Principles and Practice of Constraint Programming 25th International Conference CP 2019 Stamford CT USA September 30 October 4 2019 Proceedings Thomas Schiex download
Principles and Practice of Constraint Programming 25th International Conference CP 2019 Stamford CT USA September 30 October 4 2019 Proceedings Thomas Schiex download
https://textbookfull.com/product/principles-and-practice-of-
constraint-programming-26th-international-conference-
cp-2020-louvain-la-neuve-belgium-september-7-11-2020-proceedings-
helmut-simonis/
https://textbookfull.com/product/computational-logistics-10th-
international-conference-iccl-2019-barranquilla-colombia-
september-30-october-2-2019-proceedings-carlos-paternina-
Information and Software Technologies 25th
International Conference ICIST 2019 Vilnius Lithuania
October 10 12 2019 Proceedings Robertas Damaševi■ius
https://textbookfull.com/product/information-and-software-
technologies-25th-international-conference-icist-2019-vilnius-
lithuania-october-10-12-2019-proceedings-robertas-damasevicius/
https://textbookfull.com/product/prima-2019-principles-and-
practice-of-multi-agent-systems-22nd-international-conference-
turin-italy-october-28-31-2019-proceedings-matteo-baldoni/
https://textbookfull.com/product/algorithmic-game-theory-12th-
international-symposium-sagt-2019-athens-greece-
september-30-october-3-2019-proceedings-dimitris-fotakis/
https://textbookfull.com/product/algorithmic-decision-theory-6th-
international-conference-adt-2019-durham-nc-usa-
october-25-27-2019-proceedings-sasa-pekec/
Founding Editors
Gerhard Goos
Karlsruhe Institute of Technology, Karlsruhe, Germany
Juris Hartmanis
Cornell University, Ithaca, NY, USA
123
Editors
Thomas Schiex Simon de Givry
INRA INRA
Castanet Tolosan, France Castanet Tolosan, France
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This volume contains the proceedings of the 25th International Conference on the
Principles and Practice of Constraint Programming (CP 2019), which was held in
Stamford, Connecticut, USA, during September 30 – October 4, 2019. Detailed
information about the conference should be available at http://cp2019.a4cp.org. The CP
conference is the annual international conference on constraint programming. It is
concerned with all aspects of computing with constraints, including theory, algorithms,
environments, languages, models, systems, and applications such as decision making,
resource allocation, scheduling, configuration, planning, or automated design. These
two facets of CP are represented by the technical and application track of the
conference.
The CP community is increasingly keen to ensure CP remains open to interdisci-
plinary research at the intersection between constraint programming and other directly
related fields. The senior Program Committee of the CP 2019 edition, therefore,
included people with mixed backgrounds beyond CP – propositional logic and integer
linear programming, among others. To reach out beyond these direct ‘discrete opti-
mization’ connections, the CP 2019 edition continued to offer specialized thematic
tracks targeted at the frontier between CP and another specific area. With the current
progress and excitement around machine learning technology and the new opportu-
nities that such progress offers for modeling and solving, the CP, Data Science and
Machine Learning Track was the most successful thematic track of this edition,
attracting even more submissions than the application track. Other thematic tracks were
targeted at specific application domains that often raise specific challenges for CP,
including the Testing and Verification Track, the Multi-agent and Parallel CP Track,
the Computational Sustainability Track and the CP and Life Sciences Track. Each track
has a specific sub-committee to ensure that specialist reviewers from the relevant
domains vetted papers in their track.
We invited submissions to all tracks and we received 118 submissions (excluding
abstracts). The review process of CP 2019 relied on a multitier approach involving one
senior Program Committee with Program Committees for all tracks and additional
reviewers recruited by Program Committee members. Authors could submit either full
papers, with a maximum length of 15 pages without references and abstracts that are
not included in these proceedings. All full paper submissions were assigned to a senior
Program Committee member and three members of the relevant track Program
Committee. Authors were given the opportunity to respond to reviews, generating
discussions overseen by the senior Program Committee members and the chairs.
Abstracts were directly managed by the chairs. Meetings between the conference chairs
and all members of the senior Program Committee were held at the end of June, chaired
by the program chairs, where the reviews, author feedback, and discussions were
revisited in detail, based on meta-reviews previously prepared by senior Program
Committee members. The result of this was that the acceptance rate was 39%. The
vi Preface
senior Program Committee and the chairs awarded the Best Conference Paper Prize to
Alex Mattenet, Ian Davidson, Siegfried Nijssen, and Pierre Schaus for “Generic
Constraint-based Block Modeling using Constraint Programming,” the Best Student
Paper Prize to Rocsildes Canoy and Tias Guns for “Vehicle Routing by Learning from
Historical Solutions,” and the Distinguished Student Paper Price to Mohd Hafiz Hasan
and Pascal Van Hentenryck for “The Flexible and Real-Time Commute Trip Sharing
Problems.” The program chairs also invited two papers for direct publication in the
Constraints journal (Editor-in-Chief Michela Milano). These papers were presented at
the conference like any other paper and appeared later in the Constraints journal.
Awarded papers and nominated papers were also invited to submit an extended version
of their paper in the JAIR journal (with its own rigorous reviewing process being
applied to these extended submissions).
The conference program featured four invited talks by Ian Davidson, Bistra Dilkina,
Nina Narodystka, and Phebe Vayanos. These invited talks were selected with the senior
Program Committee with the general idea of supporting the current trend of
hybridization between CP and machine learning, and the increasing importance of
‘algorithms’ in everybody’s life. This volume includes one-page abstracts of their talks.
The conference also included four tutorials and three satellite workshops, whose topics
are listed in this volume. The doctoral program gave Ph.D. students an opportunity to
present their work to more senior researchers, to meet with an assigned mentor for
advice on their research and early career, to attend special invited talks, and to interact
with each other. Doctoral program papers went through an internal reviewing process
that allowed young scientists to familiarize themselves with reviewing, discussion, and
with the usage of EasyChair, the usual CP conference submissions management tool.
The program chairs are grateful to the many people that made this conference such a
success. First of all, we are grateful to the authors who provided the material from
which the conference is made. Then to the senior Program Committee members who
helped us in several of the crucial phases of the conference organization, be it for
tutorial and invited speaker selection or for reviews, rebuttal, and discussion man-
agement, meta-review writing, and participation in live remote meetings for final
acceptance decisions (sometimes at extreme hours). The chairs are also extremely liable
to the members of the program committees. By filtering the most novel and original
contributions and maintaining high standards of quality in rigor and writing quality,
their work is essential for both authors and the community at large. The chairs also
address a very special thanks to the authors of various additional reviews that were
needed to give all papers enough reviews from qualified persons.
Of course, there is a whole team standing around us, who directly managed specific
aspects of the conference: Pierre Schaus (Application track chair), Michele Lombardi
and Tias Guns (CP, Data Science and Machine Learning track), Arnaud Gotlieb and
Nadjib Lazaar (Testing and Verification track), Ferdinando Fioretto and William Yeoh
(Multi-agent and Parallel CP track), Michela Milano and Barry O’Sullivan (Compu-
tational Sustainability track), François Fages and Sylvain Soliman (CP and Life Sci-
ences track), Javier Larrosa (workshop and tutorial chair), and Charlotte Truchet
(publicity chair).
We would also like to thank the Association for Constraint Programming (ACP).
The ACP has been managing the conference for fifteen years now, the conference
Preface vii
benefits from very helpful organization support. The program chairs are grateful for the
help of the ACP president (Maria Garcia de la Banda) and the ACP conference
coordinator (Claude-Guy Quimper) for their support and availability when we needed
them.
The conference would not have been possible without the great job done by all the
people involved in the local organization. The program chairs heavily relied on the
local chair (David Bergman) to provide support for all the special contributors to the
CP 2019 program such as invited talks and tutorial speakers and to speedily announce
program updates on the conference website. David was supported in this endeavor by
Ugochukwu Etudo, Tamilla Triantoro, Niam Yaraghi, Mohsen Emadikhiav, Teng
Huang, Saharnaz Mehrani, and Arvind Raghunathan. We would therefore also like to
thank the institutions that supported them during the organization: the University of
Connecticut first, but also the Quinnipiac University and the Mitsubishi Electric
Research Lab.
We acknowledge the generous support of all our sponsors. They include, at the time
of this writing:
– The Artificial Intelligence Journal (Elsevier)
– Cosling (a French CP startup)
– Huawei
– IBM Research
– AIMMS
– Mitsubishi Electric Research Laboratories Inc.
– The Optimization Firm
– Springer
Thanks to these donations, the local chairs have been able to make CP even better,
especially by supporting some of the doctoral program expenses.
Tutorials
Complete Characterisation of Tractable Constraint Languages
Andrei Bulatov School of Computing Science, Simon Fraser
University, Canada
Workshops
Constraint Modeling and Reformulation (ModRef 2019)
Kevin Leo Monash University, Australia
Alan Frisch University of York, UK
Program Chairs
Simon de Givry INRA, France
Thomas Schiex INRA, France
Conference Chair
David Bergman University of Connecticut, USA
Publicity Chair
Charlotte Truchet Université de Nantes, France
Local Chairs
Ugochukwu Etudo University of Connecticut, USA
Tamilla Triantoro Quinnipiac University, USA
Niam Yaraghi University of Connecticut, USA
Website Chairs
Mohsen Emadikhiav University of Connecticut, USA
Teng Huang University of Connecticut, USA
Sponsorship Chairs
Saharnaz Mehrani University of Connecticut, USA
Arvind Raghunathan Mitsubishi Electric Research Lab, USA
Program Committee
Deepak Ajwani University College Dublin, Ireland
Özgür Akgün University of St Andrews, UK
Roberto Amadini The University of Melbourne, Australia
Carlos Ansótegui Universitat de Lleida, Spain
Gilles Audemard CRIL, Lens, France
Behrouz Babaki Polytechnique Montreal, Canada
Fahiem Bacchus University of Toronto, Canada
Sebastien Bardin CEA LIST, France
Organization xiii
Additional Reviewers
Ian Davidson
Abstract. We will give a broad overview of our decade long effort to add
constraints to machine learning. We will begin by exploring motivating exam-
ples of the need for constraints from applications in social network analysis,
medical imaging, and intelligent tutoring systems. We then discuss how con-
straints can be used such as for encoding domain knowledge, for transfer
learning, and for adding humans to the machine learning loop.
We overview our results on encoding constraints in terms of their computa-
tional difficulty in general and for encoding in different types of solvers
including SAT and CP solvers. With the benefit of hindsight, we will then
discuss successful and unsuccessful formulations we have worked on in the past
from clustering, regression, block modeling, and outlier detection.
We will conclude by overviewing future new directions and challenges for
using CP and other discrete solvers in machine learning such as encoding rules a
priori for deep learning, post-processing results for explanation, and encoding
notions of fairness.
Discrete Optimization and Machine Learning
for Sustainability
Bistra Dilkina
Nina Narodytska
Abstract. Deep neural networks are among the most successful artificial intel-
ligence technologies making an impact in a variety of practical applications.
However, many concerns were raised about the ‘magical’ power of these net-
works. It is disturbing that we are clearly lacking an understanding of the
decision making process behind this technology. Therefore, a natural question is
whether we can trust decisions that neural networks make.
There are two ways to address this problem that are closely related. The first
approach is to define properties that we expect a neural network to satisfy.
Verifying whether a neural network fulfills these properties sheds light on the
properties of the function that it represents. Verification guarantees can reassure
the user that the network has an expected behavior. The second approach is to
better understand the decision making process of neural networks. Namely, the
user can require that a neural network decision must be accompanied by an
explanation for this decision. Such explanations help the user to understand the
decision making process of the network function.
In this talk, we consider both research directions. We take a logic-based
approach to analysis of neural networks, where the network is represented in a
logical formalism, like Boolean Satisfiability (SAT) or Satisfiability Modulo
Theories (SMT). From this standpoint, we overview the progress in verification
and explainability. In particular, we will discuss recent progress in verification
of neural networks, focusing on a special class of neural networks – Binarized
Neural Networks – that can be represented and analyzed using Boolean Satis-
fiability. We discuss how we can take advantage of the training procedure and
the structure of the network to speed up verification. In particular, we demon-
strate that the choice of the training procedure can have significant impact on
scalability of the network verification procedure. For the explainability, we
present our work on producing logical explanations for machine learning model
decisions. We also explain how logic-based tools can be used to verify the
quality of explanations produced by well-known explainer tools.
AI and Robust Optimization for Social Good
Phebe Vayanos
Abstract. In the last decades, significant advances have been made in AI and
optimization. Recently, systems relying on these technologies are being tran-
sitioned to the field with the potential of having tremendous positive influences
on people and society. With increase in the scale and diversity of deployment of
AI- and optimization-driven algorithms in the open world come several chal-
lenges including the need for tractability and resilience, issues of data scarcity
and bias, information endogeneity, ethical considerations, and issues of shared
responsibility between humans and algorithms. In this talk, we focus on the
problems of homelessness, wildlife conservation, and public health in vulner-
able communities, and present research advances in AI and robust optimization
to address one key cross-cutting question: how to effectively allocate scarce
intervention resources in these domains while accounting for the challenges of
open world deployment? We will show concrete improvements over the state
of the art in these domains based on both real world data and deployments in the
LA area. We are convinced that, by pushing this line of research, AI and robust
optimization can play a crucial role to help fight injustice and solve complex
problems facing our society.
Contents
Technical Track
Industrial Size Job Shop Scheduling Tackled by Present Day CP Solvers. . . . 144
Giacomo Da Col and Erich C. Teppan
Application Track
Lemma Synthesis for Automating Induction over Algebraic Data Types. . . . . 600
Weikun Yang, Grigory Fedyukovich, and Aarti Gupta
xxvi Contents
1 Introduction
In constraint programming, each problem class is defined by a problem speci-
fication; many different specifications are possible for the same problem class.
A problem specification identifies a class of combinatorial structures, and lists
constraints that these structures must satisfy. A solution is a structure satisfy-
ing all constraints. Problem specifications usually also have formal parameters,
which are variables for which the specification does not assign values but are not
intended to be part of the search for solutions. Values for such formal parameters
c Springer Nature Switzerland AG 2019
T. Schiex and S. de Givry (Eds.): CP 2019, LNCS 11802, pp. 3–19, 2019.
https://doi.org/10.1007/978-3-030-30048-7_1
4 Ö. Akgün et al.
are provided separately, and the specification together with a particular choice
of values for these formal parameters defines a problem instance.
Instance generation is the task of choosing particular values for the formal
parameters of a problem instance, and is often a key component of published
work when existing benchmarks are inadequate or missing. Our goal is to auto-
mate instance generation. We aim to automatically create parameter files con-
taining definitions of the formal parameters of a problem specification, from the
high level problem specification itself, and without human intervention.
We automate instance generation by rewriting a high level constraint speci-
fication in the Essence language [7] into a sequence of generator instances for
the problem class. Values for the parameters of the generator specification are
chosen based on the high level types in the problem specification. A solution to
a generator instance is a valid parameter file defining a problem instance. We
use irace [15], a popular tool for the automatic configuration of algorithms, to
search the space of generator parameters for regions where “graded instances”
exist. Graded instances have specific properties; in this work they are satisfiable,
and neither too trivial nor too difficult to be solved. However, our methodology
does not depend on a specific definition of grading, and can be applied more
generally. We first prove the soundness of our rewriting scheme. The system is
then empirically evaluated over 5 different problem classes that contain differ-
ent combinations of integers, functions, matrices, relations and sets of sets. We
show the viability of our system and the efficacy of the parameter tuning against
randomised search over all problem classes.
2 Related Work
In combinatorial optimisation a wide variety of custom instance generators have
been described. These are used to construct synthetic instances for problem
classes where too few benchmarks are available. In just the constraint program-
ming literature generators have been proposed for many problem classes, includ-
ing quasigroup completion [4], curriculum planning [17], graph isomorphism [26],
realtime scheduling [11], and bike sharing [6]. Different evolutionary methods
have also been proposed to find instances for binary CSPs [18], Quadratic Knap-
sack [13], and TSP [23]. In particular, Ullrich et al. specified problem classes with
a formal language, and used this system to evolve instances for TSP, MaxSAT,
and Load Allocation [24]. Efforts have also been made to extend existing repos-
itories of classification problems via automated instance generation [19].
Instance generators are typically built to support other parts of the research,
such as verifying robustness of models. However, a generator often requires signif-
icant effort to develop, and it cannot be applied to new problem classes without
major modifications [24]. A generator is typically controlled by means of parame-
ters, and a further challenge of instance generation is to find regions of parameter
values where an instance generator can reliably create interesting instances.
Gent et al. developed parametric generators of instances for several problem
classes [8]. They developed a semi-automated prototype to produce instances for
Instance Generation via Generator Instances 5
3 Background
solver (in our case, Minion [9]) takes at least 10 s to decide the instance. To
ensure an instance is not too difficult, we exclude instances for which the solver
has not returned a solution in 5 min. Our choices of grading criteria were guided
by our computational budget and available resources, and so in this work we
have chosen to accept only satisfiable instances as graded. We do not advocate a
specific definition of grading, and other criteria for grading would be reasonable,
such as “the instance is decided or solved to optimality by at least one solver
from a portfolio of solvers in a reasonable amount of time”.
A key step of our method is a process of automatic rewriting, discussed in
more detail in Sect. 4.1. Briefly, the rewriting steps are:
1. remove all constraints (such that statements) and decision variables (find
statements),
2. replace all input parameters (given statements) with decision variables (find
statements) and type specific constraints, and
3. promote parameter constraints (where statements) to constraints.
We call the result of this process a generator specification.
Definition 1. A generator instance consists of a generator specification together
with a particular choice of generator parameters, which restrict the domains of
decision variables appearing in the generator instance.
Rewriting is one step in an iterative process. The choice of generator parame-
ters is performed automatically using the parameter tuning tool irace. Solutions
to the generator instance are then filtered according to our grading criteria,
retaining graded instances. We want the rewriting procedure to have the follow-
ing two properties: soundness (the solutions of the generator instance should
always be valid inputs for the instance), and completeness (every valid instance
should be obtainable as a possible solution of the generator instance). We now
discuss the semantics of generator instances and our approach.
Variables may represent tuples, and for clarity of presentation we take some
liberties with the corresponding Essence syntax. When referring to a specifi-
cation s with variables v, we omit the variables that occur within the scope of
a quantifier, and partition the remaining variables so that s(x | y) denotes a
specification s with formal parameters x and decision variables y, both of which
are generally tuples. With s(x := a | y) we denote the specification s (| y) (with
no formal parameters and only decision variables) which is obtained from s by
substituting the tuple of formal parameters x by a fixed tuple of values a.
Start with an Essence specification of the form
s(x | y) := given x : D where h(x) find y : E such that f (x, y)
where the specification has formal parameters x and decision variables y. We
are interested in valid inputs a, such that s(x := a | y) is a valid instance. Here
domain D may be a product of component domains, of arbitrarily nested types
as allowed in the Essence language. Now let
s (| x) := find x : D such that h(x)
Instance Generation via Generator Instances 7
be a specification obtained from s(x) by our rewriting process, which drops the
original constraints f (x, y), replaces the given by a find, and modifies where
statements into such that statements, leaving a specification with no formal
parameters but only decision variables (Note that many possible but equivalent
specifications are possible for s ). In principle we could search for a solution
x := a to this specification s (| x), as this would be a valid input for s(x | y),
yielding the instance s(x := a | y). Such search seldom finds graded instances in
a reasonable amount of time, unless more guidance is provided.
Thus, we want to introduce a new parameter p with domain P to structure
our search for instances. We then rewrite the specification s(x | y) differently, as
so that as the values assigned to the formal parameters p vary, the solutions
to the instance s (p := q | x) form valid inputs to s(x | y). The specification
s (p | x) will be our generator specification, instead of s (| x). We can then treat
P as a space of parameters, and explore this space with a parameter tuning tool.
The types or domain expressions of the formal parameters x with domain D
in the specification s (| x) may have a lot of structure and be quite complex.
Exploring such parameter spaces successfully is a challenging problem. We there-
fore aim to simplify our task of instance generation by replacing these structured
domains by the usually smaller domains D(p) in the new specification s (p | x),
and automatically incorporate this structural information into the constraints
c(p, x) instead; the constraints c(p, x) include both the constraints h(x) and also
the additional constraints to capture structural information. Like D, the param-
eter domain P is usually a product of domains, but for P these are usually just
intervals of reals, ranges of integers, or Booleans.
4 Methodology
In Fig. 1 we show how our system turns an abstract problem specification into
concrete problem instances with the use of rewriting rules and an iterated
sequence of tuned generator instances. The steps of the automated process are:
The rest of this section describes the details of this process, correctness of the
rewriting procedure, how we use tuning based on instance difficulty, the problem
classes we studied, and our experimental setup.
Discovering Diverse Content Through
Random Scribd Documents
BARUCH:
Gottes Diener und Profet ist mein Meister – Jeremias.
ZEDEKIA (ausbrechend):
Jeremias! Er, immer er! Immer der Schatten hinter meiner Tat,
immer in Aufruhr wider mich! In den Kerker habe ich ihn
verschlossen, aber noch immer schreit er zu mir wie am ersten Tage:
Friede, Friede! Was drängt er sich vor? Was will er mich verwirren,
was quert er meinen Weg? Wo ich mich wende, ist auch er, im
Palast, in der Stadt, und durch seine Boten wirft er sich auf wider
mich. Was verfolgt er mich?
BARUCH:
Du irrst, mein König! Jeremias liebt dich mehr, denn einen andern
dieser Stadt.
ZEDEKIA:
Ich brauche seine Liebe nicht, ich speie sie an und zerblase seinen
Zorn! Wer ist er, daß er wagt, mich zu lieben? Darf einer aufstehn in
der Gasse und künden, er liebet mich oder liebet mich nicht? Was
stößt er sich zwischen mich und meinen Entschluß? Will er mehr sein
als ich? Ich bin der König, ich allein! Möge er schreien: Friede,
Friede! nicht seine Hand hält Jerusalems Geschick. Ich bin der König,
und nicht rühmen soll er sich, er habe mich geschreckt mit seinen
Träumen. Eher sinke die Stadt, als daß sie gerettet sei durch
Jeremias! (Zu Baruch): Du gehst zu Nabukadnezar und sagest ihm an:
Nie wird Zedekia ein Joch tragen, nie hebt er den Vorhang des
Heiligsten. Möge er kommen mit seinen Völkern, Zedekia ist ihm
bereit.
(BARUCH, im Schrecken beide Hände hebend, will sprechen.)
ZEDEKIA:
Kein Wort! Und bringst du die Botschaft nicht, so fällt Jeremias
Haupt. Zweimal habe ich seines Lebens geschont, doch zu Ende ist
meine Milde. Nicht will ich Richter hinter mir und Schatten hinter
meiner Tat, ich will sterben als König zu Jerusalem.
(BARUCH hebt noch einmal die Hände.)
ZEDEKIA:
Ein Wort dawider, und sein Haupt sinkt hin. In deinen Händen ist
meine Botschaft, ist Jeremias Haupt. Geh! Ich befehle dir: geh!
(BARUCH bleibt noch einen Augenblick stehen, dann verhüllt er sein Antlitz und
wendet sich ab.)
ZEDEKIA (hat sich drohend aufgerichtet gegen den Zögernden. Wie Baruch
abgeht, fällt sein ausgereckter Arm nieder wie zerbrochen, sein Antlitz verdüstert
sich wieder von neuem. Plötzlich sich aufreckend):
Vorbei! Ein Ende, ein Ende! Nur nicht mehr die Qual! (Er geht wieder
auf und ab, hebt den Vorhang und sieht lange stumm sinnend auf die Stadt.
Endlich stampft er zweimal mit dem Fuße.)
ZEDEKIA:
Wer ist draußen im Gange? Ich höre einen Schritt. Ist der Späher
nicht gegangen, zögert er noch?
SCHWERTTRÄGER:
Er ist gegangen, Herr! Der draußen wacht, ist mein Bruder Nehemia.
ZEDEKIA:
Er soll nicht so laut schreiten des Nachts vor meinem Schlafgemach.
Ich will nichts hören um mich. Ich will schlafen. Auch ich will
schlafen wie die andern.
SCHWERTTRÄGER:
Es soll geschehen, Herr! (Er schlägt die Vorhänge des Pfühles auseinander
und verhüllt die Ampel. Nur ein trüber Schein von Mondlicht glänzt in den Raum.)
SCHWERTTRÄGER:
Soll ich dir noch lesen aus den heiligen Büchern, mein König, wie
gestern und ehetags?
ZEDEKIA:
Aus den Büchern?... Nein, laß die Bücher, auch sie wissen nicht Rat.
Ich will schlafen, schlafen einmal wie die andern. Meine Lider
brennen, und mein Herz brennt mit.
SCHWERTTRÄGER (hilft ihm aus dem Obergewand. Zedekia wirft sich auf das
Ruhelager):
ZEDEKIA (der reglos gelegen, richtet sich im Dunkel ganz leise auf. Wie ein Tier
im Ansprang, krümmt sich sein Körper in der Anstrengung des Lauschens, er
krampft sich immer mehr zusammen, und plötzlich schreit er heftig):
Ich höre eine Stimme. Aus der Tiefe dringt sie empor!
ZEDEKIA:
Ah, du hörst sie auch!
SCHWERTTRÄGER (schaudernd):
Es tönt wie Gesang. Die Geister der Tiefe sind wach unter dem
Haus. Es klagt und stöhnt wie ein gefesseltes Tier.
NEHEMIA:
Vielleicht ist es Wind, in eine Spalte verfangen?
ZEDEKIA:
Nein, Worte sind es, ich fühle sie, ohne sie zu fassen. Wer singt hier
nachts in meinem Haus? Ist den Sklaven so wohl, daß sie singen,
indes ich, der König, hier liege mit brennenden Lidern? Geh, Joab,
und mache ihn stumm.
(SCHWERTTRÄGER eilends ab.)
ZEDEKIA:
Wer war es, der da sprach?
SCHWERTTRÄGER (zitternd):
Ich weiß es nicht, Herr. Ich bin ihm nicht genaht. Wie ich niederstieg
zur Halle, hörte ich stärker das Singen, aus der Tiefe der Erde schien
es zu kommen, und grauenhaft tönten die Worte. Ich ging nach, wo
sie tönten, und fand doch keinen, der sang in der Halle, immer war
es tiefer als ich, immer tiefer, wie aus einem Brunnen klang es
empor oder einer Grube. Und ich hörte seine Worte, die waren
fürchterlich. Dreimal stieß ich den Speer auf die Erde. Und da
schwieg die Gehenna.
ZEDEKIA:
Was tönte die Stimme?
SCHWERTTRÄGER (schaudernd):
Ich ... ich kann es nicht sagen!
ZEDEKIA:
Ich befehle dir: sage die Worte!
SCHWERTTRÄGER:
Lästerung war es, mein König, die aufströmte vom Brunnen.
ZEDEKIA:
Was waren die Worte? Bei meinem Zorn!
SCHWERTTRÄGER (schaudernd. Seine Stimme wird psalmodierend im
Gesang):
So sang es von der Tiefe:
Ich habe mein Haus verlassen müssen
Und mein Erbe meiden,
Und was meine Seele liebet, in der Feinde Hand geben.
Meine Augen fließen mit Tränen Tag und Nacht
Und hören nicht auf,
Denn die Jungfrau, die Tochter meines Volks,
Ist greulich zerplagt.
ZEDEKIA (aufschreiend):
Jeremias! Er, immer er!
SCHWERTTRÄGER (wie begeistert weitersingend):
Du meinst ... Nein, ich fürchte ihn nicht. Ich fürchte niemanden. Und
ich weiß nicht, ob er mein Feind ist. Vielleicht war es töricht, vor ihm
zu flüchten ... Vielleicht ... (Er geht unruhig auf und ab): Schwertträger!
SCHWERTTRÄGER:
Mein König?
ZEDEKIA:
Geh hinab und schließe auf die Düngergrube. Nimm mit deinen
Bruder Nehemia, und bringet den Mann aus der Tiefe vor mich her.
Geheim muß er gebracht werden und im geheimen wieder hinab.
(DER SCHWERTTRÄGER und sein Bruder eilig ab.)
Mich?!
Du mich blenden, du Ruchloser!? Nein!
Anders hat Gottes Entschluß bestimmt!
Wohl wird einer geblendet sein,
Ehe der Tag noch sein Ende nimmt,
Doch jener, der längst schon verblendet war,
Als sein Auge noch blickte und sah: –
Höre mich, König Zedekia!
(ZEDEKIA hat ihn losgelassen und starrt ihn erschrocken an.)
JEREMIAS (mit beiden Fäusten auf ihn zu):
Dich
Werden sie fassen, des Königs Knechte
Im Hause Gottes, das du verstört,
Sie reißen die Rechte
Dir los vom Altar,
Daran sie zur Hilfe verklammert war!
Du willst dich wehren, sie brechen dein Schwert,
Umtun deine Arme mit eisernen Flechten
Und schleppen
Und schleifen dich über die Treppen,
Wie ein Opfertier mit Peitschen und Schlägen
Jenem entgegen,
Dessen Hand du verstoßen, dessen Joch du zerbrochen
Und der dir ein feuriges Urteil gesprochen!
(ZEDEKIA ist zurückgefahren und hebt wie abwehrend die Hände.)
JEREMIAS:
In die Knie
Knicken sie dich und stoßen dich sie,
Ein Feuer loht knisternd auf rundem Stein,
Vier Hände halten den Blendstahl hinein.
Heiß
Frißt die Hitze
Vom schwarzen Griff sich auf in die Spitze.
Sie glüht! Sie flammt! Sie wird rot! Sie wird weiß!
Und dann
Fassen dich rauh ihre Fäuste an,
Zischend und rauchend
Tauchen
Sie die Nacht dir in dein Auge hinein.
ZEDEKIA (aufschreiend und sich an die Augen greifend, wie ein Geblendeter):
Weh!
JEREMIAS:
Doch eh
Dir noch in einer brennenden Gischt
Von Blut und Tränen dein Blick verlischt,
Mußt du noch sehn
Deine Söhne, die drei, vor dem Henker stehn!
Doch dich halten die Knechte, dich halten die Ketten,
Du kannst sie nicht lösen, du kannst sie nicht retten,
Du kannst nur aufschrein, wie jetzt das Schwert
In den ersten! den zweiten! den letzten fährt!
Du siehst,
Wie ihr Blut, ihr junges, im Kote fließt,
Und siehst,
Eh der rote Stahl dich für immer blendet,
Wie Israels Stamm und Königtum endet.
ZEDEKIA (der, wie ein Blinder tappend, auf das Ruhebett gesunken ist, die
Hände flehend aufhebend):
Erbarmen! Erbarmen!
JEREMIAS:
Was für eine Macht ist dir gegeben, Jeremias! Die Kraft hast du
gebrochen in meinen Gliedern, und das Mark steht starr mir im
Leibe. Furchtbar sind deine Worte, Jeremias!
JEREMIAS (die Ekstase ist von ihm gefallen, der Glanz in seinen Augen
erloschen):
Arm sind meine Worte, Zedekia, Ohnmacht meine Macht. Nur wissen
kann ich und nicht wenden!
ZEDEKIA (erschüttert):
Warum bist du nicht früher vor mich getreten?
JEREMIAS:
Ich war immer zur Stelle. Doch du fandest mich nicht.
ZEDEKIA:
Es muß so Gottes Wille gewesen sein! (Schweigen. Dann steht Zedekia
langsam auf und geht auf Jeremias zu): Jeremias, höre mich an – ich ...
glaube dir! Furchtbareres hast du gekündet, denn je gekündet ward
einem König in Israel, und doch – ich glaube dir. Mit Schauer hast du
mein Herz geschlagen, und doch, ich ward dir nicht gram. Es möge
kein Streit mehr sein zwischen uns im Schatten des Todes. Geh
hinab, woher du gekommen, nicht soll es dir fehlen an Zehrung, das
letzte Brot meines Tisches will ich teilen mit dir. Und niemand wisse
um unsere Zwiesprache denn Gott allein.
(JEREMIAS wendet sich zum Gehen.)
ZEDEKIA (gequält):
Jeremias! Muß es denn sein? Oh, Jerusalem, mein Jerusalem! Kannst
du es nicht wenden?
JEREMIAS (düster):
Es muß sein! Nichts vermag ich zu wenden. Verkünden ist mein Amt.
Wehe den Ohnmächtigen!
ZEDEKIA (schweigt, dann von innen):
Jeremias, ich habe es nicht gewollt! Ich mußte Krieg künden, aber
ich liebte den Frieden. Und ich liebte dich, weil du ihn gekündet
hast. Nicht leichten Herzens hab ich den Harnisch genommen, es
war Krieg vor mir unter Gottes Angesicht und wird auch nachdem
sein. Viel habe ich gelitten, sei dessen Zeuge zu seiner Zeit. Und sei
bei mir, wenn dein Wort sich erfüllt.
JEREMIAS (ergriffen):
Ich werde bei dir sein, mein Bruder Zedekia!
(JEREMIAS wendet sich langsam, abgekehrten Gesichtes von ihm. Er ist schon bei
der Türe, da ruft noch einmal:)
ZEDEKIA:
Jeremias!
(JEREMIAS wendet sich.)
ZEDEKIA:
Tod ist über mir, und ich sehe dich zum letztenmal. Du hast mir
geflucht, Jeremias – nun segne mich auch, ehe wir scheiden.
JEREMIAS (zögert, dann schreitet er feierlich zurück und hebt die Hände über
des Königs Stirn):
Der Herr segne dich und behüte dich auf allen deinen Wegen. Er
lasse dir leuchten sein Angesicht und gebe dir den Frieden.
ZEDEKIA (träumerisch verworren nachsprechend):
Und ... gebe ... uns ... den Frieden ...
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade
Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.
textbookfull.com