Professional Documents
Culture Documents
Computer Aided Verification 32nd International Conference CAV 2020 Los Angeles CA USA July 21 24 2020 Proceedings Part I Shuvendu K. Lahiri
Computer Aided Verification 32nd International Conference CAV 2020 Los Angeles CA USA July 21 24 2020 Proceedings Part I Shuvendu K. Lahiri
https://textbookfull.com/product/computer-aided-
verification-32nd-international-conference-cav-2020-los-angeles-
ca-usa-july-21-24-2020-proceedings-part-ii-shuvendu-k-lahiri/
https://textbookfull.com/product/runtime-verification-20th-
international-conference-rv-2020-los-angeles-ca-usa-
october-6-9-2020-proceedings-jyotirmoy-deshmukh/
https://textbookfull.com/product/automated-technology-for-
verification-and-analysis-16th-international-symposium-
atva-2018-los-angeles-ca-usa-october-7-10-2018-proceedings-
shuvendu-k-lahiri/
https://textbookfull.com/product/hci-
international-2020-posters-22nd-international-conference-
hcii-2020-copenhagen-denmark-july-19-24-2020-proceedings-part-i-
Computer Aided Verification 28th International
Conference CAV 2016 Toronto ON Canada July 17 23 2016
Proceedings Part II 1st Edition Swarat Chaudhuri
https://textbookfull.com/product/computer-aided-
verification-28th-international-conference-cav-2016-toronto-on-
canada-july-17-23-2016-proceedings-part-ii-1st-edition-swarat-
chaudhuri/
https://textbookfull.com/product/hci-
international-2020-posters-22nd-international-conference-
hcii-2020-copenhagen-denmark-july-19-24-2020-proceedings-part-ii-
Shuvendu K. Lahiri
Chao Wang (Eds.)
LNCS 12224
Computer Aided
Verification
32nd International Conference, CAV 2020
Los Angeles, CA, USA, July 21–24, 2020
Proceedings, Part I
Lecture Notes in Computer Science 12224
Founding Editors
Gerhard Goos
Karlsruhe Institute of Technology, Karlsruhe, Germany
Juris Hartmanis
Cornell University, Ithaca, NY, USA
Computer Aided
Verification
32nd International Conference, CAV 2020
Los Angeles, CA, USA, July 21–24, 2020
Proceedings, Part I
123
Editors
Shuvendu K. Lahiri Chao Wang
Microsoft Research Lab University of Southern California
Redmond, WA, USA Los Angeles, CA, USA
© The Editor(s) (if applicable) and The Author(s) 2020. This book is an open access publication.
Open Access This book is licensed under the terms of the Creative Commons Attribution 4.0 International
License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution
and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and
the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this book are included in the book’s Creative Commons license,
unless indicated otherwise in a credit line to the material. If material is not included in the book’s Creative
Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use,
you will need to obtain permission directly from the copyright holder.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, expressed or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
It was our privilege to serve as the program chairs for CAV 2020, the 32nd
International Conference on Computer-Aided Verification. CAV 2020 was held as a
virtual conference during July 21–24, 2020. The tutorial day was on July 20, 2020, and
the pre-conference workshops were held during July 19–20, 2020. Due to the
coronavirus disease (COVID-19) outbreak, all events took place online.
CAV is an annual conference dedicated to the advancement of the theory and
practice of computer-aided formal analysis methods for hardware and software sys-
tems. The primary focus of CAV is to extend the frontiers of verification techniques by
expanding to new domains such as security, quantum computing, and machine
learning. This puts CAV at the cutting edge of formal methods research, and this year’s
program is a reflection of this commitment.
CAV 2020 received a very high number of submissions (240). We accepted 18 tool
papers, 4 case studies, and 43 regular papers, which amounts to an acceptance rate of
roughly 27%. The accepted papers cover a wide spectrum of topics, from theoretical
results to applications of formal methods. These papers apply or extend formal methods
to a wide range of domains such as concurrency, machine learning, and industrially
deployed systems. The program featured invited talks by David Dill (Calibra) and
Pushmeet Kohli (Google DeepMind) as well as invited tutorials by Tevfik Bultan
(University of California, Santa Barbara) and Sriram Sankaranarayanan (University of
Colorado at Boulder). Furthermore, we continued the tradition of Logic Lounge, a
series of discussions on computer science topics targeting a general audience.
In addition to the main conference, CAV 2020 hosted the following workshops:
Numerical Software Verification (NSV), Verified Software: Theories, Tools, and
Experiments (VSTTE), Verification of Neural Networks (VNN), Democratizing Soft-
ware Verification, Synthesis (SYNT), Program Equivalence and Relational Reasoning
(PERR), Formal Methods for ML-Enabled Autonomous Systems (FoMLAS), Formal
Methods for Blockchains (FMBC), and Verification Mentoring Workshop (VMW).
Organizing a flagship conference like CAV requires a great deal of effort from the
community. The Program Committee (PC) for CAV 2020 consisted of 85 members – a
committee of this size ensures that each member has to review a reasonable number of
papers in the allotted time. In all, the committee members wrote over 960 reviews while
investing significant effort to maintain and ensure the high quality of the conference
program. We are grateful to the CAV 2020 PC for their outstanding efforts in evalu-
ating the submissions and making sure that each paper got a fair chance. Like last
year’s CAV, we made the artifact evaluation mandatory for tool paper submissions and
optional but encouraged for the rest of the accepted papers. The Artifact Evaluation
Committee consisted of 40 reviewers who put in significant effort to evaluate each
artifact. The goal of this process was to provide constructive feedback to tool devel-
opers and help make the research published in CAV more reproducible. The Artifact
vi Preface
Evaluation Committee was generally quite impressed by the quality of the artifacts,
and, in fact, all accepted tools passed the artifact evaluation. Among the accepted
regular papers, 67% of the authors submitted an artifact, and 76% of these artifacts
passed the evaluation. We are also very grateful to the Artifact Evaluation Committee
for their hard work and dedication in evaluating the submitted artifacts. The evaluation
and selection process involved thorough online PC discussions using the EasyChair
conference management system, resulting in more than 2,000 comments.
CAV 2020 would not have been possible without the tremendous help we received
from several individuals, and we would like to thank everyone who helped make CAV
2020 a success. First, we would like to thank Xinyu Wang and He Zhu for chairing the
Artifact Evaluation Committee and Jyotirmoy Deshmukh for local arrangements. We
also thank Zvonimir Rakamaric for chairing the workshop organization, Clark Barrett
for managing sponsorship, Thomas Wies for arranging student fellowships, and Yakir
Vizel for handling publicity. We also thank Roopsha Samanta for chairing the Men-
toring Committee. Last but not least, we would like to thank members of the CAV
Steering Committee (Kenneth McMillan, Aarti Gupta, Orna Grumberg, and Daniel
Kroening) for helping us with several important aspects of organizing CAV 2020.
We hope that you will find the proceedings of CAV 2020 scientifically interesting
and thought-provoking!
Program Chairs
Shuvendu K. Lahiri Microsoft Research, USA
Chao Wang University of Southern California, USA
Workshop Chair
Zvonimir Rakamaric University of Utah, USA
Sponsorship Chair
Clark Barrett Stanford University, USA
Publicity Chair
Yakir Vizel Technion - Israel Institute of Technology, Israel
Fellowship Chair
Thomas Wies New York University, USA
Program Committee
Aws Albarghouthi University of Wisconsin-Madison, USA
Jade Alglave University College London, UK
Christel Baier Technical University of Dresden, Germany
Gogul Balakrishnan Google, USA
Sorav Bansal India Institute of Technology, Delhi, India
Gilles Barthe Max Planck Institute, Germany
Josh Berdine Facebook, UK
Per Bjesse Synopsys, USA
Sam Blackshear Calibra, USA
Roderick Bloem Graz University of Technology, Austria
Borzoo Bonakdarpour Iowa State University, USA
Ahmed Bouajjani Paris Diderot University, France
Tevfik Bultan University of California, Santa Barbara, USA
Pavol Cerny Vienna University of Technology, Austria
viii Organization
Steering Committee
Kenneth McMillan Microsoft Research, USA
Aarti Gupta Princeton University, USA
Orna Grumberg Technion - Israel Institute of Technology, Israel
Daniel Kroening University of Oxford, UK
Additional Reviewers
AI Verification
NNV: The Neural Network Verification Tool for Deep Neural Networks
and Learning-Enabled Cyber-Physical Systems . . . . . . . . . . . . . . . . . . . . . . 3
Hoang-Dung Tran, Xiaodong Yang, Diego Manzanas Lopez,
Patrick Musau, Luan Viet Nguyen, Weiming Xiang, Stanley Bak,
and Taylor T. Johnson
Concurrency
Model Checking
Automata Tutor v3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Loris D’Antoni, Martin Helfrich, Jan Kretinsky, Emanuel Ramneantu,
and Maximilian Weininger
ADAMMC: A Model Checker for Petri Nets with Transits against Flow-LTL . . . 64
Bernd Finkbeiner, Manuel Gieseking, Jesko Hecking-Harbusch,
and Ernst-Rüdiger Olderog
Software Verification
Stochastic Systems
Global PAC Bounds for Learning Discrete Time Markov Chains . . . . . . . . . 304
Hugo Bazille, Blaise Genest, Cyrille Jegourel, and Jun Sun
STMC: Statistical Model Checker with Stratified and Antithetic Sampling . . . 448
Nima Roohi, Yu Wang, Matthew West, Geir E. Dullerud,
and Mahesh Viswanathan
Contents – Part II xvii
Synthesis
The material presented in this paper is based upon work supported by the Defense
Advanced Research Projects Agency (DARPA) through contract number FA8750-18-
C-0089, the National Science Foundation (NSF) under grant numbers SHF 1910017 and
FMitF 1918450, and the Air Force Office of Scientific Research (AFOSR) through award
numbers FA9550-18-1-0122 and FA9550-19-1-0288. The U.S. Government is authorized
to reproduce and distribute reprints for Government purposes notwithstanding any
copyright notation thereon. Any opinions, finding, and conclusions or recommendations
expressed in this material are those of the author(s) and do not necessarily reflect the
views of AFOSR, DARPA, or NSF.
c The Author(s) 2020
S. K. Lahiri and C. Wang (Eds.): CAV 2020, LNCS 12224, pp. 3–17, 2020.
https://doi.org/10.1007/978-3-030-53288-8_1
4 H.-D. Tran et al.
1 Introduction
Deep neural networks (DNNs) have quickly become one of the most widely used
tools for dealing with complex and challenging problems in numerous domains,
such as image classification [10,16,25], function approximation, and natural lan-
guage translation [11,18]. Recently, DNNs have been used in safety-critical cyber-
physical systems (CPS), such as autonomous vehicles [8,9,52] and air traffic col-
lision avoidance systems [21]. Although utilizing DNNs in safety-critical applica-
tions can demonstrate considerable performance benefits, assuring the safety and
robustness of these systems is challenging because DNNs possess complex non-
linear characteristics. Moreover, it has been demonstrated that their behavior
can be unpredictable due to slight perturbations in their inputs (i.e., adversarial
perturbations) [36].
Table 1. Overview of major features available in NNV. Links refer to relevant files/-
classes in the NNV codebase. BN refers to batch normalization layers, FC to fully-
connected layers, AvgPool to average pooling layers, Conv to convolutional layers, and
MaxPool to max pooling layers.
for plant configuration. NNV makes use of YALMIP [27] for some optimization
problems and MatConvNet [46] for some CNN operations.
The NNV toolbox contains two main modules: a computation engine and an
analyzer, shown in Fig. 1. The computation engine module consists of four sub-
components: 1) the FFNN constructor, 2) the NNCS constructor, 3) the reach-
ability solvers, and 4) the evaluator. The FFNN constructor takes a network
configuration file as an input and generates a FFNN object. The NNCS con-
structor takes the FFNN object and the plant configuration, which describes
the dynamics of a system, as inputs and then creates an NNCS object. Depend-
ing on the application, either the FFNN (or NNCS) object will be fed into a
reachability solver to compute the reachable set of the FFNN (or NNCS) from
a given initial set of states. Then, the obtained reachable set will be passed to
the analyzer module. The analyzer module consists of three subcomponents: 1)
a visualizer, 2) a safety checker, and 3) a falsifier. The visualizer can be called to
plot the obtained reachable set. Given a safety specification, the safety checker
can reason about the safety of the FFNN or NNCS with respect to the specifica-
tion. When an exact (sound and complete) reachability solver is used, such as the
star-based solver, the safety checker can return either “safe,” or “unsafe” along
with a set of counterexamples. When an over-approximate (sound) reachability
solver is used, such as the zonotope-based scheme or the approximate star-based
solvers, the safety checker can return either “safe” or “uncertain” (unknown).
In this case, the falsifier automatically calls the evaluator to generate simulation
traces to find a counterexample. If the falsifier can find a counterexample, then
NNV returns unsafe. Otherwise, it returns unknown. Table 1 shows a summary
of the major features of NNV.
NNV recently introduced a new set representation called the ImageStar for use
in the verification of deep convolutional neural networks (CNNs). Briefly, the
ImageStar is a generalization of the star set where the anchor and generator
vectors are replaced by multi-channel images. The ImageStar is efficient in the
analysis of convolutional layers, average pooling layers, and fully connected lay-
ers, whereas max pooling layers and ReLU layers consume most of the com-
putation time. NNV implements exact and over-approximate reachability algo-
rithms using the ImageStar for serial CNNs. In short, using the ImageStar, we
can analyze the robustness under adversarial attacks of the real-world VGG16
and VGG19 deep perception networks [31] that consist of >100 million param-
eters [37].
4 Evaluation
neural networks [21] for all 45 networks.2 All the experiments were done using
4 cores for computation. The results are summarized in Table 2 where (SAT)
denotes the networks are safe, (UNSAT) is unsafe, and (UNK) is unknown.
We note that (UNK) may occur due to the conservativeness of the reachability
analysis scheme. Detailed verification results are presented in the appendix of
the extended version of this paper [44]. For a fast comparison with other tools,
we also tested a subset of the inputs for Property 1–4 on all the 45 networks. We
note that the polyhedron method [40] achieves a timeout on most of networks,
and therefore, we neglect this method in the comparison.
Verification Time. For property φ3 , NNV’s exact-star method is about 20.7×
faster than Reluplex, 14.2× faster than Marabou, 81.6× faster than Marabou-
DnC (i.e., divide and conquer method). The approximate star method is 547×
faster than Reluplex, 374× faster than Marabou, 2151× faster than Marabou-
DnC, and 8× faster than ReluVal. For property φ4 , NNV’s exact-star method
is 25.3× faster than Reluplex, 18.0× faster than Marabou, 53.4× faster than
Marabou-DnC, while the approximate star method is 625× faster than Reluplex,
445× faster than Marabou, 1321× faster than Marabou-DnC.
2
We omit properties φ1 and φ2 for space and due to their long runtimes, but they
can be reproduced in the artifact.
10 H.-D. Tran et al.
Table 3. Verification results for ACC system with different plant models, where V T
is the verification time (in seconds).
Verification Results. For linear dynamics, NNV can compute both the exact
and over-approximate reachable sets of the ACC system in bounded time steps,
while for nonlinear dynamics, NNV constructs an over-approximation of the
reachable sets. The verification results for linear and nonlinear models using the
over-approximate star method are presented in Table 3, which shows that safety
of the ACC system depends on the initial velocity of the lead vehicle. When
the initial velocity of the lead vehicle is smaller than 27 (m/s), the ACC system
with the discrete plant model is unsafe. Using the exact star method, NNV can
construct a complete set of counter-example inputs. When the over-approximate
star method is used, if there is a potential safety violation, NNV simulates the
system with 1000 random inputs from the input set to find counter examples. If
a counterexample is found, the system is UNSAFE, otherwise, NNV returns a
safety result of UNKNOWN. Figure 3 visualizes the reachable sets of the relative
distance Drel between two vehicles versus the required safe distance Dsaf e over
time for two cases of initial velocities of the lead vehicle: vlead (0) ∈ [29, 30] and
vlead (0) ∈ [24, 25]. We can see that in the first case, Dref ≥ Dsaf e for all 50
time steps stating that the system is safe. In the second case, Dref < Dsaf e in
some control steps, so the system is unsafe. NNV supports a reachLive method
to perform analysis and reachable set visualization on-the-fly to help the user
observe the behavior of the system during verification.
The verification results for the ACC system with the nonlinear model are
all UNSAFE, which is surprising. Since the neural network controller of the
ACC system was trained with the linear model, it works quite well for the linear
model. However, when a small friction term is added to the linear model to form a
nonlinear model, the neural network controller’s performance, in terms of safety,
is significantly reduced. This problem raises an important issue in training neural
network controllers using simulation data, and these schemes may not work in
real systems since there is always a mismatch between the plant model in the
simulation engine and the real system.
Verification Times. As shown in Table 3, the approximate analysis of the ACC
system with discrete linear plant model is fast and can be done in 84 s. NNV
12 H.-D. Tran et al.
60
Distance
40
20
0 5 10 15 20 25 30 35 40 45 50
Control Time Steps
Actual Distance (blue) vs. Safe Distance (red)
60
Distance
50
40
30
0 5 10 15 20 25 30 35 40 45 50
Control Time Steps
Fig. 3. Two scenarios of the ACC system. In the first (top) scenario (vlead (0) ∈
[29, 30] m/s), safety is guaranteed, Drel ≥ Dsaf e . In the second scenario (bottom)
(vlead (0) ∈ [24, 25] m/s), safety is violated since Dref < Dsaf e in some control steps.
5 Related Work
NNV was inspired by recent work in the emerging fields of neural network and
machine learning verification. For the “open-loop” verification problem (verifica-
tion of DNNs), many efficient techniques have been proposed, such as SMT-based
methods [22,23,30], mixed-integer linear programming methods [14,24,28], set-
based methods [4,17,32,33,48,50,53,57], and optimization methods [51,58]. For
the “closed-loop” verification problem (NCCS verification), we note that the
Verisig approach [20] is efficient for NNCS with nonlinear plants and with Sig-
moid and Tanh activation functions. Additionally, the recent regressive polyno-
mial rule inference approach [34] is efficient for safety verification of NNCS with
nonlinear plant models and ReLU activation functions. The satisfiability mod-
ulo convex (SMC) approach [35] is also promising for NNCS with discrete linear
Another random document with
no related content on Scribd:
Or look at the painting of another vault on the opposite page. This is
more stiff than the former, because it was executed nearly a century
later; still, there is nothing to declare its Christian character until the
eye rests on the Good Shepherd, who appears below the principal
part of the decoration.
We are not saying that the artists who executed these paintings
had no Christian meaning in them; on the contrary, we believe that
they had, and that the paintings really suggested that meaning to
those who first saw them. For we know, on the authority of Tertullian,
that “the whole revolving order of the seasons” (which are
represented in the second painting) was considered by Christians to
be “a witness of the resurrection of the dead.” This, therefore, was
probably the reason why they were painted here; and no Christian
needs to be reminded that our Lord spoke of Himself under the
image of a vine, which sufficiently explains the first painting. Still the
fact remains that the representations themselves are such as might
have been used by Christian and by Pagan artists indifferently. If any
of our readers feel disappointed that the first essays of the Christian
painter should not have had a more distinctly Christian character,
they must remember that a new art cannot be created in a moment.
If the Christian religion in its infancy was to make use of art at all, it
had no choice but to appropriate to its own purposes the forms of
ancient art, so far as they were pure and innocent; by degrees it
would proceed to eliminate what was unmeaning, and substitute
something Christian.
Some writers have supposed that Christians used at first Pagan
subjects as well as Pagan forms of ornamentation; and they point to
the figure of Orpheus, which appears in three or four places of the
Catacombs, and to that of Psyche also, which may be seen about as
often. So insignificant a number of exceptions, however, would
scarcely suffice to establish the general proposition, even if they
were in themselves inexplicable. But, in truth, the figure of Orpheus
has no right to be considered an exception at all, for he was taken by
some of the early Fathers as a type of our Lord; and it was even
believed by some of them, that, like the sybil, he had prophesied
about Him. Clement of Alexandria calls our Lord the Divine
enchanter of souls, with evident reference to the tale of Orpheus;
and the same idea will have occurred to every classical scholar, as
often as he has heard those words of the Psalmist which speak of
the wicked as “refusing to hear the voice of the charmer, charm he
never so wisely.” When, then, we find Orpheus and his lyre, and the
beasts enchanted by his song, figured on the walls or roofs of the
Catacombs, we have a right to conclude that the artist intended a
Christian interpretation to be given to his work; and a similar
explanation may be given of any other subjects of heathen
mythology which have gained admittance there.
If we were asked to name the subject which seems to have been
used most frequently in the early decorations of the Catacombs, we
should give the palm to the Good Shepherd; nor is this preference to
be wondered at. Any one who has meditated upon the words in
which our Blessed Lord took this title to Himself, will easily
understand why the first Christians, living in the midst of heathen
persecutors, should have delighted to keep so touching an image
always before them. They scratched it, therefore, roughly on the
tombstone as they laid some dear one in the grave; they carved it on
their cups, especially on the sacred chalice; they engraved it on
signet rings and wore it on their fingers; they placed it in the centre of
the paintings with which they covered the ceiling of their
subterranean chapels, or they gave it the chief place immediately
over the altar. We meet with it everywhere, and everybody can
recognise it.
There are, however, one or two peculiarities in its mode of
treatment which require a word of explanation. The shepherd is
generally represented as a young man lightly clad, with his tunic girt
high about his loins, denoting thereby his unwearied activity; he is
surrounded by sheep, or he carries one on his shoulders, bearing it
home to the fold,—the most tender act of his office. And there is
nothing in this but what we might naturally have expected. But he is
also sometimes represented with a goat instead of a sheep upon his
shoulders; and, in later paintings, he has the pastoral reed or tuneful
pipe either hanging on the tree by his side or he is playing on it. Now
this last particular has no place in the gospel parable, and the former
seems directly opposed to it, since the goat is the accepted symbol
of the wicked, the sheep only of the good. Hence these points have
been taken up by some critics, either as tokens of thoughtless
carelessness on the part of the Christian artists, or as proofs that
their work, whether consciously or unconsciously, was merely copied
from some Pagan original. Neither of these remarks appears to be
just. The images of a shepherd in Pagan art, with scarcely a single
exception, are of a very different kind; and the particular details
objected to are not only capable of receiving a Christian
interpretation, they even express consoling Christian truths. St.
Gregory Nazianzen speaks of the anxious care of the shepherd as
he sits on the hillside, filling the air with the soft notes of his pipe,
calling together his scattered flock; and he observes that in like
manner the spiritual pastor, desirous to recall souls to God, should
follow the example of his Divine Master, and use his pipe more
frequently than his staff. Then, as to the substitution of the goat for
the sheep, it was probably intended as a distinct protest against the
un-Christian severity of those heretics, who in very early times
refused reconciliation to certain classes of penitent sinners.
Not many, however, of the most ancient Christian paintings are of
the same simple and obvious character as the Good Shepherd. The
leading feature which characterises most of them is this, that they
suggest religious ideas or doctrines under the guise of artistic
symbols or historic types. One doctrine specially prominent in them,
and most appropriately taught in cemeteries, is that of the
resurrection and the everlasting life of happiness which awaits the
souls of the just after death. It is in this sense that we must
understand not only the frequent repetitions of the stories of Jonas
and of Lazarus—the type and the example of a resurrection—but
also of Daniel in the lions’ den, and the three children in the fiery
furnace. These last, indeed, very probably had reference also to the
persecution which the Christians were then suffering, and were
intended to inspire courage and a confident expectation that God
would deliver them, even as He had delivered His chosen servants
of old; but, as they are spoken of in very ancient Christian
documents (e.g., in the hymns of St. Ephrem and in the Apostolic
Constitutions) as foreshadowing the future triumph of the body over
death, whence these too had been in a manner delivered, we prefer,
in obedience to these ancient guides, to assign this interpretation to
them; at any rate, it is certain that this interpretation cannot be
excluded. Figures also of the deceased, with arms outstretched in
prayer, sometimes accompanied by their names, or standing in the
midst of a garden, or, again, figures of birds pecking at fruits and
flowers, we understand as images of the soul still living after death,
received into the garden of Paradise, and fed by immortal fruits.
Sometimes there may be a difference of opinion perhaps as to the
correctness of this or that interpretation suggested for any particular
symbolical painting; but the soundness of the principle of
interpretation in itself cannot be called in question, nor will there
often be any serious difficulty in its application, among those who
study the subject with diligence and candour. The language, both of
Holy Scripture and of the earliest Fathers, abounds in symbols, and
it was only natural that the earliest specimens of Christian art should
exhibit the same characteristic. More was meant by them than that
which met at first the outward senses; without this clue to their
meaning, the paintings are scarcely intelligible,—with it, all is plain
and easy.
Tombstone from the very ancient Crypt of St. Lucina, now united with the
Catacomb of St. Callixtus.
Ι ΗϹΟΥϹ = Jesus
Χ ΡΙϹΤΟϹ = Christ
Θ ΕΟΥ = of God
Υ ΙΟϹ = Son
Ϲ ΩΤΗΡ = Saviour
The Greek for fish is here written perpendicularly, one letter above
another, ΙΧΘΥϹ; and it is seen that these five letters are the initial
letters of five words, which, together, contain a tolerably complete
account of what Christ is. He is Jesus Christ, Son of God,
Saviour. Thus, this one word, ιχθυς, or fish, read in this way, tells a
great deal about our Lord’s name and titles; it is almost a miniature
creed, or, as one of the Fathers expresses it, “it contains in one
name a whole multitude of holy names.” It would take us too long to
enquire into the origin of this device for expressing our Lord’s name
and titles in so compendious and secret a form. Clearly, whoever
may have invented it, it was very ingenious, and specially convenient
at those times and places where men dared not speak of Him freely
and openly. We cannot say when it began, but it was in universal use
throughout the Church during the first three hundred years of her life,
and then, when she was in the enjoyment of peace and liberty, it
gradually dropped, first out of sight in Christian monuments, and
then out of mind also in Christian literature. But, during the ages of
persecution, it had sunk deep into the habits of Christian thought and
language; it became, as it were, a part of the very Catechism,—
every baptized Christian seems to have been familiar with it, whether
he lived on the banks of the Tiber or of the Po, of the Loire, of the
Euphrates, or of the Nile. In all these parts of the world, writers in
books, poets in hymns, preachers in sermons, artists in painting, the
very masons themselves on gravestones, made use of it without a
word of explanation, in a way that would utterly mystify any modern
Christian community. Who would now dream of carving or painting a
fish upon a gravestone in a Christian churchyard? yet scores of
graves in the Catacombs were so marked, and some of them with
hardly a word or an emblem upon them besides. Or what meaning
could we attach to the picture of a dove or a lamb standing on a
fish’s back, if we did not understand that the fish represented Christ,
and the dove or the lamb a Christian, so that the whole symbol stood
for a Christian soul supported by Christ through the waves and
storms of life? Or again, only imagine a Christian in these days
having buried with him, or wearing round his neck during life, a little
figure of a fish cut in ivory, or crystal, or mother of pearl, or some still
more costly material? Yet a number of those who were buried in the
Catacombs did this; and some of these fish even bear an inscription,
calling upon the fish to be a Saviour!
It was necessary to give this explanation of certain symbols, and
to justify it by sufficient examples, before we proceed to study any of
the more complex paintings in the Catacombs. But now, with these
thoughts in our minds, let us enter the Cemetery of St. Callixtus, and
look on a figure represented two or three times on a wall of one of its
most ancient chambers: a fish swimming and carrying on its back a
basket of bread, and in the midst of the loaves of bread, a glass
vessel containing a red liquid. What is this but bread and wine, the
elements of the Sacrament of Love, and Jesus Christ Its reality? St.
Jerome, when speaking of a holy bishop of Toulouse who had sold
the gold and silver vessels of his church to relieve the poor, uses
these words, “What can be more rich than a man who carries the
body of Christ in a basket of wicker-work, and the blood of Christ in a
vessel of glass?” Here are undeniably the basket of wicker-work and
the vessel of glass; and who can doubt that we have the other also,
veiled under the figure of the fish?
Consecration of the Holy Eucharist.
Sacrifice of Isaac.
Resurrection of Lazarus.
Such is the full meaning of the scene at the Sea of Tiberias, as
interpreted according to the unanimous consent of the Fathers; and
the adjuncts of this picture show that it was intended to be so
understood here also; for on one side is the figure of the
consecration already described; and on the other, the sacrifice of
Isaac by his father, which was surely a most lively type of the
sacrifice of Christ upon the altar; wherein blood is not really shed,
but the Lamb is only “as it were slain,” just as Isaac was not really
slain, but was received back from the dead, “for a parable.” Lastly, as
has been mentioned before, there follows on the third wall of the
same chamber the natural complement of the rest; the doctrine of
the Resurrection, as contained in the fact of the rising again of
Lazarus. Thus, this whole series of paintings, executed at the end of
the second century, or within the first twenty or thirty years of the
third, and repeated (as has been said) in several successive
chambers, was a continual homily, as it were, set before the eyes of
the faithful, in which they were reminded of the beginning, progress,
and consummation of their new and supernatural life.
We do not say that every modern Christian who looks at these
paintings will thus read their meaning at once; but we believe that all
ancient Christians did so, because it is clear from the writings of the
Apostles themselves and their successors, that nothing was more
familiar to the Christian mind of those days than the symbolical and
prophetical meaning of the facts both of the Old and of the New
Testaments. They believed the facts themselves to have taken place
just as they are recorded, but they believed also that they had a
mysterious signification, whereby the truths of the Christian faith
were insinuated or expressed, and that this was their highest and
truest meaning. “Perhaps there is no one recorded miracle of our
Lord,” says St. Gregory, “which is not therefore selected for
recording because it was the type of something to happen in the
Church;” and precisely the same was felt to be true also of the
histories of the patriarchal and Jewish dispensations. “All these
things had happened to them in figure, and they were written for our
correction, upon whom the ends of the world are come.”
It may not be often possible to trace as clearly as we have just
done in a single instance, the logical order and dependence of the
several subjects that were selected for representation in each
chamber of the Catacombs; they may not always have been so
admirably arranged as to be in fact equivalent, as these were, to a
well-ordered dogmatic discourse. Nevertheless it is only when read
in this way, that the decoration of the Catacombs can be made
thoroughly intelligible; and it is certain that some such meaning must
have been intended from the first. The extremely limited number of
Biblical subjects selected for representation, while such an immense
variety is really contained in the Bible (and so many of those that are
neglected might have seemed equally suitable for the purpose), and
then again, the thoroughly unhistorical way in which these few
subjects are dealt with, shows clearly that the principle of selection
was theological rather than artistic. The artists were not left to
indulge their own unfettered fancy, but worked under ecclesiastical
supervision; and the Bible stories which they depicted were not
represented according to their historical verity, because they were
not intended to be a souvenir of past facts, but to symbolise and
suggest something beyond themselves. In order, therefore, to
understand them, it is necessary to bring them face to face with the
Christian doctrines which they foreshadow.
Noe in the Ark.