling calorimeter are bunched towards
a readout box, where PMTs convert the Čerenkov light signals into electric pulses. Photograph
courtesy CERN
created by showers inside the calorimeter, they generate signals that in practice may
make the information provided by the very-forward calorimeters rather meaningless
for the physics analyses, and most definitely renders these calorimeters useless for
triggering purposes.
The solution envisaged for this problem is to replace the readout by sensors that
are less subject to this effect [16]. In a first stage, the replacement sensors will be
multi-anode PMTs with thinner glass windows. The idea is that light produced in
the calorimeter will cause similar signals in all anodes of the PMT, while the signal
from one single anode will dominate when (localized) light produced by a particle
traversing the glass window of a PMT is detected. In addition, algorithms intended to
distinguish the stray muon signals from those generated by particles absorbed in the
calorimeter are being developed. To that end, the time structure of the events is being
used, just as was done to handle the spike events discussed in Sect. 4.3 (Fig. 4.10). In
a later stage, CMS plans to replace the entire readout by a system based on silicon
photomultipliers.
References 89
References
5.1 Introduction
When designing a calorimeter system for a certain experiment, it is important that the
detector be sufficiently large to contain the showers of interest at an adequate level.
Shower particles escaping from the detector represent a source of fluctuations that
may affect the precision of the measurements. Another important design considera-
tion is the segmentation of the calorimeter into separate independent sections, which
record part of the signals generated by the developing shower. This segmentation is,
for example, intended to
• recognize and separate particles that develop showers in each other’s vicinity,
• identify the showering particle, and
• measure the direction of the showering particle
The energy deposit profiles of the showering particles are of course a crucial consid-
eration for the design decisions in that respect.
In this chapter, we present information on the absorber size needed to contain
the showers, on average, at a certain level, e.g., 95%, and on the shower profiles.
It turns out that the measured shower profiles are not necessarily equivalent to the
energy deposit profiles, but may depend on the signal generation mechanism. It
should also be emphasized that the average containment level in itself is no indica-
tion of the effects of shower leakage on the energy resolution and other aspects of
the calorimeter quality. These effects are determined by event-to-event fluctuations
about this average. It turns out that 5% longitudinal shower leakage has much larger
effects in that sense than 5% transverse shower leakage. This is further discussed in
Sect. 6.6.
Fig. 5.1 Size requirements for electromagnetic shower containment. The depth of a calorimeter
needed to contain electron showers, on average, at the 99% level, as a function of the electron
energy. Results are given for four different absorber media (a). Average lateral containment of
electron-induced showers in a copper and an aluminium based calorimeter, as a function of the
radius of an infinitely deep cylinder around the shower axis (b) [2]
5.2 Electromagnetic Showers 95
Longitudinal em shower profiles were discussed in Sect. 3.1. The lateral spread of
em showers is caused by two effects:
1. Electrons and positrons move away from the shower axis because of multiple
Coulomb scattering.
2. Photons and electrons produced in more isotropic processes (Compton scatter-
ing, photoelectric effect) move away from the shower axis. Also, bremsstrahlung
photons emitted by electrons that travel at a considerable angle with respect to
the shower axis may contribute to this effect.
The first process dominates in the early stages of the shower development, while the
second process is predominant beyond the shower maximum, particularly in high-Z
absorber media.
When the shower development in the plane perpendicular to the direction of
the incoming particle is discussed, the way data are presented frequently leads to
confusion. There are two different ways of presenting such data:
1. The energy density, i.e., the amount of energy per unit volume, is shown as a
function of the distance between that unit volume and the shower axis. We will
refer to energy distributions and shower profiles of this type as lateral energy
distributions and lateral shower profiles.
2. The energy contained in a radial slice of a certain thickness is shown as a function
of the distance between that slice and the shower axis. We will refer to energy
distributions and shower profiles of this type as radial energy distributions and
radial shower profiles.
In addition, we will use the term transverse, as well as radial or lateral, to describe
general, non-quantitative features of shower development in the plane perpendicular
to the direction of the incoming particle.
Of course, distributions/profiles of type #1 are considerably narrower than those of
type #2. In practice, both terms are being used for both types of distributions/profiles,
and it is sometimes not clear which of these two is shown in presentations. In this
book, we use the terminology defined above in a consistent manner.
Both types of distributions may either concern a certain longitudinal slice of the
absorbing structure, or the entire absorbing structure. In the latter case, the lateral or
radial profiles are said to be integrated over the full depth.
Figure 5.2 shows the radial distributions of the energy deposited by 10 GeV elec-
tron showers developing in copper, at various depths. The two mentioned components
can be clearly distinguished. Both show an exponential behavior (note the logarith-
mic ordinate in Fig. 5.2), with characteristic slopes of ∼3 mm (∼0.2ρM ) and ∼25
mm (∼1.5ρM ), respectively. The radial shower profile shows a pronounced central
core (the first component), surrounded by a halo (the second component). The central
core disappears beyond the shower maximum.
96 5 Containment and Profiles
This radial profile, integrated over the total shower depth, is shown in Fig. 5.3,
together with the equivalent profiles for 10 GeV electron showers developing in lead
and aluminium. The distance from the shower axis (plotted horizontally) is expressed
in Molière units. Scaling with ρM would imply that these three profiles are identical.
The figure shows that this is approximately true, but that there are also some clear
differences between these three profiles. And just as in the case of the longitudinal
5.2 Electromagnetic Showers 97
A
σ (E) = (5.1)
NA l
For example, in lead, the asymptotic cross section for photon interactions amounts
to 42 barns (Eq. 3.1), whereas the cross section drops to 15.6 b for E γ = 2 MeV
[5]. Therefore, the mean free path of γ s of a few MeV equals 42/15.6 × 9/7 = 3.5
X 0 . That is why radioactive 60 Co sources, which emit γ s of about 1.3 MeV, require
substantial lead shielding. These γ s are among the most penetrating ones available,
and therefore the most difficult ones to shield.
We have used Eq. 5.1 and the Tables from [5] to calculate the mean free paths
for γ s in the energy range from 1–3 MeV for a variety of absorber materials with
different Z values and have converted these into units of the Molière radius. The
results are given in Fig. 5.4. It turns out that this distribution gradually decreases
with increasing Z . In the Al–Cu region, the mean free path of these photons amounts
98 5 Containment and Profiles
to 1.6 − 1.8ρM , while for lead it has dropped to ∼1.0ρM . These values are in good
agreement with the observed slopes of the halo in Fig. 5.3.
Another effect that may contribute to the less steep halo slopes in low-Z materials
is that the absorption of low-energy bremsstrahlung photons requires, on average,
more steps than in high-Z materials. In Compton scattering, a photon of lower energy
is produced and this photon needs to be absorbed in a subsequent process. In the
energy range from 0.2–0.5 MeV, Compton scattering is the most likely process in
aluminium and iron (see Fig. 2.5), while in lead, photons in this energy range are
predominantly absorbed in a single-step process, photoelectron production.
All the results shown so far in this subsection were obtained with (EGS4) Monte
Carlo simulations. However, available experimental results confirm all the trends
mentioned above. Detailed measurements on em shower profiles were carried out
by Bathow and coworkers [6]. They studied the three-dimensional profiles induced
by 6 GeV electrons in aluminium, copper and lead. Their experimental data were
provided by arrays of tiny (a few mm3 ) silver-phosphate dosimeter glasses which
were installed at various depths inside the absorber blocks. In these glasses, lumi-
nescence centers were formed by ionizing radiation. Afterwards, the accumulated
dose could be determined, with a relative precision of about 5%, by exposing the
irradiated dosimeters to ultraviolet light.
The absorber blocks with the built-in dosimeters were irradiated with a 6 GeV
electron beam provided by the DESY synchrotron, at a rate of 5 · 1010 particles per
second. The dosimeter matrix thus accumulated a three-dimensional image, corre-
sponding to the average shower profile of a very large number of 6 GeV electrons,
that was “frozen” into the block and that could be analyzed offline.
Figure 5.5 shows some results of these experiments. In Fig. 5.5a, the longitudinal
shower profiles in aluminium, copper and lead are given, with the depth expressed
in units of X 0 . These curves show the same characteristics as the simulated ones
in Fig. 3.10: approximate scaling with X 0 , but as Z increases, the shower maximum
shifts to greater depth and the slope beyond the shower maximum becomes less steep.
In Fig. 5.5b, the lateral profile in lead is shown, at three different depths. As in the
simulated curves (Fig. 5.2), there are clearly two components visible, the steepest
of which disappears beyond the shower maximum. Figure 5.5c shows the energy
deposition, integrated over the full shower depth and plotted as a function of the
distance from the shower axis, in units of ρM . The latter curves also exhibit the two-
component structure, and the long-distance component is steeper for lead than for
copper, as in the EGS4 simulations (Fig. 5.3).
5.2 Electromagnetic Showers 99
Fig. 5.5 Experimental results on the shower profiles of 6 GeV electrons in aluminium, copper and
lead. Shown are the longitudinal profiles in these three materials (a), the lateral profiles in lead,
measured at 3 different depths (b), and the energy deposition integrated over depth as a function of
the distance to the shower axis, for different absorber materials (c). Data from [6]
It is commonly assumed that the radial energy deposit profile of em showers scales
with the Molière radius. For this reason, experiments typically choose the granularity
of their em calorimeter in terms of that parameter. For example, if a cell size with an
effective radius of 1ρ M is used, em showers deposit typically ∼80% of their energy
in one cell, if the particle enters that cell in its central region (Fig. 5.1b). In order to
increase that percentage to 90%, the effective radius of the cell has to be doubled,
i.e., the number of cells is reduced by a factor of four. Recent measurements with a
calorimeter that used a much finer granularity showed that the energy deposit profile
is very strongly concentrated near the shower axis [7].
100 5 Containment and Profiles
Fig. 5.6 Lateral profiles of electromagnetic showers in the brass-fiber dual-readout SiPM calorime-
ter, measured separately with the Čerenkov and the scintillation signals (a). The fraction of the
shower energy deposited in a cylinder around the shower axis as a function of the radius of that
cylinder, measured separately with the Čerenkov and the scintillation signals (b). One mm cor-
responds to 0.031ρ M in this calorimeter. From: Antonello, M. et al. (2018). Nucl. Instr. and
Meth. A899, 52
For example, when the granularity was increased by a factor of 20 (reducing the
effective radius of a detector cell from 1ρ M to 0.22ρ M ), one cell contained ∼45% of
the shower energy, and if the granularity was increased by another factor of 30 (to
cells with a radius of 0.04ρ M ), ∼10% of the total shower energy was still deposited
in one cell. Figure 5.6 shows the profiles measured with this detector, a dual-readout
calorimeter (see Sect. 8.3) with fibers read out by a silicon photomultiplier array. The
signals from each individual fiber were sensed by a 1 mm2 SiPM [7].
Figure 5.6a shows a remarkable difference between the profiles measured by
the two types of fibers that constitute the active material of this calorimeter. The
Čerenkov light is much less concentrated near the shower axis than the scintillation
light. This is a consequence of the fact that the early, extremely collimated component
of the developing shower does not contribute to the Čerenkov signals, since the
Čerenkov light falls outside the numerical aperture of the fibers. This phenomenon
has interesting consequences for the detection of muons with a device of this type,
since a comparison between the two signals makes it possible to distinguish the
ionization and radiative components of the energy loss by muons traversing this
calorimeter (see Sect. 11.4.1 [8]).
5.3 Hadronic Showers 101
When an experiment is designed, one of the most important decisions concerns the
total thickness of the calorimeter. Especially in a 4π geometry, this decision has
serious consequences for the cost of the experiment. Let us, as an example, consider
a spherical calorimeter that surrounds the interaction vertex, starting at a distance of 1
m. Let us also assume that the effective nuclear interaction length of this calorimeter is
20 cm. If we want to make this calorimeter 7λint thick, then its total volume amounts
to 43 π(2.43 − 1) = 53.7 m3 . Should we want to add one extra interaction length,
this volume would increase by 29% (69.4 m3 ), while the surface area of detectors
installed outside the calorimeter would increase by 17%. And since the cost of many
detectors is more or less proportional to the instrumented mass, a decision to go from
7λint to 8λint would have major financial implications.
In this subsection, we discuss the absorber thickness needed to contain hadron
showers, on average, at a certain level. We re-emphasize what was said in Sect. 5.1,
namely that the effects of shower leakage on the quality of the calorimeter data is
determined by event-to-event fluctuations about this average, and not by the average
shower containment itself. Both for em and for hadron showers, these fluctuations
are much larger for longitudinal leakage than for lateral shower leakage, at a given
level of shower containment. The reasons for this are discussed in Sect. 6.6.
Because of the practical implications mentioned above, there is plenty of ex-
perimental information about hadronic shower containment. Representative results,
obtained by WA1 for hadron absorption in iron [9], are shown in Fig. 5.7. The aver-
age shower fraction contained in the absorber material is shown as a function of the
absorber thickness, for showering pions with energies ranging from 10 GeV to 138
GeV. The absorber thickness needed to contain 95% of the shower energy ranges
from ∼3λint at 10 GeV to more than 6λint at 138 GeV. For 99% containment, the
absorber thickness has to be at least 5λint deep for 10 GeV pions and ∼9λint for 138
GeV ones.
Fig. 5.8 Average lateral profile of the energy deposited by 80 GeV π − showering in the SPACAL
detector. The collected light per unit volume is plotted as a function of the radial distance to the
impact point. From: Acosta, D. et al. (1992). Nucl. Instr. and Meth. A316, 184
In Sect. 3.2, longitudinal hadron shower profiles, as well as differences with em ones
are discussed. Hadron showers do not only start to develop until reaching a much
greater depth inside the absorber material, they are also considerably broader than
electromagnetic showers. The lateral shower profiles exhibit in most materials a
narrow core, surrounded by a halo. A representative shower profile, integrated over
the full depth of the absorber, is shown in Fig. 5.8. This profile was measured with
the SPACAL detector [10].
The narrow core represents the electromagnetic shower component, caused by
π 0 s produced in the shower development. The halo, which has an exponentially
decreasing intensity, is caused by the non-electromagnetic shower component. A
detailed comparison of lateral profiles measured with the SPACAL detector showed
that the radius of the cylinder around the shower axis needed to contain 80 GeV π −
showers at the 95% level is about 32 cm (1.5λint ), nine times larger than the 3.5 cm
(1.8 ρM ) radius for containing 80 GeV em showers at the same level [10].
Longitudinally, the difference in the amounts of material needed for containing
these two types of showers at a certain level is very similar (i.e., a factor of about
nine). Measurements showed that the average energy fraction leaking out at the back
of the 9.6λint deep SPACAL detector amounted to ∼0.3%, for showers induced by
80 GeV π − [11]. An average longitudinal containment of 99.7% for 80 GeV electron
5.3 Hadronic Showers 103
for the spatial density of thermal neutrons at a depth of 4λint inside the block of
depleted uranium.
The nuclide 99 Mo is a fission product of uranium. The threshold for neutron-
induced 238 U fission is about 1.5 MeV. Therefore, the distribution of this radioactive
nuclide is a measure for the spatial distribution of the non-thermalized, MeV-type
neutrons. Obviously, these neutrons have traveled a much smaller distance from their
point of origin and are therefore much more concentrated around the shower axis
than the thermal neutrons.
Finally, 237 U is most likely produced through the reaction 238 U (γ , n) 237 U. The
cross section for this process reaches a maximum value of ∼0.35 b, at a photon
energy of about 11 MeV [14]. Gammas of this energy are abundantly produced in
the em showers generated by π 0 s. Therefore, one expects to find 237 U concentrated
in the narrow shower core (Fig. 5.8), close to the shower axis.
Transverse shower containment results are shown in Fig. 5.11, where the energy
fraction contained in a cylinder around the shower axis is plotted as a function of
the radius of this cylinder, for pions of several energies showering in lead absorber
[10]. Unlike for em showers, the results do depend in this case on the energy of the
showering particle, in a way that at first sight seems counter-intuitive: the higher
the energy of the incoming particle, the narrower the cylinder needed to contain the
shower. For example, to contain 10 GeV pions at the 95% level, a cylinder with a
radius of ∼1.7λint is needed, while 1.4λint is enough for 150 GeV pions.
5.3 Hadronic Showers 105
Fig. 5.11 Average energy fraction contained in an infinitely long cylinder of absorber material,
as a function of the radius of this cylinder (expressed in nuclear interaction lengths), for pions
of different energy showering in lead absorber. From: Acosta, D. et al. (1992). Nucl. Instr. and
Meth. A316, 184
Fig. 5.12 Signal distributions for 300 GeV pions (a) and protons (b) in the CMS forward calorime-
ter. Average signals per GeV for protons and pions as well as the ratio of these response values
in this detector, as a function of energy (c). From: Akchurin, N. et al. (1998). Nucl. Instr. and
Meth. A408, 380
Experimental studies have confirmed these effects [15, 16]. Figure 5.12 shows
the signal distributions measured for 300 GeV pions (a) and protons (b), respec-
tively. The signal distribution for protons is much more symmetric, as indicated by
the Gaussian fit. This is because the em component of proton-induced showers is
typically populated by π 0 s that share the energy contained in this component more
evenly than in pion-induced showers. The figure also shows that the rms width of
the proton signal distribution is significantly smaller (by ∼20%) than for the pions.
Figure 5.12c shows that the average signal per GeV deposited energy is smaller for
the protons than for the pions, by about 10%. This is also a consequence of the
limitations on π 0 production that affect the proton signals. So while the response to
protons is smaller in this calorimeter, the energy resolution is better. Similar effects
are expected to play a role for the detection of kaons, where π 0 production is limited
as a result of strangeness conservation in the shower development.
The ATLAS Collaboration used the longitudinal profile data to determine the
shower leakage expected in their calorimeter system. Since the effective calorimeter
thickness increases with the pseudorapidity (η), the leakage decreases with the angle
between the particle trajectory and the beam line. Figure 5.13 shows the average
energy leakage for 100 GeV pions and protons as a function of η. The correspond-
ing effective calorimeter thickness is plotted on the top axis. The smaller leakage
observed for protons is a reflection of the fact that the shower profile for these parti-
cles is a bit shorter because of the smaller interaction length, combined with leading
particle effects [16].
5.3 Hadronic Showers 107
Only relativistic, charged shower particles contribute to the signals from calorimeters
based on Čerenkov light. Their velocity should exceed the Čerenkov threshold: v >
c/n, where n represents the refraction index of the medium in which the particles
travel. Media that are frequently used in Čerenkov calorimeters include water (n =
1.33), quartz (n = 1.46) and various types of lead-glass (n = 1.5–1.75). Shower
particles that may contribute to the signals include
• Electrons and positrons. For n values of ∼1.4, these particles emit Čerenkov light
when their kinetic energy exceeds 200 keV.
• Charged pions. In typical calorimeter media, these short-lived particles emit
Čerenkov light when their total energy is larger than about 190 MeV.
• Protons. These need to carry a kinetic of at least 400 MeV in order to generate
Čerenkov light in such calorimeters.
The signals from Čerenkov calorimeters depend completely on the extent to which
Čerenkov-capable particles are produced in the shower development. As we have
seen before, hadronic shower development involves various processes, each with
very different rates of Čerenkov-capable particle production:
• π 0 s produced in hadronic shower development give rise to em showers. Most of
the energy of these π 0 s is deposited through electrons and positrons above the
Čerenkov threshold (kinetic energy larger than 200 keV).
• Of the energy carried by the non-electromagnetic shower component, on average
∼20% is deposited by charged pions.
• The rest of the non-electromagnetic energy is deposited by protons and neutrons,
almost all of which are non-relativistic, and through release of nuclear binding
energy, which leaves no directly measurable signal.
108 5 Containment and Profiles
Since electrons and positrons produced in the showers dominate the signals from
this calorimeter, hadron showers thus register predominantly through their electro-
magnetic shower core. This has several important consequences.
One of these consequences concerns the (three-dimensional) hadronic shower
profiles. The instrumented volume needed to contain the Čerenkov-capable shower
component is substantially smaller than that required for full containment of the
entire hadron shower. This is true both in depth (longitudinal containment) and in
the transverse plane, since the shower tails in all directions are primarily composed
of non-relativistic particles (soft nucleons).
For this reason, hadron showers in Čerenkov calorimeters appear to be consider-
ably narrower than in other types of calorimeters. This may be an important advan-
tage when particle densities are very high (for example, in the high-η region of LHC
experiments), since it reduces shower overlap.
This point is illustrated in Fig. 5.14, which shows transverse characteristics of
showers, initiated by 80 GeV π − mesons in the Quartz Fiber Calorimeter (Čerenkov
light, copper absorber) and in the SPACAL calorimeter (scintillating fibers, lead
absorber, sensitive to all ionizing particles, cf. Fig. 5.8). The profiles in the Čerenkov
calorimeter are considerably narrower, even though both the Z value and the density
of the absorber in this calorimeter were considerably smaller than in the other one.
This figure illustrates one very important point, which is emphasized time and
again in this book, namely that calorimeter signals are in general not proportional to
the amount of energy deposited in the area from which they are collected. This be-
comes extremely clear in the case of the Čerenkov calorimeter. The profile measured
with the Quartz Fiber detector is not at all representative for the energy deposit profile
in the shower development. It just measures the transverse distribution of Čerenkov
light produced in this process and thus the transverse distribution of shower parti-
cles capable of emitting such light. Since most of these particles are electrons and
positrons generated in the em shower core, this light is concentrated near the shower
axis.
5.4 Application of Differences 109
The large difference between the radiation length and the nuclear interaction length in
high-Z materials can be successfully exploited for electron/pion separation. In many
calorimeter systems, this difference manifests itself in the form of a very different
energy sharing between the longitudinal calorimeter sections, for showers induced
by electrons and pions, respectively.
However, the difference between (the very early phases of) em and hadronic
shower development may also offer excellent opportunities for e/π separation in
structures that are very much thinner than a shower-absorbing calorimeter system.
This feature forms the basis of many preshower detectors. An extremely simple
preshower detector (PSD) may consist of a plate of lead, 1 cm (1.9 X 0 , 0.06 λint )
thick, followed by a sheet of plastic scintillator. When a beam consisting of a mixture
of high-energy electrons and pions is sent through this device, almost all pions (96%)
traverse it without strongly interacting. These pions produce a minimum ionizing
peak in the scintillator. On the other hand, the electrons lose a considerable fraction
of their energy by radiating large numbers of bremsstrahlung photons. Some of these
photons convert into e+ e− pairs in the PSD and thus contribute to the scintillation
signals produced by this device.
The result is a very clear separation between electrons and pions. Figure 5.15
shows the signal distributions for 75 GeV electrons and pions in the described device,
used in beam tests of the CDF Plug Upgrade calorimeter [18, 19]. Even with such
simple devices, pion rejection factors of the order of hundred are readily achieved.
Such preshower detectors are particularly helpful during beam tests, for example
for obtaining a pure sample of pion events. Especially at low energy, pion beams at
accelerators are often contaminated with electrons. By placing a device of the type
described above in front of the calorimeter to be tested, this electron contamination
can be effectively eliminated, through a cut on the PSD signals. Since pions are
110 5 Containment and Profiles
defined as events that cause a mip signal in the PSD, the pion sample is not biased
by this procedure.
Although the electron signals from the described PSD are substantial (compared
with the signal from a mip), the fraction of the energy lost in this very early shower
development phase is small. Provided that the PSD is placed directly in front of the
calorimeter, the effect on the energy resolution of the latter is even smaller [20].
References
6.1 Introduction
In the practice of a particle physics experiment, one will want to use a given calorime-
ter signal to determine the characteristics (e.g., the energy) of the particle that pro-
duced it. In order to be able to make a statement about the energy of a detected
particle, one needs to know
1. the relationship between the measured signals and deposited energy (i.e., the
detector calibration), and
2. the energy resolution of the calorimeter.
The energy resolution determines the precision with which the (unknown) energy
of a given particle can be measured. It is experimentally determined from the precision
with which the energy of particles of known energy is reproduced in the calorimetric
measurements.
The energy resolution is often considered the most important performance char-
acteristic of a calorimeter. In particle physics experiments, the energy resolution of
the calorimeter may be the factor that limits the precision with which the mass of
new particles can be determined (e.g., the top quark). It may limit the separation
between particles with similar masses (e.g., in the jet–jet decay of the intermediate
vector bosons W and Z ). And it determines the signal-to-background ratio in event
samples collected in almost every experiment.
In this chapter, the factors that contribute to and limit the energy resolution of
calorimeters are discussed. This resolution is determined by (a) the effects of fluctu-
ations in the absorption process and (b) systematic effects. In Sect. 4.2, an example of
a systematic effect was described, which was the result of the calibration procedure
for the different sections of the calorimeter in question (Fig. 4.2). Such effects are
typically ignored in physics analyses, but are nevertheless very real. At the end of
this chapter, some other systematic effects that tend to be swept under the carpet are
discussed. However, we start with the effects of fluctuations on the precision with
which calorimetric measurements can be performed.
In the shower development that takes place when high-energy particles are absorbed
in the block of matter that we call a calorimeter, the energy of the particles is degraded
to the level of atomic ionizations or excitations that may be detected. The precision
with which the energy of the showering particles can be measured is limited by
1. fluctuations in the processes through which the energy is degraded, and
2. the technique chosen to measure the final products of the cascade processes.
The fluctuations in the shower development process are unavoidable. In elec-
tromagnetic showers, they determine the ultimate limit on the achievable energy
resolution. However, because of the chosen measurement techniques, the energy res-
olutions obtained in practice with em calorimeters are usually considerably worse
than that.
The situation is quite different for hadron calorimeters. Contrary to em showers,
in which the entire energy of the incoming particle is used to “heat up” the absorber,
in hadron showers some fraction of the initial energy is used to break up atomic
nuclei. This fraction, usually referred to as the invisible energy component since it
does not contribute to the calorimeter signals, varies wildly from one event to the
next. As a result, event-to-event variations in the calorimeter signal are much larger
than for em showers of comparable energy, since this phenomenon has no equivalent
in the latter. These fluctuations, and methods to mitigate their effects, are the topic
of Chaps. 7 and 8.
Many, but not all, fluctuations contributing to the energy resolution of calorimeters
obey the rules of Poisson statistics. For example, fluctuations in the number of quanta
(scintillation or Čerenkov photons, ion–electron or electron–hole pairs, etc.) that
constitute the calorimeter signals are Poissonian, but shower leakage fluctuations are
not.
Let us assume that a particle with energy E creates a signal S that, on average,
consists of n signal quanta (e.g., photoelectrons). Event-to-event fluctuations in the
signal correspond to Poisson (or Gaussian, for n 20) fluctuations in the number n.
The relative width of the signal distribution, σ S /S, i.e., the relative
√ precision√ of the
calorimetric measurement of the energy, σ E /E, is then equal to n/n = 1/ n.
If the calorimeter is linear, it will produce a signal that consists, on average, of 4n
quanta when it absorbs a particle with energy 4E. The event-to-event fluctuations
in the detection of such particles correspond to Gaussian fluctuations in the number
4n. The relative precision
√ of the calorimetric√ measurement of the √ energy of these
particles amounts to 4n/4n = 0.5/ n, i.e., a factor of two (= 4) better than
for particles with energy E.
For linear calorimeters measuring signal quanta that obey the rules of Poisson
statistics, these considerations lead directly to the familiar relationship
√
σ E /E = a/ E (6.1)
6.2 The Effects of Fluctuations on the Calorimeter Performance 113
We will follow the conventional practice of expressing the energy resolution of the
calorimeter, σ E /E, as a dimensionless number, representing a fraction of the particle
energy E, e.g., 3.6%. Unless stated otherwise, we will assume that the particle energy
is given in units of GeV. It has become customary to characterize calorimeters, for
what concerns the precision with which they can measure the energy of the particles
they absorb, in terms of the value of a. In this convention, a thus represents the energy
resolution for a 1 GeV energy deposit. Since calorimetry is based on statistical
processes (the production of ionization charge, photons, electron–hole pairs, the
excitation of Cooper pairs, etc.), the relative precision of the energy measurement
(σ E /E) thus improves with increasing energy. For example, an em calorimeter for
which a = 10% will measure electrons of 4 GeV with a resolution of 5%, while the
resolution will improve to 2% for 25 GeV electrons.
This very attractive feature has greatly contributed to the popularity of calorimeters
in particle physics experiments. For other particle detection techniques, the relative
precision of the measurements tends to deteriorate with increasing energy. This is
most noticeably the case for momentum measurements in a magnetic field, where the
√
size of the spectrometer has to increase proportional to p to keep the momentum
resolution Δp/ p constant.
However, not all types of fluctuations contribute to the calorimetric energy resolu-
tion as E −1/2 . Some fluctuations are energy independent, e.g., fluctuations resulting
from non-uniformities in the calorimeter structure. Other fluctuations may depend
on the energy in a different way, e.g., fluctuations resulting from electronic noise
(E −1 ), or from lateral shower leakage (E −1/4 ).
Not all types of fluctuations have a symmetric probability distribution around
a mean value. As an example, we mention fluctuations in f em , the em fraction of
hadron showers. The probability of finding f em values larger than the most probable
one is larger than that of finding smaller values (Fig. 7.2b).
In many cases, several sources of fluctuations contribute to the energy resolution
of a given calorimeter. For example, Fig. 6.1 shows the energy resolution of the em
barrel calorimeter of the ATLAS experiment as a function of energy, together with
the various contributions to this energy resolution. Since their energy dependence
may be different, the relative importance of each of these sources depends on the
energy. For example, instrumental effects that cause energy-independent signal fluc-
tuations tend to dominate at high energy, where the effects of Gaussian fluctuations
has become very small. On the other hand, electronic noise, an important factor
in LAr calorimeters, dominates the energy resolution at low energy, where its E −1
dependence overtakes the E −1/2 contributions from Gaussian fluctuations.
Most sources of fluctuations that contribute to a calorimeter’s energy resolution
are mutually uncorrelated. If that is the case, as in Fig. 6.1, the uncertainties they
cause in the energy of the particles may be added in quadrature. This means that if
sources 1, 2 and 3 cause fluctuations with standard deviations σ1 , σ2 and σ3 in the
measurements of particles with energy E, then the total energy resolution amounts
to σ E /E, with
114 6 The Energy Resolution of Calorimeters
σE = (σ1 )2 + (σ2 )2 + (σ3 )2 = σ1 ⊕ σ2 ⊕ σ3 (6.2)
If the various sources are completely or partially correlated, their effects have to be
combined accordingly. Depending on the details, this may result in resolutions that
are either better or worse than expressed in Eq. 6.2.
The horizontal scale of Fig. 6.1 is one that is used as the standard throughout
this book. The scale is linear in E −1/2 , with the origin located in the bottom right
corner. In that way, the energy increases from left to right, as is common in plots that
describe energy dependence, to reach E = ∞ for E −1/2 = 0. Since the horizontal
axis is chosen in this way, experimental data that scale with E −1/2 (Eq. 6.1) will be
located on a straight line that extrapolates to the origin (0% resolution at infinite
energy). The main reason for choosing this format is that it makes it immediately
clear if and what type of deviations from E −1/2 scaling are playing a role in the
presented data. In general, the energy itself will be plotted on the top axis of plots of
this type, which will further help to appreciate their contents.
Fluctuations in the number of detected signal quanta form the ultimate limit for the
energy resolution that can be achieved with a given calorimeter. However, in most
calorimeters, the resolution is dominated by other factors, discussed in the following
sections. In this section, we describe examples of detectors in which fluctuations in
the number of detected signal quanta determine the calorimeter resolution.
6.3 Signal Quantum Fluctuations 115
Fig. 6.2 Detection of nuclear γ -rays, from the decay of 152 Eu, with a high-purity germanium
crystal. The energy resolution of this calorimeter is about 0.1% at 1 MeV. Courtesy of G. Roubaud,
CERN
The first example concerns the nuclear γ detectors based on semiconductor crystals,
such as Ge, Ge(Li) and Si(Li). It takes very little energy to create one electron–hole
pair in these crystals, e.g., only 2.9 eV in germanium. The signal generated by a 1 MeV
γ fully absorbed in such a crystal therefore consists of some 350,000
√ electrons. The
fluctuations in this number lead to an energy resolution of 1/ 350, 000, or 0.17%
(at 1 MeV!). In terms of Eq. 6.1, this means that a = 0.005%, orders of magnitude
smaller than anything we will see for “conventional” calorimeters used in particle
physics experiments.
Owing to correlations in the production of consecutive electron–hole pairs (the
so-called Fano factor [3, 4]), the limit on the energy resolution given by fluctuations
in the number of primary processes is even smaller than indicated above. In practice,
energy resolutions close to 1.0 keV at 1 MeV are indeed achieved with such detectors
(Fig. 6.2).
resolution. For example, X-rays of 6 keV have been measured with resolutions of
about 0.1% in such devices [5]. This can only be achieved if the number of primary
processes is of the order of ∼106 . In terms of Eq. 6.1, this translates into a coefficient
a for the stochastic term of the energy resolution of the order of 10−6 !
The next example of detectors in which signal quantum fluctuations may play an
important role concerns Čerenkov calorimeters. The numbers of signal quanta con-
stituting the signals in such calorimeters are many orders of magnitude smaller than in
the previous examples. Detectors of this type include lead-glass em shower counters,
water Čerenkov counters (widely used in cosmic-ray experiments and proton-decay
studies), as well as a variety of sampling calorimeters based on quartz as active
medium. All these calorimeters are based on detection of Čerenkov light emitted by
relativistic shower particles with velocities in excess of c/n.
In em showers, the signals are produced by relativistic electrons and positrons gen-
erated in the shower development. Typically, the Čerenkov threshold corresponds
to a total energy of ∼0.7 MeV for these particles (kinetic energy >0.2 MeV).
Figure 3.9 indicates that in practice most of the em shower energy is deposited
by particles capable of emitting Čerenkov light.
The Čerenkov mechanism accounts for an energy loss that is about 4 orders of
magnitude smaller than the energy loss by ionization for superluminous charged
particles (see Sect. 1.4.2). The spectrum of the emitted light has a characteristic
1/λ2 dependence, and thus has a dominating UV component. It is estimated that
in media with n ∼ 1.5 (e.g., water, quartz) the light yield in the visible part of the
spectrum amounts to ∼30, 000 photons per GeV deposited energy [1]. The light
yield in practical detectors is of course much smaller, because of light absorption,
incomplete coverage by the light detectors, inefficiencies in the conversion of photons
into photoelectrons (p.e.) and, in sampling calorimeters, the sampling fraction.
The largest light yield, ∼7, 000 p.e./GeV has been reported by Super-Kamiokande,
which has a photocathode coverage of ∼40% [6]. In lead-glass detectors, that cover-
age is much smaller and, therefore, only a small fraction of the Čerenkov photons pro-
duced are actually detected. Monte Carlo simulations lead to an estimate of ∼1,000
p.e./GeV [1]. Statistical fluctuations
√ in this number would lead to an expected energy
resolution (σ/E) of 3.2%/ E. √
The best lead-glass detector systems have reached energy resolutions of ∼5%/ E
for electromagnetic showers in the energy range from 1 to 20 GeV [7, 8]. This might
indicate that Gaussian fluctuations in the number of photoelectrons are not the only
factor determining the energy resolution of this type of calorimeter, although it is most
definitely a very important factor. Additional factors that may play a role include,
but are not limited to
6.3 Signal Quantum Fluctuations 117
Fig. 6.3 The energy resolution for electron detection with the QFCAL prototype detector, as a
function of energy. Results are given for measurements in which photomultiplier tubes with a glass
window were used and for measurements in which PMTs of the same type were equipped with a
quartz window. From: Akchurin, N. et al. (1997). Nucl. Instrum. Methods A399, 202
Fig. 6.4 Signal distributions for 10 GeV (a) and 200 GeV (b) electrons showering in the CMS
quartz-fiber calorimeter, measured with a PMT with a glass window. The curves represent Gaussian
fits to the experimental data. From: Akchurin, N. et al. (1997). Nucl. Instr. and Meth. A399, 202
The light yield of this calorimeter is very small indeed. At 10 GeV, the measured
signals were composed of 5.3 photoelectrons, on average, when read out with a PMT
with a glass window. This small number of photoelectrons caused an asymmetric line
shape (Fig. 6.4a), characteristic for a Poisson distribution Pn for a discrete variable
n with a small average value μ:
6.3 Signal Quantum Fluctuations 119
μn −μ
Pn = e (6.3)
n!
At high energy, the number of photoelectrons was so large that the Poisson fluctu-
ations led to a symmetric, Gaussian line shape. This is illustrated in Fig. 6.4b, for 200
GeV electrons, whose average signal consisted
√ of 106 photoelectrons. The energy
resolution σ/E thus corresponded
√ to 1/ 106, or 9.7%, for these high-energy elec-
trons. It improved to 1/ 174 (7.6%) when the calorimeter was read out with a PMT
with a quartz window.
One type of detector for which the energy resolution is not limited by fluctuations in
the number of signal quanta, is the scintillation counter. This may be illustrated by the
following experimental data. Measurements with NaI(Tl) crystals on 6 keV X-rays
have yielded a resolution σ E /E ≈ 15%. If we assume that this result is dominated
by fluctuations in the number of signal quanta, then this implies that the signals
consist on average of ∼40 photoelectrons. On the√basis of this result, one would then
for 1 MeV γ -rays expect resolutions of 15%/( 1,000/6) ≈ 1.2%. Yet, in reality
the best resolutions obtained at this energy are only about 5%. In NaI(Tl) crystals,
electromagnetic showers
√ are detected with resolutions of about 1% at 1 GeV, whereas
a factor of thirty ( 1,000) improvement of the 1 MeV result should be expected if
photoelectron statistics limited the resolution. Clearly, the energy resolution of this
and other scintillation calorimeters is dominated by other factors.
When factors that are not determined by Poisson statistics contribute, the energy
resolution does not scale as E −1/2 with the energy of the showering particles. This is
illustrated by Fig. 6.5, in which the data points do not extrapolate to the bottom right
hand corner, as in Fig. 6.3. These data were measured with a sampling calorimeter,
which was built in the context of the SPAKEBAB R&D project [10]. This detector
had a sandwich structure. It consisted of a large number of very thin (0.63 mm thick)
lead sheets, interleaved with 1 mm thick plastic scintillator plates. The scintillation
light was transported to the rear end of the calorimeter by means of a large number
of wavelength-shifting fibers, which were spaced by 4 mm and ran through the entire
detector structure. Because of the very high sampling frequency, this calorimeter had
an em energy resolution that was among the very best ever achieved with a sampling
calorimeter.
It turned
√ out that the contribution of Poisson fluctuations to the energy resolution
(5.7%/ E) was dominated by sampling fluctuations, and not by fluctuations in the
number of scintillation photoelectrons. As a matter of fact, this is how it should be.
Sampling fluctuations should always dominate in well designed sampling calorime-
ters based on scintillation or ionization charge collection. In Sect. 6.5, we describe
how the contributions from sampling fluctuations and signal quanta fluctuations can
be distinguished and measured.
120 6 The Energy Resolution of Calorimeters
Fig. 6.5 The energy resolution of the SPAKEBAB calorimeter (0.63 mm sampling layers) for
electrons as a function of the electron energy. For comparison, the results for σ/E = a E −1/2 , with
a = 5% and 6% (the straight lines) are given as well. From: Dubois, O. et al. (1996). Nucl. Instr.
and Meth. A368, 640
Sampling calorimeters, which are typically much cheaper than homogeneous ones,
are especially competitive at higher energies. In properly designed instruments of
this type, the energy resolution is determined by sampling fluctuations. These rep-
resent fluctuations in the number of different shower particles that contribute to the
calorimeter signals, convolved with fluctuations in the amount of energy deposited
by individual shower particles in the active calorimeter layers. They depend both on
the sampling fraction, which is determined by the ratio of active and passive material,
and on the sampling frequency, determined by the number of different sampling ele-
ments in the region where the showers develop. Sampling fluctuations are stochastic
and their contribution to the energy resolution can be described by
asamp
(σ/E)samp = √ , with asamp = 0.027 d/ f samp (6.4)
E
in which d represents the thickness of individual active sampling layers (in mm), and
f samp the sampling fraction for minimum ionizing particles (mips). This expression
describes the em energy resolution obtained with a variety of different sampling
calorimeters based on plastic scintillator or liquid argon as active material reasonably
well (Fig. 6.6) [11].
Table 6.1 lists some characteristics of a representative selection of sampling
calorimeters used in partlcle physics experiments. Above 100 GeV, the resolution of
all calorimeters mentioned above is ∼1%, and systematic factors, such as stability of
the electronic components, the effects of light attenuation, or temperature variations
of the light yield, tend to dominate the performance.
6.4 Sampling Fluctuations 121
Fig. 6.6 The em energy resolution of a variety of sampling calorimeters as a function of the
parameter (d/ f samp )1/2 , in which d is the thickness (in mm) of an active sampling layer (e.g., the
diameter of a fiber or the thickness of a liquid-argon gap), and f samp the sampling fraction for mips.
The energy E is expressed in units of GeV [11]
Table 6.1 A representative selection of electromagnetic sampling calorimeters used in past and
present particle physics experiments. The energy E is expressed in GeV, the sampling fraction f samp
refers to minimum ionizing particles. For the energy resolution, only the E −1/2 scaling term, which
dominates in the practically important energy range for these experiments, is listed
Experiment Calorimeter X 0 (cm) f samp (%) σ/E References
structure
√
KLOE Pb/fibers 1.6 17 4.7%/ E [12]
(Frascati)
238 U/scintillator
√
ZEUS 0.7 9 18%/ E [13]
(DESY)
√
NA48 Pb/LKr 1.5 23 3.5%/ E [14]
(CERN)
√
ATLAS Pb/LAr ≈3 ≈25 10%/ E [15]
(LHC)
√
PHENIX Pb/scintillator 3.1 29 7.8%/ E [16]
(RHIC)
√
AMS-02 Pb/fibers 1.3 19 10.4%/ E [17]
(ISS)
The validity of Eq. 6.4 is limited to calorimeters with plastic scintillator or liquid
argon/krypton as active material. When the active layers are very thin (in terms of
stopping power), as in calorimeters with gaseous or silicon readout, an additional fac-
tor contributes to the energy resolution: pathlength fluctuations. In such calorimeters,
122 6 The Energy Resolution of Calorimeters
Fig. 6.7 The em energy resolution of sampling calorimeters with silicon layers as active material.
From: Wigmans, R. (2018). J. Progr. Part. Nucl. Phys. 103, 109. Experimental data from [19, 20]
the energy deposited by a typical shower electron depends on its trajectory inside the
active material. For example, the energy loss of electrons in 100 µm silicon amounts
to ∼115 keV [18]. The signal from shower electrons with energies larger than
115 keV produced in Compton scattering or photoelectric effect therefore depends on
the angle at which they traverse an active layer. The larger the angle with the shower
axis, the larger the contribution of these particles to the signal. For 500 µm silicon,
the same is true for shower electrons with energies larger than 330 keV. And since the
sampling fluctuations are determined by fluctuations in the total energy deposited by
the shower particles, and since these soft electrons are an important component of the
developing em showers, these pathlength fluctuations are important for calorimeters
with very thin active layers.
This is illustrated by Fig. 6.7, which shows the energy resolution for calorime-
ters with thin silicon layers as active material. All calorimeters have approximately
the same structure. Absorber layers of tungsten with a thickness that increases with
depth from 1.5 mm to 4.5 mm are interleaved with thin layers of silicon. In the four
configurations of which the em energy resolution is displayed, the thickness of the
silicon is 100 µm, 200 µm, 300 µm and 525 µm, respectively. These calorimeters
thus have different sampling fractions, but the ratio d/ f samp used in Eq. 6.4 is approxi-
mately the same. The experimental data point comes from √ CALICE [19], which used √
525 µm silicon, and obtained a resolution of 16.5%/ E (Eq. 6.4 gives ∼12%/ E
for this configuration). The other data points concern GEANT4 simulation results
for the HGCAL upgrade calorimeter for the CMS endcap region [20], which has
6.4 Sampling Fluctuations 123
about the same d/ f samp value as the CALICE one, but uses thinner √silicon layers.
√ resolution for this device ranges from 19.9%/ E for 300 µm
The expected energy
silicon to 24.3%/ E for 100 µm silicon. These results, once again, illustrate the
importance of the contribution of soft shower particles to the signals from sampling
calorimeters, discussed earlier in Sect. 3.5.2.
Because of the very small light yield, fluctuations in the number of photoelectrons are
almost always the dominating contribution to the em energy resolution of calorime-
ters that use Čerenkov light as the source of their signals. However, in sampling
calorimeters based on scintillation or direct detection of the ionization charge, sam-
pling fluctuations usually dominate, even though the contribution from fluctuations
in the number of signal quanta cannot be neglected either. Both types of fluctuations
contribute a term that scales as E −1/2 . In the following, we describe two examples
of detectors for which the relative contributions of these two sources have been dis-
entangled. The principle of the methods used in this procedure is in both cases to
make a change in the experimental configuration that affects only the fluctuations
one wants to investigate and measure the effects of that change.
6.5.1 SPAKEBAB
fore the contribution of photoelectron statistics to the relative precision of the energy
measurement amounts to √
(σ/E)p.e. = 1/ x E (6.5)
If now the number of photoelectrons is reduced by a factor f , the new energy reso-
lution becomes
(σ/E)filter = asamp
2 /E + f /x E = (σ/E)2nofilter + ( f − 1)/x E (6.8)
When f is known, x can thus be derived from the degradation of the energy
resolution. This method yields the most accurate results at low energy, where the
contributions of non-stochastic processes to the energy resolution are small and the
differences between the filtered and unfiltered energy resolutions large.
The measurements were performed with filters that reduced the number of pho-
tons by factors ( f ) of about three and ten, respectively, for electron showers of
6, 8, 10 and 15 GeV. By combining all results, it was found that this calorimeter
produced 1,300 ± 90 photoelectrons√per GeV deposited energy. Therefore, photo-
electron statistics contributed 2.8%/ E to the em energy resolution. √ Since the total
em energy resolution of this√detector was measured to be 5.7%/ E, sampling fluc-
tuations contributed 5.0%/ E (5.0 ⊕ 2.8 = 5.7).
For sampling calorimeters, this is a high light yield, the result of a very efficient
light collection. In many other scintillation calorimeters used in particle physics
experiments, the light yield is typically one order of magnitude smaller. Therefore, the
contribution of photoelectron
√ statistics to the energy resolution of such calorimeters
is typically ∼10%/ E.
6.5.2 ZEUS
The second example concerns the ZEUS Collaboration, who measured the contribu-
tion of sampling fluctuations to the electromagnetic and hadronic energy resolutions
6.5 Measuring the Contribution of Different Types of Fluctuations 125
ΔE A ΔE B ΔE sum ΔE diff
σA = , σB = , σsum = , σdiff =
E A E B E sum E sum
Fig. 6.8 Pulse height distributions for 30 GeV hadrons obtained with the ZEUS lead/plastic-
scintillator prototype calorimeter. Diagram a shows the distributions of E sum , E diff and 2E A , mea-
sured in the configuration depicted in Fig. 6.8c, with the black tape in place. Diagram b shows the
same distributions measured in the same configuration, but with the black tape removed. See text
for details. From : Drews, G. et al. (1990). Nucl. Instr. and Meth. A290, 335
This contribution to the resolution was obtained by removing the black tape (see
Fig. 6.8c) and repeating the measurements. In this new geometry, the two signals
E A and E B resulting from a developing shower were thus measured with one and
the same sampling device, but each signal contained only (approximately) half of
the photoelectrons generated in the event. Therefore, σdiff measured in that case the
fluctuations in the number of photoelectrons.
An example of the experimental results obtained in these measurements is shown
in Fig. 6.8a, b. This figure contains pulse height distributions measured for 30 GeV
hadrons with the 10 mm lead/2.5 mm plastic-scintillator prototype calorimeter. Each
of the two diagrams shows three distributions: E sum , E diff and 2E A . The latter dis-
tribution was obtained by multiplying the A signals by a factor of two and thus has
the same central value as the distribution of the summed signals of A and B (E sum ).
Diagram a shows the results obtained in the configuration depicted in Fig. 6.8c,
with black tape, in diagram b the black tape was removed. This figure exhibits the
following features:
• When the tape was removed, σdiff became considerably smaller. This means that
fluctuations in the number of photoelectrons were much smaller than the sampling
fluctuations.
6.5 Measuring the Contribution of Different Types of Fluctuations 127
• When the tape was removed, the distributions of E sum and 2E A were practically
identical. Since these two distributions only differ in the average number of pho-
toelectrons constituting the signals, this means that fluctuations in the number of
photoelectrons did not contribute significantly to the hadronic energy resolution
of this calorimeter.
• When the black tape was in place, the distributions of E sum and 2E A were not
identical. In this configuration, the sampling fraction was different by a factor of
two for these two distributions. Sampling fluctuations were, therefore, a major
contribution to the hadronic energy resolution of this calorimeter.
• The distributions of E sum and E diff had practically the same width when the black
tape was removed. This means that sampling fluctuations were not only a major
contribution to the hadronic energy resolution, but that these fluctuations com-
pletely dominated the resolution. Less surprisingly, the same phenomenon was
observed for the electron data.
The latter observations also reveals a very important other feature. If the widths of
the distributions of E sum and E diff are essentially the same, then this implies that the
signal distributions of E A and E B are completely uncorrelated. In both calorimeter
configurations, the same showers were sampled, but the shower particles contributing
to the signals in configuration A were not the same as those contributing to the signals
in configuration B. Therefore, the typical shower particle contributing to the signals
from this calorimeter traveled a distance that was short compared with the thickness of
one sampling layer. The signal contribution from shower particles that traversed more
than one sampling layer was negligible. This was earlier concluded by Willis and
Radeka, who did similar measurements for electrons in their much-finer-sampling
Fe/LAr calorimeter [22].
These measurements confirm that the signals from em calorimeters are dominated
by electrons with energies that are much smaller than the critical energy, and that
the signals from hadron showers are dominated by protons. Pions produced in the
shower development, which could traverse many sampling layers, play indeed a very
minor role.
After a careful analysis of all available experimental information, the ZEUS
authors were able to unravel the em and hadronic energy resolutions of these two
compensating calorimeters in their different contributing components. The results of
this analysis are summarized in Table 6.2.
From these results, the following conclusions can be drawn:
1. The energy resolutions for em showers are strongly dominated by sampling fluctu-
ations in these calorimeters, with a minor contribution coming from photoelectron
statistics (increasing asamp = 16.5% to atotal = 18.5% for the uranium detector,
and asamp = 23.5% to atotal = 24.5% for the lead detector). Given the size ot the
error bars, all one can say is that the light yield is 100 p.e./GeV.
2. The sampling fluctuations for electrons are in good agreement with Eq. 6.4. For
the uranium calorimeter, this formula predicted asamp = 15.6%, while the experi-
mental value was measured to be 16.5 ± 0.5%. For lead, the prediction of 21.9%
was also close to the experimental value (23.5 ± 0.5%).
128 6 The Energy Resolution of Calorimeters
Table 6.2 The contributions of sampling fluctuations and intrinsic fluctuations to the energy
resolutions for electrons and pions in compensating uranium/plastic-scintillator and lead/plastic-
scintillator calorimeters. Listed are the values of the coefficient a (Eq. 6.1), expressed in %. Data
from [21]
Fluctuations (%) 3 mm uranium/2.5 mm plastic 10 mm lead/2.5 mm plastic
Electrons Pions Electrons Pions
σA , σB 26.6 ± 1.0 49.5 ± 1.0 36.0 ± 1.0 60.5 ± 1.0
σsum 18.5 ± 1.0 37.3 ± 1.0 24.5 ± 1.0 43.5 ± 1.0
σdiff 19.2 ± 1.0 32.6 ± 1.0 25.8 ± 1.0 42.3 ± 1.0
σsamp 16.5 ± 0.5 31.1 ± 0.9 23.5 ± 0.5 41.2 ± 0.9
σintr 2.2 ± 4.8 20.4 ± 2.4 0.3 ± 5.1 13.4 ± 4.7
3. The energy resolutions for hadrons are also strongly dominated by sampling
fluctuations in these calorimeters, especially in the lead one. As a result, the
resolutions of calorimeters A and B are, in all cases, within √ experimental errors
equal to the resolution of the complete instrument times 2.
4. The hadronic sampling fluctuations are about a factor of two larger than the em
sampling fluctuations. This indicates that the number of different shower particles
constituting the hadronic signal is much smaller. One reason for that is the fact
that the specific ionization of the spallation protons that dominate the signals is
much larger than for mips (Fig. 4.4a).
5. The contributions of “intrinsic” fluctuations, i.e., the fluctuations that remain after
subtracting the contributions from sampling fluctuations, photoelectron statistics
and other instrumental effects from the total measured hadronic energy resolution,
scale with E −1/2 . This is typical for compensating calorimeters.
6. The intrinsic fluctuations are smaller in the lead calorimeter than in the uranium
one.
Results 5 and 6 are further discussed in Chap. 8.
means that shower particles escape the calorimeter. These particles may cause signals
in other detectors, e.g., the muon system, which may disturb the performance of these
detectors.
At a given (average) leakage level, the effects of shower fluctuations on the energy
resolution are much larger if the leakage occurs longitudinally than when the energy
leaks out sideways.
130 6 The Energy Resolution of Calorimeters
Fig. 6.9 The effects of longitudinal and lateral shower leakage on the energy resolution, as measured
for 15 GeV electrons (a) and pions (b) by the CHARM Collaboration in a low-Z calorimeter [23,
24]. From: Amaldi, U. (1981). Phys. Scripta 23, 409
This was already known a long time ago, as illustrated by Fig. 6.9. This figure
shows the effects of longitudinal and lateral shower leakage fluctuations on the energy
resolution of 15 GeV electrons (Fig. 6.9a) and pions (Fig. 6.9b), measured by the
CHARM Collaboration for their low-Z (marble absorber, Z eff ∼ 13) calorimeter
[23, 24]. A 10% lateral leakage had a smaller effect on the energy resolution of this
detector than a 5% longitudinal leakage, both for em and hadronic showers.
These differences can be qualitatively understood from the very different char-
acteristics of longitudinal and lateral shower fluctuations. The longitudinal shower
fluctuations, and thus the longitudinal leakage fluctuations, are for a very important
part driven by fluctuations in the starting point of the shower, i.e., by the interaction
characteristics of one individual shower particle (the first one). On the other hand,
event-to-event fluctuations in the lateral shower development, and thus the fluctua-
tions in lateral shower leakage, are determined by the interaction characteristics of
∼100 shower particles combined. As a result, these fluctuations are much smaller
than the longitudinal ones.
Shower leakage through the front face of the calorimeter (albedo) only plays a
significant role at very low energies. The shower particles that escape in this way
are, by definition, very soft, since they must be produced in scattering processes
from which they emerged at large angles. Examples of such processes are Compton
scattering and the photoelectric effect in the case of em showers and elastic neutron
6.6 Shower Leakage 131
Fig. 6.10 A comparison of the effects caused by different types of shower leakage. Shown are
the induced energy resolutions resulting from albedo, longitudinal and lateral shower leakage as
a function of the average energy fraction carried by particles escaping from the detector. The
longitudinal and lateral leakage data concern 10 GeV γ s, the albedo data are for γ -induced showers
of different (low) energies. Results from EGS4 Monte Carlo calculations in which the detector
was represented by a block of tin. From: Wigmans, R. and Zeyrek, M.T. (2002). Nucl. Instr. and
Meth. A485, 385
scattering in hadronic showers. Since the energies of the scattering products are at
maximum a few MeV, only low-energy showers are significantly affected.
The effects of the different types of shower leakage on the em energy resolution are
compared in Fig. 6.10. The effects of lateral shower leakage on the energy resolution
are much smaller than for longitudinal leakage. According to Fig. 6.10, as much as
10% leakage can be tolerated before the induced energy resolution resulting from
lateral leakage fluctuations exceeds 1%. A similar limitation would require 99%
longitudinal containment.
The relative insensitivity of the em calorimeter resolution to lateral leakage is quite
fortunate, since em showers tend to have considerable radial tails. This is illustrated
by Fig. 5.1b, which shows the average lateral leakage fraction as a function of the
radius of an infinitely deep calorimeter centered on the shower axis. It turns out that a
radius of two Molière radii (ρ M ) is more than adequate for 90% lateral containment.
This number is almost independent of the shower energy and the absorber material, in
stark contrast with the depth requirements. However, in order to capture 99% instead
of 90% of the shower energy, the detector mass has to be increased by an order of
magnitude.
The SPACAL Collaboration has made an extensive study of various aspects of
the shower leakage issue, for hadron-induced showers. Their lead/scintillating-fiber
calorimeter measured 9.5λint in depth, with a diameter of ∼4.7λint .
The longitudinal leakage from the back of this calorimeter was on average less
than 0.5%, even for the highest-energy pions with which the calorimeter was tested
(150 GeV). When the signals from all (155) readout cells were summed together,
132 6 The Energy Resolution of Calorimeters
the lateral leakage was found to be four times as large as the longitudinal one, at 150
GeV (Fig. 6.11). At lower energies, the ratio between lateral and longitudinal shower
leakage rapidly increased, to reach a factor of thirty at 10 GeV [25, 26].
It is interesting to note that the average fraction of the shower energy that leaked out
laterally increased at lower energies. This means that high-energy pion showers were
thus better contained than low-energy ones in this calorimeter (see also Sect. 5.3.3).
This is a direct consequence of the increase of the (average) energy fraction contained
in the em shower component, f em , with energy. At 10 GeV, this core represents on
average ∼30% of the shower energy, rising to ∼60% at 150 GeV.
Shower leakage had a relatively larger impact on the calorimeter resolution as the
energy increased. It turned out that the effect of lateral shower leakage on the hadronic
energy resolution was reasonably described by a E −1/4 term, added in quadrature to
the stochastic term (a1 E −1/2 ) that described the calorimeter resolution in the absence
of shower leakage:
a1 2 x 2
σ/E = √ + √ 4
+ ... (6.9)
E E
where x represents the average lateral leakage fraction. More detailed results of these
leakage studies can be found in [25, 27]. The essence of the conclusions about the
effects of incomplete shower containment described here was also confirmed by
studies that were performed with the ATLAS TileCal [28, 29].
6.7 Instrumental Effects 133
In real life, neither the calorimeter construction nor the environment in which it has
to operate is ideal, and this has consequences for the calorimeter performance. In
this section, we investigate the consequences of some common effects on the energy
resolution.
The effects that are discussed have in common that the associated fluctuations
do not scale with E −1/2 . This means that their relative contribution to the total
energy resolution is energy dependent. Most effects, namely those that cause energy-
independent fluctuations, are dominating the energy resolution at very high energies,
where the contributions from the processes governed by Poisson statistics are small.
However, the first effect to be discussed is an exception to this rule, since it dominates
the resolution at low energies.
For sampling calorimeters with an active medium based on direct collection of the
charge produced in the ionization-chamber (e.g., liquid argon) or proportional mode
(wire chambers), the signals typically amount to a few picocoulombs of charge per
GeV of shower energy.
The signals produced by these calorimeters correspond to the charge collected
during a certain time, which we will call the “gate time” in the following. Since
the detector has a certain capacitance, there is inevitably a contribution of electronic
noise to the signals. This means that, in the absence of a showering particle, the
integrated charge collected during the gate time fluctuates from event to event. Since
the standard deviation of the electronic noise fluctuations corresponds to a certain,
fixed energy, the contribution of this noise to the energy resolution of calorimetric
shower measurements, σ/E, scales like E −1 .
The stochastic fluctuations are completely uncorrelated to the noise and the noise
in the various electronic channels is completely incoherent (uncorrelated). Therefore
the total energy resolution can be written as a quadratic sum of the two terms
anoise astoch
(σ/E)tot = ⊕ √ (6.10)
E E
Because of its E −1 dependence, the noise term dominates the energy resolution at
low energy. This is illustrated in Fig. 6.1.
Although these effects mainly play a role in calorimeters based on direct collection
of the ionization charge, they may affect other types of calorimeters, in particular
scintillator-based ones, as well. In scintillator calorimeters, the PMT signals are
digitized and analyzed by means of analog-to-digital converters (ADCs). When no
PMT signals are offered, the ADC may still accumulate a certain amount of charge
134 6 The Energy Resolution of Calorimeters
during the gate time, resulting in a “pedestal.” To find the calorimeter signals, the
pedestal has to be subtracted from the raw signals.
However, the pedestal may exhibit fluctuations, e.g., due to ground loops, improper
impedance matching and a variety of other electronic problems. These pedestal fluc-
tuations play the same role for scintillator calorimeters as electronic noise does for
ionization calorimeters. The difference is that pedestal fluctuations can in general
be made insignificant, for example by increasing the gain of the PMTs, and thus the
size of the signals. In liquid-argon calorimeters, which are based on the collection
of non-amplified ionization charge, this option does not exist.
The second source of instrumental effects that may contribute to the energy reso-
lution may manifest itself in a variety of different ways, depending on the type of
calorimeter. They all have the same common origin: event-to-event fluctuations in
the sampling fraction (and thus the response) of the calorimeter volume in which
the shower develops. These fluctuations typically lead to an energy independent
contribution to the energy resolution. Examples of such effects include:
• Dependence of the sampling fraction on the impact point of the particle. This
phenomenon plays a role, for example, in the “accordion” Pb/LAr ECAL of the
ATLAS experiment, where it is an inherent consequence of the way in which
active and passive elements have been arranged [30]). In Fig. 6.12, the average
Fig. 6.12 The em response as a function of the impact position of the particles, for 90 GeV electrons
in a prototype Pb/LAr accordion calorimeter for the ATLAS experiment. From: ATLAS (1996).
The ATLAS Calorimeter Performance, report CERN/LHCC/96-40
6.7 Instrumental Effects 135
Fig. 6.13 Scatter plot showing the SPACAL signal for 150 GeV π − (a) and 150 GeV e− (b)
versus the center of gravity of the light production in the showers. The bottom scale shows the
lateral displacement of the shower’s center of gravity with respect to the particle’s impact point.
In the top scale, this displacement is converted into the average depth z of the light production.
From: Acosta, D. et al. (1991). Nucl. Instr. and Meth. A305, 55
• The line shape. Often, the measured signal distributions exhibit non-Gaussian tails.
In that case, one should quote the σrms value as the energy resolution. However,
some authors use another variable, in order to make the results less dependent
on the tails of the signal distributions they measure, and thus look better. This
variable, called rms90 , is defined as the root-mean-square of the values located
in the smallest range of reconstructed energies that contains 90% of the total
event sample (see, e.g., Fig. 6.14). For the record, it should be pointed out that
for a Gaussian distribution, this variable gives a 21% smaller value than σrms
(i.e., σfit ). Of course, one is free to define variables as one likes. However, one
should then not use the term “energy resolution” for the results obtained in this way,
and compare results obtained in terms of rms90 with genuine energy resolutions
138 6 The Energy Resolution of Calorimeters
Fig. 6.14 The line shape of the CALICE W/Si + Fe/plastic combination for 80 GeV pion showers,
before (a) and after (b) correction procedures were applied based on the starting point of the showers
and the estimated leakage. From: Sefkow, F. et al. (2016). Rev. Mod. Phys. 88, 015003
Fig. 6.15 Distribution of the total collected charge in beam tests of a prototype calorimeter for the
NA62 experiment at CERN (a). These tests were carried out with a beam of 606 MeV electrons.
The peaks are the result of several beam particles entering the detector simultaneously. The energy
resolution derived from these measurements (b). The data points were fit with a curve of the type
σ/E = a E −1/2 ⊕ bE −1 ⊕ c. The resulting values of the coefficients a, b and c are shown in the
legend. From: Antonelli, A. (2018). Nucl. Instr. and Meth. A877, 178
References
7.1 Introduction
The development of hadronic cascades in dense matter differs in essential ways from
that of electromagnetic ones, with important consequences for calorimetry. Hadronic
showers consist of two distinctly different components:
For the purpose of calorimetry, the main difference between these components is that
some fraction of the energy contained in the non-em component does not contribute
to the signals. This invisible energy, which mainly consists of the binding energy
of nucleons released in the numerous nuclear reactions, may represent up to 40%
of the total non-em energy, with large event-to-event fluctuations. For this reason,
homogeneous hadron calorimeters offer no particular advantage, as illustrated by
Fig. 1.4. As a matter of fact, sampling calorimeters hold all the records for best
hadronic performance.
In this chapter, we investigate the consequences of the fluctuations in invisible
energy for the performance of hadron calorimeters.
Let us define the calorimeter response as the conversion efficiency from deposited
energy to generated signal, and normalize it to minimum ionizing particles (mips).
If the average signal is proportional to the deposited energy, i.e., if the calorime-
ter is linear, this definition implies that the response of this calorimeter is energy
independent. The responses of a given calorimeter to the em and non-em hadronic
© Springer Nature Switzerland AG 2019 143
M. Livan and R. Wigmans, Calorimetry for Collider Physics, an Introduction,
UNITEXT for Physics, https://doi.org/10.1007/978-3-030-23653-3_7
144 7 The Fundamental Problems of Hadron Calorimetry
Fig. 7.1 Illustration of the meaning of the e/ h and e/mi p values of a calorimeter. Shown are distri-
butions of the signal per unit deposited energy for the electromagnetic and non-em components of
hadron showers. These distributions are normalized to the response for minimum ionizing particles
(“mi p”). The average values of the em and non-em distributions are the em response (“e”) and
non-em response (“h”), respectively [1]
shower components, which we will call e and h, are usually not the same, as a result
of invisible energy and a variety of other effects. We will call the distribution of the
signal per unit deposited energy around the mean value (i.e., e or h or the response
to showers) the response function.
Figure 7.1 illustrates the different aspects of the calorimeter response schemat-
ically. The em response is larger than the non-em one, and the non-em response
function is broader than the em one, because of event-to-event fluctuations in the
invisible energy fraction. Both e and h are smaller than the calorimeter response for
minimum ionizing particles, because of inefficiencies in the shower sampling pro-
cess [2]. The calorimeter is characterized by the e/ h and e/mi p ratios, which in this
example have values of 1.8 and 0.8, respectively. Calorimeters for which e/ h = 1
are called non-compensating.
The properties of the em shower component have important consequences for
the hadronic energy resolution, signal linearity and response function. The average
fraction of the total shower energy contained in the em component, f em , was mea-
sured to increase with energy following a power law [3, 4], confirming an induction
argument made to that effect [5]:
k−1
E
f em = 1 − (7.1)
E0
Fig. 7.2 Properties of the electromagnetic fraction of hadron showers. Shown are the results of
measurements of the average value of that fraction as a function of energy, for showers developing
in lead or copper (a) [1] and the distribution of f em values measured for 150 GeV π − showers
developing in lead (b). The curves in diagram a represent Eq. 7.1. From: Acosta, D. et al. (1992).
Nucl. Instr. and Meth. A316, 184. Experimental data from [3, 4]
resolution and their asymmetric characteristics are reflected in the response function
[2]. It is often assumed that the effect of non-compensation on the energy resolution
is energy independent, and should thus be described i.e., as a “constant term” added
in quadrature to the stochastic term:
σ a1
= √ ⊕ a2 (7.2)
E E
where the value of a2 is determined by the degree of non-compensation (e/ h). This
is incorrect, since it implies that the effect is insignificant at low energies, e.g., 10
GeV, which is by no means the case. The measured effects of fluctuations in f em can
be described by a term that is very similar to the one used for its energy dependence
(7.1). This term should be added in quadrature to the E −1/2 scaling term which
accounts for all fluctuations that obey Poisson statistics:
l−1
σ a1 E
= √ ⊕ a2 (7.3)
E E E0
Fig. 7.4 The effects of non-compensation on the hadronic energy resolution. Comparison of the
effects of adding either a constant term or a term as in Eq. 7.3 in quadrature to the stochastic term
(a). Comparison of Eq. 7.3 (up to 400 GeV) and a stochastic term with a slightly larger coefficient
0.55 instead of 0.50 (b) [1]
i.e., a linear sum of a stochastic term and a constant term. However this expression
suggests that there is complete correlation between the fluctuations that contribute to
the two terms, which is nonsense. Figure 7.4b provides the solution of this apparent
mystery. The solid line in this figure is exactly the same as in Fig. 7.4a, except that it
stops at E = 400 GeV, which is the highest pion energy for which experimental data
have been reported. This curve runs almost parallel to the dotted line, which represents
a stochastic term with a coefficient a1 = 0.55 (in the solid curve, a1 = 0.50). This
means that it is practically impossible to distinguish between fits such as
σ 50% E −0.28
= √ ⊕ 30%
E E 0.7
and
σ 55%
= √ + 3.5%
E E
in the energy range for which experimental data are available. Experimental data
at very high energies would be needed to make that distinction. The observation
that experimental resolution data tend to be better described by a linear sum of a
stochastic term and a constant term rather than by a quadratic sum of such terms may
thus be interpreted as support for Eq. 7.3.
Many sets of experimental hadronic energy resolution data exhibit indeed this
characteristic, for example the results reported for ATLAS [8] shown in Fig. 7.5.
Following the representation chosen in this book, the energy resolution is plotted
on a scale linear in −E −1/2 . Scaling with E −1/2 is thus represented by a straight
line through the bottom right corner in this plot. The experimental ATLAS data are
located on a line that runs parallel to such a line, indicating that the stochastic term
(c1 ) is ≈80% and the constant term (c2 ) is ≈5% in this case.
148 7 The Fundamental Problems of Hadron Calorimetry
Fig. 7.5 Experimental consequences of non-compensation for the hadronic calorimeter perfor-
mance. The energy resolution reported by ATLAS [8], both for the Tilecal in stand-alone mode
and for the combination of the em and hadronic calorimeter sections. For comparison, the hadronic
energy resolution reported for the compensating ZEUS calorimeter [9] is shown as well. See text
for details [1]
Figure 7.5 also shows another interesting phenomenon, namely that the ATLAS
hadronic energy resolution was actually measured to be better when the hadronic
calorimeter section (Tilecal) was used in stand-alone mode, rather than in combina-
tion with the LAr ECAL. The reason for this is the fact that these calorimeter sections
have different e/ h values. For the Tilecal, an e/ h value of 1.336 ± 0013 has been
reported [10] , while the value for the Pb/LAr ECAL, which unlike the Tilecal is very
insensitive to the neutrons produced in the shower development, is estimated at ∼1.5
[1]. Typically, the energy deposited by showering hadrons and jets is shared between
these compartments, and the large event-to-event fluctuations in this energy sharing
translate into an additional contribution to the hadronic energy resolution. This con-
tribution is absent when the showers develop entirely in the Tilecal, hence the better
energy resolution. However, this better resolution is still considerably worse than
that of the compensating ZEUS calorimeter, which is shown for reference purposes
in Fig. 7.5. Especially at energies above 100 GeV, the advantages of compensation
are very substantial.
The discrepancy between the hadronic energy resolutions measured for the
hadronic section alone and for the total calorimeter system is even much larger
for the CMS experiment, where the crystal em section has a value of 2.4, while
e/ h = 1.3 for the hadronic section. The hadronic performance of this calorimeter
system was systematically studied with various types of particles (e, π, K , p, p̄),
covering a momentum range from 1 – 300 GeV/c. Results are shown in Fig. 7.3. It
turned out that the response strongly depends on the starting point of the showers
[11]. Figure 7.6 shows results for two event samples, selected on that basis: show-
7.2 The e/ h Ratio and Its Consequences 149
Fig. 7.6 The response to electrons and pions as a function of energy, for the CMS barrel calorimeter.
The pion events are subdivided into two samples according to the starting point of the shower, and
the pion response is also shown separately for these two samples. From: Akchurin, N. and Wigmans,
R. (2012). Nucl. Instr. and Meth. A666, 80
Fig. 7.7 Event-to-event fluctuations in the em fraction of 150 GeV π − showers in lead (a). From:
Acosta, D. et al. (1992). Nucl. Instr. and Meth. A316, 184.Signal distributions for 300 GeV pions
(b) and protons (c) detected with a quartz-fiber calorimeter. The curve represents the result of a
Gaussian fit to the proton distribution. From: Akchurin, N. et al. (1998). Nucl. Instr. and Meth. A408,
380
to a charged pion does not necessarily lead to a small em fraction for the shower as
a whole. The limitations on π 0 production also lead to a smaller value of f em in
baryon induced showers, compared to pion induced ones (Eq. 7.1).
Experimental studies have confirmed these effects [12, 13]. Figure 7.7 shows the
signal distributions measured for 300 GeV pions (b) and protons (c), respectively.
The pion distribution resembles that of the fluctuations in the em fraction measured
for 150 GeV π − showers developing in lead (Fig. 7.7a). The signal distribution for
protons is much more symmetric, as indicated by the Gaussian fit. This is because
the em component of proton-induced showers is typically populated by π 0 s that
share the energy contained in this component more evenly than in pion-induced
showers. The figure also shows that the rms width of the proton signal distribution
is significantly smaller (by ∼20%) than for the pions. The average signal per GeV
deposited energy is smaller for the protons than for the pions, by about 10%. This is
also a consequence of the limitations on π 0 production that affect the proton signals
in this non-compensating calorimeter (e/ h > 1). So while the response to protons
is smaller in this calorimeter, the energy resolution is better. This will be true for
all non-compensating calorimeters, but the extent of the proton–pion differences
depends of course on the degree of non-compensation.
Similar effects are expected to play a role for the detection of kaons, where π 0
production is limited as a result of strangeness conservation in the shower develop-
ment. Also in this case, the absence of a leading particle effect may be expected to
make the fluctuations in f em smaller and more symmetrically distributed than for
pions.
7.3 The e/mi p Ratio and Its Effects for Jet Detection 151
7.3 The e/mi p Ratio and Its Effects for Jet Detection
Figure 7.1 shows that the response of typical calorimeters to the em shower com-
ponent not only differs from that to the non-em component, but it also differs from
the response to minimum ionizing particles. The reasons for this are discussed in
Sect. 3.5.2. This effect, which I will refer to as e/mi p = 1, may have important con-
sequences for some aspects of the calorimeter performance, even if the calorimeter
can be made compensating (e/ h = 1.0). For example, at low energies, the probabil-
ity that a hadron is stopped in the calorimeter before it has an opportunity to initiate
a nuclear reaction, and thus start a shower, rapidly increases. The entire energy of
such hadrons is used to ionize the calorimeter material. There are thus no losses due
to invisible energy and, as a result, the calorimeter response to such hadrons is larger
than that to hadrons that do develop showers while being absorbed. The response is
actually similar to that for muons, which deposit their energy in the same way.
This is illustrated in Fig. 7.8, which shows the response of the 238 U/plastic-
scintillator ZEUS calorimeter to low-energy charged hadrons [14]. This calorimeter
had an e/ h value very close to 1.0, but since the e/mi p value was about 0.6 (see
Sect. 7.4), the hadronic response increased for energies below a few GeV, reflecting
the increasingly mip-like absorption process. This calorimeter was thus quite non-
linear for hadrons with energies less than 5 GeV, which had important consequences
for the performance for jet detection. A jet is a collection of particles (mainly pions
and γ s) produced in the fragmentation of a quark or gluon. Relatively low-energy
fragments account for a significant fraction of the energy of high-energy jets, such as
the ones produced in the hadronic decay of the W and Z intermediate vector bosons.
Figure 7.9 shows the distribution of the energy released by Z 0 s (decaying through
the process Z 0 → u ū) and Higgs bosons (decaying into a pair of gluons) at rest that
is carried by charged final-state particles with a momentum less than 5 GeV/c. The
figure shows that, most probably, 21% of the energy equivalence of the Z 0 mass is
carried by such particles, and the event-to-event fluctuations are such that this frac-
Fig. 7.9 Distribution of the fraction of the energy released by hadronically decaying Z 0 (a) and
H 0 (b) bosons at rest that is carried by charged final-state particles with a momentum less than 5
GeV/c [15]
tion varies between 13 and 35% (for a 1σrms interval). For Higgs bosons decaying
into a pair of gluons, the average fraction is even larger, 34%, with rms variations
between 23 and 45%.
As a result of the important contribution from soft jet fragments, and the large
event-to-event fluctuations in this contribution, the energy resolution for intermediate
vector bosons measured with the compensating ZEUS uranium calorimeter turned
out to be worse than expected on the basis of the single-pion resolution.
As shown in the next section, this effect would have been considerably smaller if
the calorimeter had used a lower-Z absorber material such as copper. In that case the
e/mi p value of the calorimeter would have been ∼0.85 instead of 0.6 and Fig. 7.8
indicates that the decrease of the response to low-energy hadrons would have been
a factor of three smaller than for uranium.
In this section, we briefly review the expected e/mi p and e/ h values for different
types of calorimeters.
The Z values of the active and passive calorimeter materials are the most important
factors determining the e/mi p value. This is illustrated in Fig. 7.10, which shows
results of EGS4 calculations of the sampling fraction for (10 GeV)1 electron showers
in sampling calorimeters with either plastic scintillator or liquid argon as active
material, as a function of the Z value of the absorber material [16]. The thickness
of the absorbers was chosen to be 1 X 0 in these simulations, the active layers were
2.5 mm thick. The e/mi p ratio, which follows directly from these results, gradually
decreases when the Z value of the absorber increases. The e/mi p values are also
systematically larger when liquid argon (Z = 18) readout is used instead of plastic
scintillator. It is the difference in Z values between active and passive media that
determines the e/mi p ratio. The simulations show that this ratio can even be made
larger than 1.0 if Z active > Z passive , e.g., in Al/Lar calorimeters.
Experimental results on the value of e/mi p are included in this figure. They
concern measurements on calorimeters based on iron, lead or uranium as absorber
material. The results are in reasonable agreement with the simulations.
The e/ h values are based on the responses to the various components of hadron
showers and the fractions of their contribution to the shower development. All
responses are normalized to the response to mips, so that
e e/mi p
= (7.5)
h f rel · r el/mi p + f p · p/mi p + f n · n/mi p
where the non-em component consists of relativistic particles, protons and neutrons,
which carry fractions f rel , f p and f n of the non-em energy, respectively. The expected
values are summarized in Table 7.1. The e/mi p values for the calorimeters in this
Table are taken from Fig. 7.10. The fractions of the non-em shower component carried
by relativistic particles and by protons and neutrons from nuclear breakup were taken
from Monte Carlo simulations by Gabriel [17]. It was assumed that signal quenching
by densely ionizing spallation protons does not play a role in LAr, and that evaporation
neutrons do not generate signals in this medium. For more background information
on the underlying assumptions that were used we refer to [1].
154 7 The Fundamental Problems of Hadron Calorimetry
In Sect. 3.5.2, it was shown that the sampling fraction of em showers decreases as
the shower develops. Depending on the specifics of the calorimeter structure, that
effect may be as large as 25% (Fig. 3.13). As shown in Chap. 9, this causes major
problems for the calibration of longitudinally segmented em calorimeters.
The problems are even much larger for hadron calorimeters, especially when the
different longitudinal calorimeter segments have different e/ h values, as in CMS
(Fig. 7.6). A common misconception is that the em shower fraction in the devel-
opment of hadron showers is limited to the em section of the calorimeter system
and that it therefore is OK to have longitudinal sections with different e/ h values.
However, Fig. 3.6 demonstrates that π 0 production may occur anywhere in the devel-
opment of hadron showers, and is by no means limited to the early stages. Therefore,
a calorimeter that consists of longitudinal segments with different e/ h values will
be subject to additional signal fluctuations for the detection of hadrons, compared to
a calorimeter of which all segments have the same e/ h values.
7.6 Summary
The problems that limit the performance of hadron calorimeters can be summarized
as follows.
• In the absorption of high-energy hadrons, some fraction of the energy is essentially
invisible, it cannot contribute to the calorimeter signals.
7.6 Summary 155
• This fraction fluctuates wildly from one event to the next. These fluctuations are
non-Poissonian.
• This leads to hadronic signal non-linearity, non-Gaussian response functions and
poor hadronic energy resolutions.
• The response function of the calorimeter depends on the type of hadron.
• Dividing a calorimeter into longitudinal segments with different e/ h values makes
these problems worse.
• The effects may be worse for jets than for single hadrons.
In the next chapter, the methods that have been developed to mitigate/solve these
problems are described.
References
8.1 Introduction
The root cause of the poor performance of hadron calorimeters is thus the invisible
energy. Because some fraction of the energy carried by the hadrons and released in
the absorption process does not contribute to the signal, the response to the non-
em shower component is typically smaller than that to the em shower component.
And the characteristic features of the energy sharing between these two components
lead to hadronic signal non-linearity, a poor energy resolution and a non-Gaussian
response function.
To mitigate these effects, one has thus to use a measurable quantity that is cor-
related to the invisible energy. The stronger that correlation, the better the hadronic
calorimeter performance may become. In this chapter, two such measurable quanti-
ties are discussed: the kinetic energy released by neutrons in the absorption process
(Sect. 8.2) and the total non-em energy (Sect. 8.3). In Sect. 8.4, the beneficial effects
of both methods are compared.
8.2 Compensation
The first successful attempt to mitigate the effects described in the previous section
involved a calorimeter that used depleted uranium as absorber material [1]. The
underlying idea was that the fission energy released in the absorption process would
compensate for the invisible energy losses. By boosting the non-em calorimeter
response (h) this way, the e/ h ratio would decrease and, as a matter of good fortune,
reach the (ideal) value of 1.0. This is the reason why calorimeters with e/ h = 1.0 have
become known as compensating calorimeters. However, it turned out that nuclear
fission was neither essential nor sufficient to reach the compensation condition. Sev-
eral uranium calorimeters that were built after this initial success were found to have
e/ h values that were larger [2], and in one case significantly smaller than 1.0 [3].
On the other hand, it was demonstrated that the compensation condition can also be
obtained in calorimeters that use lead as absorber material [4].
In order to understand how compensation can be achieved, one should understand
in detail the response to the various types of particles that contribute to the calorimeter
signals. Let us look at the generic definition of the e/ h value (Eq. 7.5). Because of
the invisible energy, the “natural” value is larger than 1.0, and quite a bit larger
if the calorimeter only responds to relativistic shower particles (which is the case
for Čerenkov calorimeters). In order to bring the e/ h value down, two terms are
very important: the numerator (e/mi p) and the n/mi p term in the denominator.
As shown in the previous chapter (Fig. 7.10), the e/mi p value can be significantly
smaller than 1.0 when high-Z absorber materials are used. This was definitely an
important factor contributing to the initial success of the uranium based calorimeter,
where e/mi p ≈ 0.6.
However, most important in this context are the neutrons. Neutrons carry typi-
cally not more than ∼10% of the non-em shower energy. However, their contribution
to the calorimeter signals may be much larger than that. This is because neutrons
only lose their energy through the products of the nuclear reactions they undergo.
Most prominent at the low energies typical for hadronic shower neutrons is elastic
scattering. As shown in Sect. 3.5.2 (Table 3.1), MeV-type neutrons sent into a Pb/H2
structure (50/50 in terms of numbers of nuclei), transfer 98% of their kinetic energy
to hydrogen nuclei, and only 2% to lead. Since the sampling fraction for charged par-
ticles (mips) amounts to 2.2% in this structure, the potential for signal amplification
through neutron detection (SAND) is enormous, especially when the recoil protons
produced in the active material directly contribute to the calorimeter signal.
Hydrogenous active material is an extremely efficient medium for SAND in
calorimeters. Nowhere has the role of hydrogen been demonstrated more dramati-
cally than in the L3 uranium/gas calorimeter [5]. Figure 8.1a shows the signals of
this calorimeter for pions and for electrons, as a function of energy, for two different
gas mixtures: Argon/CO2 and isobutane. For the electron signals, the choice of gas
made no significant difference. However, the pion response doubled when isobutane
(C4 H10 ) was used instead of argon/CO2 . The L3 group also tested other gas mixtures.
It turned out that by changing the hydrogen content of the gas mixture used in the
wire chambers that produced the calorimeter signals, the π/e response ratio could be
changed by as much as a factor of two. By choosing the proper mixture, the responses
to em and hadronic showers could be equalized (Fig. 8.1b).
Compensation can also be achieved in other types of calorimeters, provided that
the active material contains hydrogen. Plastic scintillator is well suited, since the
recoil protons contribute to the calorimeter signals. Because of signal quenching
(Fig. 4.3), the n/mi p ratio (and thus the e/ h value) is much less sensitive to the precise
amount of hydrogen than indicated in Table 3.2 and measured by L3. By carefully
choosing the relative amount of hydrogen in the calorimeter structure, compensation
can be achieved. This has been demonstrated experimentally for plastic-scintillator
structures with Pb or 238 U as absorber material (Fig. 8.2).
All compensating calorimeters rely on the contribution of neutrons to the hadronic
signals. By properly amplifying the neutron signals (with respect to those from
8.2 Compensation 159
Fig. 8.1 Signals for pion and electron showers in the L3 uranium/gas calorimeter, for 2 different
gas mixtures in the readout chambers (a). Pion/electron response ratio as a function of the hydrogen
content of the gas mixture (b). From: Galaktionov, Y. et al. (1986). Nucl. Instr. and Meth. A251,
258
Fig. 8.2 Experimental data on the e/ h values measured for calorimeters based on plastic scintillator
as active material and using either 238 U (a), lead (b) or iron (c) as absorber material. The e/ h values
are plotted as a function of the ratio of the thicknesses of the absorber and scintillator plates (bottom
scale), i.e., as a function of the sampling fraction for mips (top scale). The curves represent the
results of simulations based on a neutron transportation Monte Carlo program [6]
charged shower particles depositing the same amount of energy), one can com-
pensate for the invisible-energy losses. Therefore, the essential ingredients for a
compensating calorimeter are:
160 8 Methods to Improve Hadronic Calorimeter Performance
Fig. 8.3 Signal distributions for hadrons of 10, 40 and 150 GeV measured with SPACAL. From:
Acosta, D. et al. (1991). Nucl. Instr. and Meth. A308, 481
responses (e/ h = 1.0), by means of the sampling fraction. This mechanism works
because the calorimeter response to charged shower particles is much more sensitive
to a change in the sampling fraction than the response to neutrons.
(the effects of) its fluctuations. The correct hadron energy can be determined from a
combination of both signals.
A dual-readout calorimeter produces two types of signals for the showers develop-
ing in it, a scintillation signal (S) and a Čerenkov signal (C). Both signals can be
calibrated with electrons of known energy E, so that S = C = E for em show-
ers, and the calorimeter response to em showers, Rem = S/E = C/E = 1. For
a given event, the hadronic signals of this calorimeter can then be written as
1
S = E f em + (1 − f em )
(e/ h) S
1
C = E f em + (1 − f em ) (8.1)
(e/ h)C
C f em + 0.21 (1 − f em )
= (8.2)
S f em + 0.77 (1 − f em )
8.3 Dual-Readout Calorimetry 163
Fig. 8.4 Graphic representation of Eqs. 8.3 [11, 14]. The data points for hadron showers detected
with a dual-readout calorimeter are located around the straight (red) line in this diagram. The data
points for em showers in this calorimeter are clustered around the point where this line intersects
the C = S line, i.e., the point (1,1). From: Lee, S., Livan, M. and Wigmans, R. (2018). Rev. Mod.
Phys. 90, 025002
where 0.21 and 0.77 represented the h/e ratios of the Čerenkov and scintillator
calorimeter structures, respectively. They used the measured data to show that the
broad, asymmetric signal distribution typical for a non-compensating calorimeter is
in fact a superposition of many individual, narrow, Gaussian distributions for events
with the same f em but a different central value that gradually increased with f em . The
overall signal distribution just reflected the distribution of f em values in the events.
These results are shown in Fig. 8.11a, b and it is instructive to compare them with
the f em distribution from Fig. 7.7a.
Let us now look again at Eqs. 8.1, and rewrite these as
S/E = (h/e) S + f em 1 − (h/e) S
C/E = (h/e)C + f em 1 − (h/e)C (8.3)
Figure 8.4 shows that the experimental data points for hadron showers detected with
a dual-readout calorimeter are thus located around a straight (red) line in the C/E
versus S/E diagram. This line links the point [(h/e) S , (h/e)C ], for which f em = 0,
with the point (1,1), for which f em = 1. The experimental data points for electron
showers are concentrated around the latter location in this diagram, where the red line
intersects with the C = S line. By projecting these data points on the horizontal and
vertical axes, the raw scintillation and Čerenkov signal distributions are obtained,
and the shape of these distributions is indicative for the f em distribution in the event
sample. Different f em distributions will thus lead to different signal distributions, as
in Fig. 7.7b, c.
164 8 Methods to Improve Hadronic Calorimeter Performance
The slope of the red line around which the hadron data points are clustered, i.e., the
angle θ , only depends of the two e/ h values, and is thus independent of the hadron
energy. We define
1 − (h/e) S
cot θ = =χ (8.4)
1 − (h/e)C
and the parameter χ is thus also independent of energy. Because of this feature, the
scintillation and Čerenkov signals measured for a particular hadron shower can be
used to reconstruct its energy in an unambiguous way:
S − χC
E = (8.5)
1−χ
This equation implies that the data point (S, C) in Fig. 8.4 is moved up along the red
straight line until it intersects the line defined by C = S. If this is done for all hadronic
data points, the result is a collection of data points that cluster around the point (1,1),
just like the data points for electron showers. Projecting these data points on the axes
of this diagram gives the scintillation (x) and Čerenkov (y) signal distributions.
The dual-readout procedure thus effectively uses the measured signals to deter-
mine the em shower fraction, f em , and then calculates what the signals would be if
f em was 1.0, i.e., as if the shower was entirely electromagnetic. The actual f em dis-
tribution for showers produced in the absorption of a sample of hadrons of the same
type and energy is therefore not a factor that affects the energy measurement for that
event sample. A dual-readout calorimeter is therefore linear for hadron detection,
since the correct energy is reproduced in each case.
Interestingly, a dual-readout calorimeter will also produce signal distributions with
the same average value for event samples of pions, protons and kaons of the same
energy. The f em distributions are quite different for showers produced by these dif-
ferent types of hadrons, as a result of conservation of baryon number and strangeness
in the shower development. This prevents the production of a very energetic, leading
π 0 in the case of protons and kaons, respectively. These differences are illustrated
in Fig. 8.4, where the protons and pions are represented by red and blue data points,
respectively. Measurements with conventional calorimeters have clearly shown sig-
nificant differences between the response functions of protons and pions (Fig. 7.7).
Figure 8.5c illustrates that the mentioned effects do not play a role for dual-readout
calorimeters. The relationship (8.5) is universally valid for all types of hadrons, and
also for jets.
Figure 8.5 shows an example of results obtained in practice with a dual-readout fiber
calorimeter built and tested as part of the RD52 detector R&D program [15]. The fact
that θ and χ are independent of the energy and the particle type offers the possibility
8.3 Dual-Readout Calorimetry 165
Fig. 8.5 Results obtained with the RD52 lead-fiber dual-readout calorimeter. Shown are the mea-
sured signal distributions for 20 GeV π + (a) and 125 GeV protons (b), the average signal per unit
deposited energy as a function of energy, separately for pions and protons (c), and the measured sig-
nal energy resolution for protons and pions as a function of energy (d). From: Lee, S. et al. (2017).
Nucl. Instr. and Meth. A866, 76
to measure the hadronic energy with unprecedented precision, in test beams at the
CERN SPS. It should be emphasized that no information on the beam energy or the
hadron type was used in the analyses of the experimental data.
Figure 8.5a, b show signal distributions for 20 GeV pions and 125 GeV protons,
respectively. Both distributions are well described by a Gaussian function with a
central value of 20.5 GeV and 126.5 GeV, respectively. The relative width, σ/E, was
measured √ to be 6.61% for √ the pions and 2.85% for the protons. This corresponds
to 29.6%/ E and 31.9%/ E, respectively. The narrowness of these distributions
reflects the clustering of the data points around the axis of the locus in Fig. 8.4. It
should be re-emphasized that the energy of the beam particles was not used to obtain
these signal distribution. The straight line that was used to fit the experimental data
points in the scatter plot (Fig. 8.4) intersected the C = S line at approximately the
correct energy. The same was true for particles of other energies, for different types
of hadrons and also for multiparticle events, always using the same procedure and
the same value for the angle θ .
166 8 Methods to Improve Hadronic Calorimeter Performance
Figure 8.5c shows that this dual-readout calorimeter is very linear. It produces
the same response for pions and protons, within ≈ ±1%, and, most importantly, the
hadron energy was correctly reproduced in this device, of which the energy scale was
set with electron showers.
√ Finally, the hadronic energy resolution was measured to
scale with ∼30%/ E over the full energy range for which the calorimeter was tested,
without significant deviations (Fig. 8.5d). Since the calorimeter (including its system
of leakage detectors) had a mass of less than 2 tonnes, the effects of fluctuations in
(mainly lateral) shower leakage were probably not at all negligible, and even better
resolutions may be expected for larger instruments of this type.
In dual-readout calorimeters, the total non-em energy, which can be derived from
the measured total energy (Eq. 8.5) and the em shower fraction (Eq. 8.1), thus repre-
sents the measurable quantity correlated to the invisible energy. The limitations that
apply for compensation do not apply in this case. Any absorber material may be used,
as a matter of fact the dual-readout method may even be applied for homogeneous
calorimeters, such as BGO crystals [16]. The sampling fraction is not restricted and
neutron detection is not a crucial ingredient for this method. Therefore, one is con-
siderably less constrained when designing a calorimeter system of this type than in
the case of a system based on compensation.
Fig. 8.6 Energy resolutions reported for the detection of electrons(a) and hadrons (b) by RD52
[15, 17], SPACAL [8] and ZEUS [10]. From: Lee, S., Livan, M. and Wigmans, R. (2018). Nucl.
Instr. and Meth. A882, 148
resolution is extremely sensitive to the angle between the beam particles and the
fiber axis when this angle is very small [18].
2. The instrumented volume of the RD52 calorimeter (including the leakage coun-
ters) was less than 2 tonnes, while both SPACAL and ZEUS obtained the reported
results with detectors that were sufficiently large (>20 tonnes) to contain the
showers at the 99+% level. The hadronic resolutions shown for RD52 are defi-
nitely affected by fluctuations in lateral shower leakage, and a larger instrument
of this type is thus very likely to further improve the results.
The comparison of the hadron results (Fig. 8.6b) seems to indicate that the
dual-readout approach offers better opportunities to achieve superior hadronic per-
formance than compensation. Apparently, in hadronic shower development the cor-
relation with the total nuclear binding energy loss is thus stronger for the total non-em
energy (derived from the em shower fraction) than for the total kinetic neutron energy.
Intuitively, this is not a surprise, since the total non-em energy consists of other com-
ponents than just neutrons, and the total kinetic energy of the neutrons is not an exact
measure for the number of neutrons (which is the parameter expected to be correlated
to the binding energy loss).
In order to investigate the validity of this interpretation of the experimental results,
Lee et al. performed Monte Carlo simulations of shower development in a block of
matter that was sufficiently large to make the effects of shower leakage insignificantly
small. Large blocks of copper or lead were used for this purpose [19]. The simulations
were carried out with the GEANT4 Monte Carlo package [20] for pions of 10, 20,
168 8 Methods to Improve Hadronic Calorimeter Performance
Fig. 8.7 The average value of the em shower fraction and the average fraction of the energy
represented by nuclear binding energy losses, for pions absorbed in large blocks of copper and
lead, as a function of the pion energy (a). Event-to-event fluctuations in the nuclear binding energy
losses, expressed as a fraction of the total detected energy (b). From: Wigmans, R. (2018). J. Progr.
Part. Nucl. Phys. 103, 109
50, 100, 200, 500 and 1000 GeV, using the default physics list used in simulations
for the CMS and ATLAS experiments at CERN’s Large Hadron Collider [21].
Some results of these simulations are shown in Figs. 8.7, 8.8, 8.9 and 8.10. Figure
8.7a shows the average value of the em shower fraction and the average fraction of
the pion energy represented by nuclear binding energy losses as a function of the pion
energy. There are clear differences between copper and lead. The em shower fraction
increases with energy and is larger for copper (in agreement with the measurement
results shown in Fig. 7.2a), while the average binding energy losses decrease with
energy and are larger for lead. The event-to-event fluctuations in the binding energy
loss are also larger for lead, as illustrated in Fig. 8.7b. This figure represents the
ultimate precision with which the energy of the pions can be measured in a copper-
or lead-based calorimeter in which no effort is made to mitigate the effects of these
fluctuations on the hadronic energy resolution.
However, both the non-em energy (which follows directly from f em ) and the total
kinetic neutron energy turned out to be clearly correlated with the nuclear binding
energy loss (Fig. 8.8). To examine the degree of correlation, event-by-event ratios
were determined. Histograms of these ratios are shown in Fig. 8.9 for 50 GeV π −
showers in copper.
These figures confirm that the correlation between the total non-em energy and
the nuclear binding energy loss is better than the correlation between the total kinetic
neutron energy and the nuclear binding energy loss. This was found to be true both
for copper and for lead. These simulations were also used to estimate the effects of
the correlations discussed above on the energy resolution for hadron calorimeters that
are based on dual-readout or compensation. The results are summarized in Fig. 8.10,
for pions in copper (a) and lead (b). These resolutions should be considered Monte
8.4 Dual-Readout Versus Compensation 169
Fig. 8.8 Scatter plots in which the nuclear binding energy losses for 100 GeV pions absorbed in
copper are compared event by event to the non-em energy measured with the dual-readout method
(a) and the total kinetic energy of the neutrons produced in the absorption process, which is the
essential ingredient of compensating calorimeters (b). The straight lines represent constant values
(0.2 or 0.5) of the ratio between the parameters plotted on the vertical and horizontal axes. Results
of GEANT4 Monte Carlo simulations. From: Lee, S., Livan, M. and Wigmans, R. (2018). Nucl.
Instr. and Meth. A882, 148
Fig. 8.9 Distributions of the ratio of the non-em energy and the nuclear binding energy loss (a) and
the ratio of the total kinetic energy carried by neutrons and the nuclear binding energy loss (b) for
hadron showers generated by 50 GeV π − in a massive block of copper. Results from GEANT Monte
Carlo simulations. From: Lee, S., Livan, M. and Wigmans, R. (2018). Nucl. Instr. and Meth. A882,
148
Carlo predictions for the ultimate hadronic energy resolution that can be achieved
with calorimeters using either dual-readout or compensation as the method to mitigate
the effects of (fluctuations in) invisible energy. A comparison of these results with
those from Fig. 8.7b shows to what extent these methods are successful in that respect,
especially at increasing energy.
The resolution limits scale remarkably well with E −1/2 , in the energy range con-
sidered here (10–1000 GeV). Both for lead and for copper absorber, the limits are
170 8 Methods to Improve Hadronic Calorimeter Performance
Fig. 8.10 The limit on the hadronic energy resolution derived from the correlation between nuclear
binding energy losses and the parameters measured in dual-readout or compensating √ calorimeters,
√ as
a function of the particle energy. The straight lines represent resolutions of 20%/ E and 10%/ E,
respectively, and are intended for reference purposes. Results from GEANT Monte Carlo simula-
tions of pion showers developing in a massive block of copper (a) or lead (b). From: Lee, S., Livan,
M. and Wigmans, R. (2018). Nucl. Instr. and Meth. A882, 148
considerably better for dual-readout calorimeters than for compensating ones, in this
entire energy range. The ultimate limit for the energy resolution that√can be achieved
with calorimetric detection of hadron showers seems to be ∼12%/ E.
Experimental data obtained by the RD52 Collaboration also support the conclu-
sion that the correlation exploited in dual-readout calorimeters provides a more accu-
rate measurement of the invisible energy. Figure 8.11a, b show that the (Čerenkov)
signal from the DREAM fiber calorimeter is actually a superposition of many rather
narrow, Gaussian signal distributions. Each sample in Fig. 8.11b contains events
with (approximately) the same f em value, i.e., with the same total non-em energy.
The dual-readout method combines all these different subsamples and centers them
around the correct energy value. The result is a relatively narrow, Gaussian signal
distribution with the same central value as for electrons of the same energy.
Figure 8.11d shows that the DREAM (Čerenkov) signal is also a superposition of
Gaussian signal distributions of a different type. In this case, each sample consists of
events with (approximately) the same total kinetic neutron energy. The dual-readout
method may combine all these different subsamples in the same way as described
above. In doing so, the role of the total non-em energy is taken over by the total
kinetic neutron energy, and the method becomes thus very similar to the one used in
compensating calorimeters.
8.4 Dual-Readout Versus Compensation 171
Fig. 8.11 Distribution of the total Čerenkov signal for 100 GeV π − (a) and the distributions for
three subsets of events selected on the basis of the electromagnetic shower fraction (b). From:
Akchurin, N. et al. (2005). Nucl. Instr. and Meth. A537, 537. Distribution of the total Čerenkov
signal for 200 GeV multiparticle events (c) and the distributions for three subsets of events selected
on the basis of the fractional contribution of neutrons to the scintillator signal (d). From: Akchurin,
N. et al. (2009). Nucl. Instr. and Meth. A598, 422
A comparison between Fig. 8.11b and d shows that the signal distributions from
the event samples are clearly wider when the total kinetic neutron energy is chosen
to dissect the overall signal. This is consistent with our assessment that dual-readout
is a more effective way to reduce the effects of fluctuations in invisible energy on
the hadronic energy resolution.
Apart from that, dual-readout offers also several other crucial advantages:
• Its use is not limited to high-Z absorber materials.
• The sampling fraction can be chosen as desired.
• The performance does not depend on detecting the neutrons produced in the absorp-
tion process. Therefore, there is no need to integrate the calorimeter signals over
a large detector volume.
• The signal integration time can be limited for the same reason.
This is not to say that there is no advantage in detecting the neutrons produced in
the shower development. In fact, this may further improve the hadronic calorimeter
resolution, since f em and f n are correlated with the nuclear binding energy losses
in different ways, and thus may offer complementary benefits. Figure 8.12a shows
that a decrease in the Čerenkov/scintillation signal ratio (from which f em can be
derived) corresponds to an increase of the neutron component ( f n ) of the scintilla-
tion signal. However, as shown in Fig. 8.12b, this correlation is not perfect. In this
scatter plot, the f n values are plotted for two narrow bins in the distribution of the
172 8 Methods to Improve Hadronic Calorimeter Performance
Fig. 8.12 Results obtained with the DREAM copper-fiber dual-readout calorimeter, in which the
time structure of the signals was measured [22]. A scatter plot in which the measured contribution
of neutrons to the signals ( f n ) is plotted versus the measured ratio (Q/S) of the Čerenkov and
scintillation signals (a). A scatter plot in which the measured Čerenkov signal is plotted versus f n ,
for two different bins of the Q/S distribution (b). The energy resolution as a function of f n , for
events with (approximately) the same f em value (c). From: Akchurin, N. et al. (2009). Nucl. Instr.
and Meth. A598, 422
Čerenkov/scintillation signal ratio. In both cases, the f n values cover a much larger
range than the ±2% range of the f em values. Figure 8.12c shows that the energy
resolution depends rather strongly on the chosen f n value, for a given value of f em .
RD52 has shown that the complementary information provided by measurements of
f em and f n leads to a further improved hadronic energy resolution [22]. However,
even without explicitly determining f n , which involves measuring the time structure
of each and every signal, the hadronic energy resolution that can be obtained with
the dual-readout method is already superior to what has been achieved by the best
compensating calorimeters.
8.5 Exploiting the Time Structure of the Signals 173
The availability of ultrafast electronics at a reasonable price has opened new possi-
bilities for applications in calorimetry. In this subsection, we give some examples
of how relatively modest time resolutions could help with particle identification and
with the elimination of systematic sources of error in calorimetric measurements.
There is a deeply rooted belief that calorimeter systems for high-energy collider
experiments should be longitudinally subdivided into several sections. As a mini-
mum, one will usually want to have an electromagnetic and a hadronic section. A
major reason for this belief is that such a subdivision is needed for recognizing em
showers, and thus identify electrons and γ s entering the calorimeter.
This is a myth. It has been demonstrated repeatedly that there are several ways
to identify em showers in longitudinally unsegmented calorimeters, and a good time
resolution can be a wonderful tool in that respect. Figure 8.13 illustrates one of these
methods, developed by the RD52 Collaboration, which uses the starting time of the
calorimeter signals, measured with respect to the signal produced in an upstream
Fig. 8.13 The measured distribution of the starting time of the calorimeter’s scintillation signals
produced by 60 GeV electrons (a) and 60 GeV pions (b). This time is measured with respect to
the moment the beam particle traversed a trigger counter installed upstream of the calorimeter.
These data were also used to determine the distribution of the average depth at which the light was
produced in the hadron showers (c). The horizontal line represents the position resolution of this
measurement. From: Akchurin, N. et al. (2014). Nucl. Instr. and Meth. A735, 120
174 8 Methods to Improve Hadronic Calorimeter Performance
detector [23]. This method is based on the fact that light in the optical fibers travels
at a lower speed (c/n) than the particles that generate this light (∼c). The deeper
inside the calorimeter the light is produced, the earlier the calorimeter signal starts.
For the polystyrene fibers used in this detector, the effect amounted to 2.55 ns/m.
Figure 8.13 shows the measured distribution of the starting time of the signals
from 60 GeV e− (Fig. 8.13a) and π − (Fig. 8.13b). The starting time of the electron
signals has a standard deviation of 0.5 ns. Since all em showers generated light at
approximate the same depth inside the calorimeter, this time resolution translates into
a longitudinal position resolution of ∼20 cm. This pion distribution peaked ∼1.5 ns
earlier than that of the electrons, which means that the light was, on average, produced
60 cm deeper inside the calorimeter. The distribution is also asymmetric, it has an
exponential tail towards early starting times, i.e., light production deep inside the
calorimeter. This signal distribution was also used to reconstruct the average depth
at which the light was produced for individual pion showers. The result, depicted in
Fig. 8.13c, essentially shows the longitudinal profile of the 60 GeV pion showers in
this calorimeter.
Apart from particle identification, the measurement of the depth of the light pro-
duction in a longitudinally unsegmented calorimeter may also turn out to be useful
for other purposes. For example, it may be used to correct for the effects of light atten-
uation in the fibers on the calorimeter signals. Even though the attenuation lengths
are typically (considerably) longer than 5 m, uncertainties in the depth at which the
light is produced are not completely negligible in high-resolution hadron calorime-
ters. The position resolution of 20 cm (indicated by the red line in Fig. 8.13c) limits
the contribution of light attenuation effects to the energy resolution. It can of course
be further improved when better time resolution is available.
It is also possible to use the time structure of the calorimeter signals themselves,
for example for particle identification. This was demonstrated by the SPACAL Col-
laboration, who used the pulse width at 20% of the amplitude (FWFM) to this end [24]
and measured very significant differences between the distributions of this variable
for electrons and pions. In their case, the differences were considerably increased
by the fact that the upstream ends of their fibers were made reflective. Therefore,
the deeper inside the calorimeter the (scintillation) light was produced, the wider the
pulse (Fig. 8.14).
As the available time resolution further increases, other applications may become
feasible. For example, the depth measurement in several neighboring towers con-
tributing to the shower signal may provide an indication of the direction at which
the particle(s) entered the calorimeter, thus allowing measurement of the entire four-
vector. And time resolutions of ∼10 ps might even make it possible to mitigate the
effects of “pile-up”, which could seriously deteriorate the performance of calorime-
ters in the high-luminosity era of the LHC operations [25].
References 175
References
9.1 Introduction
All these effects may thus lead to a systematic mismeasurement of the energy.
This fact tends to be ignored. Calibration problems are most severe for longitudinally
subdivided calorimeter systems, especially if the calorimeter consists of sections with
very different e/ h values. However, before going into these problems, we first discuss
the easier case of a longitudinally unsegmented calorimeter.
in which E denotes the beam energy and Si j the signal from calorimeter cell i for
event j.
In calorimeters where the signals are amplified, it is often convenient to choose the
gain factors such that equal energy deposits lead to equal signals, in all calorimeter
cells. In that way, signals from different cells may be directly added without apply-
ing normalization constants. This is particularly convenient when combinations of
calorimeter signals from different cells are used to decide whether an event meets
preset selection criteria (triggering).
If the calorimeter is compensating, the calibration constants derived from em
shower detection are also valid for hadrons and jets. This is of course the most ideal
situation.
If the calorimeter is non-compensating, energy-dependent correction factors need
to be applied to derive the energy of hadrons and jets from the calorimeter signals
generated by these objects. These correction factors can be established with test
beams of hadrons and electrons, preferably spanning the entire energy range of
interest in the experiment for which the calorimeter is intended.
unusual either. Some calorimeters for PFA based systems1 have more than 50 longi-
tudinal segments [1].
Arguments typically used to design a calorimeter system in this way are:
1. It makes it possible to use available resources in the most economical way, by
constructing a fine-sampling, high-granularity electromagnetic section, followed
by a more crudely designed hadronic compartment.
2. It allows easy identification of electrons and γ s.
3. It helps to improve the hadronic performance.
Calibration complications are the price to pay for these (perceived) advantages.
These complications are such that a proper alternative title of this chapter would be:
(Mis)calibration – The pitfalls of longitudinal segmentation
In some cases (segments with different e/ h values), it is even questionable if there
is a correct way to calibrate longitudinally segmented calorimeters at all.
We start the description of these complications with a very simple case, involving
em showers in a homogeneous calorimeter, which we assume to be based on the
detection of Čerenkov light, e.g., a block of lead glass (Fig. 9.1). When electrons
are sent into this detector, Čerenkov photons are generated in the absorption of the
shower, and these photons are converted into photoelectrons in a light sensor, which
produces the signals. When this device is calibrated with 100 GeV electrons, the sig-
nal from his sensor consists, on average, of 1000 photoelectrons. We conclude that
the calibration constant for this calorimeter is 10 photoelectrons per GeV deposited
energy (10 p.e./GeV, or 0.1 GeV/p.e.). Since this calorimeter is linear, a beam of
20 GeV electrons will produce a signal that, on average, consists of 200 photoelec-
trons (20 × 10, or 20/0.1). Since we have defined the calorimeter response as the
average signal per GeV deposited energy, we can also say that the response of this
calorimeter is 10 photoelectrons (per GeV), and the average signal for a 50 GeV
electron will thus consist of 50 × 10 = 500 photoelectrons.
Now we are going to cut this detector into three parts, or rather we arrange things
in such a way that the signals produced in segments I, II and III are detected separately
(Fig. 9.1). This cut is made such that the 100 GeV electrons that were used for the
detector calibration deposit, on average, 30% of their energy in segment I, 40% in
segment II and 30% in segment III.
Čerenkov light is only produced by the charged shower particles that are suffi-
ciently relativistic, e.g., the electrons and positrons that carry at least 0.3 MeV kinetic
energy. Shower particles with energies below this cutoff value do participate in the
energy deposition process, but not in the signal generation. These soft particles are
Fig. 9.1 A hypothetical homogeneous lead glass calorimeter in which an electromagnetic shower
develops. This detector is cut in three components, which are read out separately [2]
rather rare in the early stages of the shower development, but they dominate in the late
stages. This means that if we now calibrate the three segments of the calorimeter sep-
arately, a different relationship will be found between deposited energy (in GeV) and
resulting signal (in photoelectrons) for these three segments. For example, we find
that in the first segment 15 photoelectrons are produced per GeV deposited energy.
In segment II, the signals from the 100 GeV electron showers consist, on average,
of 10 photoelectrons per GeV deposited energy, and in segment III the calibration
constant is 5 photoelectrons per GeV.
With these new, separate calibration constants, the average total calorimeter signal
for 100 GeV electron showers is still the same as before the cut was made. Because
of the 30%/40%/30% sharing of the deposited energy, segment I will contribute on
average 30 × 15 = 450 photoelectrons to the total signal, segment II 40 × 10 = 400
and segment III 30 × 5 = 150, for a total of 450 + 400 + 150 = 1000 photoelectrons.
However, if we now send electrons of another energy into this segmented calorime-
ter, the energy sharing between these three segments will be different than for the 100
GeV ones, because of differences in the longitudinal shower profile (Fig. 3.1). For
example, the energy sharing among segments I, II and III for a 20 GeV electron is,
on average, 45%/35%/20%, i.e., 9 GeV in segment I, 7 GeV in segment II and 4 GeV
in segment III. Based on the calibration constants of these segments established with
the 100 GeV electrons, the energy of the 20 GeV electrons will be underestimated.
This can be easily seen as follows. Imagine that the entire energy of the 20 GeV
shower was deposited in segment I. The shower does not know that the calorimeter
readout is split into three segments and would still produce 200 photoelectrons, as
before. However, these would now be interpreted as an energy deposit of 200/15
= 13.3 GeV, because energy deposit in segment I is converted on the basis of 15
p.e./GeV. Given the mentioned energy sharing for 20 GeV showers, the energy would
only be correctly reproduced if the shower produced 9 × 15 + 7 × 10 + 4 × 5 = 225
photoelectrons, instead of 200. The showers would thus be assigned an energy of
20 × 200/225 = 17.8 GeV in this example.
9.3 Longitudinally Segmented Systems 183
Similarly, the energy of electrons with an energy larger than 100 GeV would be
systematically overestimated in this example, since they would deposit a relatively
large fraction of their energy in segment III, which is more generous in converting
photoelectrons into GeVs than the other segments. For example, a 500 GeV electron
with energy sharing 20%/30%/50% = 100/150/250 GeV, would be reconstructed at
the correct energy if it generated 100 × 15 + 150 × 10 + 250 × 5 = 4,250 photoelec-
trons. However, since the particles generate, on average, 5,000 photoelectrons, their
signals would be interpreted as coming from a 588 GeV electron.
These problems are not limited to electrons with different energies than the
one used for calibrating the individual longitudinal segments. Also a γ of 100
GeV would be reconstructed, on average, with a different energy than a 100 GeV
electron. This is a consequence of differences between the average longitudinal
shower profiles (Sect. 3.1.1). Since γ showers deposit their energy typically deeper
inside the calorimeter than electron showers of the same energy, the energy of a
100 GeV γ would thus be interpreted as coming from a higher-energy object, e.g.,
105 GeV. Experiments that have developed elaborate calibration schemes for their
longitudinally segmented em calorimeter, based on electron detection (e.g., ATLAS,
Sect. 9.3.4) ought to keep this in mind when reconstructing γ showers.
This example illustrates the problems caused by item #3 listed in the introduc-
tion of this chapter, namely the fact that the calorimeter response (i.e., the relation
between deposited energy and resulting signal) changes as the shower develops, and
is thus a function of the depth inside the calorimeter. This implies that the calibration
constants, which relate signals to deposited energy, are different for the different
segments of a longitudinally segmented calorimeter. And since the shower profiles
change with the particle energy, this leads to signal non-linearities. This is a very
serious problem, as we will see in the following sections, which deal with real,
non-hypothetical calorimeters.
The calorimeter system for the HELIOS experiments in the heavy-ion beam at CERN
[3] consisted of thin uranium plates interleaved with 3 mm thick plastic scintillator
sheets. It was longitudinally subdivided into two sections. The first (em) section
was 6.4 radiation lengths deep and used 2 mm thick uranium plates. The second
(hadronic) section was ∼4 nuclear interaction lengths deep. The uranium plates
were 3 mm thick in that section. In the transverse plane, the calorimeter consisted
of rectangular towers, measuring 20 × 20 cm2 , read out from two sides by means of
wavelength shifting plates.
This calorimeter system was calibrated with electrons of 8, 17, 24, 32 and 45 GeV,
and also with cosmic muons. When the electron beam was steered into the center of
a calorimeter tower, showers were completely contained in one of the towers, with
comparable fractions of the energy deposited in the two longitudinal segments. This
is graphically shown in Fig. 9.2.
184 9 Calibrating a Calorimeter System
Fig. 9.2 The HELIOS modules were calibrated with beams of electrons [3]. The showers produced
by these particles produced signals of comparable strength in both sections of the calorimeter mod-
ules. Alternatively, the modules were calibrated with muons, which traversed the entire module [2]
The calibration constants A and B for the em and hadronic calorimeter sections,
respectively, were determined by minimizing the width of the total signal distribution,
i.e., by minimizing the quantity
N
n
n 2
Q = E − A Siem
j − B Sihad
j (9.2)
j=1 i=1 i=1
em had
where E is the electron beam energy and S and S are the sums of all the
signals in the towers i of the em and hadronic calorimeter sections that contributed
to the measured signal for event j. With this method, values for A and B and,
more importantly, for the intercalibration constant B/A were determined for each
calorimeter tower.
However, two fundamental difficulties were encountered when this calibration
method was applied:
1. The values of A, B and B/A were found to be energy dependent.
2. The values of B/A differed considerably (on average, more than 20%) from the
ones found with muons.
This is illustrated in Fig. 9.3, where the fractional widths of various total signal
distributions, σrms /E, are plotted as a function of the value of B/A (the back/front
weighting factor). In the following, we use B/A values that are normalized to the one
for muons that traversed both calorimeter sections. The value B/A = 1 (indicated
by the red dashed line in the figure) thus represents the calibration result derived
from the muon signals, as described below. If B/A > 1, then the signals from the
hadronic section were given a relatively larger weight for the calculation of the total
energy. If B/A < 1, then the signals from the em section were given a larger weight.
The B/A value for muons, and thus the normalization factor for all measurements,
was experimentally determined as follows by the HELIOS group. Cosmic muons
traversing both sections of a given tower were selected by means of a cosmic-ray
telescope. These muons generated the characteristic Landau signal distributions in
the em and hadronic sections of the tower. The B/A value was determined from
the ratio of the most probable signal values in the hadronic and em sections of the
9.3 Longitudinally Segmented Systems 185
tower. This ratio was compared with its expected value. The latter did not require
an experimental measurement, but could be calculated from the composition of the
calorimeter and from the specific ionization of the active and passive materials for
mips traversing both sections. The overall normalization factor was chosen such as
to equalize the measured and expected signal ratios. This procedure was repeated for
each individual calorimeter tower.
Figure 9.3 shows that the fractional widths of the signal distributions for electrons
reached a minimum for B/A values well below the value expected on the basis of the
muon measurements (B/A = 1.0). In addition, the B/A value for which this width
reached a minimum shifted upward with the electron energy, by about 20% over the
energy range for which measurements were done (8–45 GeV).
The HELIOS group also found that using B/A values different from 1.0 resulted
in signal non-linearity: the em response became energy dependent. Signal linearity
for em showers is a very fundamental calorimeter property, since the entire em shower
energy is used to ionize and excite the molecules of which the calorimeter consists.
Twice as much energy thus leads to twice as many excited and ionized molecules
and should thus lead to calorimeter signals that are twice as large. In the HELIOS
case, it seemed impossible to fulfill this fundamental requirement and to optimize
the energy resolution for em showers simultaneously.
The explanation for the described phenomena lies in the fact that the sampling
fraction, i.e., the fraction of the deposited shower energy that is converted into a
measurable signal, changes with depth (Sect. 3.5.2). In the (HELIOS) case of uranium
absorber and plastic scintillator as active material, the sampling fraction decreases
considerably as the shower develops, by as much as 25–30% over the volume in
which the absorption takes place. In the early stage of its development, the shower
still resembles a collection of mips, but especially beyond the shower maximum the
186 9 Calibrating a Calorimeter System
energy is predominantly deposited by soft (<1 MeV) γ s. The latter are much less
efficiently sampled than mips in this type of structure, where dominant processes such
as photoelectric effect and Compton scattering strongly favor the high-Z absorber
material. Therefore, a given energy deposit by the fast (more mip-like) component of
the shower in the electromagnetic calorimeter section leads to a considerably larger
signal than the same energy deposited in the soft tail in the hadronic calorimeter
section (Fig. 3.13).
The effect of this on the calibration results is energy dependent. The optimal B/A
ratio reflects the difference in (average) sampling fractions between both calorimeter
sections. As the electron energy is increased, more energetic shower particles pene-
trate the hadronic calorimeter section and the response difference between the two
calorimeter segments becomes smaller, i.e., the optimal value of the B/A weighting
factor gets closer to 1 (Fig. 9.3).
This phenomenon indicates that there is a fundamental problem. The relationship
between the energy deposited by the shower particles and the resulting calorimeter
signal (i.e., the calorimeter response) is very different for the two sections of the
calorimeter and, moreover, energy dependent. Translating the signals in the individual
calorimeter sections into deposited energy is therefore an extremely delicate issue,
no matter which calibration constants are being used.
Based on the described phenomena, HELIOS decided to use the B/A values
derived from the muon measurements as the basis of their calorimeter calibration.
The values of A (and thus automatically of B) were fixed such as to reproduce best
the electron beam energies over the measured range. In this way, signal linearity was
achieved.
In the HELIOS calorimeter, the thickness of the electromagnetic calorimeter
section was chosen such as to achieve roughly equal energy sharing for em showers.
It turned out that this choice maximized the described effects. The hadronic energy
resolution was much less sensitive to the value of B/A, because typically only a very
small fraction of the shower energy was deposited in the front section.
There are some very important lessons to be learned from this HELIOS experience:
• Showers should not be used to intercalibrate sections of a longitudinally segmented
calorimeter.
• In a longitudinally segmented calorimeter, no matter how the segments are inter-
calibrated, only the total shower energy derived from the signals is meaningful.
The signals from the individual calorimeter segments cannot be interpreted in a
straightforward way in terms of deposited energy.
Reference [2] describes a detailed Monte Carlo study of the effects of a calibration
procedure in which the width of the total signal distribution of showers that develop in
several different calorimeter segments is minimized. It is shown that such a procedure
9.3 Longitudinally Segmented Systems 187
leads inevitably to a non-linear response. Results are given for various values of the
intercalibration constant (or back/front weighting factor) B/A for a calorimeter with
the structure of the HELIOS one. Defining the degree of non-linearity as the fractional
change of the electron response in the energy interval from 1 to 100 GeV, it was found
to range from −15% for B/A = 0.7 to +5% for B/A = 1.1. Only for B/A = 1,
was the detector found to be linear for electron detection.
Now one might argue that there is in principle no reason why a calorimeter that
is non-linear for em shower detection, although somewhat inconvenient, should be
unacceptable. Any type of non-linearity could in principle be dealt with by means of
a polynomial relationship between the signals S and the corresponding energy E:
E = c0 + c1 S + c2 S 2 + c3 S 3 + ... (9.3)
and the fact that other constants than c1 have a non-zero value might be a small price
to pay for improving energy resolution. This line of reasoning is, however, crucially
flawed (see also Sect. 4.2).
This is because the average energy sharing between the calorimeter sections (and
thus the response) does not only depend on the particle’s energy but also on the
particle(s) that caused the shower. For example, the sharing is different for electrons
and γ s of the same energy (Sect. 3.1.1). A γ will deposit, on average, a larger
fraction in the deeper calorimeter section than an electron of the same energy. If the
signals from that section are given a smaller weight (which is what is happening for a
calibration scheme with B/A < 1), then the response for such γ s will be smaller than
for electrons. Also, if the em shower consists of several unresolved showers caused
by lower-energy particles, then the average energy fraction in the deeper calorimeter
section will be smaller, and this leads to an increased response. This may occur, for
example, when a K 0 decays into several π 0 s, each of which in turn decays into 2 γ s.
This is illustrated in Fig. 9.4, which shows these effects for three chosen cali-
brations (B/A = 0.7, 0.8 and 0.9, respectively). The mean response values differed
by as much as 1 GeV. Only for B/A = 1 all distributions had, within the statistical
uncertainty of the Monte Carlo simulations (∼0.05 GeV), the same mean value [4].
These effects are not small, given the fact that some of the calorimeters in particle
physics experiments are nowadays designed to achieve sub-1% energy resolution.
Figure 4.2 shows the signal distributions for 50 GeV γ s, 50 GeV π 0 s and 50 GeV
K 0 s (decaying into π 0 π 0 π 0 ) in the simulated calorimeter, calibrated on the basis of
B/A = 0.8 with 50 √ GeV electrons, and assuming an intrinsic energy resolution of
0.5% (i.e., 3.5%/ E). Based on this calibration, single photons were reconstructed
with an energy that was, on average, too low by 0.67 GeV, i.e., 2.7 times the intrinsic
energy resolution of the calorimeter. On the other hand, the energy of the kaons was
systematically overestimated, on average by 0.85 GeV (3.4 σ ). The reconstructed
energy of the π 0 s was approximately correct, because the energy sharing between
the two calorimeter segments was approximately the same for showers of 25 GeV
γ s (the decay products of the π 0 s) and 50 GeV electrons (used for the calibration).
This analysis illustrates a fundamental problem inherent to non-linear calorime-
ters. The calorimeter information is intended to determine the energy of particles
188 9 Calibrating a Calorimeter System
Fig. 9.4 The average reconstructed energy for 20 GeV γ s and for 20 GeV particles decaying
into multiple γ s in a longitudinally segmented Pb/scintillator calorimeter that was calibrated with
20 GeV electrons, for different values of the ratio of the calibration constants for the two longitudinal
segments, B/A. See text for details. From: Wigmans, R. and Zeyrek, M. (2002). Nucl. Instr. and
Meth. A485, 385
from the signals they generate. The precision with which the energy can be mea-
sured is determined by the energy resolution. However, in a non-linear calorimeter,
the energy resolution needed in this context does not correspond to the width of the
signal distribution measured for single, monoenergetic particles. This observation is
also particularly important for the evaluation of jet resolutions in a non-compensating
calorimeter.
The calibration problems described in the previous subsections become more compli-
cated when the calorimeter consists of more than two longitudinal segments. A recent
example of an experiment that has to deal with this intercalibration issue is ATLAS,
whose Pb/LAr electromagnetic calorimeter consists of three longitudinal segments.
At η = 0, the depths of these segments are 4.3X 0 , 16X 0 and 2X 0 , respectively. When
the particles enter the barrel calorimeter at a non-perpendicular angle, the total depth
of this calorimeter increases (from 22X 0 at η = 0 to 30X 0 at |η| = 0.8), and so do
the depths of these three segments. The sampling fraction for mips is the same in all
three segments.
Figure 9.5 shows how the sampling fraction for em showers evolves as a function
of depth, in an energy dependent way. The sampling fraction for muons that traverse
this detector does not change in this process and, therefore, the e/mi p value decreases
9.3 Longitudinally Segmented Systems 189
Fig. 9.5 The evolution of the sampling fraction for electron showers of different energies in the
three longitudinal segments of the ATLAS LAr calorimeter, at η = 0. From: Aharouche, M. et
al. (2006). Nucl. Instr. and Meth. A568, 601
by 20–25%, depending on the electron energy. In light of our discussions in the earlier
sections of this chapter, these data should by now look very familiar. Not surprisingly,
the problems encountered when calibrating this detector with electron showers were
also very similar to the ones experienced by HELIOS, except that there were now
three calibration constants to be determined, instead of two. When minimizing Q in
N
n
n
n 2
ij ij ij
Q = E − A S1 − B S2 − C S3 (9.4)
j=1 i=1 i=1 i=1
it turned out that the resulting calibration constants A, B and C not only depended
on the electron energy, as in HELIOS, but also on the location of the calorimeter
module (the η value). The latter dependence can be understood from the fact that the
effective depth of the longitudinal segments changes with the angle of incidence of
the particles. And just as in HELIOS, it was found that any choice of the calibration
constants resulting from such a minimization procedure introduced a response non-
linearity.
Rather than intercalibrating the different longitudinal calorimeter segments with
muons (the HELIOS solution), ATLAS decided to approach this problem in a much
more complicated way, relying heavily on Monte Carlo simulations in an attempt
to achieve the best possible combination of energy resolution and linearity. These
elaborate simulations led to a very complicated procedure for determining the energy
of a shower detected in the various segments of the calorimeter. This procedure was
based on a variety of parameters that depended both on the energy and the η value
(Fig. 9.6). It was tested in great detail with Monte Carlo events and combines excellent
signal linearity with good energy resolution [5].
The energy dependence of the various parameters derives from the fact that the
longitudinal shower profiles, and thus the energy sharing between the three segments,
190 9 Calibrating a Calorimeter System
Fig. 9.6 The formula used by ATLAS to determine the energy of a shower developing in the
longitudinally segmented ECAL. The energy dependence of the various parameters is shown in
graphs (a–d). From: Aharouche, M. et al. (2006). Nucl. Instr. and Meth. A568, 601
change with the energy of the incoming electron. In Sect. 3.1.1, we argued that
showers induced by photons are in this respect quite different from showers induced
by electrons (see for example Fig. 3.3). Figure 9.4 shows the consequences of that
for the energy measurement. Therefore, we believe that the calibration procedure
developed by ATLAS on the basis of electron showers (Fig. 9.6) would need to be
modified for the detection of γ s and particles decaying into several γ s. In a recent
paper [6], the ATLAS Collaboration describes a LAr calibration scheme based on
experimental data and simulations that takes this issue into account.
and is about 1X 0 thick. A minimum ionizing particle deposits, on average, 11.7 MeV
upon traversing such a layer. The AMS-02 collaboration initially calibrated this
calorimeter by sending a beam of muons though it and equalizing the signals from
all eighteen longitudinal segments. This seems like a very good method to calibrate
this detector, since all eighteen layers have exactly the same structure. However,
when this calorimeter module was exposed to beams of high-energy electrons, it
turned out to be highly non-trivial how to reconstruct the energy of these electrons.
Figure 9.7a shows the average signals from 20 GeV electron showers developing
in this calorimeter. These signals were translated into energy deposits based on the
described calibration. The measured data were then fitted to a Γ -function:
d E/dt ∝ t α exp(−βt)
where t is the layer number and α and β the coefficients to be fitted. Since the showers
were not fully contained, the average leakage was estimated by extrapolating this
fit to infinity. As shown in Fig. 9.7b, this procedure systematically underestimated
this leakage fraction, more so as the energy (and thus the leakage) increased. The
reason for this is the same as the one that plagued the calorimeters discussed in
the previous subsections, namely the fact that the sampling fraction of em showers
decreases as the shower develops. Therefore, a procedure in which the relationship
between measured signals and the corresponding deposited energy is assumed to be
Fig. 9.7 Average signals for 20 GeV electrons in the eighteen longitudinal sections of the AMS-
02 lead/scintillating fiber calorimeter. The curve represents the results of a fit to the experimental
data. The superimposed histogram is the expected average profile from the Monte Carlo simulations.
Average relative difference between the measured energy and the electron beam energy, after leakage
corrections based on extrapolation of the fitted shower profile were applied. From: Cervelli, F. et
al. (2002). Nucl. Instr. and Meth. A490, 132
192 9 Calibrating a Calorimeter System
the same for each depth segment will cause the energy leakage to be systematically
underestimated, more so if that leakage increases.
Based on these observations, AMS had to change its calibration procedure. Rather
than just integrating the Γ -function that describes the measured shower profile to
infinity, the energy of the particle that caused the shower is now determined on the
basis of the fraction of the total signal measured in the last two detector segments.
The deposited energy in the calorimeter derived from the Γ -function is subsequently
corrected with a factor that is determined by this fraction. The value of this correction
factor (needed to reproduce the actual electron energy) was established empirically
in testbeam measurements [8], for the available energy range (up to 250 GeV).
The very complicated issues discussed here will most definitely also affect PFA
calorimeters, which are all based on structures that are highly segmented, both lon-
gitudinally and laterally. The underlying problem is that the relationship between
deposited energy and resulting signal is not constant throughout a developing shower.
As the composition of the shower particles changes, so does the sampling fraction.
The examples of the calorimeters discussed here clearly illustrate the problems that
may be expected when this is not properly recognized and dealt with. The assumption
that the relationship between deposited energy and recorded signal remains the same
throughout the developing shower has been the modus operandi for the calorimeters
built in the context of the CALICE project [1].
Fig. 9.8 Calibration of the different longitudinal compartments of a calorimeter system with par-
ticles that deposit (almost) their entire energy in one of the compartments. Electrons are used to
calibrate the em section, while pions penetrating this section without starting a shower are used to
calibrate the hadronic calorimeter section [2]
measurement depends on the magnetic field B and the track length as (B 2 )−1 . A
large bending power thus extends the energy range of the method. This energy range
is often a weak point that may limit the usefulness of the in situ measurements.
At first sight, the idea of calibrating each calorimeter section separately, using
the particles that in practice generate the signals from these individual sections, is
very appealing. By using calibration constants that correctly reproduce the energy
deposited by electrons in the em section and by hadrons in the hadronic section, one
hopes to eliminate the problems arising from the fact that the calorimeter response
to these particles is different in non-compensating calorimeters. In all calorimeters
with e/ h > 1, the response to hadrons is smaller than that to electrons of the same
energy. For such calorimeters, the described method thus corresponds to a choice of
B/A > 1. In other words, if electrons were to be sent into the hadronic section of a
calorimeter calibrated this way, their energy would be overestimated, on average by
factor of B/A.
This calibration method works fine for those hadrons that penetrate the em section
without starting a shower. However, this sample usually represents only a small
fraction of all hadrons. Most hadrons undergo their first nuclear interaction in the em
calorimeter section. They deposit (a sometimes large) part of their energy in the em
section and the remainder in the hadronic section. For these events, this calibration
method does not produce correct results.
It turns out that in non-compensating calorimeters (with e/ h > 1), the energy
of early showering particles is systematically underestimated as a direct result of
this calibration method, to an extent that depends on the energy sharing between
the two calorimeter compartments. This calibration method thus introduces a strong
dependence on the starting point of the showers. Hadrons that happen to interact
early in the detector are attributed an energy that is, on average, considerably lower
than for hadrons that start showering after penetrating an interaction length or more.
A second undesirable consequence of this calibration method concerns the signal
linearity for hadrons. As discussed in Sect. 7.2, all non-compensating calorimeters
are intrinsically non-linear for hadron detection, since the average energy fraction
carried by the em shower component depends on the energy of the incoming particle
(Fig. 7.2a). However, this calibration method tends to make the non-linearity worse.
An example of this effect is shown in Fig. 9.9b.
This can be understood as follows. As the energy of the incident hadron increases,
a larger fraction of the shower energy is deposited in the hadronic calorimeter section.
Since the signals from this section are amplified (by a factor B/A, see above),
the hadronic response increases with energy as a direct result of this calibration
procedure. The intrinsic non-linearity, i.e., the increase of the hadronic response
because of the increasing fraction of the shower energy carried by π 0 s, is thus further
enhanced.
Although the calibration method discussed in this subsection looked at first sight
attractive and logical, it thus turned out to be fundamentally flawed. It should be
emphasized that the effects discussed here are an artifact of all calorimeters with
e/ h = 1. The degree to which the reconstructed hadron energy depends on the start-
ing point of the shower development and the degree to which the non-linearity of the
9.3 Longitudinally Segmented Systems 195
The third calibration method that we will discuss was inspired by the desire to
eliminate the non-linearities in the hadronic response that are characteristic for non-
compensating calorimeters. In this method, which is described and applied in [11],
the signals from the em section are given different weights for the detection of
electrons and hadrons. The calibration constants for em showers are determined in
the same way as described in the previous subsection, i.e., by means of electrons of
known energy that deposit all their energy in the em calorimeter section. However,
when hadrons are showering in the calorimeter, the signals from the em section are
weighted with a factor that is chosen to achieve signal linearity for these particles.
Usually, non-compensating calorimeters have an e/ h value larger than 1. In
such undercompensating calorimeters, the hadronic response increases with energy
(Fig. 7.2a). As the energy increases, the fraction of the shower energy deposited in
the hadronic calorimeter section increases, on average, as well. Therefore, restoring
linearity for hadrons in such calorimeters may be achieved either by suppressing
the signals from the hadronic compartment or by boosting the signals from the em
compartment. This is equivalent to applying a weighting factor to the intercalibration
constant B/A that is smaller than 1.
When hadrons are detected in an under-compensating calorimeter that is calibrated
in this way, the signals from the em compartment will thus be attributed a larger
energy than for electrons that produce the same signals in the same compartment.
As a result, the hadronic signal distributions obtained on the basis of this calibra-
tion method turn out to depend even more on the starting point of the showers than
those based on the calibration scheme from the previous subsection. In particular,
they depend on the fraction of the shower energy deposited in the em section. In
extreme cases, differences of a factor of three in the response have been reported for
hadrons of the same energy [12]. Especially for jets with a leading π 0 , a large frac-
tion of the energy is deposited in the em calorimeter section. As a result, the energy
of such jets is typically overestimated by a substantial factor when this calibration
method, defined on the basis of the response to single hadrons, is used.
This calibration method was developed to eliminate the energy dependence of the
hadronic calorimeter response. It succeeded in achieving that goal. However, there
is a price to be paid, as a direct consequence of this method:
196 9 Calibrating a Calorimeter System
• The hadronic response becomes dependent on the starting point of the showers.
• The response becomes dependent on the longitudinal shower development char-
acteristics. This leads to differences between the calorimeter responses to jets and
to single hadrons.
As before, these effects are avoided when e/ h = 1.0 for both calorimeter sections,
because in that case this calibration method would be equivalent to using B/A = 1.
The fourth calibration method is, just like the three previous ones, inspired by a laud-
able goal. The purpose of this method is to make the hadronic response independent
of the starting point of the shower. It was used by CDF for the calibration of their
“plug upgrade” calorimeter [13]. CDF also tried two other calibration methods and
compared the results, some of which are shown in Fig. 9.9. The method that is the
topic of this subsection is called “Method III”. Method I was based on the calibra-
tion procedure described in Sect. 9.3.7, where the calibration constant (“B I ”) of the
hadronic compartment2 was determined by the response to pions that penetrated the
em compartment without starting a shower. As shown in Fig. 9.9a, this calibration
constant increased with the pion energy, because of the increased average em shower
component. This was not the case with Method II, in which the calibration constant
of the hadronic compartment was determined in the same way as for the em com-
partment, namely with electrons. And since both compartments were linear for em
shower detection, both A I I and B I I were independent of the particle energy. This
is in essence the method discussed in Sect. 9.3.11, and which we call the “B/A =
1” method. CDF only used Method III to determine the calibration constants with
pions of 10 GeV, because of concerns that shower leakage might affect the results at
higher energies.
Figure 9.9b shows the response to pions as a function of energy, using all events,
i.e., not only the pions that deposited their entire shower energy in the hadronic
compartment. For all three methods, the pion response increased with energy, as
a result of the fact that e/ h > 1 in both calorimeter compartments. However, the
non-linearity was clearly worst for Method I. As the pion energy increased, a larger
fraction of the shower energy was deposited in the hadronic calorimeter compartment.
Boosting the signals from that compartment, which is the essence of Method I, thus
tended to increase the already existing non-linearity. For Method III, the opposite
effect occurred. By giving less weight to the signals from the hadronic compartment,
the “natural” response non-linearity was reduced, albeit it not to the extent seen in
Sect. 9.3.8, where the suppression of the signals from the hadronic compartment was
so large that it resulted in response linearity.
2 Note that CDF expressed calibration constants in ADC cts/GeV, whereas we use the inverse quantity
Fig. 9.9 Results from three different methods investigated by CDF to calibrate their forward (“plug
upgrade”) calorimeter. The methods are described in the text. Shown are the calibration constants
for the hadronic calorimeter compartment (a) and the response to single pions (b) and to jets (c) as
a function of energy. From: Albrow, M. et al. (2002). Nucl. Instr. and Meth. A487, 381
CDF also studied the effects of these three calibration methods for jets, using
a semi-empirical procedure to determine the jet response. Figure 9.9c shows the
calorimeter response to these jets, as a function of energy. The same trends are
observed as for single pions, but the non-linearities are clearly smaller for all methods.
This is due to two factors:
1. Part of the jet signal, on average one-third, comes from γ s which develop em
showers. The calorimeter is linear for that component of the jet signal. The non-
linearity only affects the remaining portion of the jet fragments.
2. The non-linearity is not determined by the jet energy itself, but by the average
energy of the jet fragments. Since the multiplicity increases with energy, this
average energy of the fragments increases more slowly than the jet energy itself.
Yet, the differences between the non-linearities observed for the three calibration
methods indicate that the effects described for single pions propagate into the energy
measurement of jets.
Even though Method III eliminates the starting point dependence of the hadronic
response, it introduces at the same time other problems. For example, the response to
muons becomes different for the two compartments of the calorimeter system, to an
198 9 Calibrating a Calorimeter System
on a specific goal that, at least at first sight, looked quite reasonable and is briefly
summarized below:
1. Minimization of the width of the total signal distribution (Sect. 9.3.6)
2. Correct energy reconstruction of pions penetrating the em compartment without
starting a shower (Sect. 9.3.7)
3. Hadronic signal linearity (Sect. 9.3.8)
4. Independence of hadron response on starting point shower (Sect. 9.3.9)
5. Equal response to electrons and pions (Sect. 9.3.10)
We also saw that each of these approaches introduced specific additional prob-
lems. We now conclude this section on the calibration of longitudinally segmented
calorimeter systems with a statement that could summarize this entire chapter and
therefore deserves to be printed in bold face.
The correct way to intercalibrate the different sections of a longitudinally
segmented, non-compensating calorimeter system is by using the same particles
for all individual sections. If these particles develop showers, then they can only
be used to calibrate sections in which these showers are completely contained.
Only in this way is the relationship between the deposited shower energy (in GeV)
and the charge (in picoCoulombs) generated as a result established unambiguously.
We have referred to this as the B/A = 1 method. The use of a beam of muons
to intercalibrate the eighteen segments of the AMS-02 electromagnetic calorimeter
(Sect. 9.3.5) definitely qualifies as a viable method in this respect. The mistake made
in that case did not concern the calibration method itself, but the interpretation of the
results.
An example of a good B/A = 1 method for calibrating calorimeters consisting of
separate em and hadronic compartments is to determine the calibration constant of the
em compartment with electrons of known energy and to intercalibrate the different
longitudinal segments with a beam of muons. In non-compensating calorimeters,
the energy of showering hadrons is not correctly reproduced in this way. However,
that problem can be handled by applying an overall correction factor, which can be
measured independently. An example of a calorimeter that was calibrated in this way
is the CCFR Target calorimeter that operated in Fermilab’s neutrino beam [16].
Another correct (B/A = 1) calibration method is to expose the individual longi-
tudinal segments separately to an electron beam and derive the calibration constants
from the signals recorded in that way. In many cases, constructional details prohibit
such a procedure from being applied. Yet, this method was used for the calibration
of the prototype modules of the CMS calorimeter system, results of which are shown
in Figs. 7.3 and 7.6.
The CMS calorimeter system suffers from a particularly nasty complicating fac-
tor, namely the fact that the em and hadronic compartments have very different
e/ h values. The em section of the barrel calorimeter is made of PbWO4 crystals
(e/ h ≈ 2.5), the hadronic section consists of brass plates interleaved with plastic
scintillator (e/ h ≈ 1.4). For a systematic study of the hadronic performance of this
calorimeter system, both compartments were calibrated with a 50 GeV electron beam.
200 9 Calibrating a Calorimeter System
Figure 7.6 shows that the response to pions, represented by the black dots, is very
non-linear. This non-linearity is especially evident below 10 GeV, which is impor-
tant since pions in this energy range carry a large fraction of the energy of jets at the
LHC. More troublesome is the fact that the response depends on the starting point of
the showers. The figure shows results for two event samples, selected on that basis:
showers starting in the em section () or in the hadronic section ( ).
In CMS, the energy dependence of the response shown in Fig. 7.6 is the basis of
the correction for the observed non-linearity; measured signals are simply multiplied
with the inverse of the response value for that particular energy value. The use of
this energy dependent overall correction factor restores hadronic signal linearity on
average, even though the starting point dependence remains. If the starting point of
the hadron shower can be determined, then the inverse of the or curves from
Fig. 7.6 could be used as the correction factor instead.
Just like in CMS, the em and hadronic sections of the ATLAS calorimeter system
also have different e/ h values, but the differences are much smaller in that case.
The consequences of these differences are qualitatively the same, but less dramatic
than indicated in Figs. 7.3 and 7.6. Compensating calorimeter systems offer more
calibration possibilities than ATLAS and CMS, because several other methods are
equivalent to B/A = 1 in that case.
9.3.12 Validation
When all is said and done, it is important to check the correctness of the chosen
calibration scheme with experimental data. For this, one needs to have sources of
known experimental energy deposits. One such source is a particle of precisely
known mass, whose decay products are detected by the calorimeter. This method
offers excellent possibilities to validate the calibration of em calorimeters, because
of the availability of a variety of particles that are abundantly produced in today’s
accelerator experiments and which cover a large mass (i.e., energy) range: π 0 (mass
135.0 MeV/c2 ), η (547.9 MeV/c2 ), J/ψ (3.097 GeV/c2 ), Υ (9.460 GeV/c2 ) and Z 0
(91.19 GeV/c2 ) all decay into particles (e+ e− or γ pairs) that develop electromagnetic
showers. Examples of mass peaks reconstructed from these decay products are shown
in Fig. 9.10 [17].
Unfortunately, there are no such clearcut calibration sources that can be used for
hadron calorimeters. Yet, there are certainly possibilities. At low energies, one could
use the decay K S0 → π + π − for this purpose and at high energies the hadronic decay
modes of the intermediate vector bosons W and Z . In the latter case, the problem
is the QCD background, which makes it hard to extract a sample of boson decay
events from the di-jet invariant mass distributions. The only experiment that has
managed to do so was UA2 [18]. However, at the LHC one could take advantage of
the high production rate of t t¯ events, and select a much cleaner sample of hadronically
decaying W s from the dominant decay mode t → W b.
9.3 Longitudinally Segmented Systems 201
Fig. 9.10 Invariant mass distributions of two γ s and four γ s detected in the same event by the
KLOE em calorimeter. The mass peaks of the π 0 (a), η (b) and K S0 (c) are located within 1% of
their established values. From: Adinolfi, M. et al. (2002). Nucl. Instr. and Meth. A482, 364
Fig. 9.11 Energy sharing between the electromagnetic and hadronic calorimeter sections for pions
detected with the ALEPH and ZEUS calorimeters [2]. Diagram a shows the fraction of the energy
deposited in the em calorimeter section, as a function of energy. Experimental data from [11, 21].
Diagram b shows the energy fraction deposited by 10 and 25 GeV pions in iron absorber as a
function of the absorber thickness. Experimental data from [22]
calorimeter was strongly non-compensating, with e/π signal ratios ranging from 1.7
at 2 GeV to 1.3 at 30 GeV. This means that pions produced signals that were con-
siderably smaller than those from electrons at the same energy. To get to the correct
energy, the calibration constants ALEPH used for pions were therefore considerably
larger than for electrons.
The philosophy of the ALEPH calibration method, which was essentially the
(B/A < 1) one discussed in Sect. 9.3.8, was to achieve a situation in which
1. pions and electrons of the same energy produce, on average, the same signals,
and
2. the average pion signal is proportional to the energy (i.e., hadron linearity).
The price that had to be paid for reaching that goal was a considerable mismea-
surement of the energy of particles or jets whose signals did not resemble those
of the average pion. By overestimating the energy deposited in the em calorimeter
section by more than a factor of two, the energy of particles or jets that happened
to deposit a considerable fraction of their energy in this calorimeter section was
similarly overestimated. And the energy of particles that penetrated the em section
without undergoing a strong interaction was systematically underestimated by a con-
siderable factor.
The most serious problems resulting from this calibration strategy may be
expected for jets. First, since jets contain some fraction of π 0 s, which always deposit
their entire energy in the em calorimeter section, the average energy sharing between
the two sections may be quite different for charged pions and for jets of the same
energy. The latter deposit, on average, a larger fraction of their energy in the em
calorimeter section. As a result, the jet energy is, on average, overestimated.
Second, the jet energy measurement is strongly correlated with the jet topology. If
the leading particle in the fragmentation process that generated the jet was a π 0 , then
all its energy would be deposited in the em calorimeter section. If it were charged,
then most of its energy would go to the hadronic section. The ALEPH detector would
measure a very different value for the energy of these two jets. Assuming that the
non-leading jet particles would generate the same signals in both cases, the difference
between the energies measured for these two jets could be as much as 35% [23]. The
signal distribution for mono-energetic jets detected with the ALEPH calorimeter
would thus be considerably broader than the signal distribution for mono-energetic
pions of the same energy.
Possible evidence for the problems that this calibration scheme may have caused
for jet detection may be derived from Fig. 9.12. This figure shows the total energy dis-
tribution for hadronically decaying Z 0 particles measured with the ALEPH calorime-
ter system, and may thus be considered a jet measurement at 90 GeV.
Of course, some of the jets initiated by b and c quarks may have contained neu-
trinos and muons which were not (completely) absorbed by the calorimeters. Also,
some energy may have leaked out due to the fact that the detector did not cover
the complete 4π solid angle. Such phenomena cause a low-energy tail in the signal
distribution, which can actually be observed in Fig. 9.12.
204 9 Calibrating a Calorimeter System
When this tail was ignored and the distribution in the area E/E LEP = 0.85 − 1.5
(the black experimental points in Fig. 9.12) was fitted to a Gaussian function, a
fractional width σ /mean of 14.3% was obtained. This may be compared with a
width of 8.8% that one would have expected √ on the basis of the energy resolution
quoted for single hadrons (σ/E = 84%/ E [11]).
The fact that not one, but a multitude of hadrons (an unknown and fluctuating
mixture of π 0 s and other particles) were measured simultaneously
√ by the calorimeter,
introduced thus an extra contribution of ∼11% (i.e., 14.32 − 8.82 ) to the energy
resolution in the case of 90 GeV jets.
It should be emphasized that this discussion is by no means intended to criti-
cize ALEPH. This example was chosen because of its educational value. It clearly
illustrates the problematic consequences of a wrong calibration strategy.
In the LEP studies of Z 0 decay, hadronic calorimetry played only a very modest
role. The ALEPH experiment was equipped with very powerful tracking capabil-
ities and offered, therefore, excellent alternatives to calorimetry for studying the
Z 0 physics. The ALEPH Collaboration demonstrated that the distribution shown in
Fig. 9.12 could be considerably improved when the calorimetric information on the
charged hadrons was replaced by the measured track momenta [24]. If, in addition,
the event sample was limited to Z 0 s that produced jets in the central detector region,
then the resolution improved from 14.3% to 6.9%.
However, it should also be emphasized that the very clean nature of the e+ e− →
0
Z events, which made such improvements possible, is by no means representative
for today’s experiments in particle physics, which increasingly rely on excellent and
correctly calibrated calorimeter systems for their physics analyses.
9.5 Off-Line Compensation 205
In the previous sections we have shown several examples of problems that may arise
when a calorimeter system consisting of two longitudinal segments is incorrectly
calibrated. Similar problems occur when the calorimeter consists of three, four, five
or even more longitudinal segments. Also for such systems, the only correct way to
intercalibrate the various segments is with particles that behave, on average, identi-
cally in each and every segment.
Only in this way is the energy deposited in the active calorimeter material mea-
sured unambiguously. Alternative calibration methods, designed to correct the unde-
sirable effects of e/ h = 1 by means of weighting factors, optimized for an “average
shower”, tend to amplify these effects for non-average events.
Sometimes, a longitudinal subdivision into many segments is applied in the hope
of achieving the advantages offered by an intrinsically compensating calorimeter, for
example the superior hadronic energy resolution. However, this cannot be achieved
through general weighting factors, since resolutions are determined by event-to-
event fluctuations, not by mean values. If and only if one succeeds in measuring
the em shower content event by event can one reduce the effect of event-to-event
fluctuations in the em shower content, which tend to dominate the hadronic energy
resolution of non-compensating calorimeters. The dual-readout method discussed in
Sect. 8.3 successfully achieves that goal. Prior to that, attempts were developed to
use differences between the profiles of the em and non-em components of hadron
showers to that end. This method, that became known as off-line compensation, was
pioneered by the WA1 Collaboration [25].
The H1 Collaboration [26] applied a calibration method in which large local
energy deposits were selectively suppressed. This also eliminated some of the con-
sequences of event-to-event fluctuations in the em content of hadronic showers. Their
LAr calorimeter consisted of an em compartment subdivided into five longitudinal
segments and a hadronic compartment subdivided into six such segments. The total
energy was calculated as
5
6
E = Cew Q i (1 − ηe Q i ) + Chw Q i (1 − ηh Q i ) (9.5)
i=1 i=1
Fig. 9.13 WA1 results on offline compensation. The signal distributions for 140 GeV pions (a) and
the hadronic energy resolution as a function of energy (b), before and after the weighting procedure
described in the text was applied to the experimental data. From: Abramowicz, H. et al. (1981).
Nucl. Instr. and Meth. 180, 429
large concentrations of energy deposit responsible for the high-end tails of pionic
signal distributions in a non-compensating calorimeter.
The experimental data (Fig. 9.13) showed that procedures of this type worked
especially well at high energies, where π 0 s and non-em forms of energy deposit led
to distinctly different signals in individual longitudinal segments of the detectors. At
low energies, where these differences were much less spectacular, the improvements
in energy resolution were found to be marginal, at best. Also the hadronic signal
non-linearity did not disappear at energies below 30 GeV. Neither WA1 nor H1 have
demonstrated any beneficial effects of these “off-line compensation” methods for
jets, or more generally for a situation in which energy is deposited in the calorimeter
system by a collection of particles with unknown composition and energies.
The ATLAS experiment has been inspired by the technique developed by H1 to
correct their hadronic energy measurements for the effects of non-compensation. The
ATLAS barrel calorimeter consists of a LAr ECAL plus an iron/plastic-scintillator
HCAL, each of which is subdivided into three longitudinal compartments. In addi-
tion, there is a presampler intended to recover energy lost in the material upstream
of the calorimeter system. In the beam tests on the basis of which their “cell weight-
ing method” was developed, the hadronic calorimeter consisted of four longitudinal
compartments. In addition, a “mid sampler” was installed to account for energy lost
in the cryostat walls separating both calorimeter systems [27].
Also in their approach, the reconstruction of the energy of the showering hadron
relies on upwards corrections of relatively small signals. The signals from cells that
are relatively small compared to those from cells in which em shower components
deposit energy are given a weight intended to equalize the response of those two
types of cells. The total energy is determined with a formula similar to (9.5):
9.5 Off-Line Compensation 207
E= Wem (E cell , E beam )E cell + Whad (E cell , E beam )E cell + E cryo (9.6)
em cells had cells
The weight factors applied to the signals from each individual calorimeter cell are a
function of the compartment in which the calorimeter cell is located and of the beam
energy:
The parameters A E , B E , A H and B H were taken from a fit and all depend on the
hadron energy. Since finding the hadron energy was the purpose of the entire exercise,
the parameter values had to be determined in an iterative process. The term E cryo is
intended to account for the loss of energy in the cryostat and is proportional to the
geometric mean of the energy lost in the last em compartment and the first hadronic
one. It turned out that E cryo defined in this way was nicely correlated with the energy
detected in the “mid sampler”, so that there was no need to include the latter device
in the design of the final detector.
As stated before, procedures of this type may be applied with some success in
beam tests, where the properties of the particle that generates the signals are precisely
known. However, the benefits in the messy environment of a high-luminosity hadron
collider are not clear, and in any case remain to be demonstrated.
9.6 Conclusions
3. The third strategy aimed for hadronic signal linearity in combination with recon-
struction of the beam energy for electrons and pions simultaneously. This was
done by increasing the signals from the em calorimeter compartment for hadron
showers by a weighting factor (Sect. 9.3.8). A similar strategy was used to make
the hadron response independent of the starting point of the showers (Sect. 9.3.9).
4. The fourth strategy aimed to achieve equal responses for electrons and pions by
multiplying the signals from the (8X 0 thick) first calorimeter segment with zero
(Sect. 9.3.10).
All four strategies looked, at first sight, reasonable and logical. They were all
designed to achieve a specific beneficial result: optimization of the energy resolution
(method 1), signal linearity for hadrons (method 3), equalization of the response to
electrons and pions (method 4). Yet all four methods turned out to be fundamentally
flawed at more or less severe levels. In each case, the benefit for which the calibration
method was designed was offset by the introduction of new, sometimes very severe,
problems such as
• Signal non-linearity for em showers (1,4)
• Systematic mismeasurement of the energy of early (2) or late (3) showering par-
ticles
• Systematic mismeasurement of the energy of hadrons that penetrate the em
calorimeter compartment without starting a shower (3)
• Systematic mismeasurement of the energy of jets (2,3,4)
• A general degradation of the energy resolution and the line shape (3,4)
• Systematic mismeasurement of the energy of γ s and π 0 s (1,4)
The underlying reason for all these problems is the fact that the mentioned cali-
bration strategies only work on average. The envisaged goals are only achieved for a
particular subset of events, for example the pions that penetrate the em compartment
without interacting (method 2), or electrons of a given energy (method 1), or pions
that deposit 44 ± 1% of their energy in the em calorimeter compartment (method 3).
At the same time, new problems are introduced for events that do not resemble the
average for which the calibration method was designed. Manipulating calibration
constants may thus seem to work out fine for one particular subset of events, but may
have very negative consequences for other, possibly more relevant, subsets.
The only way to avoid such problems is to calibrate the individual sections of a
longitudinally segmented calorimeter system in exactly the same way (B/A = 1).
If the individual compartments can be separated and are deep enough to contain
electron showers, then one may use an electron beam for this purpose. If that is not
possible, then these sections may be intercalibrated with muons traversing the entire
depth of the calorimeter. Only in this way is the relationship between the deposited
shower energy and the resulting signal established unambiguously, identically for all
calorimeter sections.
In doing so, the conditions that exist automatically in longitudinally unsegmented
calorimeters and that make these devices trivial to calibrate are reproduced. This leads
us to the conclusion that longitudinal segmentation, in general, does not serve any
9.6 Conclusions 209
purpose with regards to the precision with which the energy of showering particles
can be measured in a calorimeter.
Only if the segmentation makes it possible to determine the em shower content
event by event could some benefit be expected. However, even the feasibility of this
is questionable in environments other than a testbeam where particles of precisely
known energy and type are delivered to the detector.
The above conclusion may be illustrated by considering a perfectly compensating
calorimeter, i.e., a calorimeter in which event-to-event fluctuations in the em shower
content do not contribute to the energy resolution. If such a calorimeter were longi-
tudinally segmented, then the extra information on the shower development would
not allow one to improve the energy resolution. In fact, one could only deteriorate
the excellent resolution of the unsegmented device by calibrating the segments in a
way that differs from the only correct one: B/A = 1.
We have noticed that there is a general belief that collecting more information
about the absorption process inside calorimeters, which is the goal of having mul-
tiple longitudinal segments, must necessarily lead to achieving better performance
characteristics. We think that this is an illusion, for the same reason as it is an illusion
to think that by measuring the four-vectors of all the individual molecules of the gas
in a certain container one would be able to make a more precise determination of the
temperature inside that container. It would just have the effect of complicating the
issues to a degree that would make it almost impossible to avoid making mistakes.
Something very similar is happening here. Simple is better, and money for readout is
much better spent on lateral segmentation (granularity!) than on an increased number
of longitudinal compartments.
A fine lateral segmentation could also be very helpful for the identification of
electrons and γ s, one of the other issues that is often used as an argument in favor
of longitudinal segmentation. This is further discussed in the context of applications
of calorimeter information for particle identification (Chap. 11).
References
The calorimeters used in experiments at colliders do not operate in the ideal con-
ditions that are typical for beam tests of prototype modules. Instead, they face a
challenging environment that tends to make the extraction of the envisaged physics
information, and sometimes the very operation of the instrument, problematic. This is
increasingly true for experiments at each new generation of colliders, since the cross
sections of the interesting processes and the background obscuring these processes
increases with the center-of-mass energy of the collisions.
In this chapter, we discuss three sources of problems that play a role in this context:
1. Magnetic fields
2. Radiation damage
3. High luminosity
The use of magnetic materials, such as iron or nickel, should be avoided in the con-
struction of the calorimeters. The forces exerted on these materials when the magnetic
field is switched on could cause severe damage to the experimental equipment. But
even if everything were so rigidly constructed that nothing physically moves when
the field is switched on, the iron could severely distort the magnetic field itself,
causing problems for the reconstruction of the trajectories of charged particles.
Operation in a magnetic field also imposes limitations on the readout technology that
can be used. If the calorimeter produces light signals, then it is often not possible to
use photomultiplier tubes as light detecting elements.
Most PMTs are so sensitive to magnetic fields that already fields as weak as the
Earth’s magnetic field (∼5 · 10−5 T) may have a significant effect on the gain. It can
be easily checked whether or not this is the case for a certain setup by changing the
orientation, since only the non-axial component of the field affects the trajectories
of the photoelectrons.
To avoid such effects, PMTs are usually equipped with shields made of material
with a high magnetic susceptibility (“μ-metal”), which surround the sensitive areas
(mainly the photocathode). Some PMTs have been specially designed to be able to
operate in magnetic fields. This is mainly achieved through a very compact dynode
structure, as in close proximity focusing tubes [1]. Also compact structures with
only one or two dynodes (known as vacuum phototriodes or phototetrodes) have
been demonstrated to be capable of operating in moderate magnetic fields, up to ∼1
T. Both OPAL [2] and DELPHI [3] have used such devices to read out their lead-
glass em calorimeters. CMS uses phototriodes for reading out the PbWO4 crystals of
the endcap sections of their electromagnetic calorimeter [4], where the (solenoidal)
magnetic field is almost axial. These particular devices have an anode of very fine
copper mesh (10 µm pitch), which allows them to operate with <10% gain loss in
the 4 T magnetic field of CMS [5].
Alternative light detectors capable of operating in (strong) magnetic fields include
avalanche photodiodes (APDs) and silicon photomultipliers (SiPMs). However, each
of these alternative solutions may have disadvantages, compared with PMTs. For
example, these devices may be very unstable under small changes in temperature
or bias voltage. APDs are also very sensitive to charged shower particles crossing
them, which may lead to catastrophic effects (Sect. 4.3), and SiPMs are intrinsically
non-linear devices (Sect. 4.2.2).
The effects of magnetic fields on detectors depend strongly on the orientation of the
magnetic field with respect to the electric field that accelerates the (photo)electrons
towards the collecting electrode. The effects are of course minimal when B E.
They reach a maximum when B ⊥ E, because the slow (photo)electrons are bent
away from the direction they should follow to produce a signal.
One way to deal effectively with the possible problems caused by magnetic fields
in light detectors is to transport the light signals from their source to an area where
the magnetic field is so low that it does not prohibit readout with PMTs. Clear
plastic fibers allow one to do so with minimal light losses. Tests done in the context
10.1 Operation in a Magnetic Field 213
Apart from the effects that magnetic fields have on the functioning of detectors
and electronics that handle calorimeter signals, they may also affect these signals
themselves. We mention two examples.
Increased Light Yield
The light yield of some plastic scintillators changes when placed in a magnetic field.
Typically, the light yield increases, by some 5–10%, for magnetic fields up to 3 T
[7–11]. For larger fields, no further increase was observed by Bertoldi and coworkers
[11], who tested a variety of different scintillators and wavelength shifters for fields
up to 20 T. The same authors also reported that the increased light yield was only
observed when the scintillators were excited with ionizing particles. In particular,
no significant changes in the light yield were observed when the excitation was
performed with UV light.
This is a strong indication that the effects are due to an increased production
of UV light in the excitation of the polymer base material (polystyrene or polyvinyl
toluene), since the excitation of the scintillating fluors with UV light was not affected
by the magnetic field.
The light yield is not a simple function of the field strength, and there is no
experimental indication that the effects depend on the orientation of the magnetic
field or on the nature of the ionizing radiation.
Effects on Shower Profiles
Magnetic fields may also affect the shower profiles, since the paths of the charged
shower particles are subject to the Lorentz force. This may have consequences for the
early, non-isotropic shower component. These consequences depend on the strength
and the orientation of the magnetic field. If the field is perpendicular to the shower
axis, then the lateral shower profile will be broadened. A large fraction of the signals
from electromagnetic showers comes from soft electrons, produced in Compton
scattering and in photoelectric processes (Sect. 3.4). For example, in a copper-based
214 10 Operational Challenges
calorimeter, more than half of the shower energy is deposited by electrons softer
than 4 MeV (Fig. 3.9). The trajectory of these soft electrons is very sensitive to
magnetic fields, since they are subject to the Lorentz force. For example, the radius
of curvature of an electron with a momentum of 4 MeV/c in a 2 T magnetic field
oriented perpendicular to its direction of motion is only about 6 mm.
These soft electrons only contribute to the signals from sampling calorimeters if
they are produced in a very thin boundary layer. The range of 4 MeV electrons in
copper is only 3 mm (Fig. 3.12a). If a 4 MeV electron escaped from the copper and
traversed a 5 mm thick plastic scintillator layer perpendicularly, then it would lose
typically 1 MeV. However, if the calorimeter were placed in a strong magnetic field
oriented parallel to the sampling layers, then this escaping electron would describe
a curved trajectory and lose a considerably larger fraction of its energy in the scin-
tillator. Depending on the gap width between the absorber plates and the strength
of the magnetic field, it could even reverse direction and deposit its entire energy in
the scintillator plate, in a way similar to the “cork-screws” that were the signature of
electrons in bubble-chamber pictures (Fig. 10.1a).
This effect results in an increased calorimeter response. It increases with the
strength of the magnetic field, since the fraction of shower particles trapped in the
gaps between the absorber plates increases with the field strength. It would not play
a role for magnetic fields oriented perpendicular to the sampling layers.
Fig. 10.1 Trajectories of few-MeV electrons contributing to the signals of a sampling calorimeter
in the absence and presence of a magnetic field oriented parallel to the sampling layers, pointing into
the plane of the figure (a). The relative increase in the response of the CMS copper/plastic-scintillator
calorimeter, as a function of the strength of a magnetic field oriented parallel to the sampling layers
(b). The response dependence is given for showers induced by 100 GeV electrons and by 100 GeV
pions. In order to eliminate the effects on the specific light yield of the scintillator, all responses
have been normalized to those for muons traversing the calorimeter in the same direction as the
showering particles. From: Kunori, S. (1997). Proc. 7th Int. Conf. on Calorimetry in High Energy
Physics, Tucson, Arizona (Singapore: World Scientific, Singapore), p. 224
10.1 Operation in a Magnetic Field 215
The described effects were observed by the CMS Collaboration in prototype tests
of their hadron calorimeter [12]. Since the copper/plastic-scintillator CMS calorime-
ter has to operate in a very strong magnetic field, the effects of this field on the
calorimeter performance were studied in great detail. Some results of these studies
are shown in Fig. 10.1b, where the relative increase in the calorimeter response is
plotted as a function of the magnetic field strength, for showers induced by 100 GeV
electrons and 100 GeV pions. The field was oriented parallel to the sampling layers
in these studies. In order to eliminate the “scintillator brightening” effects discussed
in Sect. 10.1.2, the calorimeter responses to these particles were normalized to those
for muons that traversed the calorimeter in the same direction. The em response was
measured to increase by about 20% when the field strength reached 3T, the maximum
value for which measurements were performed in these studies.
The effect of the magnetic field on the hadronic response was clearly smaller,
commensurate with what should be expected if only the em shower component was
affected by the field. This makes perfect sense, since hadronic shower particles with
similar sensitivity to the Lorentz force as the electrons from the example discussed
above are extremely non-relativistic and would thus deposit their entire kinetic energy
in the scintillator, with or without a magnetic field. Therefore, the field does not affect
the response to the non-em shower component. Since only the em shower component
is affected, the e/ h value of the calorimeter is thus increased when it operates in a
magnetic field.
The CMS tests showed no measurable effect on the calorimeter response (over
and above the scintillator brightening discussed above), if the magnetic field was
oriented perpendicular to the sampling layers, i.e., parallel to the shower axes.
Fig. 10.2 Effects of radiation damage on the performance of the CMS em calorimeter. The left
diagram shows the light transmission in the PbWO4 crystals after irradiation with γ s and protons.
The overlaid black dotted line represents the lead tungstate light emission spectrum. The right
diagram shows the (simulated) effect on the scintillation signals from 50 GeV showers in the
crystals as a function of the pseudorapidity, for various values of the integrated luminosity. From:
CMS Collaboration (2015). Technical proposal for the phase-II upgrade of the Compact Muon
Solenoid, CERN-LHCC-2015-10
concern as well. As an example, I mention the case of the ATLAS Liquid Argon
Calorimeter [14]. In preparation for the increased LHC luminosities foreseen for the
future, ATLAS will replace the ASICs1 that handle the calorimeter signals by more
radiation hard ones. ASICs based on IBM’s 130 nm CMOS (8RF) technology meet
the requirements in that respect. The new ASICs will also be better adapted to handle
the trigger rates expected at the higher luminosities. The analog on-detector Level-1
pipeline will be replaced by a system in which all calorimeter signals are digitized at
40 MHz and sent to the off-detector front-end electronics. This approach is expected
to remove all constraints imposed by the calorimeter readout on the ATLAS trigger
system.
10.3 Pileup
Radiation damage is not the only problem LHC experiments face as a result of the
increasing luminosity. At the present time, ATLAS and CMS have to deal with, on
average, 55 events per 25 ns bunch crossing (at a luminosity of 2 · 1034 cm−2 s−1 ).
This rate is increasing proportionally with the luminosity and in the High Luminosity
LHC era (after 2024), one will have to cope with a luminosity of 7 · 1034 cm−2 s−1 , at
which point interesting events accompanied by more than 200 “pileup” events will
be no exception. Of course, the overwhelming majority of these “underlying events”
are uninteresting, and involve relatively few high-p⊥ particles that contribute to the
detector signals. Yet, pileup induced background is expected to be a factor that
seriously deteriorates the detector performance. High-precision timing is considered
one of very few options to mitigate these effects, and this has given rise to a number
of dedicated experimental studies [15–17].
The elapsed time for an LHC bunch crossing has an rms spread of 170 picosec-
onds, which means that the 50–100 ps time resolution commonly achieved in the
time-of-flight systems used for particle identification purposes is not adequate for
solving this problem. One expects to need time resolutions of at least 20–30 ps to
make a significant difference in this respect. A major complicating factor is that this
performance has to be achieved in a very-high-rate environment. The approach fol-
lowed by the mentioned R&D projects focuses on instantaneous light signals, such as
those produced by the Čerenkov mechanism, combined with ultrafast photo detec-
tors, such as Avalanche Photo Diodes (APDs) [15], Microchannel Plates (MCPs)
[16], or micromegas [17].
Figure 10.4 shows results recently obtained with MCPs. These detectors are either
operated in the standard PMT-MCP mode, or in the i-mode, in which the photocath-
ode is removed and the signal is produced by secondary emission of electrons from
the MCP layers crossed by the ionizing particles [18]. The figure shows the distri-
bution of the time difference between signals from 491 MeV electrons measured in
two MCPs, one operated in i-mode and the other in PMT-MCP mode. This Gaussian
distribution has a width (σ ) of 17±2 ps. A time resolution of 75 ps was reported for
single photoelectrons by [17]. These are encouraging developments, but there is of
course still a long way to go before systems capable of assigning signals to different
events occurring in the same bunch crossing will be available for implementation in
the extremely high-rate environment of LHC-type experiments.
In a more traditional approach to the pileup problem, it is treated as an additional
source of electronic noise. Using a series of signal samples collected at intervals
of 25 ns, the properties of the true signal are estimated on the basis of a variance
minimization of the noise covariance matrix, a method known as optimal filtering
[19]. The bipolar pulse shaping, introduced by Radeka [20], is crucial for the success
of this method with the LAr signals from the ATLAS calorimeters. This method
10.3 Pileup 219
Fig. 10.5 Cell energy distribution reconstructed by the Constrained Optimal Filter √ (COF) and by
the Optimal Filtering (OF2) algorithms, using 2012 ATLAS pp collision data at s = 8 TeV and
25 ns bunch spacing. The average number of interactions per bunch crossing was 11.3 for this
event sample (around 25 millions entries). The COF method is resilient to out-of-time signals and,
therefore, leads to a better energy resolution than OF2. Its design is luminosity independent and
requires only the information of the pulse shape and pedestal value to compute the 7 amplitudes
associated to the 7 samples of the read-out. In this figure only the central sample reconstruction is
shown. From: J.M. Seixas (ATLAS), private communication
works best for Gaussian noise. Pile-up has the tendency to add positive or negative
tails to the noise distribution, which thus becomes non-Gaussian. The extent of these
effects depends on the number of underlying events, and thus leads to a luminosity
dependence.
Recently, a method has been proposed that is in principle independent of the
luminosity. It is based on a deconvolution process of the same type used in digi-
tal processing for communication channel equalization, and aims to fully recover
the target signal, rather than estimate its amplitude from pulse sampling [21]. This
method has been tested with experimental data obtained with the ATLAS TileCal
hadronic calorimeter. Figure 10.5 shows some results from this work.
References
Calorimeters exist in a wide variety and are used in very different types of experi-
ments. The design of a particular calorimeter system is usually driven by require-
ments stemming from the physics goals of the experiment, and by the available
budget. These factors lead to large differences in performance between the various
calorimeter systems used in particle physics experiments.
In this chapter we review some of the aspects of the performance of calorimeter
systems. The factors that determine and limit various aspects of this performance
are described, and results from some representative calorimeters are presented. The
possibilities for using calorimeter information for particle identification are also
discussed.
In Chap. 6, the factors that contribute to and determine the energy resolution of
calorimeters are discussed in detail. It is clear what is needed to set new records
for calorimetric energy resolution, both for what concerns the detection of elec-
trons/photons and for hadrons/jets. However, in modern collider experiments, design
choices are often driven by considerations other than the desire to obtain the best
possible energy resolution. Assuming that the performance is adequate for achieving
the scientific goals of the experiment, the most important practical considerations
when choosing a calorimeter system include:
• The cost,
• The size, which may affect the cost of other components of the detector system,
and
• The expected lifetime, in view of radiation and other environmental conditions.
The last point puts systems based on scintillating crystals at a disadvantage, espe-
cially at colliders that involve hadrons. These crystals are, apart from very expensive,
also very sensitive to ionizing radiation. The CMS experiment, which uses PbWO4
© Springer Nature Switzerland AG 2019 223
M. Livan and R. Wigmans, Calorimetry for Collider Physics, an Introduction,
UNITEXT for Physics, https://doi.org/10.1007/978-3-030-23653-3_11
224 11 Calorimeter Performance
crystals as their em calorimeter experienced this in a big way. Even though these
crystals were initially chosen based on their supposed radiation hardness, CMS has
to replace a substantial fraction of them after having received only a few percent of
the envisaged total integrated luminosity. Crystal calorimeters are still used in some
experiments at e+ e− colliders, such as KEKB, where the event rates are many orders
of magnitude smaller than at colliders involving hadron beams. Figure 11.1 shows
the energy resolution for photons detected in the Belle-II calorimeter, which consists
of CsI(Tl) crystals [1].
A good and increasingly popular alternative is provided by liquified noble gases.
Both the ionization charge and the scintillation produced in the absorption process
may be used as the source of the signals in this case. Radiation damage is not much
of a concern since the liquid can be easily replaced, but the fact that these devices
have to operate in cryogenic conditions, as well as the VUV nature of the scintillation
(∼100 nm), are the cause of challenging complications.
An example of a homogeneous em calorimeter that detects the ionization charge
generated by the shower particles can be found in Novosibirsk, where a 70-ton
liquid-krypton detector (KEDR) operates at the electron–positron collider VEPP-
4M [2]. The energy resolution of this device was measured with positrons on a 400
kg prototype. The measured σ/E values ranged from ∼5.7% at 0.13 GeV to 1.7% at
1.2 GeV [3].
The NA62 experiment operates a 27X0 (1.25 m) deep LKr calorimeter at CERN’s
SPS, which earlier served the predecessor experiment NA48 [4]. Its em energy res-
olution was measured with electrons at high energies (10–80 GeV) [5]. The results
are shown in Fig. 11.2.
Also LAr and LXe are used in this mode. The CMD-3 Collaboration operates a
400-liter LXe detector at the e+ e− Collider √ in Novosibirsk [2]. For a 40-liter LXe
device, a resolution was reported of 3.4%/ E, for electrons in the energy range
1–6 GeV [6]. Because of the small size of this detector, shower leakage probably
contributed significantly to this result. The latter problem does not play a role for the
multi-ton LAr detectors ICARUS, MicroBooNE and DUNE. Despite the very long
radiation length (14 cm), these detectors are sufficiently large to fully contain em
11.1 Energy Measurement 225
showers (Fig. 1.8b). A first attempt to measure the energy of em showers developing
in liquid argon was performed by members of the ICARUS collaboration, who used
signals from such showers in their 600-ton detector to select π 0 → γ γ events [7].
Using a restricted sample of “clean” events with an average energy of 700 MeV, they
measured the π 0 mass with a resolution of 16%.
By far the best performance for calorimetric hadron detection is achieved with
sampling calorimeters. The reasons for this are spelled out in detail in Chap. 7. The
overwhelmingly dominating role of fluctuations in the em shower content (and the
related fluctuations in invisible energy) on the hadronic energy resolution of calorime-
ters is crystal clear. This is perhaps most dramatically illustrated with Fig. 11.3a,
which shows the energy resolution for pions in a very large, fully homogeneous
calorimeter consisting of 60 tons of mineral oil doped with scintillating agents [9].
All other sources of fluctuations, such as sampling fluctuations, have been completely
eliminated in this device. The energy resolution turned out to be only weakly depen-
dent on the energy of the pions and did not drop below ∼10%, even at the highest
energies (150 GeV) at which this detector was tested. For comparison, the resolution
of SPACAL, a calorimeter in which only 2.3% of the pion energy was sampled, was
about three times better at this energy (Fig. 11.3b). Even better performance was
reported by RD52, for their prototype dual-readout fiber calorimeter (Fig. 8.5).
Experimental data make it very clear that there is a price to be paid, in terms of
degraded hadronic performance, when the calorimeter system is designed for optimal
em energy resolution. This is because of the large e/h values typical for good em
calorimeters. A case in point is CMS, where the em calorimeter section has an e/h
value of 2.4. The result of this choice is not only an exceptionally poor hadronic
energy resolution, but also very substantial signal non-linearities, and large response
differences depending on the starting point of the showers (Fig. 7.6).
226 11 Calorimeter Performance
The most frequently used method to determine the position of a particle that showers
in a calorimeter is by reconstructing the center of gravity (x̄, ȳ) of the energies Ei
deposited in the various detector cells (with coordinates xi , yi ) that contribute to the
signal:
xi Ei
x̄ = i (11.1)
i Ei
individual calorimeter cell, and the exact percentage is only weakly dependent on
the location of the impact point.
This effect is illustrated in Fig. 11.4, which shows the signals generated by
80 GeV showering electrons in neighboring cells of the SPACAL calorimeter [13].
Only in events in which the impact point was located close to the boundary between
different calorimeter cells was the shower energy really shared between different
cells. In these cases, the particle’s position was most precisely and most accurately
measured. In other situations, one cell received a lion’s share, and the remaining
“halo” was shared among the surrounding cells.
Consider an event (E1 ) in which the showering particle entered the calorimeter
halfway between the center of a cell (say cell #1) and the boundary with another cell
(cell #2), e.g., at the position y ≈ 20 mm in Fig. 11.4. For this event, the fraction of
the shower energy deposited in cell #1 would not be very different from the fraction
recorded for another event (E2 ), in which the particle entered in the center of cell #1
(y ≈ 40 mm). In both cases, the contribution of cell #1 to the sum in Eq. 11.1 would
be about the same. In both cases, the coordinates of the particle’s impact point would
be calculated (with Eq. 11.1) under the assumption that the entire energy recorded in
cell #1 was deposited in its center (point C).
Small differences between the signals recorded in the various surrounding cells
would in events of the type E1 thus be the only basis for finding an impact point
deviating from the center (C) of the hit calorimeter cell. However, the reconstructed
impact point would always be located too close to C, because the bulk of the signal
was mis-attributed as originating from this point.
The effects described above are responsible for the peculiar patterns in Fig. 11.5.
This figure shows scatter plots for 80 GeV electrons detected with SPACAL, in which
the impact point of the particles, reconstructed from the calorimeter data, is plotted
versus the “true” impact point, measured with wire chambers installed upstream of
the calorimeter [13].
Figure 11.5a, in which this information is given for the x coordinate (see Fig. 11.6c
for the definition of the coordinate system in these hexagonal cells), shows that the
impact point was only correctly reconstructed when the particles entered either in
the center of a calorimeter cell (x = 0), or near the point where three hexagonal cells
228 11 Calorimeter Performance
Fig. 11.5 Scatter plots for 80 GeV electrons detected with the SPACAL calorimeter, showing
the relations between the coordinates of the particle’s impact point, measured with wire chambers
upstream of the calorimeter (horizontal), and determined from the calorimeter data on the basis of
the center-of-gravity method (vertical). The data shown in a and c concern the x coordinate, b and
d refer to the y coordinate. The plots in c and d were obtained after applying the corrections from
Eq. 11.2 to the calorimeter data. See text for details. From: Acosta, D. et al. (1991). Nucl. Instr. and
Meth. A305, 55
joined (x ≈ 40 mm). In these cases, the points in the scatter plot cluster around the
dashed line, which represents the equality of the x values reconstructed on the basis
of the calorimeter data on the one hand and the upstream wire chambers on the other.
However, if the electron entered the calorimeter anywhere else than in the men-
tioned areas, then the impact point was always reconstructed too close to the cell’s
center. Figure 11.5a shows that practically all events with (true) impact points in a
cylinder with a radius of about 25 mm around the cell’s center were reconstructed
as entering very close to the calorimeter center (xCAL = 0). These events constitute
the horizontal band in this figure. The systematic mismeasurement of the particle
position could thus be as large as 30 mm when the calorimeter data were used at face
value.
11.2 The Other Components of the Four-Vector 229
Fig. 11.6 The position resolution for 80 GeV electrons in SPACAL as a function of x (a) and y
(b). The coordinates (x , y ) used for the hexagonal SPACAL geometry (c), and their relationship
to the cartesian (x, y) coordinates. From: Acosta, D. et al. (1991). Nucl. Instr. and Meth. A305, 55
Figure 11.5b shows similar data for the y coordinate. Also here, the experimental
points cluster in a horizontal band (around the calorimeter center, yCAL = 0), and
only near the boundaries with neighboring modules were the impact points correctly
reconstructed. In the area between the center and these boundaries, the y coordinate
was also systematically mismeasured when the calorimeter data were used at face
value.
This problem can be solved in several ways. In one method that is sometimes
applied, a larger weight is given to cells in which only a small fraction of the shower
energy is deposited. In another method, the position of the center of gravity found
with Eq. 11.1 is shifted, using an empirical algorithm. For example, an algorithm of
the type
xcorr = A arctan(Bx ) (11.2)
applied to the data from Fig. 11.5a, reproduced the x coordinate of the impact points,
on average, very well (Fig. 11.5c). A similar algorithm changed the picture for the y
coordinate (Fig. 11.5b) to the pattern shown in Fig. 11.5d.
The position resolution is given by the widths of the bands in Fig. 11.5c (for the x
coordinate) and d (y ). Not surprisingly, this width was found to depend on the impact
point of the particles. The smallest resolutions were obtained in the boundary areas
between different calorimeter cells (Fig. 11.6). Averaged over one cell, the position
230 11 Calorimeter Performance
resolutions σx and σy were found to be 1.8 mm and 1.6 mm, respectively, for 80
GeV electrons.
√
The position resolution may be expected to scale with 1/ E on the basis of the
following arguments. The energy deposit Ei in √ each cell i has a relative precision
σi /Ei . This relative precision improves as 1/ E with the total shower energy E,
provided that the (average) shower profile stays the same. In that case, the energy
sharing between the various calorimeter cells i is, on average, independent of the
shower
√ energy. If all the terms in Eq. 11.1 have a relative precision that scales with
1/ E, and if the relative contributions of the individual terms to the sum are energy
independent, then the relative precision of the final result (i.e., the sum
√ of all the
terms, or the value of the position coordinate) must also scale with 1/ E.
This is indeed in agreement with experimental observations. For electrons entering
the detector in the center of a cell, SPACAL measured a position resolution
17.1 mm
σy = √ (11.3)
E (GeV)
and a similar result was obtained for the x coordinate. The cell center represents
the worst possible case for what concerns the position resolution. Averaged over
the entire surface, the position resolution was found to be about 20% better than
indicated in Eq. 11.3 [13].
The position resolution is not only determined by the energy resolution, but also
by the cell size. The smaller the cell size (measured in the relevant units of the
Molière radius, ρM ), the more cells contribute to the signals, and the more accurately
the shower’s center of gravity can be determined.
The cell size of SPACAL, which had an effective radius of 1.9ρM , was by no
means optimized for electron impact-point determination, since an electron hitting
a cell in its central region deposited typically ∼95% of its shower energy in this
one cell. The position resolution of the RD1 projective prototype fiber calorimeter,
which had a cell size with an effective radius of 1.1ρM , and an em energy resolution
√ to SPACAL, was found to be smaller by more than a factor of two: σx,y = 7.5
similar
mm/ E [14].
The precision with which the mass can be measured is thus not only determined by
the energy resolution, i.e., the measurement uncertainty on the γ energies E1 and
E2 , but also by the relative uncertainty on the angle (θ12 ) between the directions of
these γ s. The latter uncertainly is of course directly affected by the precision of the
mentioned localization procedure.
Figure 9.10 shows the invariant mass distributions of γ γ and γ γ γ γ combinations
recorded in the e+ e− collisions at the φ resonance in DAPHNE. The masses of the
π 0 , η and KS0 , which decay in these exclusive modes, were reconstructed, on average,
to within 1% of their established values, with good resolutions [17].
A good localization of the γ s is thus very important to identify the parent particle.
While CMS emphasized excellent energy resolution for em showers in its design of
the experiment, at the expense of degraded hadronic performance, ATLAS concen-
trated its efforts also on the localization issue. As a result, the mass resolution for
the Higgs bosons, measured in the H 0 → γ γ decay channel, turned out to be very
similar in both experiments.
Fig. 11.8 Typical calorimeter signals for 150 GeV electrons (a) and pions (b) measured with the
SPACAL calorimeter. The pion signal exhibits a clear exponential tail with a time constant of ∼10
ns (c). The t = 0 point is arbitrary and the bin size is 1 ns. From: Acosta, D. et al. (1991). Nucl.
Instr. and Meth. A302, 36
Fig. 11.9 Distribution of the E/p values of a sample of candidate electrons recorded in the E-70
experiment at Fermilab, for particles with momenta ranging from 55 to 60 GeV/c (a) and from 65
to 70 GeV/c (b). From: Appel, J.A. et al. (1975). Nucl. Instr. and Meth. 127, 495
Fig. 11.10 The ratio of the energy measured by the ECAL to the momentum measured by the
magnetic spectrometer in the LHCb experiment (a). Invariant mass plots for the e+ e− pairs in
a sample of Bs0 → J /ψφ signal events, showing the effect of a cut pT > 0.5 GeV/c for the e±
candidates (b) on the reconstructed J /ψ peak. From: Alves, A.A. et al. (2008). JINST 3, S08005
is mainly due to pion tracks with low transverse momentum, and could be efficiently
removed with a pT cut that barely affected the electrons. It turns out that the average
efficiency for identifying electrons from J /ψ → e+ e− decays in such events is 95%,
with a pion mis-identification fraction of 0.7%.
Fig. 11.11 Effects of four different shower characteristics that may be used to distinguish between
electron and hadron showers in the longitudinally unsegmented RD52 lead-fiber calorimeter. Shown
are the fraction of the total signal recorded by the tower in which the particle entered (a), the ratio
of the Čerenkov and scintillation signals of the event (b), the starting time of the signal in the PMT,
measured with respect to an upstream trigger signal (c), and the ratio of the total integrated charge
and the amplitude of the signal (d). Data obtained with 60 GeV particle beams. From: Akchurin,
N. et al. (2014). Nucl. Instr. and Meth. A735, 120
face of the calorimeter. Figure 11.11c shows distributions of the starting time of
the PMT signals for 60 GeV electron and pion showers.
4. The same phenomenon also leads to a larger width of the hadron signals, since the
light is produced over a much larger region in depth than for electrons. Therefore,
the ratio of the integrated charge and the signal amplitude is typically larger
for hadron showers. Figure 11.11d shows distributions of that ratio for showers
induced by 60 GeV electrons and pions.1
One may wonder to what extent the different methods mentioned above are cor-
related, in other words to what extent the mis-identified particles are either the same
or different ones for each method. It turned out that by combining different e/π sep-
1 This result may be compared with Fig. 8.14, which shows the difference between the widths of
electron and pion signals in the SPACAL calorimeter. However, in that case, the differences were
greatly enhanced by the aluminized upstream ends of the fibers. This had a much larger effect on
the signal structure of the hadron showers than for the electron ones.
11.3 Particle Identification 237
Čerenkov light has a number of specific properties that offer unique possibilities for
application in calorimetry:
1. It is directional.
2. It is instantaneous.
3. Its spectrum is quite different from that of typical scintillators.
One unexpected consequence of the directionality of the Čerenkov light concerned
the detection of muons with the dual-readout fiber calorimeter discussed in Chap. 8.
Simultaneous detection of the scintillation and Čerenkov light produced in this instru-
ment turned out to have unique beneficial aspects for the detection of these parti-
cles. Figure 11.12 shows the average signals from muons traversing the DREAM
calorimeter along the fiber direction [23]. The gradual increase of the response with
the muon energy is a result of the increased contribution of radiative energy loss
(bremsstrahlung) to the signals. The Čerenkov fibers are only sensitive to this energy
loss component, since the primary Čerenkov radiation emitted by the muons falls
outside the numerical aperture of the fibers. The constant (energy-independent) dif-
ference between the total signals observed in the scintillating and Čerenkov fibers
238 11 Calorimeter Performance
Fig. 11.12 Average values of the scintillation and Čerenkov signals from muons traversing the
DREAM calorimeter, as a function of the muon energy. Also shown is the difference between these
signals. All values are expressed in units of GeV, as determined by the electron calibration of the
calorimeter. From: Akchurin, N. et al. (2004). Nucl. Instr. and Meth. A533, 305
thus represents the non-radiative component of the muon’s energy loss. Since the
signals from both types of fibers were calibrated with em showers, their responses
to the radiative component were equal. This is a unique example of a detector that
separates the energy loss by muons into radiative and non-radiative components.
The difference in the time structure of the signals is an important characteristic
that can be used to distinguish between the scintillation and Čerenkov components
of the light produced by high-energy particles in crystals. And of course, the larger
the difference in the time structure, the better the separation works. The RD52 col-
laboration managed to improve the applicability of PbWO4 crystals for dual-readout
calorimetry by doping them with small amounts, O(1%), of molybdenum [24]. This
had two beneficial effects: it increased the decay time of the scintillation light and it
shifted the spectrum of the emitted scintillation light to larger wavelengths.
The effects of that are illustrated in Fig. 11.13, which shows the calorimeter signals
generated by 50 GeV electrons traversing a crystal of this type. This crystal was
oriented such as to maximize the relative fraction of Čerenkov light in the detected
signals. By selecting the UV light by means of an optical filter, almost the entire
detected signal was due to (prompt) Čerenkov light, while a yellow transmission
filter predominantly selected scintillation light, which had a decay time of ∼26 ns as
a result of the Mo-doping. Whereas the differences in angular dependence were very
suitable to demonstrate that some of the light generated in these crystals is actually
the result of the Čerenkov mechanism, the combination of time structure and spectral
characteristics provides powerful tools to separate the two types of light in real time.
One does not even have to equip the calorimeter with two different light detectors
for that. This was demonstrated with a calorimeter consisting of bismuth germanate
(Bi4 Ge3 O12 , or BGO) crystals [25].
11.4 Tricks to Obtain Useful Details of the Shower Development 239
Even though Čerenkov radiation represents a very tiny fraction of the light pro-
duced by these crystals, it is relatively easy to separate and extract it from the sig-
nals. The much longer scintillation decay time (300 ns) and the spectral difference
are responsible for that.2 Figure 11.14 shows the time structures of signals from a
BGO calorimeter recorded with a UV filter. The “prompt” component observed in
the ultraviolet signal is due to Čerenkov light. A small fraction of the scintillation
light also passes through the UV filter. This offers the possibility to obtain all needed
information from only one signal. An external trigger opens two gates: one narrow
(10 ns) gate covers the prompt component, the second gate (delayed by 30 ns and
50 ns wide) only contains scintillation light. The latter signal can also be used to
determine the contribution of scintillation to the light collected in the narrow gate.
In this way, the Čerenkov/scintillation ratio can be measured event-by-event on the
basis of one signal only [25].
Figure 11.15 illustrates how this Čerenkov light yield can be measured in practice
[26]. It concerns measurements on a PbWO4 crystal doped with 0.3% of molybde-
num. This crystal was placed at an angle θ = 30◦ with the beam line. One PMT (R)
was equipped with a UV filter, in order to select the Čerenkov light, for which the
detection efficiency is largest at this angle. At the other side of the crystal only scin-
tillation light was detected. EGS4 calculations indicated that the beam particles (50
GeV electrons) deposited on average 0.578 GeV in this crystal, which was slightly
thicker than 2X0 in this geometry. This made it possible to calibrate the scintillation
signals, the distribution of which is shown in Fig. 11.15a. This distribution was sub-
divided into 20 bins. For each bin, the signal distribution on the opposite side of the
crystal, i.e., the Čerenkov side, was measured. The fractional width of this distribu-
tion, σrms /Cmean , is plotted in Fig. 11.15b versus the average scintillator signal in this
bin, or rather versus the inverse square root of this signal (S −1/2 ). It turned out that
2 The BGO scintillation spectrum peaks at 480 nm, while Čerenkov light exhibits a λ−2 spectrum.
240 11 Calorimeter Performance
Fig. 11.14 The time structure of a typical shower signal measured in the BGO em calorimeter
equipped with a UV filter. These signals were measured with a sampling oscilloscope, which took
a sample every 0.8 ns. The UV BGO signals were used to measure the relative contributions of
scintillation light (gate 2) and Čerenkov light (gate 1). From: Akchurin, N. et al. (2009). Nucl. Instr.
and Meth. A610, 488
Fig. 11.15 The scintillation signal distribution for 50 GeV electrons traversing a PbWO4 crystal at
θ = 30◦ (a) and the fractional width of the Čerenkov signal distribution as a function of the amount
of energy deposited in the crystal, as derived from the scintillation signal (b). The crystal was doped
with 0.3% Mo. See the text for more details. From: Akchurin, N. et al. (2010). Nucl. Instr. and
Meth. A621, 212
11.4 Tricks to Obtain Useful Details of the Shower Development 241
this fractional width scaled perfectly with this variable, i.e., with E −1/2 . Since the
relationship between the energy E and the scintillation signal S is given by the calibra-
tion described above, it was also possible to indicate the energy scale in Fig. 11.15b.
This is done on the top horizontal axis. The observed scaling of σrms /Cmean with
E −1/2 means that the energy resolution is completely determined by stochastic pro-
cesses that obey Poisson statistics. In this case, fluctuations in the Čerenkov light
yield were the only stochastic processes that played a role, and therefore the average
light yield could be directly determined from this result: 55 photoelectrons per GeV
deposited energy. For an energy deposit of 1 GeV, this led to a fractional width of
13.5%, and√therefore the contribution of Čerenkov photoelectron statistics amounts
to 13.5%/ E. This is not much better than what could be achieved in a dedicated
fiber sampling calorimeter.
11.4.2 A Caveat
The directionality of Čerenkov light is also the reason for an important caveat. As
long as the shower particles are isotropically distributed with respect to the direc-
tion of incidence of the showering particle, this has no consequences. However,
if (some fraction of) the charged particles produced in the absorption process are
non-isotropically distributed, the calorimeter signal may depend on this angle of inci-
dence. An extreme consequence of this phenomenon was shown in Fig. 11.12. Muons
traversing the dual-readout DREAM fiber calorimeter only produced Čerenkov sig-
nals resulting from the radiative shower losses, since light emitted by the muon itself
was not trapped within the numerical aperture of the fibers.
This phenomenon may also affect the response of a Čerenkov calorimeter to elec-
tromagnetic showers. Figure 11.16a shows the angular distribution of the shower
particles through which the energy of a high energy electron is deposited in a (lead)
absorber [27]. This distribution contains a sizeable component of more or less isotrop-
ically distributed relativistic electrons, i.e., electrons capable of emitting Čerenkov
light. These electrons are predominantly produced in Compton scattering. They have
“forgotten” the direction of the high-energy particle that initiated the shower and are
more or less randomly oriented with respect to that direction. This component leads
to a Čerenkov signal in fibers oriented at 0◦ , or any other angle with respect to the
flight path of the showering particles. When the fibers are oriented at the Čerenkov
angle (θC = arccos (n−1 ) ≈ 46◦ ) with respect to that flight path, the signals from this
type of calorimeter contain, in addition, contributions from most of the electrons and
positrons produced in the early phase of the shower development, which is dominated
by γ → e+ e− processes.
Figure 11.16b shows the em response as a function of the angle of incidence of the
showering particles for a quartz-fiber calorimeter [28]. This response does show an
angular dependence. As expected, the highest response is indeed obtained when the
angle of incidence corresponds to the Čerenkov angle. At that angle, the fibers are
sensitive to particles that travel in the same direction as the incoming particle. If all
242 11 Calorimeter Performance
Fig. 11.16 Angular distribution of the shower particles (electrons and positrons) through which the
energy of a 1 GeV electron is absorbed in a lead-based calorimeter. Results of EGS4 Monte Carlo
simulations. The angular regions contributing to the signals from calorimeters with quartz fibers
oriented at 0◦ , 45◦ and 90◦ are shaded and indicated by arrows (a). From: Acosta, D. et al. (1990).
Nucl. Instr. and Meth. A294, 193. The electromagnetic calorimeter response of a fiber calorimeters
based on the detection of Čerenkov light, as a function of the angle of incidence of the particles,
measured with respect to the fiber axis (b). From: Ganel, O. and Wigmans, R. (1995). Nucl. Instr.
and Meth. A365, 104
shower particles that emit Čerenkov light traveled in that direction, then the response
of this calorimeter would be zero except for angles of incidence of 46◦ ± Δθ , where
Δθ is determined by the numerical aperture of the fibers and typically has values in
the 10◦ –20◦ range [29].
However, a large fraction of the shower particles through which high-energy
electrons and photons deposit their energy in an absorbing structure do not travel in
that direction (see Fig. 11.16a). The Čerenkov light emitted by such electrons thus
travels at different angles with the fiber axes as well, and the angular dependence of
the calorimeter response provides information about the angular distribution of the
shower particles with energies above the Čerenkov threshold.
For example, the signal measured at 0◦ , the orientation needed for calorimeters in
colliding-beam experiments, is caused by shower particles traveling at angles of about
45◦ with the fiber axes, in the forward direction. The response at this angle is smaller
than the maximum one, by about a factor of two. Even at an angle of 90◦ , where
the signal from this calorimeter (partially) depends on particles traveling at an angle
of about 135◦ with the fiber axis, i.e., at 45◦ in the backward direction, there is still
a significant response. The different angular regions that contribute to the signals
from a Čerenkov fiber calorimeter, depending on its orientation, are indicated in
Fig. 11.16a. The experimental response curve (Fig. 11.16b) is in excellent agreement
with the one derived from Monte Carlo simulations, which follows from the angular
distribution of the shower particles [28].
The message from all this is that the calibration of the signals from em showers
detected in a Čerenkov fiber calorimeter is, strictly speaking, only valid for particles
11.4 Tricks to Obtain Useful Details of the Shower Development 243
that enter the calorimeter at angles smaller than ∼20◦ with the fiber direction. For
particles entering at larger angles, a correction factor will be needed. However, a word
of caution is necessary concerning results such as those shown in Fig. 11.16b. The
angular dependence of the em response may in reality be less strong than indicated
in this figure. That is because measurements performed at angles larger than 20◦
typically concern only the first part of the developing shower. It has been shown
that the relative contributions of Compton and photoelectrons to the signals are
much larger beyond the shower maximum than in the early part [30]. Therefore, the
asymmetry between the response at 0◦ and at 90◦ is probably considerably smaller
than it appears to be in Fig. 11.16b. Nevertheless, the arguments listed above for the
angular dependence of the Čerenkov response to em showers are perfectly valid.
References
12.1 Introduction
A completely different method that has been proposed to meet the challenging
requirements of recognizing, reconstructing and separating hadronically decaying
W and Z bosons in future particle physics experiments is the so-called Particle Flow
Analysis (PFA). This method is based on the combined use of a precision tracker
and a highly-granular calorimeter. The idea is that the charged jet particles can be
precisely measured with the tracker, while the energy of the neutral particles is mea-
sured with the calorimeter. Such methods have indeed successfully been used to
improve the mass resolution of hadronically decaying Z 0 s at LEP [1], and the jet
energy resolution using γ -jet pT balancing events at CDF [2] and at CMS [3]. Two
detector concepts studied in the context of the experimental program for the proposed
International Linear Collider (ILC) are based on this method as well [4].
The problem that limits the success of this method is of course that the calorimeter
does not know or care whether the particles it absorbs are electrically charged. There-
fore, one will have to correct the detected calorimeter signals for the contributions of
the charged jet particles. Proponents of this method have advocated a fine granularity
as the key to the solution of this “double-counting” problem [5]. However, it has been
argued by others that this, for practical geometries, is an illusion [6]. Especially in
jets with leading charged particles, the overlap between the showers from individual
jet particles makes the fine granularity largely irrelevant.
In order to increase the spatial separation between showers induced by the various
jet particles, and thus alleviate the double-counting problem, the concept detectors for
the ILC that are based on the PFA principle count on strong solenoidal magnetic fields
(4–5 T). Such fields may indeed improve the validity of PFA algorithms, especially at
large distances from the vertex, since they open up a collimated beam of particles. It is
important to be quantitative in these matters. After having traveled a typical distance
of one meter in a 4 T magnetic field, the trajectory of a 10 GeV pion deviates by 6 cm
from that of a straight line, i.e., less than one third of a nuclear interaction length
(the characteristic length scale for lateral hadronic shower development) in typical
calorimeters. The field is not always beneficial, since it may also have the effect of
bending jet particles with a relatively large transverse momentum with respect to the
jet axis into the jet core.
Of course, in the absence of reliable Monte Carlo simulations1 the only way to
prove or disprove the advocated merits of the proposed PFA methods is by means of
dedicated experiments in realistic prototype studies.
The first statement in any talk about PFA mentions that 2/3 of the final-state particles
constituting a jet are electrically charged, and that the momenta of these particles
can be measured extremely precisely. This is of course true, but the implication that
the calorimeters of PFA based detector systems don’t have to be very good, since
they only have to measure one third of the jet energy, is incorrect. In the absence of
calorimeter information, based on tracker information alone, the jet energy resolution
would be determined by the fluctuations in the fraction of the total jet energy that
is carried by the charged fragments. This issue was studied by Lobban et al. [6],
who found that these event-to-event fluctuations are very large. Depending on the jet
fragmentation algorithm, the σrms of the energy fraction carried by charged particles
was found to be 25–30% of the average value, independent of the jet energy.
One may wonder why these fluctuations do not become smaller at higher energies,
given the fact that the number of jet fragments increases. The reason for this is that
the observed increase in multiplicity is uniquely caused by the addition of more soft
particles. The bulk of the jet energy is invariably carried by a small number of the most
energetic particles. This means that the fraction of the jet energy carried by charged
particles is strongly dependent on the extent to which these particles participate in
the “leading” component of the jet. Therefore, the event-to-event fluctuations in this
fraction are large and do not become significantly smaller as the jet energy increases.
As an aside, we mention that the same argument thus necessarily also applies
for the event-to-event fluctuations in the fraction of the neutral particles (mainly
π 0 s). These fluctuations are responsible for the poor jet energy resolution of non-
compensating calorimeters, especially at high energy, since the response of such
calorimeters is usually considerably larger for em showers than for non-em ones.
In the absence of a calorimeter, one should therefore not expect to be able to
measure jet energy resolutions better than 25–30% on the basis of tracker information
alone, at any energy. And since the contributions of showering charged particles to
the calorimeter signals have to be discounted properly for the PFA method to work,
the quality of the calorimeter information is in practice very important, if one aims to
achieve the performance needed to separate hadronically decaying W and Z bosons.
1 Concern about the absence of reliable simulations for hadronic shower development was the main
reason for a special workshop held at Fermilab in 2006 [7]. To our knowledge, the fundamental
problems addressed at this workshop, e.g., with regard to the hadronic shower widths that are crucial
for PFA, still exist.
12.3 PFA at LEP, the Tevatron and the LHC 247
One of the conclusions of the analysis described in [6] was that the PFA approach
may result in improving the jet energy resolution of “poor” calorimeter systems to
become “mediocre,” but that it will do little for the performance of calorimeters with
“mediocre” or “good” resolution. This can be understood by considering the extreme
cases: for a perfect calorimeter, there is nothing left for a tracker to improve upon,
while for no calorimeter at all, the tracker would still give 30% resolution for the
jets.
Practical experience so far seems to confirm this assessment. The first experiment
in which PFA was elaborately applied was ALEPH [1], one of the LEP experiments.
The hadron calorimeter was not considered a very important component of the LEP
detectors, which had, on the other hand, excellent tracking systems. Using a specific
(biased) subsample of hadronically decaying Z 0 s at rest,2 the authors exploited the
properties of this tracking system to the fullest extent and achieved an energy resolu-
tion of 6.2 GeV, an improvement of about 25% with respect to the reported hadronic
energy resolution of the stand-alone calorimeter system.
The CDF experiment at the Tevatron also used PFA techniques to improve their
jet energy resolution. Figure 12.1 shows the effects of including information from
the tracking system and the shower max detectors on the measured jet energy, for jets
produced at central rapidities, i.e., fragments entering the calorimeter in the barrel
region [2].
The CMS experiment took advantage of their all-silicon tracking system, plus a
fine-grained ECAL, to improve their jet energy resolution. Figure 12.2a shows the
2 Any event in which energy was deposited within 12◦ of the beam line, as well as any event in
which more than 10% of the total energy was deposited within 30◦ of the beam line, was removed
from the event sample used for this analysis.
248 12 Particle Flow Analysis
Fig. 12.2 Simulated jet energy resolution in CMS as a function of transverse momentum, effect of
PFA techniques (a). Measured jet energy resolution using PFA techniques as a function of energy
(b). From: CMS Collaboration (2009). Note CMS-PAS-PFT-09-001
expected improvement of the jet energy resolution if PFA techniques would be used,
which decreases from ∼30% at 50 GeV to ∼20% at 100 GeV and ∼10% at 500 GeV
in the energy range for which these predictions could be tested. The experimental
data (Fig. 12.2b) show a somewhat smaller improvement at the lowest and highest
energies measured for this purpose. However, the improvement that resulted from
the use of PFA was also here certainly significant.
The ATLAS calorimeter system measures jets with much greater precision than
CMS and, therefore, the replacement of the calorimeter information using the tracker
data did not lead to significant benefits. However, the tracker data did help mitigating
pileup effects. By subtracting the energy (momenta) of tracks that did not point to
the jet vertex from the measured calorimeter energy, a better measurement of the jet
energy was obtained [8].
Encouraged by the observed improvements in the jet performance, CMS has
decided to replace its entire endcap calorimeter system with a dedicated PFA detector
[9]. This system, which is designed to comprise about six million electronic channels,
is scheduled to replace the current endcap calorimeters in the forward region around
2025. It is intended to mitigate the problems of radiation damage and event pile-
up, which are expected to have rendered the current system (consisting of PbWO4
crystals, backed up by a brass/plastic-scintillator hadronic section) ineffective by
then. The new system will consist of 5λint deep fine-grained calorimetry, 40 sampling
layers with 1–1.5 cm2 silicon pads as active material, backed up by another 5λint of
“conventional” calorimetry. Also the tracking system upstream of this calorimeter
will be replaced, with upgrades foreseen both in granularity and in η-coverage.
12.4 PFA Calorimeter R&D 249
A large collaboration, called CALICE, has set out to test the viability of the PFA
ideas. In the past 10–15 years, they have constructed a variety of calorimeters, both
for the detection of em showers as well as hadronic ones. The replacement of the
CMS endcap calorimeters, mentioned in the previous subsection, is based on and
inspired by the work of this collaboration [5].
The calorimeters constructed by CALICE have one thing in common: a very
high granularity. The calorimeter modules have a very large number of independent
electronic readout channels, O(104 ) in most modules, up to half a million in one
specific case. The active elements are either:
1. Silicon pads, typically with dimensions of 1 × 1 cm2 ,
2. Small scintillator strips, read out by SiPMs,
3. Resistive Plate Chambers with small readout pads, operating in the saturated
avalanche mode.
4. As an alternative, micromegas and GEMs are being tested.
These readout elements are interspersed between layers of absorber material. Typi-
cally, tungsten is used for the detection of em showers. Its Molière radius (9.3 mm)
is the smallest of all practical absorber materials (only platinum is better!), so that
the lateral development of the em showers is limited as much as possible. This is
important for separating showers from several particles that enter the calorimeter in
close proximity. For the deeper sections of the calorimeter, typically stainless steel is
being used. Figure 12.3 is often shown to illustrate the advantage of using tungsten.
The nuclear interaction length and especially the Molière radius, which determine the
extent of the lateral shower development for hadronic and electromagnetic showers,
respectively, are considerably smaller than for steel. Of course, what really matters
here is the effective value of these parameters in the calorimeter, which also includes
low-Z materials such as plastic, silicon and air. A second thing to keep in mind is
that the particles in reality are not colored. The difference between the colored and
bitmap versions of the tungsten image illustrates that the task to assign calorimeter
hits to individual jet fragments may in practice be quite daunting indeed, even in the
densest possible absorber structures.
The largest calorimeter that was specifically designed for em shower detection is
a tungsten/silicon device [10]. It has an active surface area of 18 × 18 cm2 and is
20 cm deep, subdivided longitudinally into 30 layers. The first 10 layers are 0.4X 0
thick, followed by 10 layers of 0.8X 0 and finally another 10 layers of 1.2X 0 , for a
total absorption thickness of 24X 0 . The active layers consist of a matrix of PIN diode
sensors on a silicon wafer substrate. The individual diodes have an active surface area
of 1 × 1 cm2 , and there are thus 18 × 18 = 324 calorimeter cells per layer, 9,720
in total. These are read out by means of a specially developed ASIC. Some results
of measurements of em showers with this detector are shown in Figs. 6.7 and 4.1.
CALICE also built and tested a large hadron calorimeter, a sandwich structure
based on 38 layers of 5 mm thick plastic scintillator, interleaved with absorber plates
250 12 Particle Flow Analysis
Fig. 12.3 Simulated shower development of jet fragments in a calorimeter based on iron (left) or
tungsten (center, right) as absorber material. From: Wigmans, R. (2018). J. Progr. Part. Nucl. Phys.
103, 109
[11]. For this instrument, they either used 17 mm thick steel or 10 mm thick tung-
sten plates. This absorber material thus represents a total thickness of about 4λint in
both cases. The active layers are housed in steel cassettes with 2 mm cover plates on
both sides. This increased the total depth of the instrumented volume to ∼5.3λint .
The transverse dimensions of the active layers are 90 × 90 cm2 . Figure 12.4 shows
a picture of one of the active layers. The layer is subdivided into tiles, small ones
in the central region and larger ones in the outer regions (and also in the rear of
the calorimeter module). The smallest tiles measure 3 × 3 cm2 . Each tile has a cir-
cular groove in which a wavelength shifting fiber is embedded. This fiber collects
the scintillation light produced in the tile, re-emits the absorbed light at a longer
wavelength and transports it to a SiPM, which converts it into an electric pulse. In
total, this calorimeter contains 7,608 tiles (i.e., electronic channels). This was the
first large-scale application of SiPMs in a particle detector.
Another CALICE module has a lateral cross section of ≈1 m2 and a similar
depth as the previous one. The effective depth can be varied through the choice of
the absorber material and the thickness of the absorber plates. In between each two
plates an array of RPC cells with dimensions of 1 × 1 cm2 is inserted, i.e., about
10,000 per plane [12]. In total, there are 54 independent longitudinal segments, so
that the total number of active elements is about half a million. These RPCs operate in
the saturated avalanche mode, and thus provide a “yes” or “no” signal when a particle
develops a shower in this device. This is thus a “digital” calorimeter. An event display
in this detector consists of a pattern of RPC cells that fired when the particle that
created it was absorbed. These patterns may be very detailed, as illustrated by the
example shown in Fig. 4.6 [13].
12.4 PFA Calorimeter R&D 251
Fig. 12.5 Non-linearity in the CALICE hadron calorimeters. Diagram a shows the average signal
for positrons in the CALICE analog hadron calorimeter, as a function of the beam energy. Shown
are the measured data points, before and after corrections for saturation in the SiPM readout were
applied, as well as the Monte Carlo prediction. The shaded area represents the systematic uncertainty
in the corrections. From: Adloff, C. et al. (2011). JINST 6, P04003. Diagram b shows the average
signals of the “digital” calorimeter for electrons and hadrons as a function of energy, in the energy
range of 1–10 GeV. For comparison, the dependence for a linear calorimeter is given as well.
Experimental data from [5, 20]
This is because, as a matter of principle, signal saturation implies that the fluctuations
that determine the energy resolution are partially suppressed [19].
The signal saturation phenomenon reaches very substantial proportions in the
“digital” calorimeter built by CALICE. The resulting non-linearity (Fig. 12.5b) is
even so large (already for particle energies that are much smaller than expected in
the experiments for which this device is intended), that it leads to apparent over-
compensation [5, 20]. Because of the large suppression of fluctuations in the shower
development process, the quoted energy resolutions are not very meaningful [19]. The
large signal-nonlinearity observed for small signals is important since high-energy
jets, such as the ones from the hadronic decay of intermediate vector bosons and the
Higgs boson, consist of a considerable number of low-energy final-state particles,
which together represent a significant fraction of the total jet energy. Quantitative
information on this point is given in Fig. 7.9.
The CALICE Collaboration has apparently also realized these problems and has
embarked on equipping the RPCs with a 2-bit readout system. This provides the
possibility to subdivide the signals into three categories, on the basis of different
threshold levels. This is called the “semi-digital” option [21]. However, the RPCs
still operate in avalanche mode, and the relationship between the different thresholds
(corresponding to bit settings 1/0, 0/1 and 1/1, respectively) and the deposited energy
is not a priori clear.
12.4 PFA Calorimeter R&D 253
Fig. 12.6 Event displays for a 70 GeV pion (left) and a 70 GeV electron (right) shower in the
semi-digital hadron calorimeter built by CALICE. The different colors indicate the amplitude of
the RPC signals (see text for details). From: Deng, Z. et al. (2016). JINST 11, P04001
Some results on beam tests performed with this so-called SDHCAL are reported
in [22]. The three thresholds were set at 0.11, 5 and 15 picoCoulombs in these tests.
Since the average signal produced by a mip in the RPCs was 1.2 pC, the second
and third thresholds corresponded to the simultaneous passage of at least 4 and 12
mips, respectively. Figure 12.6 shows event displays for a 70 GeV pion and a 70 GeV
electron in which the cells exceeding the different threshold levels are indicated with
different colors, red for level 3, blue for level 2, green for level 1. Not surprisingly,
the average multiplicity of tracks is clearly larger for the electron shower, and the
highest concentration of red cells in the pion event is found near the shower axis,
where most of the π 0 production takes place.
The additional information provided in this way was used to reconstruct the shower
energy. To that end the signals observed in the RPCs were given weight factors
intended to compensate for the saturation effects:
E reco = α N1 + β N2 + γ N3 (12.1)
in which the values Ni represent the number of hit cells with signals above the
thresholds i, and the weight factors α < β < γ . It turned out that the values of
the weight factors had to vary with energy in order to reconstruct the pion energy
correctly. This is illustrated in Fig. 12.7. Moreover, the weight factors had to be
given different values to reconstruct the energy of other types of particles (especially
electrons).
Figure 12.8 shows the effects of the additional information from the RPC signals
on the distribution of the reconstructed energy for event samples of pions at 20
GeV (Fig. 12.8a) and 70 GeV (Fig. 12.8b). The solid lines represent the distributions
254 12 Particle Flow Analysis
Fig. 12.7 Evolution of the coefficients α, β and γ as a function of the total number of RPC hits
recorded in the CALICE digital calorimeter. From: Deng, Z. et al. (2016). JINST 11, P04001
Fig. 12.8 Distribution of the reconstructed energy with the binary mode (red dashed line) and
with the three-threshold mode (solid black line), for pions of 20 GeV (a) and 70 GeV (b) in the
semi-digital hadron calorimeter built by CALICE. From: Deng, Z. et al. (2016). JINST 11, P04001
obtained with this information, while the dashed lines were obtained in the “binary
mode”, i.e., without information on the size of the RPC signals. The difference is
only significant at the highest energy.
Proponents of the PFA approach argue that all these shortcomings of their
calorimeters are not very important, because of the limited role played by the
calorimeter in the measurement of jet properties. The most crucial feature of the
calorimeters developed in this context is the very high granularity, intended to unravel
the structure of the jets. Events such as the ones shown in Figs. 4.6 and 12.6, which
were obtained with a cell size of 1 × 1 cm2 , are used to illustrate this point.
12.4 PFA Calorimeter R&D 255
Fig. 12.9 Event display for an event in which three particles enter a dual-readout calorimeter
equipped with SiPM readout simultaneously [23]. Shown are the signals from the scintillating
fibers, each of which was connected to a SiPM sensor. The Čerenkov fibers were fed through holes
in the white fields and read out by a second SiPM array located directly behind the one used for the
scintillation signals. The calorimeter area covered by this event display has transverse dimensions
of only 1.2 × 1.2 cm2 . From: Antonello M. et al. (2018). Nucl. Instr. and Meth. A899, 52
However, this is by no means a feature that is unique for calorimeters of this type.
Calorimeters that were developed to achieve excellent performance in stand-alone
mode are also capable of measuring the detailed structure of events in which several
particles enter the calorimeter in close proximity. Figure 12.9 shows an event dis-
play in which three particles enter a dual-readout calorimeter equipped with SiPM
readout simultaneously [24]. In an area with transverse dimensions of 1.2 × 1.2 cm2
(i.e., almost the same as the size of one RPC cell in the CALICE digital calorimeter),
the three peaks are clearly separated from each other. This feature, which is a conse-
quence of the extremely narrow shower profile (Fig. 5.6a) illustrates the benefits of
a lateral detector granularity that is much finer than one would choose based on the
value of the Molière radius (31 mm in this particular calorimeter).
These benefits include the possibility to distinguish between em showers caused
by (the two γ s from) a π 0 and by a single electron or γ . Figure 12.10 shows simulated
event displays of a 50 GeV electron shower and of a 100 GeV π 0 produced in a vertex
2 m upstream of the entrance window of such a calorimeter. The two γ s, which
develop showers that are separated by only a few mm, are clearly recognizable as
such.
Because of this feature and other unique advantages of the dual-readout approach
(Sect. 8.3), several experiments that are being planned for future e+ e− colliders are
considering a calorimeter system of this type [25].
256 12 Particle Flow Analysis
Fig. 12.10 Simulated event displays of a 50 GeV electron shower (a) and the showers produced
by the γ s from the decay of a 100 GeV π 0 (b) produced 2 m upstream of a calorimeter of the type
that was used to measure the event from Fig. 12.9. From: Pezzotti, L. (2018). Dual-readout fiber
sampling calorimetry with SiPM light sensors, Talk at the CALOR18 conference, Eugene (OR),
May 2018
References
Calorimeters have played an important role in the recent major discoveries that
have shaped and improved our understanding of the fundamental structure of matter
and the ways it behaves. The enrichment of the arsenal of experimental techniques
brought about by calorimetry is illustrated by the fact that different features of the
calorimetric performance made the crucial difference in these discoveries. As exam-
ples, we mention
• The discovery of the intermediate vector bosons [1, 2] was possible thanks to the
capability of the calorimeters to recognize and select events in which large- p⊥
particles (including neutrinos!) were produced on-line, amidst a background that
outnumbered these events by some seven orders of magnitude.
• The observation of ν-oscillation effects in Super-Kamiokande [3, 4] was possible
thanks to some special features that derived from the use of Čerenkov light as the
source of experimental information in this calorimeter, namely (a) the possibility
to distinguish between low-energy electrons and muons, and (b) the capability to
reconstruct the direction of the particles that generated the light cones.
• The experimental observation of the Higgs boson [5, 6] became possible thanks
to the capability of the calorimeters to detect neutral particles (γ s) and to measure
their four-vectors with very high precision.
Other important discoveries in the past half century also benefited in essential
ways from calorimetry. For example, the discovery of the top quark by D0 was made
possible by the capability of the calorimeter and its auxiliary systems to recognize
charged leptons that were part of jets [7]. And the unraveling of the details of direct
C P-violation in the K 0 system was made possible by the excellent mass resolution
for π 0 s in the calorimeters of the NA31 [8], KTeV [9] and KLOE [10] experiments.
In the 1950 and 1960s, a large number of new “elementary particles” were found
in bubble chamber experiments, as resonances in invariant-mass plots (see Eq. 11.4).
Searches for new particles were conducted by carefully measuring the four-vectors
of all particles produced in the interactions and combining the four-vector of one
particular particle (e.g., a K 0 ) with that of other particles (e.g., two pions of opposite
charge), according to the chosen decay mode. Such searches were often successful
(e.g., K 1 (1270) → K 0 π π ).
This technique is still being used successfully, for example in the spectroscopy of
particles containing a charmed and/or bottom quark, in which the LHCb experiment
currently excels [11]. However, as the energy at which we study the structure of
matter is further increased, exclusive hadronic decay modes become very compli-
cated (they involve many final-state particles) and each represent only a tiny fraction
of the decays. For example, the decay of some charmed and bottom mesons pro-
ceeds through more than one hundred different channels with measurable branching
fractions.
Increasingly, the primary process at the quark level is of great interest, rather than
the precise composition of the final state. For example, the hadronic decay of a W
boson proceeds through the process W → q q̄. In order to demonstrate that hadroni-
cally decaying W s were produced in certain reactions, it is much more important to
be able to reconstruct these two jets and determine their invariant mass than to know
the characteristics of all individual particles making up the final state.
In the 1960s, K 0 s and Λs could be identified by the fact that the π + π − hypothesis
produced the correct mass value for the first one, while the pπ − hypothesis repro-
duced exactly the 1115.68 MeV/c2 on the books for the Λ. The higher the energy,
the more precise the experimental measurements of the four-vectors had to be to
distinguish these two particles.
History repeats itself, but now the individual particle tracks have been replaced by
jets. Figure 13.1 shows the jet–jet invariant mass distribution measured by the UA2
Fig. 13.1 Two-jet invariant mass distributions from the UA2 experiment. Diagram a shows the
measured data points, together with the results of the best fits to the QCD background alone (dashed
curve), or including the sum of two Gaussian functions describing W, Z → q q̄ decays. Diagram
b shows the same data after subtracting the QCD background. The data are compatible with peaks
at m W = 80 GeV and m Z = 90 GeV. The measured width of the bump, or rather the standard
deviation of the mass distribution, was 8 GeV, of which 5 GeV could be attributed to non-ideal
calorimeter performance. From: Alitti, J. et al. (1991). Z. Phys. C49, 17
13.1 Calorimeters and Physics Discoveries 259
Collaboration [12]. The peak corresponding to the decay of the intermediate vector
bosons W and Z can clearly be distinguished from the QCD background. This is a
direct result of the measurement accuracy obtained with the calorimeter used in this
experiment. An even better resolution might make it possible to observe W and Z
decays as separate peaks in this mass spectrum. Calorimeters are ideal instruments
for this type of measurement.
Thirty years ago, these intermediate vector bosons were the objects of intense
experimental searches. Now, they may become the key to discovering new phenom-
ena at the 0.1–1.0 TeV mass scale, and multi-quark spectroscopy may turn out to be
an invaluable tool for this.
The quality of the hadronic calorimetry may well turn out to be very important for
the study of phenomena beyond the Standard Model, assuming that there is physics
beyond this Model that can be studied with the available tools. The discovery of
Supersymmetry hinges on detection of the escaping LSP (Lightest Supersymmetric
Particle), which manifests itself through missing (transverse) energy in the collisions.
Establishing the existence of other hypothesized new objects, such as leptoquarks,
requires the explicit detection of fragmenting quarks and precise measurements of
their properties.
(1 − z)α
D(z) = (α + 1) (13.1)
z
in which D(z) denotes the probability that a jet fragment carries a fraction z of the
energy of the fragmenting object. The parameter α can be chosen as desired. It has
been demonstrated that a function of this type gives a reasonable description of the
fragmentation processes measured at LEP and at the Tevatron, for parameter values
α = 3 and 6, respectively [14].
Jet fragments are generated with energies z E jet , with the values of z chosen from
a distribution representing Eq. 13.1. Each fragment is assigned a mass m, a charge
and a transverse momentum p⊥ . Ten percent of the particles are assumed to be
kaons and ninety percent pions. One third of the particles are electrically neutral, the
rest are charged. The transverse momentum is chosen from an exponentially falling
distribution with a mean value of 0.3 GeV/c. If the chosen parameters yield an
energy z E jet ,
unphysical result, e.g., if the chosen mass is larger than the fragment’s
or if the transverse momentum is larger than the total momentum (z E jet )2 − m 2 ,
the fragment is discarded and a new one is selected. The selection of jet fragments is
continued until the jet energy is exceeded. In that case, the energy of the last fragment
is reduced so that the total energy of all fragments combined equals the jet energy.
The number of particles constituting the jets is quite large and increases with
the jet energy. However, only relatively few particles contribute substantially to the
total energy. For example, in α = 3 jets the 10 most energetic particles carry 90%
of the total jet energy. For α = 6 jets, that takes 15 particles, on average. This is
true at all energies, which is of course a direct consequence of the very concept of a
fragmentation function that depends only on z.
The authors defined the cone parameter R that formed the basis of the applied jet
algorithm as
13.2 Lepton Versus Hadron Colliders 261
R= (Δφ)2 + (Δη)2 (13.2)
where Δφ and Δη denote the spread around the nominal direction of the fragmenting
object in the azimuthal and polar angles, respectively. The fate of a jet fragment was
decided on the basis of the ratio of its transverse and longitudinal momenta, p⊥ / p .
If
arctan( p⊥ / p ) > R/2,
then the fragment fell outside the cone, otherwise it was considered to contribute to
the measured jet characteristics (energy, momentum, composition).
Figure 13.2a shows the average fraction of the jet energy that was found to be
contained in a jet-defining cone, as a function of the jet energy, for two different cone
sizes, R = 0.3 and 0.5, respectively. The error bars on the data points indicate the
spread in the jet containment resulting from the difference in the parameter α used
in the fragmentation function.
These data show that for jets of about 20 GeV, on average some 30% of the energy
was carried by fragments that travelled outside the cone. However, as the jet energy
increases, the containment rapidly improves. For energies above 100 GeV, typically
less than 10% of the energy is unaccounted for when the chosen jet algorithms are
applied.
Fig. 13.2 The average fraction of the jet energy contained in the jet-defining cone as a function
of the jet energy. Results are given for cones with R = 0.3 and R = 0.5. The error bars indicate
the spread in the results caused by the choice of the value of the fragmentation function parameter
α (a). The hadronic energy resolution of two different calorimeter systems and the contribution of
a jet-defining cone with R = 0.3 to the jet energy resolution, as a function of energy (b). From:
Lobban, O., Sriharan, A. and Wigmans R. (2002). Nucl. Instr. and Meth. A495, 107
262 13 Outlook
Calorimeters are instruments for measuring energy. The history of physics in general,
and of nuclear and particle physics in particular, is filled with examples that prove
that measurement precision pays off. A better, more accurate instrument allows more
precise measurements. More precise measurement results make it possible to discover
new phenomena, and/or to better understand old ones. Better understanding of the
physical world has always been a crucial element in the evolution of mankind.
The history of calorimetry itself illustrates this process in a nutshell. Calorimeters
were originally invented as crude, cheap instruments for some specialized appli-
cations (for example, detection of neutrino interactions). The original literature is
testimony to the fact that their performance was often perceived as somewhat mys-
terious by their users. Only after the physics on which calorimeters are based was
understood in detail did it become possible to develop these detectors into the pre-
cision instruments that they are nowadays and which form the centerpiece of many
modern experiments in particle physics.
13.3 The Future of Calorimetry 263
References
The following table summarizes the notation for variables used in this book, and the
units in which these variables are usually expressed.
Table A.1 lists parameters relevant for shower development in materials that are
frequently used in sampling calorimeters: the critical energy (εc ), the radiation length
(X 0 ), the Molière radius (ρM ), the nuclear interaction length (λint ) and the specific
ionization (d E/d x) for mips. The critical energy was determined on the basis of
d E/d x data using Rossi’s definition (see Sect. 2.2.1). The nuclear interaction length
is given for protons, the values for pions are up to 50% larger.
The tables in this subsection concern materials that are used for particle detection
in physics experiments. Relevant properties of some scintillating crystals and plastic
scintillators used for this purpose are listed in Tables A.2 and A.3, respectively.
Table A.2 Properties of scintillating crystals that are used in particle physics experiments
NaI(Tl) CsI(Tl) CsI BaF2 CeF3 BGO PbWO4
Density (g cm−3 ) 3.67 4.51 4.51 4.89 6.16 7.13 8.30
Radiation length (cm) 2.59 1.85 1.85 2.06 1.68 1.12 0.89
Molière radius (cm) 4.8 3.5 3.5 3.4 2.6 2.3 2.0
Interaction length (cm) 41.4 37.0 37.0 29.9 26.2 21.8 18.0
d E/d xmip (MeV cm−1 ) 4.79 5.61 5.61 6.37 8.0 8.92 9.4
Refractive index (at λpeak ) 1.85 1.79 1.95 1.50 1.62 2.15 2.2
Hygroscopicity Yes Slight Slight No No No No
Emission spectrum, λpeak
Slow component (nm) 410 560 420 300 340 480 510
Fast component (nm) 310 220 300 510
Relative light yield
Slow component 100 45 5.6 21 6.6 9 0.3
Fast component 2.3 2.7 2.0 0.4
Decay time (ns)
Slow component 230 1300 35 630 30 300 50
Fast component 6 0.9 9 10
268 Appendix: Some Data Relevant to Calorimetry
Table A.3 Properties of some plastic scintillators and wavelength shifters applied in particle physics
experiments. Listed are the wavelengths at which the absorption (for WLS) and emission spectra
peak, the characteristic decay time and, when available, the absorption length (WLS)
Absorption (nm) λabs (µm) Emission (nm) Decay time (ns)
Scintillators
SCSF-38 428 2.3
(Kuraray)
SCSF-81 437 2.4
(Kuraray)
SCSF-3HF 530 7
(Kuraray)
BCF-10 (Bicron) 432 2.7
BCF-20 (Bicron) 492 2.7
BCF-60 (Bicron) 530 7
Wavelength shifters
Y-7 (Kuraray) 440 100 490 10
Y-11 (Kuraray) 435 80 476 10
BCF-91A 420 494 12
(Bicron)
BCF-92 (Bicron) 410 492 2.7
Appendix: Some Data Relevant to Calorimetry 269
Table A.4 lists relevant properties of noble liquids used in particle physics experi-
ments.