Professional Documents
Culture Documents
Bounded Variable Logics and Counting A Study in Finite Models 1st Edition Martin Otto
Bounded Variable Logics and Counting A Study in Finite Models 1st Edition Martin Otto
Bounded Variable Logics and Counting A Study in Finite Models 1st Edition Martin Otto
https://ebookmeta.com/product/models-in-microeconomic-theory-she-
edition-martin-j-osborne/
https://ebookmeta.com/product/martin-rauch-refined-earth-
construction-design-of-rammed-earth-otto-kapfinger-editor/
https://ebookmeta.com/product/sociology-of-the-renaissance-1st-
edition-alfred-wilhelm-otto-v-martin-gertrud-lenzer-wallace-k-
ferguson/
https://ebookmeta.com/product/agrarian-socialism-the-cooperative-
commonwealth-federation-in-saskatchewan-a-study-in-political-
sociology-seymour-martin-lipset/
A Spin and Momentum Resolved Photoemission Study of
Strong Electron Correlation in Co Cu 001 1st Edition
Martin Ellguth
https://ebookmeta.com/product/a-spin-and-momentum-resolved-
photoemission-study-of-strong-electron-correlation-in-co-
cu-001-1st-edition-martin-ellguth/
https://ebookmeta.com/product/modal-logics-and-philosophy-second-
edition-rod-girle/
https://ebookmeta.com/product/study-skills-for-international-
postgraduates-bloomsbury-study-skills-2nd-edition-martin-davies/
https://ebookmeta.com/product/r-calculus-v-description-
logics-1st-edition-li/
https://ebookmeta.com/product/statistical-analysis-of-graph-
structures-in-random-variable-networks-v-a-kalyagin/
Bounded Variable Logics and Counting
Since their inception, the Perspectives in Logic and Lecture Notes in Logic series
have published seminal works by leading logicians. Many of the original books in the
series have been unavailable for years, but they are now in print once again.
In this volume, the 9th publication in the Lecture Notes in Logic series, Martin
Otto gives an introduction to finite model theory that indicates the main ideas and lines
of inquiry that motivate research in this area. Particular attention is paid to bounded
variable infinitary logics, with and without counting quantifiers, related fixed-point
logics, and the corresponding fragments of Ptime. The relations with Ptime exhibit
the fruitful exchange between ideas from logic and from complexity theory that is
characteristic of finite model theory.
This series serves researchers, teachers, and students in the field of symbolic
logic, broadly interpreted. The aim of the series is to bring publications to
the logic community with the least possible delay and to provide rapid
dissemination of the latest research. Scientific quality is the overriding
criterion by which submissions are evaluated.
Editorial Board
Jeremy Avigad,
Department of Philosophy, Carnegie Mellon University
Zoe Chatzidakis
DMA, Ecole Normale Supérieure, Paris
Peter Cholak, Managing Editor
Department of Mathematics, University of Notre Dame, Indiana
Volker Halbach,
New College, University of Oxford
H. Dugald Macpherson
School of Mathematics, University of Leeds
Slawomir Solecki
Department of Mathematics, University of Illinois at Urbana–Champaign
Thomas Wilke,
Institut für Informatik, Christian-Albrechts-Universität zu Kiel
More information, including a list of the books in the series, can be found at
http://www.aslonline.org/books-lnl.html
L E C T U R E N OT E S I N L O G I C 9
MARTIN OTTO
Rheinisch-Westfälische Technische Hochschule,
Aachen, Germany
www.cambridge.org
Information on this title: www.cambridge.org/9781107167940
10.1017/9781316716878
First edition © 1997 Springer-Verlag Berlin Heidelberg
This edition © 2016 Association for Symbolic Logic under license to
Cambridge University Press.
Association for Symbolic Logic
Richard A. Shore, Publisher
Department of Mathematics, Cornell University, Ithaca, NY 14853
http://www.aslonline.org
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
A catalogue record for this publication is available from the British Library.
ISBN 978-1-107-16794-0 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party Internet Web sites referred to in this publication
and does not guarantee that any content on such Web sites is, or will remain,
accurate or appropriate.
Preface
With respect to the work presented here, I personally owe much to two
sources that I would like to mention. One is the Freiburg logic group where
I had the opportunity to participate in an intensified engagement in finite
model theory after finishing my doctorate with Professor Ebbinghaus. Pro-
fessor Flum and Professor Ebbinghaus created an encouraging atmosphere
for taking up this new field actively. The second important source is the col-
laboration with Professor Gradel which has had much impact on the research
presented here. I am grateful to Erich Gradel for this collaboration and also
for his advice and support.
The text itself is a revised version of my Habilitationsschrift, presented
at the RWTH Aachen in 1995. I am grateful to Professor Ebbinghaus for
numerous comments on the earlier version of this text.
Preface V
0. Introduction 1
0.1 Finite Models, Logic and Complexity 1
0.1.1 Logics for Complexity Classes 1
0.1.2 Semantically Defined Classes 4
0.1.3 Which Logics Are Natural? 7
0.2 Natural Levels of Expressiveness 7
0.2.1 Fixed-Point Logics and Their Counting Extensions . . . 8
0.2.2 The Framework of Infinitary Logic 9
0.2.3 The Role of Order and Canonization 11
0.3 Guide to the Exposition 12
3. The Invariants 79
3.1 Complete Invariants for Lk and Ck 80
3.2 The ^-Invariants 81
3.3 The L*-Invariants 85
3.4 Some Applications 87
3.4.1 Fixed-Points and the Invariants 87
3.4.2 The Abiteboul-Vianu Theorem 90
3.4.3 Comparison of ICk and ILk 91
3.5 A Partial Reduction to Two Variables 93
Bibliography 177
Index 181
0. Introduction
ory. This topic is most pronounced in the search for semantic matches be-
tween levels of computational complexity and logics. The search for logics for
complexity classes, also suggestively described as capturing complexity classes
[Imm87b] or tailoring logic for complexity [Gur84], has lead to an active re-
search programme in finite model theory.
Consider any of the standard classes in computational complexity as a
class of problems for finite structures, say for finite graphs. The class of all
properties of finite graphs that can be recognized by PTIME algorithms is a
typical example. A logical characterization of PTIME on finite graphs would
have to provide a logic for PTIME on graphs in the sense that exactly all
PTIME properties of finite graphs are definable by sentences of this logic. For
the present purposes we work with a slightly informal notion of a logic for
PTIME. The exact definition underlying our treatment is due to Gurevich
[Gur88]. A more detailed discussion will be provided with Definition 1.7 in
the next chapter. It is also shown in [Gur88] that the restriction to graphs
rather than finite structures of arbitrary type is inessential for the present
issue.
Definition 0.1 (Sketch). A logic C is a logic for PTIME if exactly those
properties of finite graphs that are PTIME recognizable are C-definable.
Minimal requirements on candidate logics to be imposed are the following:
C has recursive syntax and recursive semantics that associates with each C-
sentence a PTIME algorithm for checking its truth in finite models.
It is a central open problem in the field whether there is a logic for PTIME.
Relationships between computational complexity classes and definability
in logical systems are interesting for a number of reasons:
(a) The potential for theoretical transfer between different fields. Techniques
from complexity theory may be brought to bear on logical and model
theoretic issues and vice versa. To give an example, several of the out-
standing open problems of complexity theory like the PTIME = NPTIME?
or PTIME = PSPACE? questions have found appealing non-trivial model
theoretic reformulations in terms of semantic equivalences of particular
logical systems over finite structures (compare [AV91, DLW95, Daw95b]).
Short of solving the original problems this offers new perspectives, and
investigations of related logical issues may at least lead to a better under-
standing of these problems.
(b) Logical analysis of the required kind may yield deeper insights into the
fundamental notion of complexity. Definability in logical systems can be
viewed as a kind of complexity in itself. Whereas computational com-
plexity controls the computational resources required in the solution of
a problem, definability considerations control the logical or descriptive
resources required in the specification of the problem: hence the term
descriptive complexity as used in [Imm89]. The relationship between the
0.1 Finite Models, Logic and Complexity 3
Consider the class of all PTIME recognizable graph properties — for the
moment denote it graph-PτiME. It serves as a typical example of a complexity
class on finite structures.
Why is it difficult to find a logic for graph-PτiME?
Recall that ordinary PTIME is the class of all problems that can be solved by
polynomially time bounded Turing machines. A priori Turing machines work
with words or strings as inputs. As far as recognition (i.e. decision) problems
are concerned a problem is a set of words over some alphabet. Words over
this alphabet are rejected or accepted, according to membership in the set,
in time polynomial in their length.
In particular a Turing machine does not work with abstract graphs as
inputs but rather with encodings of these. The standard encoding scheme for
finite graphs uses adjacency matrices for the input representation. The ad-
jacency matrix of a graph whose vertices are labelled v\,..., υn is the n x n
boolean matrix with entries α^ = 0 or 1 according to whether (v»,Vj) is
an edge. But obviously different adjacency matrices may encode the same,
more precisely isomorphic, graphs. Any rearrangement of the vertices in a
different order induces an equivalent representation that is different from the
given one unless the rearrangement happens to be an automorphism of the
abstract graph. Any graph algorithm, i.e. any algorithm that recognizes a
graph property, must therefore satisfy a non-trivial semantic invariance con-
dition: a graph algorithm must produce the same result on any two inputs
that represent isomorphic abstract graphs. In other words it may not reject
one graph and accept an isomorphic copy of that same graph. We adopt
the terminology of complexity theory as presented in [Pap94] to distinguish
0.1 Finite Models, Logic and Complexity 5
This crucial difference between the ordered and the unordered case is
at the root of the apparent mismatch with respect to capturing complexity
classes in the case of ordered structures or in the general case of not nec-
essarily ordered structures. For the standard complexity classes it is almost
trivial to see that the induced classes over ordered structures are presentable
as syntactic classes and therefore can be captured by logics. The point of
the corresponding capturing results indeed rather is that moreover they are
captured by very natural logical systems.
As mentioned above, no complexity class below NPTIME has been cap-
tured or shown to be recursively presentable in the general case. NPTIME here
marks a threshold because in NPTIME and above, the invariance problem can
be side-stepped as follows. Consider the class graph-NPτiME of all NPTIME
graph properties. From a graph property Q € graph-NPτiME we may pass to
its ordered version <2<, tne class °f all ordered graphs that possess the given
property:
Note that capturing results for complexity classes in the general case of
not necessarily ordered structures are not merely of theoretical interest. The
challenge is well motivated by the potential applications in database theory.
Natural abstract databases often are not ordered. Their realizations at the
machine level may involve an implicit linear ordering for representational
purposes (naively: a numbering of memory cells). Even though an ordering is
present then, it is not considered part of the intended data. A sound database
query in this case corresponds to a property of unordered relational struc-
tures. In the query specification it is desirable to hide this ordering. Logically,
one would have to have a query language corresponding to a logic for PTIME
on unordered structures in order to achieve semantic completeness within
PTIME and simultaneously to guarantee soundness — soundness in the sense
of independence of a linear ordering that is an artifact of the realization.
0.2 Natural Levels of Expressiveness 7
The first defect is taken care of in the extension to fixed-point logics. The
adjunction of fixed-point operators leads to logics that capture certain levels
of relational recursion. Least or inductive fixed-point logic FP in particular
is a very natural logic that has been studied extensively. Inductively defined
and increasing relational processes are captured by FP. The generation of the
transitive closure is a simple but typical example for the expressive power of
FP above that of first-order logic. An important point is that the increasing
nature of these relational processes guarantees termination in a stationary
value within polynomially many steps. A further extension in terms of rela-
tional recursion for arbitrary rather than increasing processes (that therefore
may or may not terminate in a stationary value) is partial fixed-point logic
PFP. The interest in FP is justified because by the theorem of Immerman
and Vardi it captures PTIME for ordered structures. Similarly, PFP captures
PSPACE in the presence of order [Var82, AV89]. In particular, in the presence
of order, FP and PFP automatically remedy the second shortcoming of first-
order logic: on ordered structures FP and PFP also capture all counting and
PTIME, respectively PSPACE, cardinality properties of definable predicates.
In the absence of order, however, this is not at all true. In the extreme case
of pure sets (graphs without edges) it is easy to see that relational recursion
and all of FP and PFP collapse to first-order. Simple cardinality properties
of the size of sets like evenness of the number of vertices are not definable
in FP or PFP. Moreover, all the simple examples of properties that are not
FP-definable but may be recognized in PTIME involve such cardinality prop-
erties.
One of the themes underlying our present investigations is the attempt
to treat these two most apparent shortcomings of first-order logic over finite
structures — recursion and counting — on an equal footing and to consider
the extensions FP and PFP to a framework that incorporates counting.
We thus obtain fixed-point logic with counting FP+C and partial fixed-
point logic with counting PFP+C. Roughly speaking we deal with two-sorted
variants of the given finite structures, augmented by a second ordered arith-
metical sort. A link between the sorts is induced by counting terms that
associate cardinality values with formulae that define sets. The usual fixed-
0.2 Natural Levels of Expressiveness 9
the other hand they provide the setting for the comparative analysis of the
expressive powers of FP+C, PFP+C, FP and PFP.
The main asset of the fragments C^ω and L^ω is in fact a methodologi-
cal point. Both possess very elegant and tractable Ehrenfeucht-Fraϊsse style
characterizations in terms of games. Such games that capture the expres-
sive power of a logic are an important tool in classical model theory. The
classical Ehrenfeucht-Fraϊsse theorem relates first-order equivalence of two
structures to the existence of a strategy in a game played on these struc-
tures. See [EFT94] for a textbook treatment of this technique in the classical
context. Variants of these games have been found and employed for various
other logics besides first-order. It is remarkable that such games are among
the few tools from classical or abstract model theory that are fully available
in restriction to finite structures without any alteration. See also [EF95].
A large part of the present work is devoted to the detailed analysis of
corresponding games for the C^ω and L^. The games are due to Barwise
[Bar77] and Immerman [Imm82], and Immerman and Lander [IL90] respec-
tively. The analysis of the games leads to the abstraction of concise PTIME
computable structural invariants that characterize finite relational structures
exactly up to equivalence in C^ω or L^ω. As mentioned above such invari-
ants were first considered by Abiteboul and Vianu in [AV91] in the context
of a computational model for relational recursion and applied to the study
of fixed-point logics. A formalization in terms of the underlying fragments
of infinitary logic has been presented in Dawar's dissertation [Daw93] and in
[DLW95] for the L^ω and in [G093, Ott96a] for the C^ω. The relationship
between the expressive power of FP-hC and the invariants for the C^ω is
also one of the main topics here. We shall investigate this relationship in
comparison with FP and the invariants for the L^ω as well.
problem itself is in PTIME is not known. Note that the entire problem of
capturing PTIME is solved trivially if there should be PTIME normalization.
Any algorithm applied to the standard encoding of a normalized version of
the input graph becomes a graph algorithm. PTIME normalization would in
fact reduce the capturing of PTIME in the general case to the ordered case.
Some of our investigations concern the variant of this approach in re-
striction to the framework of the C^ω and L^, respectively. We consider
canonization up to equivalence in these logics rather than up to isomorphism.
There is a direct connection between PTIME canonization for these rougher
equivalences and the capturing of the PTIME fragments of these fragments
of infinitary logic. Linking these considerations with the above-mentioned
PTIME invariants for the C^ω we find appealing sufficient criteria that FP+C
indeed captures PTIME in restriction to all of C^ω. It remains a challeng-
ing open problem whether these conditions are fulfilled. A reduction proce-
dure furthermore shows that the general cases hinge on the three-dimensional
cases, i.e. on canonization up to equivalence in the three-variable fragments
Of LOGO;.
A main result that will be treated in full detail in the last chapter concerns
the two-variable case [Ott95a, Ott95b]. For L\oω and C^ω we exhibit a strong
form of PTIME canonization and thus obtain non-trivial capturing results.
The classes of all PTIME properties that are C^ω- or L^3θω-definable are
indeed recursively presentable and may be captured naturally in terms of the
complete invariants for C^ω or L^, respectively.
• Chapter 7 finally is concerned with the positive results related to the two-
variable fragments. In particular there are detailed proofs that the PTIME
fragments of both Li^Qω and C^ω can be captured.
A major part of this chapter serves to review and fix notation and terminol-
ogy. The material is standard. Readers familiar with the notions addressed
might therefore only want to refer back to particular definitions at later
points. The main issues of the individual sections are the following:
• Section 1.1 sums up the basics about structures, global relations, logics
and types that are relevant for our purposes.
• Section 1.4 contains some preliminary material about types and definability
in the relevant fragments of infinitary logic.
• Section 1.7 fixes some terminology with respect to the notion of can-
onization and of complete invariants for arbitrary equivalence relations. We
also sketch some technicalities and conventions concerning orderings and pre-
orderings.
1.1.3 Logics
21 |= <p[α] says that φ is satisfied over 21 when the free variables are interpreted
as indicated. In this usage the notation φ(x\,..., Xk) does not imply that the
displayed Xi must all be syntactically free in φ, but that the free variables of φ
are among those displayed. We speak of a formula in free variables x\,..., Xk
with this meaning: free(<^) C {χ 1 ? ... ,#*.}. For instance, we allow to regard
the formula x\ —x^ also as a formula in free variables #ι,£2 ? #3 5 and write
φ(xι, x 2 , #3) = xι=X2 if this view is intended.
Similar conventions apply to second-order variables (predicate variables)
where such occur. In particular notation like φ(X,x) G £[τ] indicates that
given a r-structure plus additional interpretations for the second-order vari-
ables X by extra predicates and for the x by elements, φ evaluates to a
boolean value. 21 |= φ[P,a] expresses that φ is satisfied in 21 with the indi-
cated interpretations for X and x.
1.1 Structures and Types 19
A Ξ £2
denotes this semantic equivalence over finite relational structures. The possi-
ble weakening of this requirement, that the two logics define the same classes
of finite structures is explicitly indicated. We write "£ι = £2 for sentences "
or "£ι = £2 for boolean queries ".
Observe that L\ Ξ £2 says that for every formula of C\ there is a formula
of £2 that is equivalent over finite structures, and vice versa. The weaker
notion of equivalence expresses the same requirement in restriction to sen-
tences. The distinction between the two notions of equivalence is of a purely
formal nature for our considerations. Most natural logics admit a faithful re-
duction from definable global relations to definable boolean global relations
so that their expressive power is fully determined by their strength in defining
classes, i.e. by their sentences.
The notation = for semantic equivalence extends with analogous meaning
to classes of queries that are not specified by logics. For instance if C is a class
of queries and C a logic then C = C says that every query in C is definable
by a formula of C and that conversely all £-definable queries are in C.
1.1.4 Types
£(a) = (φ(x) € C(
£
Tp (τ; k) is the class of all £[r]-types in variables xι,...,xk:
Some of the logics that play a central role in the following possess only
a bounded supply of variable symbols. If C only has variables xi, . . . , Xk we
agree to apply the notion of £-type only to tuples α of length at most k. We
adopt the convention that Tp £ (τ;r) = 0 for r > k in this context.
The most basic types considered are the atomic or quantifier free types.
They are obtained in the above formalism if C is chosen to be the quantifier
free fragment of first-order logic. We write atp^ (α) for the collection of all
quantifier free formulae that hold true of a in 21. Note that each such type
can be fully represented by the set of atomic formulae contained in it. In
this sense, and for finite relational vocabularies, each atomic type is finite
and we may identify an atomic type θ with a single quantifier free formula:
the conjunction over all atomic formulae contained in θ together with the
negations of all those not contained in θ. Some such syntactic normal form is
tacitly assumed when we deal with sets of atomic types. The set of all atomic
r-types in variables a?ι, . . . , Xk is denoted by Atp(τ; k):
A: stan[r] —ϊ stan[σ]
Definition 1.5. (%) The algorithm ,4:stan[τ] ->• {0,1} computes the bool-
ean query Q C fin[r] if for all 21 G stan[τ]: .4(21) = 1 if and only if
21 G Q. We also say that A recognizes the class Q.
(ii) An algorithm A: stan[τ; r] -» {0,1} computes the r-ary query R on fin[τ]
if for all (21, a) G stan[r;r]: .4(21, α) = 1 if and only if a G Rm.
(Hi) An algorithm .4:stan[τ] -> stan[σ] computes the functor F:fin[τ] ->
fin[σ] if for all 21 G stan[r]: .4(21) ^ F(%).
(iv) An algorithm A: stan[τ] -> S computes the functor F: fin[τ] -> S whose
range is some domain of standard objects S if for all 21 G stan[τ]:
,4(a)=F(a).
algorithms, φ »->• Λφ such that Λφ evaluates (the query defined by) φ over
fin[r].
£ is a logic for C or captures C if C coincides with the class of queries that
are definable in £, C = C, and if the recursive semantics φ *-+ Aφ maps £[τ]
to algorithmic realizations within C.
Of course the same notion applies to other classes of queries that are
defined in terms of algorithmic criteria, in particular to subclasses of com-
plexity classes of queries. The well known theorem of Immerman and Vardi
for instance says that the class of all PTIME queries on ordered structures is
captured in exactly this sense by fixed-point logic.
Sometimes it is useful to strengthen these requirements so that other
important data also become recursive in terms of the formulae of £, compare
[Gur88, EF95]. For instance one may require that data describing complexity
bounds on Aφ be recursive in φ. While such strengthenings are crucial for
certain arguments we can here stick to the basic notion.
It is worth to note the essential equivalence between the notions of cap-
turing by some logic and that of a recursive presentation. It is clear that a
logical representation as in the last definition provides a recursive presenta-
tion through {Aφ \ φ £ £} . Conversely, any recursive set of algorithms may
be regarded as a logic with recursive syntax in the abstract sense; for the
semantics choose the obvious one embodied in the algorithms. In this way
any recursive presentation of C can essentially be regarded as a logic that
captures C. There are some fine points to be considered if as usual we want
abstract logics to satisfy some appropriate regularity criteria as outlined in
[Ebb85]. For this it is obviously necessary that C itself as a set of queries
satisfies corresponding closure criteria. At least for classes C that are natural
in such respects it follows that indeed the two notions are equivalent. We
do not here enlarge on this issue, in fact an informal concept of 'logics for
complexity classes' will be quite sufficient for our purposes.
Full infinitary logic is the logic LOOW that has the usual first-order rules
for the formation of formulae and in addition is closed under infinitary con-
junctions and disjunctions: if # is any set (!) of formulae of L(X)ω then /\ #,
the conjunction over #, and V ^> the disjunction over Ψ, are also formulae of
LOOU,. Their semantics is the obvious one: /\ Ψ evaluates to true if all formulae
in \P evaluate to true and V & evaluates to true if at least one formula in Ψ
does. Note that one often has to deal with families of formulae (t/Όίe/ and
then writes for instance /\ ίe/ ψi instead of f\{ψi | i € /}.
As mentioned in the introduction any query on finite structures is defin-
able in LOOU This follows from the observation that any finite structure 21 is
characterized up to isomorphism by some first-order sentence φ^. If Q is a
boolean query, for instance, then the infinite disjunction ψ = \/ φ<% over all
φ% for 21 £ Q Π stan[τ] clearly defines Q. Recall, however, that φ<& typically
requires n -f 1 variables if the size of 21 is n. This motivates the introduction
of the finite variable fragments of LOOU
Definition 1.8. L^ω is the fragment of Looω that consists of formulae us-
ing only variable symbols from {#ι , . . . , Xk}- The union of the L^ω is denoted
Lt^oω . It consists of all formulae of Looω that use finitely many variable sym-
bols (from the standard supply {xi\i ^ I}).
We also consider the corresponding bounded variable fragments of Lωω :
let L^ω denote first-order logic with variable symbols {#ι, ...,#&}.
The union of the L^ω is full first-order logic Lωω. In actual formalizations
we often use variable symbols x, ?/, z, ... instead of the standardized X{
for the sake of easier readability. The official restriction to standard sets of
variables is convenient, however, to have syntactic closure under conjunctions
and disjunctions for each L^ω. We give some examples for the expressive
power of the L^ω. Formalizations with few variable symbols typically require
clever re-use of already quantified variables. Examples 1.9 and 1.11 are from
[KVa92a], Example 1.16 plays an important role in [DLW95] in a context
that will also concern us here later.
Example 1.9. Over linear orderings (A, <) two different variable symbols
suffice to produce first-order formulae ψi(x), for i ^ 0, which express that x
is the z'-th element with respect to <. Equivalently, for the standard linear
orderings (n, <):
(n, <) \= φi[τn] exactly for ra = i.
To obtain these formulae put ψo(x) := -~>3y y<xto define the bottom element
in any linear ordering (A, <). Inductively let
ψi+ι(x):= f\ ^ψj(x) Λ
26 1. Definitions and Preliminaries
where ψj (y) is the result of exchanging x and y throughout the formula ψj (x) .
Example 1.10. The class of acyclic directed graphs (A, E) is definable by
a sentence of L\oω. Observe that a finite graph is acyclic if it has no infinite
-E-paths, or equivalently if there is some finite bound on the length of E-
paths. Put ξo(x) := -~<3yEyx to characterize those vertices that have no
E-predecessors. Inductively let
Then (A, E) \= ξi[υ] if and only if there is no E-path of length greater than
i reaching v. It follows that
¥>ord := ξ1 Λ φQ ,
whereto := VxVy(x=yVx<yVy<x),
1.3 Some Particular Logics 27
defines the class of all finite linear orderings. It remains to argue that φord
enforces
(i) irreflexivity — Vx-ix < x : it does since £' forbids loops.
(ii) antisymmetry — VxV2/-»(x < y Λ y < x) : it does since ξ' also forbids
cycles of length two.
(iii) transitivity — VxVyVz (x < y Λ y < z -»> x < z ) : if x < y and y < z
then x Φ z (forbidden cycle of length two) and z £ x (forbidden cycle
of length three), whence ΨQ forces x < z.
In fact, over not necessarily finite structures and for any ordinal λ the class
of all well-orderings of order type less than λ is L^-definable. It follows
that for instance the class of all countable well-orderings is L^-definable
over arbitrary structures. This claim is justified just as in the last example,
if we replace £' by a sentence that expresses that < is a well-founded rela-
tion of rank less than λ (compare the remark following Example 1.10). ξ' in
particular forbids loops and cycles, so that irreflexivity, antisymmetry and
transitivity follow as above.
Returning to finite structures, we have the following:
Corollary 1.13. Let <£ r and let τ have no relation symbols of arity greater
than 2. Then any query of linearly ordered finite r -structures Q C ord[τ] is
L^-definable.
Sketch of Proof. Consider without loss r = {<,E} with one extra binary
relation E besides <. Let 21 = (n, <,£?) have standard domain with the
natural interpretation of <. Put, for formulae ψi as in Example 1.9 that
characterize the ΐ-th element of the ordering,
Λ /\
Note that connectedness is implied by (i) and (ii) if there cannot be cycles.
Now cycles can be forbidden in L\oω according to Example 1.10. (i) can even
be formalized in L^ω through
(ii) actually needs a third variable for mere counting, as for instance in the
formalization
Λ Ezx -+y = z.
A binary tree is a tree in which all nodes have out-degree 0 or 2. A full
binary tree is a binary tree in which all leaves (nodes of out-degree 0) are at
the same distance from the root (at the same height).
Example 1.16. The class of all full binary trees is definable in L^ω. The
condition on equal height of all leaves is formalized in three variables us-
ing auxiliary formulae ψi that define the set of vertices at height i. These
are constructed inductively similar to the formulae used in Example 1.9.
ΨQ(X) := -iByEyx defines the root. Inductively ψi+ι(x) := 3y(ψi(y) Λ Eyx)
is as desired. The L^-sentence \J ^x(-^yExy -» ψi(x)) forces all leaves to
be at the same height.
It might look surprising that the condition on out-degree 2 should also
be expressible in just three variables. It is however, and quite simply, in the
context of trees. The sentence
expresses that there are no three vertices with common direct predecessors
for any two of them. If the in-degree can at most be 1 then this is equivalent
with a bound of 2 on the out-degree. The condition that the out-degree of
all vertices apart from the leaves must be at least 2 is trivially first-order
expressible in three variables.
In Example 1.15 some variables had to be spent for mere counting in the
conditions on the degree of vertices. In general the explicit formalization of
3^πιxφ(χ) requires at least m variables as for instance in
Indeed, over the empty vocabulary it follows easily from game arguments that
are treated in Chapter 2, that no sentence of L^~l can express the existence
of at least m distinct elements. It is therefore natural to consider fragments
of first-order and infinitary logic with a bounded supply of variables that
admit, however, arbitrary counting quantifiers
1.3 Some Particular Logics 29
to express regularity.
Example 1.19. Even C%ω does not have the finite model property, i.e. there
are sentences of C%ω without finite models that are satisfiable in infinite
structures. For instance it is expressible in C*ω that a binary relation E is
the graph of an injective total function that is not surjective:
\/x3=lyExy Λ Vy3^lxExy Λ 3yVχ-^Exy.
First-order logic with two variables l?ωω on the other hand is known to have
the finite property by a result of Mortimer's, see Theorem 7.35.
Example 1.20. Any finite equivalence relation is characterized up to iso-
morphism among all finite equivalence relations by its C^ -theory. The ax-
iomatization of equivalence relations, however, needs at least three variables.
Let 21 = (A, ~) where ~ is an equivalence relation over A. A complete in-
variant that characterizes any finite equivalence relation up to isomorphism
is its spectrum, the set of pairs (i,Vi) such that there are exactly Vi classes
of size ί in (A, ~). This information is expressed in two variables by
That two variables even with counting quantifiers do not suffice to express
transitivity of a binary relation E is shown easily with games discussed in
Chapter 2, see Example 2.6.
Example 1.21. The classes of all trees and of all full binary trees are de-
finable in C^ou In f act we observed in Example 1.15 and 1.16 that a third
variable was only needed for counting purposes. For instance (ii) of Exam-
ple 1.15 now simply becomes
30 1. Definitions and Preliminaries
With this sequence the partial fixed point of F is defined to be either the
stationary value of this sequence if such exists or the empty set otherwise:
Xi if -Xi+1 — Xi
PFP[F] := I
0 if Xi+1 φ Xi for all i.
The inductive or inflationary fixed point associated with F. Induc-
tively, the following increasing sequence of r-ary predicates is generated:
Note that this sequence may alternatively be obtained through iterated ap-
plication of a modified operation Fίnfl to the empty predicate. F infl is the
inflationary variant of F, defined by Finfl(P) := P U F(P); generally an op-
eration G on sets is called inflationary if always G(P) D P.
Since the domain A is finite, the sequence of the Xi must become station-
ary within polynomially many steps of the iteration. The limit reached is the
inductive or inflationary fixed point:
(2t,Γ)μV[δ] if α
Definition 1.23. Fixed-point logic FP is the smallest logic closed under first-
order constructors and the FP- constructor.
Theorem 1.24 (Immerman, Vardi). Over the class of linearly ordered fi-
nite structures FP captures PTIME: a class of linearly ordered structures is
recognized by a PTIME algorithm if and only if it is the class of finite mod-
els of some sentence of FP; a global relation on ordered structures is PTIME
computable if and only if it is definable in FP.
Theorem 1.25 (Abiteboul, Vardi, Vianu). Over linearly ordered finite
structures PFP captures PSPACE: a class of linearly ordered structures is
recognized by a PSPACE algorithm if and only if it is the class of finite models
of some sentence of PFP; a global relation on ordered structures is PSPACE
computable if and only if it is definable in PFP.
Theorem 1.24 was obtained in [Imm86] and [Vaτ82], Theorem 1.25 com-
bines results from [Var82] and [AV89] (via equivalences with relational While-
queries).
Let us briefly indicate how the fixed-point processes available in FP and
PFP can be used to code PTIME, respectively PSPACE, computations over
linearly ordered structures — an observation that is the key to the preceding
theorems.
Example 1.26. Polynomially space bounded configurations of a Turing ma-
chine can be encoded by predicates over the ordered domains of the input
structures. The arity of the predicate depends on the degree of the space
bound. A definable indexing of the tape cells is provided by the lexicographic
ordering on an appropriate power of the universe. In terms of these encodings
by predicates it is easily seen that the successor step from one configuration
to the next becomes first-order definable.
For a PSPACE machine that computes a boolean query say, consider the
partial fixed-point process based on this first-order formalization of the com-
putational successor (with a corresponding definition of the initial configura-
tion). Provided that we adapt the transition function of the machine in such
a way that an eventual halting configuration is formally repeated indefinitely,
the limit value of this partial fixed-point process is an encoding of the halting
configuration if the machine halts on the given input structure.
For a PTIME machine we may pass from the encoding of individual con-
figurations to a cumulative encoding of initial segments of the computation
path. We use a fixed-point variable with additional entries for a lexicographic
indexing also for a polynomial number of time steps. This leads to an induc-
tive fixed-point process whose i-th level describes the computation path up
to step i. The limit reached in this process is an encoding of the entire com-
putation path on the given input structure.
In either case the actual result of the computation (acceptance or rejection
in the case of a boolean query) is first-order definable in terms of the final
configuration.
The original and intuitively also very appealing definition of fixed-point
logic FP in terms of a least fixed-point operator LFP rather than the in-
1.3 Some Particular Logics 33
ι=I,....m
Iteration in the spirit of IFF or PFP yields inflationary or partial fixed points
for such systems of interrelated relational transformations. Standard tech-
niques for the encoding of tuples of relations through a single relation of
higher arity can be employed to show that fixed-point logic FP and partial
fixed-point logic PFP are closed under the formation of respective fixed points
for systems.
The fixed-point logics PFP and FP both are sublogics of L^ω. Since FP itself
is a sublogic of PFP it suffices to review the argument that yields PFP C
L^oω. Since we have already seen that Lt^oω expresses queries of arbitrary
complexity, whereas PFP-definable queries obviously are in PSPACE, it follows
that PFP £ L^ω.
For a convenient statement of the following arguments it is useful to elimi-
nate first-order parameters in fixed-point processes. This is easily done at the
expense of an increase in the arity of the second-order fixed-point variable.
Lemma 1.28. Any formula in PFP is equivalent with one in which fixed-
point operators are applied only in the form PFPχ$φ, where all free first-
order variables of φ are among those in x. The same holds o f F P .
Sketch of Proof. Consider φ with free first-order variables of, ~z. Assume that
x and ~z together consist of pairwise distinct variables and that no variable in ~z
is bound in φ. Let φ1 be the result of replacing any atom Xΰ in φ by the atom
Z~zϋ, where Z is a new second-order variable whose arity is that of X plus
arity of z. It is easily seen that {zx \ [PFPx^φ]x} = {zx \ [PFPz^Έφ']zx}.
D
Road from
The road from Feshn and Maghagha to Baharia
Feshn and leaves the edge of the Nile Valley cultivation at Qasr el
Maghagha Lamlum Bey, which bears 51½° west of true north from
to Baharia.Maghagha railway station, and is distant 15·4
kilometres. From this point the road is well-defined and easily
followed right into the oasis. In the following description the
distances are given from the edge of the Nile Valley cultivation.
The road at first leads over a strip of drift sand, half a kilometre
broad, with short prickly scrub, passing a white mosque on the left
and then turning off somewhat to a direction 26° south of west, and
continuing in a straight line for 15 kilometres over an undulating
gravelly plain. The high prominent cliffs, about 7 or 8 kilometres to
the north-west, are the flanks of Jebel Muailla, and a valley known as
Wadi Muailla leads through them to the Wadi Rayan in the Fayum
depression.[18] At 19 kilometres the valley scarp, with a number of
isolated peaks, is approached on the right, while ridges and low
mounds form the plain below, well-marked lines of drainage running
from here in a south-east direction towards the cultivation. At 23
kilometres the scarp runs back, enclosing a large bay, across which
the road runs and ascends to the plateau beyond at 27·8 kilometres.
Numerous isolated parallel sand-dunes in the form of small ridges
are seen running out into the bay from the cliff at the far end, all lying
slightly west of north and east of south, or parallel to the normal wind
direction.
The escarpment bounding the Nile Valley at this point is only
some 15 metres in height, being thus quite insignificant compared
with the cliffs on the east side. The plateau here was found to be
about 140 metres above the cultivation, the road having risen
gradually throughout. The latter continues for about 1½ kilometres
across the strip of plateau when it again descends, making a slight
detour to the left for easy descent. It then continues 9° south of west,
slightly winding, over gravelly undulating ground. At 31-32 kilometres
a line of low hills is passed on the right, while a dark well-marked
range lies 6-7 kilometres to the left.
A ridge of sandstone, known as Jebel el Ghudda, is passed on
the right at 45 kilometres, from the end of which a small dune runs
out; beyond, the plain resumes its monotonous undulating character,
a low ridge being crossed at 61 kilometres. There, the road,
consisting of a number of more or less parallel well-marked narrow
paths worn by camels, which have a somewhat general habit of
marching in line one behind the other, changes its direction to 36°
south of west, falling gradually in level until a patch of scrub is
reached 6½ kilometres further on. This scrub was dead at the time of
the visit, and furnished a useful supply of fuel. From this point the
course is 7° south of west (true), which direction is maintained for the
next 42 kilometres over a remarkably monotonous undulating gravel-
covered desert, the typical “serir” of the Arabs. At 92 and 93½
kilometres some more patches of dead scrub were passed on the
right, while logs of silicified wood were noticed strewing the plain on
the left. An Arab grave was met with at 96 kilometres, while
skeletons of camels lie about near the roadside at frequent intervals;
at 110 kilometres the eye of the traveller is relieved by a small grove
of green thorny flat-topped acacia trees (Acacia nilotica, or “sunt” of
the Arabs) with a patch of coarse grass; four gazelle (probably
Gazella dorcas) were observed browsing on the scrub here.
The course now continues 12½° south of west, over gently
undulating gravelly “serir”, until the eastern scarp of El Bahr is
reached at 125 kilometres from the cultivation of the Nile Valley. The
“serir” or undulating gravelly type of desert then ceases.
El Bahr is a depression, some 60-70 metres deep, cut out in
white limestone rocks; its breadth at the point crossed by the road
was 8 kilometres. Within it are several high prominent hills, one of
which near the centre on the left side of the road is called Jebel Gar
Marzak. The bottom of the depression was quite green with
vegetation; sufficient water is said to fall every year to keep these
plants alive, and in 1894 rain is said to have fallen to such an extent
that a pool of considerable size was formed; the silt deposited by this
is plainly visible at the present time. A good deal of blown sand
occurs within the depression. El Bahr evidently corresponds to the
Bahr Bela Ma, (river without water) figured on some authors’ maps,
which has been frequently but erroneously referred to as an old
river-course; although this idea was shown to be untenable by
Zittel[19] and Ascherson[20] it has subsequently been maintained by
non-scientific writers. No traces of any river deposit occur in the
depression, which consists simply of a series of unconnected
depressions, eroded by wind-borne sand.[21]
The track leaves the depression at 134 kilometres, rising over
heavy sand; it then continues 3° north of west. The character of the
desert has now completely changed, and instead of the smooth
undulating gravelly “serir,” its surface is rough and hummocky, being
formed of hard bare limestone, cut up into sharp knobs and grooved
into furrows by the powerful action of wind-borne sand; it resembles
closely the surface of the rough open sea. This type of desert is
spoken of as “kharafish” by the Arabs. While the “serir” forms an
ideal surface for travelling over, the “kharafish” is the worst
imaginable, the innumerable hillocks necessitating incessant small
deviations, while the hard rough surface is in some places very
troublesome to camels; moreover, an extensive view is out of the
question and no tracks are visible on the surface, so that the road is
easily lost except where marked by frequent cairns built of loose
stones.
Occasional patches of blown sand are here met with, and the first
well-marked dunes were crossed at 141 kilometres. From here
onwards for kilometres the whole area was more or less sandy with
occasional narrow well-marked ridges or dunes, running almost due
north and south, and varying in breadth from that of a single line to a
number of parallel ridges side by side half a kilometre broad. The
largest dune of this group at 146½ kilometres is known as Ghard el
Shubbab. The steepest sides are those facing west where the angle
may reach 30°.
At the particular locality crossed by this road the sand area is
very easily crossed, a circuitous route being followed in order to take
advantage of the flatter dunes with the easiest slopes when crossing
the steeper ridges. Probably the road crosses at one of the easiest
points. This remarkable line of dunes, known as the Abu Moharik,
has its origin in the neighbourhood of the oasis of Mogara and runs
southward, almost without a break, across the desert until Kharga is
reached, whence with a slight break owing to the broken character of
the ground it continues southward within the oasis-depression. The
total width of the sand-belt on the road under description is about 6
kilometres.
At 153 to 156 kilometres a number of black conical hills, Gar el
Hamra, are situate from 1 to 2 kilometres from the road on the south
side. One or two more sand-dunes were crossed and then the road,
maintaining its direction of 2°-3° north of west, lay over a more or
less uneven dark-coloured limestone desert broken up into a number
of small hills. At 169 kilometres a broad ridge of sand-dunes was
encountered, running 18° west of true north. These light yellow
dunes afford a beautiful and remarkable sight, running northwards
away to the horizon over a dark brown-coloured desert in an almost
perfectly straight line and with a sharply maintained junction-line
between the edge of the sand and the desert surface adjoining.
Within a few hundred metres of the western side of the sand-
dunes the road commences the descent from the plateau into the
oasis-depression. The road enters at the most northerly extremity of
the oasis, the descent being particularly easy at this point, passing
the large dark-coloured hill, Jebel Horabi (or Morabi?), on the right
almost immediately afterwards.
A fine view of the depression is obtainable from the top of the
escarpment, a broad low-lying expanse, bounded by steep
escarpments or walls, stretching away to the south, its monotony
relieved by several large flat-topped hill-masses, near which, on the
lowest portions of the floor, dark areas, the cultivated lands and
palm-groves can be distinguished. The road crosses the depression
in a south-westerly direction, passes a spring known as Ain el Gidr,
the first watering place, and divides in front of the great hill-masses
separating the two groups of villages, the eastern branch keeping
close under the eastern scarp of Jebel Mayesra, to avoid a large
area of soft salty ground, and leading to the villages of Zubbo and
Mandisha, While the western branch continues its course to the
cultivation surrounding El Qasr and Bawitti. The distance by this road
from Qasr el Lamlum Bey to the village of Zubbo is 190 kilometres
and to Bawitti 195 kilometres.
Geology of
Having now described the topographical features of
the Feshn- this road, the chief geological characters may be
Baharia noticed. The plain between the Nile Valley cultivation
and the scarp of the plateau is covered with sandy
gravel, partly downwash from the higher ground in Recent times, and
partly the remains of definite gravel deposits belonging to the Nile
Valley Pleistocene series.[22] The pebbles now found strewn over the
plain consist chiefly of flints, doubtless derived from the Eocene
limestones forming the deserts on both sides of the Nile Valley, and
occasional pebbles of hard felspar porphyry which must have
originally been derived from the igneous massifs of the Red Sea
Hills. Both are well rounded, although the former are frequently
broken up into angular fragments by temperature changes. White
granular beds of gypsum, of various degrees of impurity, crop out on
the plain in places, and in all probability there was in Pleistocene
times an extensive deposit of this mineral all over the surface of the
low-lying country. In the desert lying between the Fayum and the Nile
Valley further to the north, these gypseous beds occur of great
thickness and wide extent, and the deposits crossed on this road are
doubtless part of the same series.
The cliffs of Jebel Muailla to the north are capped by a hard dark
bed of limestone, which weathers with a vertical face, while the more
gentle slopes, generally more or less hidden with sand, are
doubtless formed of softer limestones, marls, and clays. During the
survey of the Fayum (see foot-note, p. 17) the hills surrounding Wadi
Muailla were found to be formed of Lower Mokattam beds (Middle
Eocene) and the hills seen from this road are doubtless composed of
the same beds. The ridges crossed at 20 kilometres are formed of
hard, compact, close-grained crystalline limestone, covered with
more or less gypsum and flint gravel; the limestone beds forming
these ridges show dips which suggest the existence of a fault
running N.E.-S.W., parallel to the trend of the cliff behind, and this
may be part of the extensive faults and folds of the Nile Valley. In
one small hill (22 kilometres) shales with Ostrea were noticed at the
base, with occasional hard oyster-limestone bands; the upper part
was formed of 10 metres of gravel consisting of well-rounded
limestone pebbles. This superficial deposit must be classed as
Pleistocene and may be a sea-beach, though no conclusive
evidence was obtainable on this point. The escarpment passed at 23
kilometres is capped by a bed of white limestone, shales forming the
slope, but was not examined at close quarters. The floor of the bay
formed by the receding cliff shows outcropping brown limestone with
Ostrea, and the escarpment on the far side is capped by a hard
white crystalline limestone with much flint, the latter forming bands.
On the surface is a thin calcareous gypseous gravel deposit,
doubtless of the same age as the gypseous beds already mentioned
as occurring on the plain below. The flanks of the scarp are hidden
by downwash. The cliff bounding this strip of plateau, 1½ kilometres
further on, is composed of the same beds, the limestone being here
silicified, with large silicified Conidæ. With regard to the age of these
limestones and clays they are probably equivalent to part of the
Lower Mokattam series already mentioned as forming the hill-
masses round Wadi Rayan, although no Nummulites gizehensis
beds were observed in the sections examined. A conspicuous black
knob among the low gravelly hills left two kilometres on the right at
32 kilometres, was found to be a neck of hard dark andesitic basalt,
one of the few occurrences of igneous rocks in the Western Desert.
Several other similar looking dark hills were in sight, but time did not
admit of their examination. The dark well-marked range 6-7
kilometres to the left of the road is probably identical with a range of
hills occurring 10 kilometres west of Bahnessa, which was
mapped[23] during the survey of the Nile Valley in 1899, and found to
consist of a mass of andesitic basalt similar to that forming the small
neck on this road. Doubtless they are both parts of the same
intrusion. The surface of the plain is still composed in part of
gypseous deposits, with occasional outcrops of the underlying
limestone, the surface being covered with a certain amount of loose
sand with rounded flints and their broken fragments. In the
neighbourhood of Jebel el Ghudda the plain consists of limestone
with numerous individuals of the large Nummulites gizehensis, and
are thus of Lower Mokattam age. Much of the limestone is
crystalline. The hills of Jebel el Ghudda are formed by younger
overlying beds consisting of hard silicified sandstones and grits
(quartzites), which lithologically are very similar to the beds of Jebel
Ahmar near Cairo, of Oligocene age. They may, however, belong to
the Upper Eocene series, so well developed above the Upper
Mokattam in the escarpments to the north of the Fayum, as this
series contains similar beds with similar silicified wood. They enclose
bands of coarse conglomerate and have a peculiar blackish burnt
colour. Occasional patches of these grits are here and there met with
right up to the depression of El Bahr, and the silicified wood passed
at 92 kilometres belongs to the same series of beds. From the scrub
area at 93½ kilometres till the road approached El Bahr there was
hardly a sign either of the grits or of section anywhere met with on
this road. The depression is cut down through the upper series of
sandstones and grits into the fossiliferous white limestones and
sandstones below. The eastern scarp showed the following beds:—
Top.
Soft white Sandstone.
⎧ Limestone with Nummulites gizehensis. ⎫
Middle Lower
⎨ Limestone with Ostrea, Exogyra and Lucina, ⎬
Eocene. Mokattam.
⎩ etc. ⎭
Road from
Road from Minia to Baharia.—The cultivated land
Minia to extends for some 9 or 10 kilometres west of Minia town,
Baharia. and at the time of the Survey expedition (early in
October, 1897) this area was mostly covered by flood-
water. Boats were therefore taken to the edge of the desert, and the
march west was commenced from Nasl Nadiub Lengat, a small
village on the cultivation-limit in lat. 28° 6′ 24″, long. 30° 39′ 45″ E.,
bearing from Minia railway station 9° north of west, distance very
nearly 10 kilometres. The road westward being ill-defined, and the
party knowing that any course followed a little north of west would
surely lead to the oasis, no attempt was made to follow the track
precisely, the course being rather chosen to take advantage of
commanding points so as to map as much topography as possible
en route. As will be seen further on, however, the road was struck
some distance before reaching the oasis, and was thence followed
to Zubbo.
On leaving Nasl Nadiub Lengat the course first taken was about
32° north of west. The desert here rises very gradually from the
flood-level, (i.e. about 40 metres above sea-level), there being no
cliff bounding the Nile Valley on the West at this point. For about 6
kilometres the ground was sandy, with occasional patches of flood-
water and some grass; then came a stretch of level gravelly ground,
and at 11½ kilometres some low sand-dunes were crossed. The
course was now changed to about 3° south of west, and at 14
kilometres a descent was made from the gravel-plain into a
limestone-depression, with gravel-capped eminences on either side.
At 22 kilometres the gravel-strewn plain was again come on, a few
low mounds, also gravelly, being passed at 31 kilometres, and
camel-tracks, doubtless part of the Minia-Baharia road, being noticed
coincident with the survey line at 37 kilometres. Continuing following
these tracks, another camel-road was found to branch off at 46
kilometres to the north-west. The march was continued along the
same track going a little more south of west, over a monotonous
gravelly plain, with limestone showing through it in small patches, till
at 78 kilometres from Nasl Nadiub Lengat a conspicuous though
small mass of fissile gritty limestone was come on, which afforded a
good survey-station. North and west of this are low gravel-covered
hills. Turning a little north of west, a large extent of low table-like hills
was seen on the south. At about 90 kilometres low plateaux of
limestone, capped by gravel, closed in so as to form a defile about a
kilometre wide, through which the line of survey passed, and at 98
kilometres a large sandy and gravelly depression, bounded by
limestone scarps and containing numerous large limestone hills, was
entered. Within the depression the course followed was about 30°
north of west, so as to map roughly the escarpments on either hand.
About 10 kilometres from the point where the depression is entered
a long stretch of heavy sand was crossed beyond which tracks going
west were come on, and the edge of the limestone plateau on the
left curved round so as to cross the road. Ascending this low scarp at
116 kilometres, the course lay over white limestone, with sharp
angular flints on its surface, and low limestone hills on either side. At
124 kilometres the broad belt of sand-dunes, the Abu Moharik, was
reached. The dunes, which are of considerable height, run nearly
north and south, and have a total width of some 4 kilometres; they
formed the most serious obstacle of the entire journey. Beyond the
dunes is a hard limestone plain, crossed by a low ridge (forming part
of a higher plateau) at 136 kilometres. Beyond this ridge are low
limestone-hills on either side, the floor being generally of hard white
limestone. At 151 kilometres an oval depression in the limestone
plateau, with its longer axis (about 4 kilometres) running W.N.W.,
was entered; this depression is full of hills of white chalk, which also
forms the floor here. Ascending the opposite scarp of the depression
at 159 kilometres, a narrow strip of higher plateau of much harder
limestone was crossed, the edge of the scarp of the oasis of Baharia
being reached at 160 kilometres, after marching for 57 hours from
Nasl Nadiub Lengat. Up to this point there had been a gradual rise of
the ground, from 40 metres at the Nile Valley to 265 metres above
sea-level at the edge of the oasis.
The descent into the oasis is not difficult, the scarp being sandy
and the fall gradual. At about 4 kilometres W.N.W. from the edge of
the scarp is the first well or spring of the oasis, Ain Gelid (lat. 28° 19′
28″ N., long. 29° 8′ 40″ E. of Greenwich), a small pool surrounded by
grasses and with a tree growing near it. The barometric observations
gave for this point the level 134 metres above sea, so that the total
drop from the edge of the scarp is about 130 metres. From Ain Gelid
the village of Zubbo bears about 15° north of west, and is some 17
kilometres distant, but owing to the hills between the two places
necessitating a slight detour, the actual distance to be traversed is a
few kilometres greater.
On the western part of the hills the gravel covering is absent, the
succession here seen being:—
Top.
1. Sandy marls, 60 centimetres.
2. Very hard pink calcareous grit, 1 metre.
3. White sandy marl, fissile, 2 metres.
4. Soft red and white marls, 10 metres.
One of the hills within the depression, to the left of the track about
6 kilometres further on, showed in a face of about 28 metres the
following beds:—
Top.
1. Hard white limestone, sand-eroded.
2. Brown fossiliferous limestone.
3. White limestone with many small fossils.
The slopes of this hill were covered with large black Ostrea
shells.
The floor of the depression further west showed limestones of
varying character, frequently highly fossiliferous. Beyond the
depression the ground passed over was mainly white limestone,
strewn with large sharp angular flints, and many spheroidal flint-
masses derived from the limestones. The surrounding hills show that
we have here two white limestones with brown beds between, the
floor and the hillcaps being white, while the feet of the hills are
brown.
The plateau-rock near the great belt of sand-dunes is a hard
white limestone with large Conidæ. Just beyond the dunes this rock
is very siliceous, the exposed surfaces showing a smoky-black
colour; the rock is however quite white on fracture. The hills rising
from the plateau here consist entirely of limestone, beds of white
chalk alternating with harder yellowish and brown limestones, in
which no fossils were noticed. Further on the plateau rises, so that at
the entrance to the depression the surface is formed of the same
brown limestone-beds which are seen in the hills behind. The hills
within the oval depression here are composed of alternations of
chalky with harder limestones. The narrow ridge which separates the
depression just described from the oasis-depression is of very hard
horny siliceous limestone.
On the descent into the oasis the following beds are passed
through:—
Top.
1. Very hard yellowish and reddish-brown siliceous limestone, in part
crystalline.
2. Fissile sandy limestone.
3. Soft yellow ochreous marly limestone.
4. Greyish-white chalky limestone, with ferruginous layers.
5. Sand-covered slope, mainly consisting of clays.
Baharia-
The road from Baharia to Farafra, traversed by
Farafra Cailliaud in 1820,[26] and by Jordan in 1874[27], was
road. taken by a party of the Geological Survey in proceeding
to Farafra after surveying the west side of Baharia. The start was
made from the same point as the return traverse to Minia, viz., the
point of ascent of the Farafra road at the south scarp of Baharia (lat.
27° 48′ 13″ N., long. 28° 32′ 49″ E. of Greenwich), and the general
course taken was in a direction 30°-40° west of south. A second,
almost disused, road from Baharia to Farafra ascends the western