Journal of Geodesy (1998) 72: 71±100
Generalized inverses of nonlinear mappings
and the nonlinear geodetic datum problem
A. Dermanis
Department of Geodesy and Surveying, The Aristotle University of Thessaloniki, University Box 503, GR-54006 Thessaloniki, Greece
Phone: +30 31 996111; fax: +30 31 996137; email: dermanis@topo.auth.gr
Received: 11 April 1996 / Accepted: 19 April 1997
Abstract. Motivated by the existing theory of the
geometric characteristics of linear generalized inverses
of linear mappings, an attempt is made to establish a
corresponding mathematical theory for nonlinear gen-
eralized inverses of nonlinear mappings in ®nite-
dimensional spaces. The theory relies on the concept
of ®berings consisting of disjoint manifolds (®bers) in
which the domain and range spaces of the mappings are
partitioned. Fiberings replace the quotient spaces gen-
erated by some characteristic subspaces in the linear
case. In addition to the simple generalized inverse, the
minimum-distance and the x0 -nearest generalized inverse
are introduced and characterized, in analogy with the
least-squares and the minimum-norm generalized in-
verses of the linear case. The theory is specialized to the
geodetic mapping from network coordinates to observ-
ables and the nonlinear transformations (Baarda's S-
transformations) between di€erent solutions are de®ned
with the help of transformation parameters obtained
from the solution of nonlinear equations. In particular,
the transformations from any solution to an x0 -nearest
solution (corresponding to Meissl's inner solution) are
given for two- and three-dimensional networks for both
the similarity and the rigid transformation case. Finally
the nonlinear theory is specialized to the linear case with
the help of the singular-value decomposition and
algebraic expressions with speci®c geometric meaning
are given for all possible types of generalized inverses.
Key words. Generalised inverse á Nonlinearity á
Pseudoinverse á Inverse problems á Datum problem
1 Introduction
The datum problem or zero-order design problem
(Grafarend 1974) arising in the adjustment of observa-
tions related to geodetic networks has received consid-
erable attention since the pioneering work of Meissl
(1965, 1969) and its popularization by Blaha (1971). The
nature of the problem has been clari®ed in two
important papers by Grafarend and Scha€rin (1974,
1976) while the relation among various solutions, as well
as Meissl's ``inner'' solution have been established with
the introduction of the famous Baarda S-transformation
(Baarda 1973; see also Mierlo 1980; Koch 1982). It
might therefore seem that every aspect of this problem
has been fully investigated and well understood for quite
some time now, though some recent work (Xu 1995)
might point to the contrary. However, with the excep-
tion of some simple cases, such as that of levelling in a
small area, the mathematical treatment of the problem is
con®ned to its linearized version, although it is well
understood as a nonlinear problem. Nonlinear adjust-
ment has been extensively studied in the geodetic
literature, especially from the viewpoint of nonlinear
least squares, see, e.g., Teunissen (1985, 1989a, 1989b,
1990), Grafarend and Scha€rin (1991) and Lohse
(1994), where further references can be found. A
completely di€erent approach is that of Dermanis and
Sanso (1995), where optimal nonlinear estimators have
been investigated from a strictly probabilistic point of
view. In both approaches the datum problem is solved in
an implicit way. In the nonlinear least-squares adjust-
ment solved by an iteration scheme, the choice of datum
depends on the choice of the initial parameter values
used for starting the iteration process, as well as on the
principle used for ``improving'' these parameters. In the
nonlinear estimation case, the need to adopt a Bayesian
point of view in order to obtain a meaningful estimation
independent of the unknown ``true'' parameter values
also solves the datum problem in an implicit way, where
the datum choice is hidden in the choice of the prior
probability distribution of the parameters.
The solution to the nonlinear datum problem pre-
sented here is based on the concept of the S-transfor-
mation and has the form of such a similarity (or rigid)
transformation with parameters which come from the
solution of a system of nonlinear equations.
Another important aspect of the geodetic datum
problem in its linear form is its relation to generalized
inverses of matrices (linear operators) which led Bjer-
72
hammar (1951) to an independent introduction of the
Moore-Penrose generalized inverse, later than Moore
(1920) but before Penrose (1955). The question arises
whether the nonlinear version of the geodetic datum
problem bears a similar relation to some type of gen-
eralized inverses of nonlinear operators. An attempt will
be made to look mainly into the geometric aspects of
such nonlinear generalized inverses, although the
building of a concrete mathematical theory requires a
more rigorous treatment which is beyond our present
scope. We shall base our investigation on the represen-
tation theory of various types of linear generalized in-
verses which has been introduced by Takos (1976) from
an algebraic point of view and especially by Teunissen
(1985) from a geometric point of view.
The datum problem is always a part of the problem of
the adjustment of redundant observations which are re-
lated to a set of parameters (coordinates in the geodetic
case) which in fact cannot be determined from the
available observations. The reason is that the informa-
tion contained in the angle and distance (angle) obser-
vations relates only to the shape and size (shape) of the
network, while coordinates relate in addition to its posi-
tion (position and size) with respect to a certain reference
frame. The problem of placing the network in relation to
a given reference frame can be also seen from the inverse
point of view of placing a reference frame in relation to a
given (i.e., physically existing) network. This choice of
reference frame (or datum in geodetic terminology) poses
a problem, the datum problem, which must be solved in an
arbitrary but consistent way based in the introduction of
additional information not contained in the observations.
The usual approach to the description of the adjust-
ment and choice of datum problem is to consider the n
observed parameters (n > r) as the coordinates of a
linear space Y , called the observation space, in which the
r-dimensional manifold M modeling the physical sys-
tem, called the model manifold, is lying. (We con®ne
ourselves to the case of discrete and ®nite observations.)
The model manifold is described as the range of a
nonlinear operator f from an m-dimensional parameter
space X (m > r) into the observation space Y . The
mapping f is established by the mathematical equations
which relate all observables to the unknown parameters.
In a ``normal'' situation, which is almost never the case
in geodesy, the number of parameters and the dimension
of the model manifold are equal (m ˆ r), in which case
the restriction fjM of f to M has an inverse which may
serve as a coordinate mapping from M to X ˆ Rm . In
other words the chosen parameters can serve as a par-
ticular system of coordinates on the model manifold and
the only problem to be solved is the adjustment prob-
lem. In the linear case the rank of f is r ˆ m and the
corresponding model is called a full-rank model.
The unavoidable observational errors are added to
the true values of the observables (which correspond to
a point on the model manifold) and give observations
corresponding to a known point outside the manifold.
The adjustment problem can be simply de®ned as the
problem of ®nding an optimal way to ``return to the
model manifold''. In the case where m > r (model with-
out full rank in the linear case), the determination of
parameter values is not trivial anymore, because there is
an in®nite set of parameter values which f maps on the
same manifold point corresponding to the adjusted ob-
servations. The datum problem is exactly the problem of
choosing one out of all possible parameter sets.
Of course rank de®ciency in an observational model
is not exclusively related to the datum problem, as
demonstrated, e.g., in Dermanis and Grafarend (1981).
Additional rank de®ciency may result from the inability
of the available observations to recover the shape (or
shape and size) of a geodetic network. Such cases, al-
though partly covered by the treatment in Sect. 3, need
individual treatment and cannot be part of the general
approach taken here, where our main subject is the
common datum defect resulting only from the use of
coordinates as parameters.
We must point out that the above point of view is not
coordinate-free, since it depends on the choice of a
speci®c set of m unknown parameters. This is perhaps of
little concern to geodesy where the choice of coordinates
as parameters imposes itself as a matter of convenience.
It is possible, however, to establish a theory where both
the adjustment and the datum problem are treated in a
coordinate-free way in the spirit of modern di€erential
geometry.
A more in-depth introduction to the (nonlinear) da-
tum problem, especially in relation to the modelling
problem, can be found in Dermanis (1991).
2 Geometric characteristics of generalized
inverses of linear operators
A natural point of departure for the study of the
nonlinear datum problem is our knowledge of the
simpler linear case. For this reason we shall review some
of the geometric characteristics of the theory of
generalized inverses of linear mappings and point out
those which have proven to be more appropriate for
generalization to the nonlinear case. Our exposition will
be more casual than in the rest of this work, since a
rigorous and thorough study exists already in the
geodetic literature (Teunissen 1985).
A linear mapping f : X ! Y is characterized geo-
metrically by its range
R…f † ˆ fy 2 Y jy ˆ f …x† for some x 2 X g
which is a linear subspace of Y and its null space
N …f † ˆ fx 2 X jf …x† ˆ 0g
which is a linear subspace of X . To any element y ˆ f …x†
of R…f † corresponds an ane subspace x ‡ N …f † of X
with elements which f maps to the same element
y ˆ f …x†. The parallel translates of N …f † are thus the
``solution spaces'' of f which correspond one-to-one to
the elements of R…f †.
If X is m-dimensional, Y is n-dimensional and rank
…f † ˆ dim R…f † ˆ r then d ˆ m ÿ r is the injectivity de-
fect of f , while f ˆ n ÿ r is its surjectivity defect (usually

called degrees of freedom). Furthermore N …f † is a d-
dimensional subspace of X and R…f † is an r-dimensional
subspace of Y .
Coming to a linear generalized inverse g of f , it fol-
lows from the de®ning property f ˆ f  g  f that both
the linear operators p ˆ f  g and q ˆ g  f , are idem-
potent and thus linear projectors. As such they are
completely characterized by their range (invariant linear
subspace) and their complementary null-space (linear
subspace along which they project), see e.g. Halmos
(1974, Sect. 41), Scha€rin et al. (1977).
We shall ®rst show that R…p† ˆ R…f †. For every
y 2 Y ; y^  p…y† ˆ f …g…y†† 2 R…f †, implying that R…p†
 R…f †. On the other hand, for any y^ 2 R…f † there exists
x 2 X such that y^ ˆ f …x† and p…^ y † ˆ …f  g†…f …x††
ˆ …f  g  f †…x† ˆ f …x† ˆ y^, implying that R…f †  R…p†.
Therefore R…p† ˆ R…f †.
R…p† ˆ R…f † is an r-dimensional (linear) subspace of
Y . The linear projector p is completely determined if also
the f -dimensional subspace C  Y is given such that
Y ˆ R…f †  C and y ÿ p…y† 2 C for every y 2 Y . Thus p
is a linear projection on R…f † along C.
The elements of C are projected by p to the zero
element of Y , i.e., C ˆ N …p† ˆ N …f  g†.
Indeed for any y 2 C it must hold that
y^ ˆ p…y† 2 R…f † while y ÿ y^ 2 C. But C is a linear sub-
space, and if y 2 C and y ÿ y^ 2 C so is their di€erence,
i.e., y ÿ …y ÿ y^† ˆ y^ 2 C.
Now y^ 2 C and y^ 2 R…f †, while C \ R…f † ˆ f0g, and
therefore y^ ˆ 0.
From the fundamental property f ˆ f  g  f and the
well-known property for the rank of matrix product
(mapping composition) it follows that rg ˆ rank…g†
cannot be less than the rank r of f , i.e., rg  r. N …g† is
thus a subspace of Y with dimension fg ˆ n ÿ rg
 n ÿ r ˆ f.
We shall show that in fact N …g†  C: if y 2 N …g† then
g…y† ˆ 0 and p…y† ˆ f …g…y†† ˆ f …0† ˆ 0, i.e., also
y 2 N …p† ˆ C and therefore N …g†  C.
For every y^ 2 R…f † the ane subspace y^ ‡ C consists
of all elements of Y which p maps on the same element y^.
The elements of any particular ane subspace y ‡ N …g†
are mapped on the same element x ˆ g…y† 2 R…g†  X .
Further mapping of this particular element x by f to
y^ ˆ f …x† has as a consequence that all elements of
y ‡ N …g† are mapped by p ˆ f  g on the same element
y^ ˆ p…y†. Therefore y ‡ N …g†  p…y† ‡ C ˆ y ‡ C, and
furthermore N …g†  C as already shown. A similar sit-
uation holds for the linear projector q. For any
x 2 X ; x^ ˆ q…x† ˆ …g  f †…x† 2 g…R…f ††  R…g†.
The subspace S ˆ R…q† ˆ g…R…f ††  R…g† is the sub-
space onto which q maps elements of X . If two elements
x1 , x2 of X are mapped onto the same element x^ 2 S,
then x^ ˆ q…x1 † ˆ q…x2 † implies that g…f …x1 † ÿ f …x2 †† ˆ 0
and f …x1 † ÿ f …x2 † 2 N …g†  C. But due to Y ˆ R…f †  C
we have R…f † \ C ˆ f0g and R…f † \ N …g† ˆ f0g. Since
f …x1 † ÿ f …x2 † ˆ f …x1 ÿ x2 † belongs to both R…f † and
N …g† it must hold that f …x1 † ÿ f …x2 † ˆ 0 and
x1 ÿ x2 2 N …f †.
Conversely, if x1 ÿ x2 2 N …f † then f …x1 † ÿ f …x2 † ˆ
f …x1 ÿ x2 † ˆ 0, implying q…x1 † ÿ q…x2 † ˆ g…f …x1 ÿ x2 †† ˆ
73
Fig. 1. The geometry of the generalized inverse g of a linear mapping
f . Here x ˆ g…y†; y^ ˆ f …x† ˆ p…y†, while x^ ˆ g…^
y † ˆ q…x†. All the
elements of y^ ‡ N …g† are mapped into x, while all the elements of
y ‡ C are mapped into x^
g…0† ˆ 0 and all elements of any ane subspace
x ‡ N …f † are projected by q on the same element x^.
Therefore q projects elements of X on S ˆ g…R…f ††
along N …f † and X ˆ N …f †  S. Since N …f † is d-dimen-
sional, S will be an r-dimensional subspace of X .
An important property of S is that it intersects any
solution space (ane subspace) x ‡ N …f † of f in a single
element.
To see this let both x^1 and x^2 belong to S \ ‰x ‡ N …f †Š.
Since x^1 2 ‰x ‡ N …f †Š; x ÿ x^1 2 N …f †; f …x ÿ x^1 † ˆ 0 and
q…x ÿ x^1 † ˆ g…f …x ÿ x^1 †† ˆ g…0† ˆ 0, implying that
q…^x1 † ˆ q…x†. Since x^1 2 S it holds that q…^ x1 † ˆ x^1 and
thus x^1 ˆ q…x†. With similar reasoning x^2 ˆ q…x† and
therefore x^1 ˆ x^2 .
R…f † and S are subspaces of the same dimension
which are in a one-to-one correspondence. If y^ 2 R…f †
then y^ ˆ f …x† for some x 2 X and g…^ y † ˆ g…f …x††
ˆ q…x†  x^ 2 S. Also x^ 2 ‰x ‡ N …f †Š and x^ ˆ g…^ y † ˆ q…x†
is the unique element in the intersection of x ‡ N …f † and
xg ˆ ‰x ‡ N …f †Š \ S. For any element x^ 2 S there
S, i.e., f^
corresponds a unique element y^ 2 R…f † such that
y^ ˆ f …^x† and x^ ˆ g…^y †. We may say that the restriction
of g to R…f † is the ordinary inverse of the restriction of f
on S: gjR…f † ˆ …fjS †ÿ1 . In symmetry with S ˆ g…R…f †† it
holds also that R…f † ˆ f …S†.
Once S is given g is de®ned on R…f †. It remains to be
de®ned outside R…f †. Let y 2 R…f † and y^ ˆ p…y†. Then
y ÿ y^ 2 C. Taking into account that N …g†  C we can
decompose
y ÿ y^ ˆ …y ÿ y^†N ‡ …y ÿ y^†C 0
into a part …y ÿ y^†N 2 N …g† and a part …y ÿ y^†C0 2 C 0
where C 0 is a complement of N …g† with respect to
C : C ˆ N …g†  C 0 . This decomposition is not unique but
depends on the speci®c choice of the subspace C 0  C
74

which is of dimension dimC 0 ˆ dimC ÿ dimN …g† ˆ

f ÿ …n ÿ rg † ˆ n ÿ r ÿ …n ÿ rg † ˆ rg ÿ r  Dr.
On the basis of the decomposition
y ˆ y^ ‡ …y ÿ y^† ˆ y^ ‡ …y ÿ y^†N ‡ …y ÿ y^†C0
and the linearity of g we have:
y † ‡ g……y ÿ y^†N † ‡ g……y ÿ y^†C0 †
g…y† ˆ g…^
ˆ g…^ y † ‡ g……y ÿ y^†C 0 †:
The g…^ y † part is uniquely de®ned: the subspaces R…f †
and C determine y^ and g…^ y † is the unique element in the
intersection of S with the ane subspace x ‡ N …f † which
f maps into y^ (the solution space of y^). It remains to
determine how g acts on the subspace C 0 . This is the only
required information to de®ne completely the action of g
in addition to a description of S and C. If a basis is
known for N …g† and a basis for C 0 they constitute
together a basis of C which is thus de®ned. The
generalized inverse g is speci®ed if (in addition to a
basis for S) we know how it acts on all the rg ÿ r
elements of the basis of C 0 .
A unique choice of C 0 can be speci®ed without am-
biguity (if so desired) by requiring in addition that C 0 is
the orthogonal complement of N …G† with respect to C,
in which case C assumes an orthogonal decomposition
C ˆ N …g† ? C 0 .
Let D ˆ g…C† be the image of C under g. If y 2 C we
have the unique decomposition y ˆ yN ‡ yC0 where
yN 2 N …G† and yC 0 2 C 0 . In this case g…y† ˆ g…yN †‡
g…yC 0 † ˆ g…yC0 † and g…C† ˆ g…C 0 † ˆ D. If y 2 C ˆ N …p†
Fig. 2. More geometry of the generalized inverse of a linear mapping:
then p…y† ˆ 0 ˆ …f  g†…y† ˆ f …g…y†† and g…y† 2 N …f † so the intersection of y ‡ R…f † and y ‡ C 0 determines the unique point y
that g…C† ˆ D  N …f †. which g (the generalized inverse of f ) maps into x ˆ g…y†. The image
The generalized inverse g is a one-to-one mapping g…y† is uniquely determined from the intersection of g…y† ‡ S
from C 0 to D and dim D ˆ dim C 0 ˆ Dr  d ˆ dim N …f †, (=image of y ‡ C 0 ) and g…y† ‡ D (=image of y ‡ R…f † under C)
the last inequality following from the fact that
D  N …f †.
The range R…g† ˆ g…Y † can be decomposed according C…y ‡ R…f †† then g…y† is determined from the inter-
to Y ˆ R…f †  C as R…g† ˆ g…R…f ††  g…C†, and since section
g…R…f †† ˆ S and g…C† ˆ D, we have that R…g† ˆ S  D. ‰g…yR † ‡ DŠ \ C…y ‡ R…f †† ˆ fg…y†g
We introduce now an alternative perspective to the
geometry of a generalized inverse which might seem at In order to specify a unique g out of the class of all
®rst to be unnecessarily complicated, but it will prove possible generalized inverses of f , we must specify the
helpful in the study of the nonlinear case. subspaces C complementary to R…f †, S complementary
To de®ne g on Y it is sucient to de®ne it on a to N …f † and D  N …f †. Furthermore we must specify the
subspace M complementary to N …g†, since for restricted linear mapping gjC 0 : C 0 ! g…C 0 † ˆ D which
Y ˆ N …g†  M we have a unique decomposition brings the elements of C 0 into a one-to-one correspon-
y ˆ yN ‡ yM with yN 2 N …g†, yM 2 M and g…y† ˆ g…yM †. dence with those of D.
Such a subspace M is provided by M ˆ C 0  R…f †. If Let us look next into the special case where g is a
y 2 M with unique decomposition y ˆ yC 0 ‡ yR , re¯exive generalized inverse, i.e., when g  f  g ˆ g
yC0 2 C 0 ; yR 2 R…f † then g…y† ˆ g…yC 0 † ‡ g…yR †, where also. In this particular case rg ˆ r necessarily …rg  r
g…yC 0 † 2 D and g…yR † 2 S. The ane subspace y ‡ C 0 ˆ follows from the fundamental property and rg  r from
yR ‡ C 0 of M is mapped into g…yR † ‡ D ˆ g…y† ‡ D which the re¯exivity property). It follows that S ˆ R…g† and
is an ane subspace of R…g†. The ane subspace C ˆ N …g†. Thus p ˆ f  g is a projector on R…f † along
y ‡ R…f † of M is mapped into an ane subspace N …g† and q ˆ g  f is a projector on S ˆ R…g† along
g…y† ‡ S ˆ g…yC 0 † ‡ S of R…g†. The image g…y† is the N …f †. Let us notice that a re¯exive generalized inverse is
unique intersection of these two ane subspaces of R…g†. completely determined once the subspaces C and S are
Since g…y† ‡ D ˆ g…yR † ‡ D is uniquely determined from known.
the yR component of y, it remains to specify a corre- A least-squares generalized inverse is one with
spondence C from the ane subspaces y ‡ R…f † of M to C ˆ R…f †? , in which case we have the orthogonal
the ane subspaces g…y† ‡ S of R…f †. If g…y† ‡ S ˆ decomposition Y ˆ R…f † ? R…f †, and y^ ˆ p…y† is the

closest element to y from R…f †. The projector p is the
orthogonal projector on R…f †.
A minimum-norm generalized inverse is one for
which S ˆ N …f †? , in which case g…y† is the element of
the solution ane subspace corresponding to y^ ˆ p…y†
which has the smallest norm. The projector q is the or-
thogonal projector on N …f †? .
From all the preceding geometric characteristics of a
generalized inverse g : Y ! X of a linear mapping
f : X ! Y , we summarize those which will be helpful in
dealing with nonlinear mappings.
The space Y is ``sliced'' by p ˆ f  g into ane sub-
spaces (p-slices) which are parallel to a speci®c (linear)
subspace C. Each slice corresponds to a particular ele-
ment y^ of R…f †  Y , since p ˆ f  g maps all the ele-
ments of the slice onto that y^.
The mapping f ``slices'' the space X into ane sub-
spaces (f -slices) which are parallel to the subspace N …f †.
Each slice corresponds to a particular element y^ of R…f †
since f maps all the elements of the slice on that y^, we
may therefore call this slice ``solution space of y^''.
The generalized inverse g maps R…f † to a subset S of
R…g† which has the property that it intersects each so-
lution space (f -slice) in a single element.
The mapping q ˆ g  f slices X into q-slices, each
slice corresponding to a particular element x^ of S, since q
maps all the elements of the slice on that x^. The q-slices
are identical to the f -slices.
Finally g ``slices'' Y into ane subspaces which are
parallel to the subspace N …g†. These g-slices have the
property that each is contained in one of the p-slices.
Thus each p-slice is itself further sliced in a more ``re-
®ned'' set of g-slices. Only in the case of a re¯exive
generalized inverse are the p-slices and g-slices
identical.
These slices, all of which are ane subspaces,
can be described by a single ``generating'' linear sub-
space (f-slices by N …f †, g-slices, and q-slices by N …g†, p-
slices by C). Of course, this possibility cannot survive in
the nonlinear case.
3 Generalized inverses of nonlinear operators
Let X ; Y be ®nite-dimensional spaces and f : X ! Y a
(nonlinear) mapping from X to Y . The question arises
whether it is possible to de®ne, under some appropriate
conditions, a class of mappings g : Y ! X which can
serve as generalized inverses of the mapping f . The
similar theory developed for linear mappings may serve
as a guide to a certain extend, although the nonlinearity
of f poses problems which do not allow for a direct
extension of the existing theory.
De®nition. A partition of a set A into disjoint sets At ,
A ˆ [t At it is called a ®bering of A, and its elements At
are called ®bers. Such a partition can arise from an
equivalence relation on A, each ®ber consisting of
equivalent elements of A (Loomis and Sternberg 1980.)
A mapping f de®ned on X gives rise to a ®bering F
of X through the equivalence relation x1  x2 if
75
f …x1 † ˆ f …x2 †. To every element y 2 R…f † in the range
R…f † of f corresponds a ®ber Fy 2 F de®ned by
Fy ˆ fx 2 X jf …x† ˆ yg …1†
The function f gives rise to a bijection
f: R…f † ! F : y ! Fy
Obviously for every x 2 X ; x 2 Ff …x† .
De®nition. The mapping p : X ! F : x ! Ff …x† is called
the projection mapping from X to the ®bering F. It holds
that p ˆ f  f .
When more than one ®berings are involved we denote
p by pF . Thus pF …x† is the unique ®ber from the ®bering
F which passes through a given point x.
De®nition. A section of a ®bering F of X is the range
S ˆ R…s† of a mapping s : F ! X such that p  s ˆ idF ,
where idF is the identity mapping in F.
A section S of a ®bering F intersects any one of its
®bers in a single element: for any x 2 S; S \ pF …x† ˆ fxg.
De®nition. Let F and K be two ®berings of the same set
A. K is called a re®nement of F if every ®ber Kt of K is
a subset of some ®ber Fs of F : 8 Kt 2 K : 9 Fs 2 F such
that Kt  Fs .
If K is a re®nement of F then pK …z†  pF …z† for
any z.
De®nition. Two ®berings F and H of the same space X
are called complementary if every ®ber of F is a section
of H, and vice versa: pF …x† \ pH …x† ˆ fxg; 8 x 2 X .
De®nition. A ®bering F of X induces a ®bering
F0 ˆ FjM on a manifold M  X with ®bers pF0 …x† ˆ
M \ pF …x† for every x 2 M. The ®bering F0 ˆ FjM is
called the restriction of the ®bering F on M.
Example 1. We shall use the simplest possible geodetic
network in order to illustrate some of the results.
Following Grafarend and Scha€rin (1974) we consider a
three-point horizontal network P1 P2 P3 with coordinates
x ˆ ‰x1 y1 x2 y2 x3 y3 ŠT
where all sides a ˆ P2 P3 ; b ˆ P1 P3 ; c ˆ P1 P2 , and all
angles A; B; C at points P1 ; P2 ; P3 , respectively, have been
observed. The observation vector
y ˆ ‰a b c A B CŠT
is a function y ˆ f …x† of the coordinates, explicitly given
by
q
a ˆ …x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2
q
b ˆ …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2
q
c ˆ …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
76

…x3 ÿ x1 † …x2 ÿ x1 † These three conditions reduce the dimension of Fy to

A ˆ arctan ÿ arctan 6 ± 3 ˆ 3. If x0 is a ®xed element of X the solution space
…y3 ÿ y1 † …y2 ÿ y1 †
…x2 ÿ x1 † …x3 ÿ x2 † through x0 is the ®ber pF …x0 † and x 2 pF …x0 † if it
B ˆ arctan ÿ arctan satis®es the three conditions
…y2 ÿ y1 † …y3 ÿ y2 †
…x3 ÿ x2 † …x3 ÿ x1 † …x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 ˆ …x03 ÿ x02 †2 ‡ …y03 ÿ y02 †2
C ˆ arctan ÿ arctan
…y3 ÿ y2 † …y3 ÿ y1 † …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2 ˆ …x03 ÿ x01 †2 ‡ …y03 ÿ y01 †2
or …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 ˆ …x02 ÿ x01 †2 ‡ …y02 ÿ y01 †2
…x3 ÿ x1 †…x2 ÿ x1 † ‡ …y3 ÿ y1 †…y2 ÿ y1 † which correspond to f …x† ˆ f …x0 † expressed by a ˆ a0 ;
A ˆ arcos qq
…x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 b ˆ b0 ; c ˆ c0 . An alternative equivalent set of condi-
tions is
…x3 ÿ x2 †…x1 ÿ x2 † ‡ …y3 ÿ y2 †…y1 ÿ y2 †
B ˆ arcos qq bc cos A ˆ b0 c0 cos A0
…x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 …x1 ÿ x2 †2 ‡ …y1 ÿ y2 †2
ac cos B ˆ a0 c0 cos B0
…x3 ÿ x2 †…x3 ÿ x1 † ‡ …y3 ÿ y2 †…y3 ÿ y1 † ab cos C ˆ a0 b0 cos C0
C ˆ arcos qq
…x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2
which in terms of coordinates are
We now consider the range R… f †. When y ˆ f …x† for …x3 ÿ x1 †…x2 ÿ x1 † ‡ …y3 ÿ y1 †…y2 ÿ y1 †
some x, i.e., when y 2 R…f †, then the six elements of y are
not independent. Only three parameters, say a; b; c; are ˆ …x03 ÿ x01 †…x02 ÿ x01 † ‡ …y03 ÿ y01 †…y02 ÿ y01 †
independent. The remaining ones, the angles, are …x2 ÿ x1 †…x3 ÿ x1 † ‡ …y2 ÿ y1 †…y3 ÿ y1 †
uniquely determined from well-known trigonometric ˆ …x02 ÿ x01 †…x03 ÿ x01 † ‡ …y02 ÿ y01 †…y03 ÿ y01 †
relations, e.g.,
…x3 ÿ x2 †…x1 ÿ x2 † ‡ …y3 ÿ y2 †…y1 ÿ y2 †
b2 ‡ c2 ÿ a2 ˆ …x03 ÿ x02 †…x01 ÿ x02 † ‡ …y03 ÿ y02 †…y01 ÿ y02 †
cos A ˆ ;
2bc
a2 ‡ c2 ÿ b2 We now return to the theory. Instead of speaking of
cos B ˆ ; the ®bering F of X induced by a certain mapping f , we
2ac
may take the inverse approach and describe a given
a2 ‡ b2 ÿ c2 ®bering by means of a mapping. For a mapping to in-
cos C ˆ
2ab duce a ®bering on its domain space it is necessary that it
This means that R…f † is of dimension 6 ± 3 ˆ 3, since is not injective, otherwise we obtain the ``trivial ®bering''
the three cosine conditions satis®ed by its elements where each ®ber consists of a single element of X !
reduce the number of independent parameters by three. The use of a mapping which, as our mapping of in-
We may think of q ˆ ‰abcŠT as three curvilinear terest f , is not surjective, is not the most ecient way for
coordinates on R…f † which in this case obtains the describing a ®bering. More straightforward is the use of
direct representation y ˆ y…q†, explicitly given by a surjective mapping, say h : X ! Rq , where to every
element d 2 Rq corresponds a ®ber Hd ˆ fx 2 X jh…x†
a ˆ a; b ˆ b; cˆc ˆ dg from the ®bering H induced by h. The ®bers Hd
are manifolds of dimension m ÿ q ˆ dim X ÿ q. A more
general way of describing a ®bering is by means of a
b2 ‡ c2 ÿ a2
A ˆ arcos ; mapping
2bc
a2 ‡ c2 ÿ b2 v : X  Rq ! Rq : …x; d† ! v…x; d†
B ˆ arcos ;
2ac with q < m, such that vd …
†  v…
; d†  is surjective for
a2 ‡ b2 ÿ c2 every d 2 Rq . This is guaranteed when od
ov
6ˆ 0 at every x
C ˆ arcos
2ab and d. The ®bers are the (m ÿ q)-dimensional manifolds
We now move on to the solution space Fy for a ®xed Hd ˆ fx 2 X jv…x; d† ˆ 0g
y 2 R…f †; x 2 Fy whenever f …x† ˆ y. Since A; B; C are
determined from a; b; c; the only independent conditions The previous case corresponds to the choice
on x are those related to the given values of a; b; c : v…x; d† ˆ h…x† ÿ d:
Another approach to the description of ®berings of a
q space X is by means of mappings having X as their range
a ˆ …x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 rather than their domain as before. A bijective mapping
q
b ˆ …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2 / : Rq  Rmÿq ! X : …d; p† ! /…d; p†
q
describes a ®bering H with ®bers Hd ˆ fx 2 X jx
c ˆ …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 ˆ /…d; x†g. Each ®ber is the range Hd ˆ R…/d † of the
 77

mapping /d …
† ˆ /…d;
†. The inverse mapping /ÿ1 is in
fact a coordinate mapping on X where the coordinates
…d; p† ˆ /ÿ1 …x† are ``adapted'' to the ®bering: each ®ber
corresponds to a ®xed value of d, while the ``free''
parameters p may serve as (intrinsic) coordinates on the
®ber.
We assume for the moment that f is de®ned on the
whole of X , it is continuous, and has a di€erential
mapping dfx at each x which is a linear mapping from
the tangent vector space Tx at x to the tangent vector
space Tf …x† at f …x† (Choquet-Bruhat et al. 1977, p. 121).
The mapping dfx is de®ned by
dfx : Tx ! Tf …x† : v ! u
…2†
u…f  /† ˆ v…/† for every f : X ! R
(where the tangent vectors v and u are visualized as
``directional derivatives'' acting on functions).
We further assume that dfx has constant rank over X ,
rank…dfx † ˆ r  min…n; m†; where m ˆ dim X and
n ˆ dim Y . Consequently R…f † is an r-dimensional sub-
manifold of Y .
We now come to the possibility of de®ning a gener-
alized inverse g of the mapping f by a similar way as in
the linear case.
De®nition. A mapping g : Y ! X is called a generalized
inverse of a given mapping f : X ! Y when
f gf ˆf …3† …G1†
[We denote Eq. (3) by (G1).] This means that for every
x 2 X ; …f  g  f †…x† ˆ f …x†, so that for every y ˆ f …x†
2 R…f † it holds that …f  g†…y† ˆ f …g…y†† ˆ y. As a
consequence g…y† 2 Fy , i.e., g must map every element
of R…f † into an element of its corresponding ®ber. In
other words g maps R…f † onto a section S of the ®bering
F where S  R…g†. The restriction of g on R…f † is a
bijection between R…f † and S which are both r-dimen-
sional manifolds.
The relation (G1) satis®ed by the generalized inverse
g implies a corresponding relation between the di€er-
entials dfx and dgf …x† of f and g respectively:
dfx  dgf …x†  dfx ˆ dfx …4†
which follows by implementing the implicit function
theorem (Choquet-Bruhat et al. 1977, p. 91).
r  rg  min…n; m† …5†
Once S is given, g is de®ned on R…f †  Y : The question
is how this particular g is extended outside R…f †. A step
in this direction is to note that as a consequence of (G1)
both q ˆ g  f and p ˆ f  g are idempotent mappings
…q2 ˆ q; p2 ˆ p†
f  g  f ˆ f ) …g  f †  …g  f † ˆ g  f and
…f  g†  …f  g† ˆ f  g …6†
which can be considered as ``nonlinear projections''
q ˆ g  f : X ! R…g  f †  R…g†  X
p ˆ f  g : Y ! R…f  g† ˆ R…f †  Y …7†
We have set R…g  f †  R…g†, which is obvious, and
R…f  g† ˆ R…f † according to the previous lemma.
Elements belonging to the range of idempotent
mappings are invariant under the mapping. Idempotent
mappings induce also a ®bering of the space they act on.
We denote by P, Q the ®berings induced by p and q,
respectively, with corresponding elements Py ;y 2 R…p†
ˆ R…f † and Qx ;x 2 R…q†  R…g†.
The mapping g gives rise to a ®bering G of Y with
elements Gx ˆ fy 2 Y jg…y† ˆ xg. Obviously for every
element y of Y it holds that y 2 Gg…y† ˆ pG …y†. If
y 2 R…f † then y 2 Gx ˆ pG …f …x††, where fxg ˆ Fy \ S:
Lemma. For the ®berings of the space Y it holds that the
®bering G induced by g is a re®nement of the ®bering P
induced by p ˆ f  g.
Proof. For any z 2 Y ; pG …z† 2 G and pP …z† 2 P. For
every y 2 pG …z†; g…y† ˆ g…z† ) p…y† ˆ f …g…y†† ˆ f …g…z†† ˆ
p…z† and y 2 pP …z†: Consequently pG …z†  pP …z† for any
z 2 Y and thus G is a re®nement of P. (

Lemma. R…f  g† ˆ R…f †.

Proof. y 2 R…f † ) 9 x 2 X:
y ˆ f …x† ˆ …f  g  f †…x† ˆ …f  g†‰f …x†Š
ˆ …f  g†…y† 2 R…f  g† ) R…f †  R…f  g†
which combined with the obvious relation
R…f  g†  R…f † implies that in fact R…f  g† ˆ R…f †: (
Assuming that g has a constant rank rg ˆ rank…dgy † Fig. 3. The geometry of the generalized inverse g of a nonlinear
for every y, application of the property of the rank of a mapping f . All elements of pG …y† are mapped into the same element
linear map composition r…AB†  min‰r…A†; r…B†Š to g…y†. The elements of pP …y† are mapped into elements of pF …x† and
Eq. (3), gives r  min…r; rg †  rg so that thus projected by p on the same y^
78

Example 1 (continued). We claim that the mapping g ®xed y 2 R…f † must simultaneously satisfy the condi-
de®ned by x ˆ g…y†, explicitly tions for x 2 Fy

c ÿ b sin A a ÿ c ÿ b cos A …x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 ˆ a2

x1 ˆ y1 ˆ
3 3 …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2 ˆ b2
c ÿ b sin A a ‡ 2c ÿ b cos A …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 ˆ c2
x2 ˆ y2 ˆ
3 3
and the conditions for x 2 R…g†
c ‡ 2b sin A a ÿ c ‡ 2b cos A x1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ y 2 ÿ y 1
x3 ˆ y3 ˆ
3 3
The combination gives after some manipulation the ®ve
is a generalized inverse of the previously de®ned conditions
mapping f . To prove this we need to show that, for
any x; y ˆ f …x† is identical to y 0 ˆ …f  g  f †…x† ˆ x1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ c; y2 ÿ y1 ˆ c
…f  g†…y† ˆ f …x0 †; where x0 ˆ g…y† is explicitly given by
c ÿ b sin A a ÿ c ÿ b cos A y3 ÿ y1 ˆ b cos A; x3 ÿ x1 ˆ b sin A
x01 ˆ y10 ˆ
3 3
which are those satis®ed by any x 2 Fy0 ˆ Fy \ R…g†:
c ÿ b sin A a ‡ 2c ÿ b cos A The dimension of Fy0 is 6±5 ˆ 1. The ®bers Fy as y runs
x02 ˆ y20 ˆ through R…f † constitute a ®bering F of X , while the
3 3 ®bers Fy0 constitute the ®bering F0 induced by F on
R…g†:
c ‡ 2b sin A a ÿ c ‡ 2b cos A We now move on to the section S ˆ g…R…f ††. When
x03 ˆ y30 ˆ
3 3 y 2 R…f †; A; B; C are determined from a; b; c and so too
Since for an arbitrary x; y ˆ f …x† 2 R…f †, the angle A x ˆ g…y†. From the de®nition of g and the fact that
satis®es the cosine condition and thus y 0 ˆ f …x0 † be- p
comes b2 ‡ c2 ÿ a2 4b2 c2 ÿ …b2 ‡ c2 ÿ a2 †
cos A ˆ ; sin A ˆ
qÿ 2 bc 2 bc

2 ÿ
2
a0 ˆ x03 ÿ x02 ‡ y30 ÿ y20 ˆ b2 ‡ c2 ÿ 2 b c cos A ˆ a
x becomes explicitly
qÿ q

2 ÿ
2
b0 ˆ x03 ÿ x01 ‡ y30 ÿ y10 ˆ b c 4b2 c2 ÿ …b2 ‡ c2 ÿ a2 †2 a c b2 ÿ a2
x1 ˆ ÿ ; y1 ˆ ÿ ÿ
3 6c 3 2 6c
q
ÿ 0
2 ÿ
2
c0 ˆ x2 ÿ x01 ‡ y20 ÿ y10 ˆ c q
c 4b2 c2 ÿ …b2 ‡ c2 ÿ a2 †2 a c b2 ÿ a2
x2 ˆ ÿ ; y2 ˆ ‡ ÿ
Since y 0 2 R…f †, the angles A0 ; B0 ; C 0 are determined from 3 6c 3 2 6c
a0 ˆ a; b0 ˆ b; c0 ˆ c through the cosine conditions and q
are therefore the same as those in y, i.e.,
c 4b2 c2 ÿ …b2 ‡ c2 ÿ a2 †2 ; a b2 ÿ a2
A0 ˆ A; B0 ˆ B; C 0 ˆ C. Therefore y 0 ˆ y and g is in fact x3 ˆ ‡ ; y3 ˆ ‡
a generalized inverse of f . 3 3c 3 3c
Turning to the range R…g† of the generalized inverse
g: for an arbitrary y 2 = R…f †; x ˆ g…y† is given from the The elements of x 2 S are functions of three parameters
de®ning equations of g which show that x satis®es at a; b; c: Therefore S has dimension three and it is
least one condition, namely x1 ˆ x2 . The remaining ®ve described by x ˆ x…q† where q ˆ ‰a b cŠT are curvilinear
equations depend on four parameters, a; b; c; A; so there coordinates on S. The section S can also be described by
must be one more condition which results from the three conditions on the coordinates of its points. The
easily derived relations y2 ÿ y1 ˆ c and x1 ‡ x2 ‡ x3 ˆ c: obvious one is x1 ˆ x2 , the second results from
Therefore x 2 R…g† if its elements satisfy x1 ‡ x2 ‡ x3 ˆ c ˆ y2 ÿ y1 . The remaining one comes
from y1 ‡ y2 ‡ y3 ˆ a. Thus x 2 S when
x 1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ y2 ÿ y1
x1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ y 2 ÿ y 1
Thus R…g† has dimension 6±2 ˆ 4. The parameters
q ˆ ‰abcAŠT may serve as curvilinear coordinates for …y1 ‡ y2 ‡ y3 †2 ˆ …x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2
R…g† since the relation x ˆ g…y† degenerates into
x ˆ x…q†: The ®rst two conditions are those satis®ed by the
We now consider the ®bering F0 induced by F on elements of range R…g† and therefore S ˆ g…R…f ††
R…g†. The elements of R…g† which satisfy f …x† ˆ y for a  R…g† ˆ g…Y †.
 79

For the mapping p ˆ f  g and the ®bers pP …y†: set- p T

sin A ˆ 1 ÿ cos2 A ˆ
ting y 0 ˆ p…y† ˆ f …g…y†† ˆ f …x†, where x ˆ g…y†, and bc
using the de®nitions of f and g we obtain the explicit  2 2
2 2 2 b ‡ c2 ÿ a2
representation of p: where T ˆ b c ÿ
p 2
a0 ˆ b2 ‡ c2 ÿ 2 bc cos A Using the explicit form of y ˆ f …x† we evaluate
b0 ˆ b
b2 ‡ c2 ÿ a2
c0 ˆ c ˆ …x3 ÿ x1 †…x2 ÿ x1 † ‡ …y3 ÿ y1 †…y2 ÿ y1 †
2
A0 ˆ A T ˆ …x3 ÿ x1 †…y2 ÿ y1 † ÿ …x2 ÿ x1 †…y3 ÿ y1 †
c ÿ b cos A c ÿ b cos A Using the explicit form for f and the relations for cos A
cos B0 ˆ ˆ p
a0 b2 ‡ c2 ÿ 2 bc cos A and sin A just given we obtain the explicit form of
x0 ˆ f …y†:
b ÿ c cos A b ÿ c cos A  
cos C 0 ˆ ˆ p 0 c T 0 a ÿ c 1 b2 ‡ c2 ÿ a2
a 0 x1 ˆ ÿ ; y 1 ˆ ÿ
b ‡ c2 ÿ 2 bc cos A
2
3 3c 3 3c 2
 2 
If we replace b ˆ b0 ; c ˆ c0 ; A ˆ A0 we obtain the three 0 c T 0 a ‡ 2c 1 b ‡ c2 ÿ a2
x2 ˆ ÿ ; y 2 ˆ ÿ
cosine conditions so that y 0 2 R…f † and in fact 3 3c 3 3c 2
R…p† ˆ R…f †, since y 0 depends on three parameters  2 
c 2T a ÿ c 2 b ‡ c ÿ a2 2
b; c; A, and therefore R…q† has the same dimension, x03 ˆ ‡ ; y30 ˆ ‡
three, as R…f †: 3 3c 3 3c 2
For a ®xed element y0 ˆ ‰a0 b0 c0 A0 B0 C0 ŠT , the ®ber which in terms of the elements of x become
pP …y0 † induced by p through y0 consists of all the ele-
ments y such that p…y† ˆ p…y0 † ˆ y 0 , which means that …x03 ÿ x02 †2 ‡ …y30 ÿ y20 †2 ˆ a2 ˆ …x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2
y 2 pP …y0 † when its elements satisfy the condition …x03 ÿ x01 †2 ‡ …y30 ÿ y10 †2 ˆ b2 ˆ …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2
p…y† ˆ p…y0 ), which in view of the preceding description
of p takes the explicit form …x02 ÿ x01 †2 ‡ …y20 ÿ y10 †2 ˆ c2 ˆ …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
p q Comparing these expressions to the identical ones for Fy ,
b2 ‡ c2 ÿ 2 bc cos A ˆ b20 ‡ c20 ÿ 2 b0 c0 cos A0 we conclude that if x0 2 pQ …x† then also x0 2 Ff …x† and
vice versa. Thus the ®bers of Q are identical to the
b ˆ b0 corresponding solution spaces.
c ˆ c0 In order to express x0 ˆ q…x† 2 pQ …x† as a function of
x we must replace a; c; T and 12 …b2 ‡ c2 ÿ a2 † with their
A ˆ A0 expressions in terms of x and thus obtain
q
c ÿ b cos A c0 ÿ b0 cos A0
p ˆ q 0
…x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
2 2
b ‡ c ÿ 2 bc cos A b20 ‡ c20 ÿ 2 b0 c0 cos A0 x1 ˆ
3
…x3 ÿ x1 †…y2 ÿ y1 † ÿ …x2 ÿ x1 †…y3 ÿ y1 †
b ÿ c cos A b0 ÿ c0 cos A0 ÿ q
p ˆ q
2 2
b ‡ c ÿ 2 bc cos A 3 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
b20 ‡ c20 ÿ 2 b0 c0 cos A0 q q
…x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 ÿ …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
Since the ®rst one and the last two are a direct y10 ˆ
consequence of the remaining three, the only conditions 3
needed so that y 2 pP …y0 † are …x3 ÿ x1 †…x2 ÿ x1 † ‡ …y3 ÿ y1 †…y2 ÿ y1 †
ÿ q
a ˆ any; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ any; C ˆ any 3 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
q
This means that pP …y0 † has dimension 6 ± 3 ˆ 3 and the
…x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
three parameters a; B; C can serve as a system of three 0
x2 ˆ
curvilinear coordinates on pP …y0 †, which assumes the 3
direct representation y ˆ y…q† ˆ y…q; y0 † with explicit …x3 ÿ x1 †…y2 ÿ y1 † ÿ …x2 ÿ x1 †…y3 ÿ y1 †
ÿ q
form
3 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
a ˆ a; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ B; C ˆ C q q
2 2
…x 3 ÿ x 2 † ‡…y 3 ÿ y 2 † ‡ 2 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
For the mapping q ˆ g  f and the ®bers pQ …x† we set y20 ˆ
x0 ˆ q…x† ˆ g…f …x†† ˆ g…y†; where y ˆ f …x† 2 R…f †, and 3
therefore the angles A; B; C are functions of 2 the sides …x3 ÿ x1 †…x2 ÿ x1 † ‡ …y3 ÿ y1 †…y2 ÿ y1 †
2 2 ÿ q
a; b; c: We need in addition to cos A ˆ b ‡c ÿa
the cor-
2bc 3 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
responding relation
80
q
The re®nement of the ®bering P by the ®bering G,
…x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
x03 ˆ i.e., for every ®ber Py 2 P its own ®bering with
3 members Gx corresponding to all x 2 Fy .
2…x3 ÿ x1 †…y2 ÿ y1 † ÿ …x2 ÿ x1 †…y3 ÿ y1 † For a given section M of the ®bering G and every
‡ q
y 2 R…f †, a corresponding mapping
3 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
cy : M \ pP …y† ! R…g† \ Fy
q q
which is subject to the following restriction:
…x3 ÿ x2 †2 ‡…y3 ÿ y2 †2 ÿ …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2
y30 ˆ if pG …y† \ M ˆ fy0 g then cy …y0 † 2 Fy \ S ˆ fg…y†g.
3
2‰…x3 ÿ x1 †…x2 ÿ x1 † ‡ …y3 ÿ y1 †…y2 ÿ y1 †Š Proof. Consider any element z 2 Y . Then the ®bering P
‡ q
3 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 determines a unique ®ber pP …z† containing z, while
y ˆ p…z† 2 R…f † is determined by R…f † \ pP …z† ˆ fyg
From the relations giving x0 in terms of a; b; c it is and pP …y† ˆ pP …z†. Since y uniquely determines Fy ; g…y†
obvious that the range R…q† of q is identical with the is also uniquely determined by Fy \ S ˆ fg…y†g. The
section S. ®bering G determines a unique ®ber pG …z† ˆ Gg…z†
As regards the ®bering G included by g: let containing z and furthermore pG …z† ˆ Gg…z†  pP …z†
y0 ˆ ‰a0 b0 c0 A0 B0 C0 ŠT be a ®xed element and consider ˆ pP …y† due to the fact that G is a re®nement of P.
the ®ber pG …y0 † (through y0 induced by g). For any other Since M is a section of G it holds that pG …z† intersects M
point y 2 pG …y0 † it must hold that g…y† ˆ g…y0 † ˆ x and at a unique element pG …z† \ M ˆ fz0 g. The known
using the explicit form of g we arrive at the conditions mapping cy de®nes an element cy …z0 † 2 R…g† and we
de®ne g…z†  cy …z0 †. It remains to show that the
3x1 ˆ c ÿ b sin A ˆ c0 ÿ b0 sin A0 mapping g de®ned for every z 2 Y by this procedure is
in fact a generalized inverse of f . For an arbitrary x 2 X
3y1 ˆ a ÿ c ÿ b cos A ˆ a0 ÿ c0 ÿ b0 cos A0
consider z ˆ f …x†. Since z 2 R…f † it follows that in this
3x2 ˆ c ÿ b sin A ˆ c0 ÿ b0 sin A0 case y ˆ z, and by the restriction imposed on cy for its
3y2 ˆ a ‡ 2c ÿ b cos A ˆ a0 ‡ 2c0 ÿ b cos A0 action on elements of R…f † it must hold that g…z†
 cy …z† 2 Fz \ S implying that g…z† 2 Fz and therefore
3x3 ˆ c ‡ 2b sin A ˆ c0 ‡ 2b0 sin A0
3y3 ˆ a ÿ c ‡ 2b cos A ˆ a0 ÿ c0 ‡ 2b0 cos A0
The condition for x2 has been discarded because it is
identical to that for x1 . The remaining are not inde-
pendent due to the relation between cos A and sin A.
Thus
y1 ‡ y2 ‡ y3 ˆ a ˆ a0
…x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2 ˆ b2 ˆ b20 ) b ˆ b0
2x1 ‡ x3 ˆ c ˆ c0
Using a ˆ a0 ; b ˆ b0 and c ˆ c0 in the above conditions
yields cos A ˆ cos A0 ; sin A ˆ sin A0 , so that A ˆ A0 .
Therefore, y 2 pG …y0 † for a given ®xed y0 , when its
elements satisfy
a ˆ a0 ; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ any; C ˆ any
The ®ber pG …y0 † has dimension 6 ± 4=2. The elements
q ˆ ‰B CŠT may serve as a system of curvilinear coordi-
nates so that pG …y0 † is described by y ˆ y…q† ˆ y…q; y0 † or
explicitly
a ˆ a0 ; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ B; C ˆ C

Proposition 1. A generalized inverse g of f is uniquely

de®ned if the following are speci®ed:
The range R…g†  X of the generalized inverse g.
The section S ˆ g…R…f ††  R…g† of the ®bering F of
X induced by f .
The ®bering P to be induced by p ˆ f  g, i.e., all
the ®bers Py corresponding to every y 2 R…f †. Fig. 4. An illustration of Proposition 1
 81

f …g…z†† ˆ z or f …g…f …x††† ˆ …f  g  f †…x† ˆ f …x† and Since y0 2 R…f † is ®xed, the image g…pP0 …y0 †† consists
since x has been arbitrarily chosen f  g  f ˆ f and g of elements x having x1 ; x2 ; x3 constant, while y1 ; y2 ; y3
is indeed a generalized inverse of f . ( depend on a single parameter a. It has therefore di-
mension 1. Since x0 ˆ g…y0 † is ®xed, x ˆ g…y† may be also
Example 1 (continued). We now consider the choice of a expressed in terms of x0 :
section M of G. We have seen that y 2 pG …y0 † when
x1 ˆ x01 ; y1 ˆ y01 ‡ aÿa
3
0

a ˆ a0 ; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ any; C ˆ any x2 ˆ x02 ; aÿa0

y2 ˆ y02 ‡ 3
In order to create a section M of G we need only to pick x3 ˆ x03 ; y3 ˆ y03 ‡ aÿa
3
0

up one element from each one of its ®bers, and this can or
be done by assigning values to both B and C. If in a ÿ a0
addition we want R…f †  M, we must assign the known x ˆ x0 ‡ ‰0 0 0 1 1 1ŠT
3
values
a2 ‡ c20 ÿ b20 a2 ‡ b20 ÿ c20 We can also express g…pP0 …y0 †† by ®ve conditions.
B0 ˆ arcos 0 and C0 ˆ arcos 0 One possible choice is
2a0 c0 2a0 b0
to B and C, respectively, whenever y0 2 R…f †. x1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ c0 ; y2 ÿ y1 ˆ c0
Thus M may be determined from the two conditions x3 ÿ x2 ˆ b0 sin A0 ; y3 ÿ y1 ˆ b0 cos A0
a2 ‡ c 2 ÿ b2 a 2 ‡ b 2 ÿ c2 Comparison with the identical conditions for x 2 Fy00
cos B ˆ ; cos C ˆ
2ac 2ab shows that in fact g…pP0 …y0 †† ˆ Fy00 ˆ Fy0 \ R…g† is an
and it has dimension 6±2=4. M contains the three- element of the ®bering F0 induced by F on R…g†.
dimensional R…f †, which satis®es these conditions and in Returning once again to the theory, we would like the
addition the similar one for cos A. The independent pa- generalized inverse g of a continuous mapping f to be
rameters a; b; c; A may serve as curvilinear coordinates itself continuous, and this means that the mappings cy
for M which is then described by y ˆ y…q†, explicitly cannot be independent. Loosely speaking, they must
a ˆ a; b ˆ b; c ˆ c; A ˆ A map neighboring elements of Y belonging to neighbor-
ing ®bers of P into neighboring elements of X . Thus cy
a2 ‡ c2 ÿ b2 a 2 ‡ b 2 ÿ c2 may be considered as the restriction of a continuous
B ˆ arcos ; C ˆ arcos
2ac 2ab mapping c : R…f †  Y ! X , de®ned by c…y; z† ˆ cy …z†,
which is continuous in both y and z.
We now consider the intersection M \ pP …y0 † ˆ pP0 …y0 †. We can overcome the need to use a family of map-
If we combine the requirements for y 2 M pings cy , one for each y 2 R…f † with the introduction of
a ˆ any; b ˆ any; c ˆ any; A ˆ any additional ®berings for Y and X and a single mapping
between them. Let M be a section of the ®bering G. In
a2 ‡ c2 ÿ b2 a 2 ‡ b 2 ÿ c2 order to de®ne g on pP …y† for every y 2 R…f † it is su-
B ˆ arcos ; C ˆ arcos
2ac 2ab cient to de®ne it on M \ pP …y† since the ®bering G, being
with those for y 2 pP …y0 † a re®nement of P, naturally extends g on the whole of
pP …y†: for any z 2 pP …y† 2= M there exists a unique ®ber
a ˆ any; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ any; C ˆ any pG …z†  pP …z† from G such that pG …z† \ M ˆ fz0 g
 M \ pP …z† and g…z† ˆ g…z0 †. Although any section M
we conclude that y 2 M \ pP …y0 † ˆ pP0 …y0 † when
of G is appropriate, it is more convenient to choose M so
a ˆ any; b ˆ b0 ; c ˆ c0 ; A ˆ A0 that it contains R…f †. This is possible in view of the next
lemma.
a2 ‡ c20 ÿ b20 a2 ‡ b20 ÿ c20
B ˆ arcos ; C ˆ arcos Lemma. For every y 2 R…f †, it holds that pG …y†
2ac0 2ab0
\R…f † ˆ fyg.
This means that pP0 …y0 † ˆ M \ pP …y0 † is of dimension 1,
where the parameter a may serve as the single curvili- Proof. Since G is a re®nement of P; pG …y†  pP …y†.
near coordinate. When, in particular, y0 2 R…f † then Since R…f † is a section of P it holds pG …y† \ R…f † ˆ fyg
b2 ‡c2 ÿa2
A ˆ A0 ˆ arcos 0 2b00c0 0 . which combined with pG …y†  pP …y† yields the desired
From the de®nition of g we see that when y 2 pP0 …y0 † result. (
where y0 2 R…f † then x ˆ g…y† ˆ cy0 …y† is given by The ®bering P induces a ®bering P0 ˆ PjM on M
c0 ÿ b0 sin A0 a ÿ c0 ÿ b0 cos A0 with ®bers pPjM …z† ˆ M \ pP …z†. R…f † is a section of P
x1 ˆ y1 ˆ and since R…f †  M by choice, it is also a section of P0 .
3 3
It is possible to consider a ®bering R of M containing
c0 ÿ b0 sin A0 a ‡ 2c0 ÿ b0 cos A0
x2 ˆ y2 ˆ R…f † which is complementary to P0 . Thus every z 2 M is
3 3 the unique element pR …z† \ pP0 …z† ˆ fzg. The general-
c0 ‡ 2b0 sin A0 a ÿ c0 ‡ 2b0 cos A0 ized inverse g must be a mapping whose restriction to M
x3 ˆ y3 ˆ
3 3 is a one-to-one mapping gjM from M to R…g†, since
82

dim M ˆ dim R…g† ˆ rg . A corresponding choice of

complementary ®berings of R…g† is possible. One is
provided by the ®bering F which induces a ®bering
F0 ˆ FjR…g† on R…g† with ®bers pFjR…g† …x†ˆ pF …x† \ R…g†.
For g to be a generalized inverse it must hold that
g…z† 2 Fp…z† \ R…g† so that each ®ber of P0 is mapped into
the corresponding element of F0 . Since z 2 M is deter-
mined from one ®bering of P0 and one ®bering of R, it
remains to introduce a ®bering S of R…g† complemen-
tary to F0 , such that g…z† is determined by the inter-
section of a ®ber of S with the already speci®ed ®ber of
F0 . More precisely, to the ®ber pP0 …z† 2 P0 in M cor-
responds the ®ber Fp…z† \ R…g† 2 F0 in R…g†, and it
remains to assign to the ®ber pR …z† 2 R in M a ®ber
S 0 2 S in R…g† so that g…z† ˆ ‰Fp…z† \ R…g†Š \ S 0 . In par-
ticular, S must contain the particular section S of F
which must be assigned to R…f † 2 R.

Proposition 2. A generalized inverse g of f , with

rank…g† > rank(f ) is uniquely de®ned if the following
are speci®ed:
A ®bering P of Y having R…f † as a section.
A re®nement G of the ®bering P.
A section M of the ®bering G containing R…f †.
A manifold R…g† of X with dimension rg > r on
which F induces a ®bering F0 of dimension
d ˆ m ÿ r ˆ dim X ÿ rank…f †.
A ®bering S of R…g† complementary to F0 .
A ®bering R of M complementary to the ®bering P0
induced by P on M, which contains R…f †.
A mapping C : R ! S. Fig. 5. An illustration of Proposition 2
The generalized g is de®ned on M by
fg…z†g ˆ ‰Fp…z† \ R…g†Š \ C…pR …z†† where p…z† is
de®ned by fp…z†g ˆ pP0 …z† \ R…f † ˆ pP …z† \ R…f †.
a ˆ k ‡ a0 ; b ˆ b0 ; c ˆ c0 ; A ˆ A0
Proof. We must show that the de®ned mapping g is a
a2 ‡ c2 ÿ b2 a2 ‡ b2 ÿ c2
generalized inverse of f . For an arbitrary x 2 X , y ˆ B ˆ arcos ; C ˆ arcos
f …x† 2 R…f †  M. Since y 2 R…f † we have that p…y† ˆ y, 2ac 2ab
pR …y† ˆ R…f † and C…pR …y†† ˆ C …R…f ††  S 2 S. where y0 ˆ p…y†. From the de®nition of the projection p,
Applying the above de®nition, we have that y0 ˆ p…y† is given by
fg…y†g ˆ ‰Fp…z† \ R…g†Š \ C…pR …y†† p
a0 ˆ b2 ‡ c2 ÿ 2bc cos A; b0 ˆ b; c0 ˆ c; A0 ˆ A
ˆ ‰Fy \ R…g†Š \ C…R…f †† ˆ ‰Fy \ R…g†Š \ S  Fy c ÿ b cos A
B0 ˆ arcos p
and consequently f …g…y†† ˆ y. Since y ˆ f …x† it follows b2 ‡ c2 ÿ 2bc cos A
that f …g…f …x††† ˆ …f  g  f †…x† ˆ f …x† and since x is b ÿ c cos A
arbitrary f  g  f ˆ f . ( C0 ˆ arcos p
b ‡ c2 ÿ 2bc cos A
2

Replacing these values in the preceding description of Rk

Example 1 (continued). We begin with the ®bering R of
it follows that any y 2 Rk satis®es
M containing R…f †. A ®bering R of M can be introduced
p
by ®xing the value of the parameter a along the ®bers a ˆ k ‡ b2 ‡ c2 ÿ 2bc cos A; b ˆ b; c ˆ c
pP0 …y0 † ˆ M \ pP …y0 †, where we can assume without loss
of generality that y0 2 R…f †. If we want R to include a2 ‡ c2 ÿ b2 a2 ‡ b2 ÿ c2
R…f † the value of a ˆ a…y† assigned to y 2 pP0 …y0 † must A ˆ A; B ˆ arcos ; C ˆ arcos
2ac 2ab
be such that a…y0 † ˆ a0 . A simple choice is to set
a ˆ k ‡ a0 , in which case k replaces a as coordinate Rk is therefore of dimension 3, since it satis®es the
along pP0 …y0 †, while k ˆ 0 gives the intersection y0 of 6 ± 3 ˆ 3 conditions
pP0 …y0 † with R…f †. With this choice a ®ber Rk from R is
described by the elements y with …a ÿ k†2 ˆ b2 ‡ c2 ÿ 2 b c cos A
 83

a2 ‡ c 2 ÿ b2 a2 ‡ b2 ÿ c2 …a0 ‡ k†2 ‡ c20 ÿ b20

cos B ˆ ; cos C ˆ B ˆ arcos
2ac 2ab 2…a0 ‡ k†c0
In order to describe the image Sk  g…Rk † we need …a0 ‡ k†2 ‡ b20 ÿ c20
®rst C ˆ arcos
2…a0 ‡ k†b0
b2 ‡ c2 ÿ …a ÿ k†2 which uniquely determine y.
cos A ˆ
2rb c The image g…y† of the y is given, according to the
h i de®nition of g, by
4b2 c2 ÿ b2 ‡ c2 ÿ …a ÿ k†2
) sin A ˆ c0 ÿ b0 sin A0 a0 ‡ k ÿ c0 ÿ b0 cos A0
2bc x1 ˆ y1 ˆ
3 3
which used in the de®nition of g yield
r
h i c0 ÿ b0 sin A0 a0 ‡ k ‡ 2c0 ÿ b0 cos A0
2 2 2
4b c ÿ b ‡ c ÿ …a ÿ k† 2 2 x2 ˆ y2 ˆ
c 3 3
x1 ˆ ÿ
3 6c
c0 ‡ 2b0 sin A0 a0 ‡ k ÿ c0 ‡ 2b0 cos A0
a c b2 ÿ …a ÿ k†2 x3 ˆ y3 ˆ
y1 ˆ ÿ ÿ 3 3
3 2 6c
r
h i Let x 2 Sk \ Fy0o ˆ g…Rk † \ g…pP0 …y0 ††. It must simulta-
4b2 c2 ÿ b2 ‡ c2 ÿ …a ÿ k†2 neously satisfy the conditions for x 2 Sk
c
x2 ˆ ÿ
3 6c x1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ y2 ÿ y1
a c b ÿ …a ÿ k†2
2
…x3 ÿ x2 † ‡ …y3 ÿ y2 †2 ˆ …y1 ‡ y2 ‡ y3 ÿ k†2
2
y2 ˆ ‡ ÿ
3 2 6c
and those for x 2 Fy00
r
h i
2 4b2 c2 ÿ b2 ‡ c2 ÿ …a ÿ k†2 x1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ c 0 ; y2 ÿ y1 ˆ c0
c
x3 ˆ ‡ x3 ÿ x2 ˆ b0 sin A0 ; y3 ÿ y1 ˆ b0 cos A0
6 3c
2 2
a b ÿ …a ÿ k†
y3 ˆ ‡ The combination gives the six conditions
3 3c
Sk has dimension 3 since it is determined by three x1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ c0 ; y2 ÿ y1 ˆ c0 ;
parameters a; b; c (k is ®xed). It can also be determined x3 ÿ x2 ˆ b0 sin A0 ; y3 ÿ y1 ˆ b0 cos A0 ;
by three conditions, e.g.,

x 1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ y2 ÿ y1 …x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 ˆ …y1 ‡ y2 ‡ y3 ÿ k†2

which uniquely determine x. Indeed solving this system
…x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 ˆ …y1 ‡ y2 ‡ y3 ÿ k†2 we get

Let y 2 Rk \ pP0 …y0 †: The y must satisfy simultaneously c0 ÿ b0 sin A0 c0 ‡ 2b0 sin A0
the conditions for y 2 Rk x1 ˆ ˆ x 2 ; x3 ˆ
3 3
a0 ‡ k ÿ c0 ÿ b0 cos A0
…a ÿ k†2 ˆ b2 ‡ c2 ÿ 2 b c cos A y1 ˆ ;
3
a0 ‡ k ‡ 2c0 ÿ b0 cos A0
a2 ‡ c 2 ÿ b2 a2 ‡ b2 ÿ c2 y2 ˆ ;
cos B ˆ ; cos C ˆ 3
2ac 2ab a0 ‡ k ÿ c0 ‡ 2b0 cos A0
y3 ˆ
and the conditions for y 2 pP0 …y0 † 3
b ˆ b0 ; c ˆ c0 ; A ˆ A0 But the above components of x determined by
fxg ˆ Sk \ Fy00 are in fact identical to those of the image
a2 ‡ c20 ÿ b20 a2 ‡ b20 ÿ c20 x ˆ g…y† where y is determined by fyg ˆ Rk \ pP 0 …y0 †.
cos B ˆ ; cos C ˆ Therefore g is uniquely determined from the mapping
2 a c0 2 a b0
C : Rk ! Sk , which to each ®ber Rk 2 R assigns a ®ber
The combination gives the six conditions Sk 2 S, the correspondence having been established
through the common parameter k. The mapping
a ˆ a0 ‡ k; b ˆ b0 ; c ˆ c0 ; A ˆ A0 C : R ! S, maps
84

This means that pP …y 0 † \ R…f † ˆ fyg and R…f † is indeed
Rk ˆ yj…a ÿ k†2 ˆ b2 ‡ c2 ÿ 2bc cos A; a section of P, as required. We can therefore describe
 the ®bers of P using y0 2 R…f †, in which case y 2 pP …y0 †
a2 ‡ c2 ÿ b2 a 2 ‡ b 2 ÿ c2 when
cos B ˆ ; cos C ˆ
2ac 2ab
a ˆ any; b ˆ any; c ˆ c0 ; A ˆ A0 ; B ˆ B0 ; C ˆ any
to

From the preceding conditions it follows that for an
Sk ˆ xjx1 ˆ x2 ; x1 ‡ x2 ‡ x3 ˆ y2 ÿ y1 ; arbitrary y its projection y0 ˆ p…y† ˆ f …g…y†† is given by

2 2 2 sin A sin B
…x3 ÿ x2 † ‡ …y3 ÿ y2 † ˆ …y1 ‡ y2 ‡ y3 ÿ k† a0 ˆ c ; b0 ˆ c ; c0 ˆ c
sin…A ‡ B† sin…A ‡ B†
The other required correspondence between pP0 …y0 † A0 ˆ A; B0 ˆ B; C0 ˆ p ÿ …A ‡ B†
and Fy00 is naturally introduced through the common
element y0 used in the de®nitions of the ®bers The next step is to introduce a re®nement G of P
pP0 …y0 † 2 P0 and the corresponding ones Fy00 2 F0 , such that each ®ber of G intersects R…f † at most one
where F0 is thus a ®bering of R…g† complementary to element. Our choice is y 2 pG …y 0 † when
S. For given y0 2 R…f † the ®ber Fy00 depends only on the
a ˆ a0 ; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ B0 ; C ˆ any
mapping f and not on the generalized inverse map-
ping g. Obviously y 2 pG …y 0 † ) y 2 pP …y 0 † and therefore
pG …y 0 †  pP …y 0 †, so that G is indeed a re®nement of P.
Example 2. We shall now give an example where the If y 0 2= R…f † then obviously for any y 2 pG …y 0 † also
generalized inverse is not de®ned directly but follows y2= R…f †, i.e. pG …y 0 † \ R…f † ˆ [: If y 0 ˆ y0 2 R…f † then
from the choice of ®berings and subspaces as well
as a ®bering mapping C, along the lines of Propo- sin A0 sin B0
a 0 ˆ c0 ; b0 ˆ c0 ;
sition 2. sin…A0 ‡ B0 † sin…A0 ‡ B0 †
We consider the same mapping f as in Example 1, C0 ˆ p ÿ …A0 ‡ B0 †
but we will choose a di€erent description of the range
R…f † and the solution spaces Fy , using this time c; A; B as and y 2 pG …y0 † when
the free parameters. In this case y ˆ ‰a b c A B CŠT
sin A0 sin B0
2 R…f † when a ˆ c0 ; b ˆ c0 ; c ˆ c0
sin…A0 ‡ B0 † sin…A0 ‡ B0 †
sin A sin B
aˆc ; bˆc ; c ˆ any A ˆ A0 ; B ˆ B0 ; C ˆ any
sin…A ‡ B† sin…A ‡ B†
A ˆ any; B ˆ any; C ˆ p ÿ …A ‡ B† in which case pG …y0 † \ R…f † ˆ fy0 g as required.
In order to choose a section M of G we must pick up
The conditions in this above description are in fact the a single element y from each ®ber pG …y 0 †. Since the ele-
well-known angle and sine conditions ments a; b; c; A; B are already determined from the equal
a b c ones of y 0 , we are left with the possibility of choosing
A ‡ B ‡ C ˆ p; ˆ ˆ one value of C for every ®ber of G. Our choice is
sin A sin B sin C
C ˆ p ÿ …A ‡ B† so that y 2 M when
We will introduce a ®bering P having R…f † as a section
by y 2 pP …y 0 † when a ˆ any; b ˆ any; c ˆ any
0 0 0
a ˆ any; b ˆ any; c ˆ c ; A ˆ A ; B ˆ B ; C ˆ any A ˆ any; B ˆ any; C ˆ p ÿ …A ‡ B†
0
If y 2 pP …y † \ R…f † then y must satisfy simultaneously M is obviously a section of G since y 2 M \ pG …y 0 † is
the conditions for y 2 pP …y 0 †: uniquely determined from
c ˆ c0 ; A ˆ A0 ; B ˆ B0 a ˆ a0 ; b ˆ b0 ; c ˆ c0 ; A ˆ A0 ; B ˆ B0 ; C ˆ p ÿ …A0 ‡ B0 †
and the conditions for y 2 R…f †:
The ®bering P induces a ®bering P0 on M with ®bers
sin A sin B pP0 …y0 † ˆ pP …y0 † \ M. Combining the conditions for
aˆc ; bˆc ; C ˆ p ÿ …A ‡ B† y 2 M with those for y 2 pP …y0 † we conclude that
sin…A ‡ B† sin…A ‡ B†
y 2 pP0 …y0 † when
which, combined, show that y is uniquely determined
from a ˆ any; b ˆ any; c ˆ c0 ; A ˆ A0 ; B ˆ B0 ;
C ˆ p ÿ …A0 ‡ B0 †
sin A0 sin B0
a ˆ c0 ; b ˆ c0 ; c ˆ c0
sin…A0 ‡ B0 † sin…A0 ‡ B0 † For the ®bering F of the solution spaces Fy …y 2 R…f ††
0 0 0 0
we need a representation which uses c; A; B as free
AˆA; BˆB; C ˆ p ÿ …A ‡ B † parameters: x 2 Fy whenever f …x† ˆ y; i.e.,
 85

…x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 ˆ c2 x1 ˆ a; x2 ˆ a ‡ c0 ; y1 ˆ b; y2 ˆ b
x3 ˆ a ‡ b0 cos A0 ; y3 ˆ b ‡ b0 sin A0
…x3 ÿ x1 †…x2 ÿ x1 † ‡ …y3 ÿ y1 †…y2 ÿ y1 †
qq ˆ cos A Since Sa;b \ Fy00 ˆ fxg, the ®bering S is complementary
…x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 to F0 within R…g†:
In order to obtain ®nally an explicit description of
…x3 ÿ x2 †…x2 ÿ x1 † ‡ …y3 ÿ y2 †…y2 ÿ y1 †
qq ˆ cos B the generalized inverse mapping g : y ! x ˆ g…y†, let
…x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 y ˆ ‰a b c A B CŠT be an arbitrary element of Y . Then
y 2 Ra;b \ pP0 …y0 † when y0 ˆ p…y† is given by
For R…g† ˆ g…Y † we must choose a subset of X which
sin A sin B
has the same dimension 5 as M, since R…g† ˆ g…M† also. a0 ˆ c ; b0 ˆ c ; c0 ˆ c
Our choice is to set x 2 R…g† when sin…A ‡ B† sin…A ‡ B†

y2 ˆ y1 A0 ˆ A; B0 ˆ B; C0 ˆ p ÿ …A ‡ B†
The ®bering F0 induced by F on R…g† has ®bers and x ˆ g…y† is uniquely determined from x 2 Sa;b \ Fy00 .
Fy0 ˆ Fy \ R…g†. Thus taking y2 ˆ y1 into account, the This means that x ˆ g…y† if we replace a ˆ a; b ˆ b and
preceding conditions for x 2 Fy we obtain, after some the values of the elements of y0 in the expression giving
algebraic manipulation, that x 2 Fy0 when x 2 Sa;b \ Fy00 . This results in
y3 ÿ y1 y3 ÿ y2 x1 ˆ a; x2 ˆ a ‡ c; y1 ˆ b; y2 ˆ b
y2 ˆ y1 ; x2 ÿ x1 ˆ c; ˆ tan A; ˆ tan B
x3 ÿ x1 x3 ÿ x2
sin B sin B
Since for every y the ®ber pP0 …y† will be mapped onto the x3 ˆ a ‡ c cos A; y3 ˆ b ‡ c sin A
0 sin…A ‡ B† sin…A ‡ B†
®ber Fp…y† , it remains to introduce a ®bering R of M
which is complementary to P0 . We correspond to each thus bringing to an end Example 2.
pair of positive numbers a; b a ®ber Ra;b 2 R which is A special case of interest is the case of a minimal rank
de®ned by y 2 Ra;b when generalized inverse g where rg ˆ r. In this case
a ˆ a; b ˆ b; c ˆ any; A ˆ any; B ˆ any; rg ˆ dim…R…g†† ˆ r ˆ dim…R…f ††
C ˆ p ÿ …A ‡ B† which combined with S  R…g† implies that S ˆ R…g†:
In order to show that R is complementary to P0 (within
M), let y 2 Ra;b \ pP0 …y0 †. Combining the conditions for Lemma. When rg ˆ r the ®bering G induced by g
y 2 Ra;b with the ones for y 2 pP0 …y0 † we conclude that y coincides with the ®bering P induced by p ˆ f  g:
is uniquely determined from
Proof. Since S ˆ R…g†, the restriction gjR…f † of g to the
a ˆ a; b ˆ b; c ˆ c0 ; A ˆ A0 ; B ˆ B0 ; range of f is in this case a bijection between R…f † and
C ˆ p ÿ …A0 ‡ B0 † R…g†. For every z 2= R…f †; g…z† 2 R…g† ˆ S and since S is
a section of F, fg…z†g ˆ S \ Fy for a unique ®ber Fy
Therefore Ra;b \ pP0 …y0 † ˆ fyg and R is complementary corresponding to a unique y 2 R…f †. Therefore
to P0 . y ˆ f …g…z†† ˆ …f  g†…z† ˆ p…z† and z 2 pP …y† ˆ
One more step remains to complete the de®nition of pP ……f  g†…z††. Consequently the ®bering G induced by
the generalized inverse g of f , that is to introduce a g coincides with the ®bering P induced by p ˆ f  g. (
mapping
For the particular choice rg ˆ r, the last two re-
C : R ! S : Ra;b ! Sa;b quirements in Proposition 1 can be relaxed. The same
simpli®ed situation can arise in a di€erent way by re-
where S is a ®bering of R…g† complementary to F0 . quiring that in addition to g being a generalized inverse
We choose Sa;b ˆ C…Ra;b † by requiring that x 2 Sa;b of f , at the same time f is a generalized inverse of g.
when Repeating property (G1) by interchanging the roles of f
x1 ˆ a; y1 ˆ b; y2 ˆ b and g leads to the following de®nition.

Obviously y1 ˆ y2 and thus Sa;b  R…g†. If x 2 Sa;b \ Fy00 De®nition. A generalized inverse g of a given mapping f
then combining these conditions for x 2 Sa;b with those is called a re¯exive generalized inverse if in addition to
for x 2 Fy00 we obtain (G1) it satis®es the following condition, henceforth
x1 ˆ a; y1 ˆ b; y2 ˆ b denoted (G2):

y3 ÿ y1 y3 ÿ y2 gf gˆg …8†

x 2 ÿ x1 ˆ c0 ; ˆ tan A0 ; ˆ tan B0
x3 ÿ x1 x3 ÿ x2
These six conditions can be solved for the elements of x Lemma. A generalized inverse g of f is a re¯exive
which is then uniquely determined by generalized inverse of f if and only if rg ˆ r:
86

We choose P in such a way that for any given

y0 2 R…f † an element y 2 pP …y0 † when
cos A‡CÿB
2 cos C
a0 ˆ a; b0 ˆ a ; c0 ˆ a
cos B‡CÿA
2 cos B‡CÿA
2

p ÿ …A ‡ B ‡ C† p ÿ …A ‡ B ‡ C†
A0 ˆ A ‡ ; B0 ˆ B ‡ ;
2 2
C0 ˆ C
These equations describe also the projection y0 ˆ p…y†
for any given y.
For the corresponding ®ber Fy0 of F we need the
appropriate description x 2 Fy0 when

…x2 ÿ x1 †…x3 ÿ x1 † ‡ …y2 ÿ y1 †…y3 ÿ y1 †

q q
…x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2

Fig. 6. The geometry of a re¯exive generalized inverse g of a nonlinear

mapping f . All elements of pP …y† ˆ pG …y† are mapped into the same
element x^ 2 S and thus projected by p onto the same element y^
Proof. Necessity: an immediate consequence of g being a
generalized inverse of f is that R…g  f † ˆ R…g†, and
rg  min…r; rg †, so that rg  r  min…n; m†; which com-
bined with r  rg from Eq. (5) implies that rg ˆ r:
Suciency: if r ˆ rg then dim…R…g†† ˆ dim…R…f †† ˆ
dim…S† and S ˆ R…g†: If x 2 S then fxg ˆ S \ Ff …x† and
g…f …x†† ˆ x. Let y be an arbitrary element of Y , then
g…y† 2 R…g† ˆ S and setting g…y† in the place of the
previous x we obtain g…f …g…y††† ˆ g…y† or …g  f  g†
…y† ˆ g…y† and since y is arbitrary g  f  g ˆ g. ( Combining the conditions for x 2 S and x 2 Fy0 , we
Proposition 3. A re¯exive generalized inverse g of f is
uniquely de®ned if the following are speci®ed:
(1) The section S ˆ g…R…f †† ˆ R…g† of the ®bering F of
X induced by f .
(2) The ®bering P to be induced by p ˆ f  g, i.e., all
the ®bers Py corresponding to every y 2 R…f †.
The question of what the ®bers Gx of a re¯exive
generalized inverse are, is answered by the following
lemma.
ˆ ÿ cos B0
…x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 ˆ a20
…x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 ˆ c20
The de®nition of the generalized inverse g is ®nalized
with the selection of the section S of F by x 2 S when
x1 ‡ x2 ‡ x3 ˆ 0; y1 ‡ y2 ‡ y3 ˆ 0; x2 ˆ x3
For an arbitrary element y of Y the corresponding ®ber
pP …y† ˆ pP …y0 † is determined from the projection
y0 ˆ p…y†. Since R…g† ˆ S while pP …y0 † is mapped onto
Fy0 , the image x ˆ g…y† must belong to both, x 2 S \ Fy0 .
obtain a system of six equations in six unknowns
…x2 ÿ x1 †…x3 ÿ x1 † ‡ …y2 ÿ y1 †…y3 ÿ y1 †
q q
…x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 …x3 ÿ x1 †2 ‡ …y3 ÿ y1 †2
ˆ ÿ cos B0
…x3 ÿ x2 †2 ‡ …y3 ÿ y2 †2 ˆ a20 ; …x2 ÿ x1 †2 ‡ …y2 ÿ y1 †2 ˆ c20
x1 ‡ x2 ‡ x3 ˆ 0; y1 ‡ y2 ‡ y3 ˆ 0; x2 ˆ x3

Lemma. For a re¯exive generalized inverse g, the …f  g†- The solution is easily veri®ed to be
induced ®bers are identical to the g-induced ®bers, i.e.,
P ˆ G: x1 ˆ 23 c0 sin B0 ; y1 ˆ ÿ 13 a0 ‡ 23 c0 cos B0

Proof. Consider any y 2 R…f †: z 2 pP …y† ) y ˆ …f  g† x2 ˆ ÿ 13 c0 sin B0 ; y2 ˆ ÿ 13 a0 ÿ 13 c0 cos B0

…z† ) g…y† ˆ …g  f  g†…z† ˆ g…z† ) z 2 pG …y† ) pP …y†
 pG …y†. But since G is a re®nement of P, it must also
hold that pG …y†  pP …y† and thus pG …y† ˆ pP …y† for any
y 2 R…f †. Therefore G ˆ P. (
Example 3. We shall give an example of a re¯exive
generalized inverse which is de®ned indirectly through
the choice of the ®bering P  G having R…f † as a section
and the section S of the ®bering F.
x3 ˆ ÿ 13 c0 sin B0 ; y3 ˆ 23 a0 ÿ 13 c0 cos B0
These expressions give in fact not g but its restriction
gjR…f † of g to R(f). This can be completed to g ˆ gjR…f †  p
from the known projection mapping p, by simply re-
placing B0 ; a0 ; c0 with their expressions in terms of
a; b; c; A; B; C. Thus we obtain the explicit form x ˆ g…y†
of the re¯exive generalized inverse g
 87

cos AÿB‡C
2
De®nition. The unique re¯exive generalized inverse g of
x1 ˆ 23 a cos C ; a given mapping f which is also a minimum distance
cos B‡CÿA
2
and x0 -nearest generalized inverse is called the pseudo-
sin AÿB‡C
2 inverse of f .
y1 ˆ ÿ13a ‡ 23a cos C
cos B‡CÿA
2 Various other nonunique generalized inverses can be
de®ned by combination of (G1) with the some of the
cos AÿB‡C
2
other properties (G2), (G3), and (G4). The inverses
x2 ˆ ÿ13a cos C ; within the same class, i.e., the ones satisfying the same
cos B‡CÿA
2 set of properties, may di€er in the following aspects:
sin AÿB‡C
2
y2 ˆ ÿ13a ÿ 13a cos C (1) In the section S when (G4) is not satis®ed.
cos B‡CÿA
2 (2) In the ®bering P induced by p ˆ f  g when (G3) is
not satis®ed.
cos AÿB‡C
2
(3) In the re®nement G of P and the mapping
x3 ˆ ÿ13a cos C ; c…y; z† ˆ cy …z† extending g outside R…f † when (G2) is
cos B‡CÿA
2 not satis®ed.
sin AÿB‡C
2
y3 ˆ 23a ÿ 13a cos C ( In the geodetic case property (G3) is a ``must'', since
cos B‡CÿA
2 it solves the adjustment problem. Re¯exivity (G2) is not
necessary but convenient. The section S is speci®ed ei-
For re¯exive generalized inverses we have a possibi- ther directly by a set of minimal constraints or by in-
lity of choices, corresponding to the choices of the sec- troducing (G4). (G4) is introduced, either directly
tion S of F and of the ®bering P of Y over the elements (pseudo-inverse solution) or indirectly by a set of inner
of R…f †. In analogy to the uniquely de®ned pseudo-in- constraints which describe the particular section S cor-
verse of a linear operator we seek two conditions, ad- responding to (G4).
ditional to (G1) and (G2), such that (G3) speci®es the
®bering P and (G4) speci®es the section S of F. De®nition. A set of (nonlinear) equations v…x; d† ˆ 0, is
We introduce the minimum-distance generalized in- called a set of minimal constraints (with respect to f ) if
verse. Let qY be the distance function of the metric space the corresponding mapping v : X  Rmÿr ! Rmÿr gives
Y . For any ®xed element y^ 2 R…f † de®ne rise to a ®bering H of X , such that its member
Hd ˆ fx 2 Hd †jv…x; d† ˆ 0g, corresponding to a ®xed d,
Ry^ ˆ fz 2 Y ; qY … z; y^† ˆ min qY …z; y†g …9† is a section of the ®bering F induced by f .
y2R…f †

Eq. (9) we name (G3). In the particular case that the
subsets Ry^ are ®bers of Y , as y^ varies over R…f †, then we
can employ the following de®nition: g is a minimum
distance generalized inverse of f , if P ˆ R, i.e., when the
®bering P induced by p ˆ f  g is identical with the
®bering R with ®bers Ry^.
If Y is an inner-product vector space with metric
induced by a quadratic norm, we may use the term
``least squares'' in place of ``minimum distance''.
We now de®ne the x0 -nearest generalized inverse.
Let qX be the distance function of X ; x0 2 X be ®xed and
let
xFy ˆ min qX …x0 ; z† …10†
z2Fy
This equation we denote (G4). In the particular case
that there exists only one element on every Fy which
satis®es (G4), then the set of all xFy constitutes a section
S of F, and we can employ the following de®nition: g is
an x0 -nearest generalized inverse of f if S is taken as the
image of R…f † under g, i.e., S ˆ g…R…f ††:
Remark. If X is an inner-product vector space with
metric induced by the norm, while x0 ˆ 0, we may use
the term ``minimum norm'' in place of ``0-nearest''.
However in the geodetic case 0 is a ``prohibited'' point,
since it corresponds to a network with all its points
coinciding.
De®nition. A set of minimal constraints h…x† ˆ d are
called inner constraints (with respect to a given point x0
of X ) when the ®ber Hd coincides with the section of F
speci®ed by the property (G4), i.e., when for every
Fy 2 F the unique element x such that Fy \ Hd ˆ fxg is
the one closest to x0 among all elements of Fy .
Remark. In the special case that f and its generalized
inverse are linear or ane mappings, R…f †, the ®bers
Fy ; Py ; Gx ; Qx , and the section S are all ane subspaces
of the corresponding spaces Y and X . The ®bers
F; P; G; Q consist of parallel ane subspaces, i.e., they
are equivalent to quotient spaces of Y and X . Each
quotient space is uniquely determined by its member
passing through zero, i.e., by their ``modulo'' linear
subspace in X or Y (Halmos 1974, Sect. 21) . The ®ber
F, e.g., is equivalent to the quotient space of ane
subspaces parallel to the null space of f . Thus the linear
of ane generalized inverses are determined by a set of
linear subspaces, instead of the ®berings (see Teunissen
1985). By the way, there is no reason to restrict the class
of generalized inverses of a linear operator to be itself
linear, as is usually done within the linear theory. A
linear operator may well have a nonlinear generalized
inverse. A geodetic example is the ane generalized
inverse obtained implicitly by imposing a set of inho-
mogeneous linear constrains Hx ˆ d; d 6ˆ 0; on the
solution of the least-squares adjustment problem.
88
The approach followed here can be characterized as
naive, and falls short of a proper mathematical theory of
generalized inverses of nonlinear mappings, even in the
relatively simple case of ®nite-dimensional spaces. The
reason is that the assumptions made are too strong for
the purpose of building which to build a general theory,
a fact which will become obvious even in the simple
geodetic case. Some of the assumptions that do not
generally hold are the following:
(1) f may not be de®ned at all points of the space X but
only on an open subset U  X .
(2) g may not be possible to de®ne at all points of the
space Y but only on an open subset V  Y :
(3) The minimum distance property (G3) may not suf-
®ce for the determination of a ®bering of Y or even
V . This has to do with the di€erential geometric
properties of R…f † as a submanifold of Y and in
particular with its curvature tensor. There might be
points in V such that their minimum distance from
R…f † does not correspond to a unique point of R…f †
but either on a set of discrete points or even a non
discrete subset of R…f †. (Think, e.g., of the case
where R…f † is a ball in Y and consider its center
which belongs to any ®ber induced by the minimum
distance principle.)
(4) The x0 -nearest property (G4) may not suce for the
determination of a ®bering of X or even U . This has
to do with the di€erential geometric properties of
the ®bers Fy as submanifolds of X . There might be
one or more ®bers Fy on which there is more than
one point attaining the minimum distance from x0 .
(5) The rank of f (=rank of dfx ) may not be the same at
all points x of U . The consequence is that R…f † is no
longer a smooth submanifold of Y and its behavior
at singular points (points with rank (dfx † < r) may
pose additional problems.
We shall keep these problems in mind when study-
ing the particular case of the mappings f arising in
geodetic problems.
The last three problems have to do with points in Y
or X which do not behave ``nicely'' with respect to the
de®nitions employing (G3), (G4), and the de®nition of
the rank r of f , r ˆ rank…dfx ). Starting from the latter,
we note that the set of points y ˆ f …x† 2 R…f †, for which
rank…dfx † < r, forms within R( f ) a set of measure zero.
This is in fact a consequence of Sard's theorem (Stern-
berg 1964, ch. II.3, Theorem 3.1), provided we impose
some mild assumptions on the di€erentiability of the
manifolds X and Y . In our case, where dim Y > dim X , it
suces to require that they are C 1 manifolds and f is
also a mapping of class C 1 .
A similar fact holds for the ``problematic'' points of
Y which happen to belong to more than two sets Ry^
which thus intersect and fail to be ®bers of Y . If such
points form a subset A ˆ Ry^1 \ Ry^2  Y which is not of
measure zero, then it will be possible to ®nd a y0 2 A and
a positive constant R such that BR …y0 †  A, where BR …y0 †
is the ball with center y0 and radius R. We shall show
that the assumption that A has nonzero measure leads to
a contradiction: let y be a point on the geodesic joining
y0 with y^2 such that y 2 BR …y0 †, in which case
q…y0 ; y^2 † ˆ q…y0 ; y† ‡ q…y; y^2 † and q…y; y^1 † ˆ q…y; y^2 †
) q…y0 ; y^1 † ˆ q…y0 ; y† ‡ q…y; y^1 †, i.e., y belongs also to
the geodesic joining y0 and y^1 , which is obviously im-
possible.
For the de®nition based on (G4) we can ascertain
that the set of problematic points forms a subset of X
with measure zero, provided that the ®bers Ff …x† do not
coincide locally with the boundaries of balls having x0 as
center.
In any case our de®nitions hold almost everywhere in
X and Y , i.e., except on subsets of measure zero. From a
physical point of view we can always work in a neigh-
borhood of X and the corresponding neighborhood of Y
which do not contain critical (problematic) points.
4 The geodetic nonlinear mapping
The nonlinear mapping f in the geodetic case arises
from n individual mappings fk ; k ˆ 1; . . . n, which map
coordinates of points into various types of observables.
We consider a geodetic network of N points with
Cartesian coordinates xi ; i ˆ 1; . . . ; N , where xi is a
vector of dimension either 2 (plane networks) or 3
(spatial networks). These vectors constitute a coordinate
vector
2 3
x1
6 .. 7
xˆ4 . 5 …11†
xN
of dimension 2N or 3N , which is considered an element
of the m-dimensional Euclidean space (m ˆ 2N or
m ˆ 3N )
ÿ
N ÿ
N
X ˆ E2N ˆ E2 or X ˆ E3N ˆ E3
equipped with the simple inner product …xa ; xb † ˆ xTa xb .
The geodetic observables we will examine here can be
distinguished as ``angular'' observables yijk and ``dis-
tance'' observables dik , which di€er with respect to their
invariance characteristics under coordinate transforma-
tions. Angular observables are invariant under similarity
coordinates transformations x0 ˆ S…x† while distance
observables are invariant under rigid coordinate trans-
formations x0 ˆ R…x†, which are de®ned pointwise by
x0i ˆ S…xi † ˆ k R xi ‡ t; x0i ˆ R…xi † ˆ R xi ‡ t …12†
t is a displacement vector, k > 0 a scale parameter and R
a proper orthogonal matrix …RT R ˆ R; RT ˆ I and
jRj ˆ 1) depending on a single rotational parameter h
in the two-dimensional case or on three rotational
parameters h1 ; h2 ; h3 in the three-dimensional case. The
observables are functions of the coordinates
yijk ˆ y…xi ; xj ; xk † ˆ yijk …x†
dik ˆ d…xi ; xk † ˆ dik …x† …13†
which have the invariance properties

y…xi ; xj ; xk † ˆ y…S…xi †; S…xj †; S…xk ††
…14†
d…xi ; xk † ˆ d…R…xi †; R…xk ††
or
yijk …x† ˆ yijk …S…x††; dik …x† ˆ dik …R…x†† …15†
The mapping f which maps the network coordinates x
to the observables y ˆ f …x† 2 R…f †  Y consists in
general of n1 angular mappings and n2 distance map-
pings …n1 ‡ n2 ˆ n† and we can distinguish two cases
with respect to its invariance properties. When n2 ˆ 0
(only angular observations) then
f …x† ˆ f …S…x†† …16†
while when n2 6ˆ 0 (observations of distances or both
distances and angles) then
f …x† ˆ f …R…x††
Remark. We must emphasize here that f is a nonlinear
mapping and S or R are general transformations not
necessarily close to the identity (represented by
k ˆ 1; R ˆ I; t ˆ 0†. It has been pointed out by Grafa-
rend and Kampmann (1996) that there are other groups
of transformations, such as the ten-parameter conformal
group G ˆ C10 …3†, in three dimensions; although it does
not satisfy Eq. (16), i.e., f …x† 6ˆ f …G…x††, when it is close
to the identity (G  Id), it satis®es the corresponding
relation dfx0 …x† ˆ dfx0 …G…x†† for the di€erential map-
ping dfx0 of f at any Taylor point x0 . In other words
when G is close to the identity (``small'' transformation)
it leaves linearized angular observations invariant, a
property shared of course with S. Therefore the C10 …3†
group, although of essentially no relevance to the strictly
nonlinear approach followed here, should be taken into
consideration in numerical applications, where we have
to resort, explicitly or implicitly (iterative solutions), to
linearization and solutions x; x0 close to x0 , in which
case the transformation from any such x to any such x0
is necessarily close to the identity.
The presence of angular observations has as a con-
sequence that f is not de®ned at every point of X . For
an angular observable yijk ˆ y…xi ; xj ; xk ) to be de®ned it
is necessary that the relevant network points are distinct,
i.e., we must exclude from X points x with xi ˆ xj , or
xj ˆ xk , or xk ˆ xi . Although the same problem does
not appear in the case of distance observations, we will
stick to the restriction that those network points which
are joined by observations must be distinct. Let J be the
index set of those pairs …i; k† of point indices corre-
sponding to points joined by observations and let Dik be
the diagonal subspace of X de®ned by
Dik ˆ fx 2 X jxi ˆ xk g …18†
We must exclude all inappropriate diagonal subspaces,
which leaves as the domain of de®nition of f the open
subset V  X de®ned by its complement V C with respect
to X
VC ˆ [ Dik ˆ D …19†
…i;k† 2 J
89
Since all Dik are closed, the same holds for D ˆ V C and
V is an open subset of X .
Turning to the range space Y we note that distances
are mappings from X to R‡ , since they obtain only
positive values, while angles are mappings from X to the
unit circle S. Thus, strictly speaking, f is a mapping
n2
f : V ! …S†n1  …R‡ † …20†
However it is standard practice to replace S with an
interval of R, our choice being I ˆ …0; 2pŠ, in which case
f becomes a mapping
n
f : V ! U  …I†n1  …R‡ † 2  Y ˆ En1 ‡n2 ˆ En …21†
Since I does not contain its limit point 0 and the same
holds true for R‡ , U is an open subset of Y . We have
assumed that Y ˆ En is Rn equipped with the euclidean
inner product, though a more general inner product
…ya ; yb † ˆ yTa Pyb may be used, in which case Y ˆ EPn .
The restriction of f from a mapping f : X ! Y to a
mapping f : V ! U , poses some serious problems as far
as the properties (G3) and (G4) of generalized inverses
are concerned. Now the range R…f † is restricted from
f …X † to U \ f …X † and further to U \ f …V †. The ad-
justment problem may have a solution y^ 2 f …X † closest
to a given y 2 U , such that y^ 2 = U \ f …V †. This means
that adjusted observations y^ may include negative or
zero distances, or angles outside the prescribed interval.
Turning to the domain V we must distinguish be-
tween the similarity transformation and the rigid
transformation case. When distances are observed and
y^ 2 f …X †, the ®bers Fy^ may be initially de®ned with the
help of any element x 2 X , such that ^ y ˆ f …x0 † by means
of
Fy^ ˆ fx 2 X jx ˆ R…x0 †g …22†
This means that if x0 2 V the same holds for any
x ˆ R…x0 † and therefore Fy^  V . Thus Fy^ is either
included in V or lies completely outside V . This means
that the problem of ®nding an element x^ of a ®ber closest
to a given element x0 2 V has a solution in V provided
that Fy^  V , i.e., that y^ 2 f …V †. When distance mea-
surements are not included the similarity transformation
S gives rise to ®bers
Fy^ ˆ fx 2 X jx ˆ S …x0 †g …23†
which are no longer closed subsets of X , because they
``converge'' at the zero element 0 2 X . Although the
scale is restricted to positive values k > 0, it is possible to
consider a sequence ki ! 0 , giving rise to a sequence of
similarity transformations Si such that
xi ˆ Si …x0 † 2 Fy^ ! 0 2
= Fy^ (convergence in norm). As a
consequence, the problem of ®nding an element x^ of a
®ber closest to a given element, x0 2 V may have no
solution (consider the case where 0 is the closest to x0
element from the closure F y^ of the ®ber Fy^).
These are problems of great concern to the mathe-
matician, but of little or no concern at all to the geo-
desist. The reason is that in geodetic applications we are
con®ned to a small neighborhood Ny  U \ f …V † of the
90
true observables y 2 U \ f …V †, such that the observa-
tions y and the adjusted observations y^ also belong to
Ny . In the same way it is possible to choose an element
x0 2 V such that Fy^ crosses a small neighborhood
Nx0  V of x0 and furthermore the element x^ of Fy^
closest to x0 also lies in the same neighborhood Nx0 .
Although we con®ne ourselves here to the deter-
ministic aspects of generalized inverses, we are tempted
to say that such a restriction to neighborhoods can be
also justi®ed when a probabilistic criterion is used for
the ``optimal'' choice of a generalized inverse, which
provides an estimator of x based on the observations y
(outcomes of random variables). In this case a Bayesian
point of view can be followed with a noninformative
prior distribution for x, such that it is homogeneous
within the desired neighborhood in X , and zero outside.
For a glimpse into the possible connection between
the deterministic approach taken here and probabilistic
aspects, consider the case where y is the outcome of a
random variable Y with likelihood function L…y; x† ˆ
pY …yjx†. If it so happens that L has the form L…x; y† ˆ
L…q…y; f …x††† where L…q† has a maximum when q is
minimized, while q quali®es as a metric, then maximum
likelihood estimate can be identi®ed with the solution
following from the use of a minimum-distance general-
ized inverse.
With these considerations in mind we turn to the
nonlinear datum problem. We assume that the adjust-
ment problem has already been solved and its solution y^
de®nes a speci®c ®ber Fy^  V  X , described by either
Eq. (22) or Eq. (23).
5 Solution to the nonlinear geodetic datum problem
The datum problem can be viewed as the problem of
completing the projection mapping p : Y ! R…f † which
solved the adjustment problem into a (minimum-dis-
tance) generalized inverse g ˆ c  p of f , with the choice
of a complementary mapping c : R…f † ! S, where S is a
section of the ®bering F induced by f in V . In fact it is
sucient to determine only S, in which case c (and thus
g) is automatically de®ned as the mapping of any
y 2 R…f † into c…y† ˆ x, where x is the unique element of
S \ Fy . One way of describing the d-dimensional man-
ifold S is by means of a set of d ˆ m ÿ r minimal
constraints v…x; d† ˆ 0, where d is a ®xed vector of d
parameters, which usually appear in the explicit form
h…x† ˆ d. The alternative possible description of S by the
m constraints x ˆ /…d; q†, where d is again a ®xed vector
of d parameters and q a vector of r free parameters, is
hardly used in geodesy.
Strictly speaking, a set of minimal constraints con-
sists of two things. First a family of mappings v…x;
†, or
/…
; q†, which corresponds to the choice of the type of
constraints, and second a set of speci®c values for the
parameters d. For example, in a horizontal network we
may decide to ®x the two coordinates of a particular
point and the azimuth of a particular side and thus
choose the function v…x;
† (or h…x†, or /…
; q†, analo-
gously). When we decide on the values to be assigned to
the ®xed coordinates and azimuth, we specify the value
of the parameter set d. The ®rst step of this procedure
corresponds to the choice of a ®bering S complemen-
tary to F, where each ®ber corresponds to a particular
set of values for d. In the second step the assignment of
speci®c values for d picks up a particular element S from
S which will serve as the section of F into which g will
map R…f †.
A particular choice for S is as the set of points x such
that if S \ Fy ˆ fxg then x is the closest point to a given
x0 2 V among all elements of Fy . The description of this
particular section by the corresponding set of minimal
constraints (inner constraints) is not as easy as in the
linear case. Instead we shall follow an approach similar
to that of Baarda's S-transformation in the linear case,
where a known reference ``minimal'' solution is trans-
formed into the desired ``inner'' solution, i.e., the x0 -
nearest solution.
In order to proceed with the solution we need an
analytical description of the solution ®bers as subman-
ifolds of X , e.g., by appropriate curvilinear coordinates.
Let G denote the group of applicable transformations
(i.e., either similarity or rigid) with elements g : X ! X .
If z belongs to a speci®c ®ber z 2 Fy then the same holds
true for x ˆ g…z† 2 Fy for every g 2 G.
This establishes for every ®ber Fy of X a natural
correspondence h : Fy  Fy ! G :
g ˆ h…z; x† , x ˆ g…z† …24†
This means that by ®xing one of the arguments of h
we can obtain a mapping hz : x ! hz …x† h…z; x† :
Fy ! G which is invertible. If further fG : G ! Rd is a
coordinate system on G (i.e., a parametrization of G),
the composite mapping
f  fG  hz : x ! p…x† : Fy ! Rd …25†
establishes a coordinate system on Fy . We have already
introduced the coordinate systems f and the corre-
sponding coordinates p in the descriptions, Eq. (12), of
the similarity and rigid transformations. The dimension
d of Fy depends on the applicable transformation group
and the dimension of the network. For two-dimensional
networks d ˆ 4 for the similarity case (coordinates
h; t1 ; t2 ; k) and d ˆ 3 for the rigid case (coordinates
h; t1 ; t2 ). For three-dimensional networks d ˆ 7 for the
similarity case (coordinates h1 ; h2 ; h3 ; t1 ; t2 ; t3 ; k) and
d ˆ 6 for the rigid case (coordinates h1 ; h2 ; h3 ; t1 ; t2 ; t3 ).
In fact we have introduced not one but a family of
coordinate systems depending on the ``reference point'' z
chosen for any ®ber Fy . This choice can be made by
introducing a section Z of the ®bering F of X , e.g., by
the use of minimal constraints which are easy to handle
in computations, e.g., of the form xik ˆ 0 (trivial
constraints), where k is the network point index and i
the coordinate index (i ˆ 1; 2 for plane and i ˆ 1; 2; 3 for
three-dimensional networks).
Once a reference solution z is established, all other
solutions in the same ®ber x ˆ x…z; p† depend on the
coordinates p which are identical to the parameters of
the transformation g…p† : z ! x. To obtain the solution
 91

speci®ed by a given set of d minimal constraints three-dimensional networks and similarity or rigid
v…x; d† ˆ 0 (usually of the form h…x† ˆ d† we have to transformations. The results are:
solve a system of d nonlinear equations with d un- Three-dimensional networks ± similarity transforma-
knowns tion:
P
v…p† ˆ v…x…z; p†; d† ˆ 0 …26† …x0i ÿ x0 †T R…zi ÿ z†
or in the usual form in geodesy xi ˆ x 0 ‡ i P R…h†…zi ÿ z† …30†
…zi ÿ z†T …zi ÿ z†
i
h…p† ˆ h…x…z; p†† ˆ d …27†
where
The solution p of the preceding system speci®es the
transformation gp ˆ g…p† which maps the reference 1X 1X
solution z into the minimal constraint solution z  zi ; x0  x0i …31†
N i N i
x ˆ gp …z†. This transformation is called the nonlinear
Baarda S-transformation in the similarity, and the and h is the solution of the three nonlinear equations
nonlinear Baarda R-transformation in the rigid case.
The solution x closest to a given element x0 2 X is 1X
‰…x0i ÿ x0 †ŠR…h†…zi ÿ z† ˆ 0 …32†
obtained either by minimizing N i
F …p† ˆ ‰x…z; p† ÿ x0 ŠT ‰x…z; p† ÿ x0 Š …28† Three-dimensional networks ± rigid transformation:
or by imposing the condition that the vector x0 ÿ x is xi ˆ x0 ‡ R…h†…zi ÿ z† …33†
orthogonal to the tangent space to the ®ber Fy^ at the
point x. Following the ®rst approach we are led to a where h is the solution of the three nonlinear equations
system of d nonlinear equations with d unknowns 1X
 T ‰…x0i ÿ x0 †ŠR…h†…zi ÿ z† ˆ 0 …34†
oF ox N i
…p† ˆ 0 ) …z; p†‰x…z; p† ÿ x0 Š ˆ 0 …29†
op op
Two-dimensional networks ± similarity transformation:
The solution p^ of this system speci®es the transfor- P
mation gp^ ˆ g…^ p† which maps the reference solution z …x0i ÿ x0 †T R…zi ÿ z†
i
into the solution x ˆ gp^…z† closest to x0 . gp^ is the non- xi ˆ x0 ‡ P R…#†…zi ÿ z† …35†
linear Baarda S-transformation (or R-transformation) to …zi ÿ z†T …zi ÿ z†
i
the x0 -nearest solution corresponding to the inner so-
lution of Meissl in the linear case. where # is the solution of the single nonlinear equation
What about the famous inner constraints of Meissl
which appear in the linear case? Well, oF 1X oR
op ˆ 0 can be …x0i ÿ x0 †T …zi ÿ z† ˆ 0 …36†
solved in principle to give p ˆ p…x0 ; z†, which substituted N i o#
 
in x ˆ ‰g…p†Š…z† ˆ x…z; p† gives an expression of the form cos # sin #
x ˆ x…x0 ; z† ˆ x…z†, since x0 is ®xed a priori. Recalling When R…#† ˆ
that z is determined by the intersection of a solution  ÿ sin# cos #
1 b a
®ber Fy 2 F with a section S0 of F determined by a set xi ˆ x0 ‡ 2 …zi ÿ z† …37†
of minimal constraints v…z; d† ˆ 0 or z ˆ /…d; q†, it is sz ÿa b
possible in principle to have a description of the r-di- where
mensional manifold S0 in the form z ˆ z…q†, where q is a  
vector of r free parameters serving in fact as coordinates 1X 0 1
for S0 . If this representation is replaced in the inner a ‰…x0i ÿ x0 †T …zi ÿ z† …38†
N i ÿ1 0
solution x ˆ x…z† we obtain x ˆ x…z…q†† ˆ w…q†. The
resulting relation x ˆ w…q† is the set of inner constraints 1X
which to every q assigns the element of the ®ber b ‰…x0i ÿ x0 †T …zi ÿ z† …39†
N i
pF …z…q†† of F passing through z ˆ z…q† its element
closest to x0 . Unlike the linear case, the inner constraints 1X
are not independent of the reference minimal constraints s2z  ‰…zi ÿ z†T …zi ÿ z† …40†
because they depend on their parametrization in terms N i
of q. The above description of the inner constraints is a
Two-dimensional networks ± rigid transformation:
conceptual rather than a practical one since the analyt-
ical solution of nonlinear equations or the derivation of xi ˆ x0 ‡ R…#†…zi ÿ z† …41†
desired alternative descriptions of sections are not pos-
sible as a rule. where # is the solution of the single nonlinear equation
The detailed derivation of the speci®c form of the 1 Xh oR i
nonlinear system given by Eq. (29) is given in Appendix …x0i ÿ x0 †T …zi ÿ z† ˆ 0 …42†
A for the four special cases corresponding to two- or N i o#
92
 
cos # sin #
When R…#† ˆ F0 FT0 vi ˆ l2i vi ) F0 FT0 VMVT …50†
ÿ sin  # cos#
1 b a V ˆ ‰v1 v2 . . . vn Š; Mik ˆ dik l2i
xi ˆ x0  p …zi ÿ z† …43†
2
a ‡b 2 ÿa b
The eigenvectors fui g; fvi g form orthonormal systems
where a and b are de®ned as in the previous case. and they can be seen as the representations with respect
to the initial bases of the new bases, in which case U and
V become the matrices of change of bases, already
6 Application to the linear case introduced with the same notation. Furthermore the
number of non-zero eigenvalues is r ˆ rank…f † for both
We have outlined an approach to the representation of k2i and l2i , which are also identical, i.e., l2i ˆ k2i ;
generalized inverses of nonlinear operators, and the i ˆ 1; 2; . . . r. In this case, if K is the r  r diagonal
question arises how this approach agrees with the matrix with elements
existing representation theory for (linear) generalized
inverses of linear operators (Takos 1976; Teunissen Kik ˆ dik ki …51†
1985). Since linear operators are simply special cases of the representations with respect to the new bases are
nonlinear operators we will proceed to show that the  
existing theory can be derived as a special case of our K 0
nonlinear case. FT F ˆ UT …FT0 F0 †U ˆ L ˆ …52†
0 0
If f : X ! Y is a linear operator, we seek a linear
 
operator g : Y ! X which is a generalized inverse of f , T T T K 0
i.e., it satis®es property (G1). In terms of given ``initial'' FF ˆ V …F0 F0 †V ˆ M ˆ …53†
0 0
orthonormal bases feX Y0
i g for X and fei g for Y , f is
0

represented by an n  m matrix F0 and we seek the m  n  

T K 0
matrix G0 representing g. We call these bases initial F ˆ V F0 U ˆ …54†
0 0
because we are going to carry out our investigation in
terms of ``new'' more convenient orthonormal bases
The simplicity of this representation is the key to the
feXi g, feYi g, with respect to which f and g are repre-
speci®c choice of the new bases. This is essentially the
sented by matrices F and G, respectively. It is well
same representation as in Takos (1976), since if the
known that if U and V are the orthogonal matrices of
normality requirement is removed we can obtain a
the change of bases, i.e.,
representation with respect to a new orthogonal (but not
X
m X
n necessarily orthonormal) base
eXi ˆ Uik eXk 0 ; eYi ˆ Vik eYk 0 …44†  ÿ1=2     
kˆ1 kˆ1
F0 ˆ K 0 VT F U Kÿ1=2 0 ˆ
Ir 0
0
0 Inÿr 0 Imÿr 0 0
it holds that
…55†
F ˆ VT F0 U; G ˆ UT G0 V …45†
We shall stick to the representation of Eq. (54) and will
F0 ˆ V F U T ; G0 ˆ U G VT …46† relate the section S, the ®berings P, its re®nement G
which are needed for the determination g, to the
For vectors x 2 X and y 2 Y it holds for their represen- submatrices of
tations that    ÿ1 
T H K H
x ˆ U T x0 ; y ˆ VT y0 …47† Gˆ ˆ …56†
J K J K
x0 ˆ Ux; y0 ˆ Vy …48†
where T ˆ Kÿ1 follows directly from the property (G1):
The choice of the new bases will be based on the so- FGF ˆ F.
called singular value decomposition (Rao and Mitra From the discussion in Sect. 2 it is obvious that in the
1971, p.6) where fexi g and feyi g are the eigenvectors of linear case the relevant ®berings consist of ane sub-
the mappings …f   f † : X ! X and …f  f  † : Y ! Y ; spaces which are parallel translates of ``generating''
respectively, where f  : Y ! X is the adjoint mapping linear subspaces. They can be identi®ed with the concept
of f . Recall that if …; †X and …; †Y are the inner products of quotient spaces. A quotient space of a linear space Z is
in X and Y then f  is de®ned by …y; f …x††Y ˆ …f  …y†; x†X a space Z=M; M being a subspace of Z, consisting of
for every x 2 X and y 2 Y (see, e.g., Halmos 1974, equivalent classes of elements of Z, where two elements
Sect. 44 and Sect. 68). In terms of the initial bases f  is are equivalent, x1  x2 , if x1 ÿ x2 2 M. The ®bering
represented by FT0 ; …f   f † by FT0 F0 ; …f  f  † by F0 FT0 P ˆ Y =C is generated by a subspace C, its re®nement
while the eigenvector problems are expressed by G ˆ Y =N …g† by N …g† and the ®bering Q ˆ F ˆ X =N …f †
by N …f †. The determination of g is completed by its
FT0 F0 ui ˆ k2i ui ) FT0 F0 ˆ ULUT …49† action from C to the subspace D ˆ g…C†.
Let us note ®rst that the idempotent mappings
U ˆ ‰u1 u2 . . . um Š; Lik ˆ dik k2i p ˆ f  g and q ˆ g  f are in this case linear projections
 93
 
ÿKH
on R…f † and S, respectively; p is represented by the basis for R…f † and the columns of a basis for C.
If
matrix product FG and q by GF, which explicitly are  
     I JK
K 0 Kÿ1 H I KH Since X ˆ S  N …f † the columns of r will rep-
FG ˆ ˆ r …57† 0 Id
0 0 J K 0 0   
I JK x1
 ÿ1     resent a basis for X and we may set x ˆ r .
K H K 0 Ir 0 0 Id x2
GF ˆ ˆ …58†  
J K 0 0 JK 0 Ir ÿKH
Since Y ˆ R…f †  C, the columns of will
  0 If
Ir represent a basis
Lemma. The subspace R…f † is represented by R    for Y and we may set
0 Ir ÿKH y1
  yˆ .
0 0 If y2
while N …f † is represented by R .
Id
Upon replacing these representations in x ˆ Gy we
Proof. x 2 N …f † , Gx ˆ 0 , obtain
          
K 0 x1 Kx1 Ir JK x1 Ir ÿKH y1
ˆ ˆ 0 , x1 ˆ 0 ˆG and
0 0 x2 0 0 Id x2 0 If y2
             
0 0 0 x1 Ir JK ÿ1 Ir ÿKH y1
,xˆ ˆ x2 , x 2 R : ˆ G
x2 Id Id x2 0 Id 0 If y2
 " ÿ1 #  
y 2 R…f † , 9 x 2 Rm : y ˆ Fx , Ir 0 K H Ir 0 y1
         ˆ
K 0 x1 Kx1 Ir Ir ÿJK Id J K 0 If y2
yˆ ˆ ˆ Kx1 2 R
0 0 x2 0 0 0 and after carrying out the computations
(    ÿ1  
x1 K 0 y1
ˆ …59†
x2 0 Q y2
Lemma. The section S ˆ g…R…f †† is represented by the
  where
I
range R .
JK Q  K ÿ JKH …60†
Proof. If x^ 2 S ˆ R…q† there exists x 2 X such that From the explicit form of Eq. (59) x1 ˆ Kÿ1 y1 ; x2 ˆ Qy2
x^ ˆ q…x† or it follows that the matrix Q determines the action of g on
     C, since if y 2 C then y2 ˆ 0; x1 ˆ 0 and
Ir 0 x 1 x1
x^ ˆ GFx ˆ ˆ        
JK 0 x2 JKx1 0 0 0 0
    xˆ ˆ ˆ Qy2 2 R
I I x2 Qy2 Id Id
ˆ x1 2 R (
JK JK The subspace D ˆ g…C† is represented by R…D†; D being
any of the m  f matrices given by
 ÿ1    
K H ÿKH 0
Lemma. The subspace C  Y de®ned by C ˆ N …p† Dˆ Tˆ T …61†
  J K If K ÿ JKH
ÿKH
ˆ N …f  g† is represented as the range R . where T is any f  f nonsingular matrix.
If
Summarizing these results we may say that the sub-
matrix J determines the section S, the submatrix H de-
Proof. y 2 C , p…y† ˆ …f  g†…y† ˆ 0 , FGy ˆ 0 termines the subspace C ˆ N …p† and thus the ®bering P
       and the remaining submatrix K determines the subspace
Ir KH y1 y1 ‡ KHy2 0
, ˆ ˆ D ˆ g…C†. The linear projection p represented by FG is a
0 0 y2 0 0 projection on R…f †, represented by R…F†, along the
 
, y1 ‡ KHy2 ˆ 0 , y1 ˆ ÿKHy2 , ÿKH
        nullspace N …p† ˆ C represented by R . The
y1 ÿKHy2 ÿKH ÿKH If
yˆ ˆ ˆ y2 2 R linear projection q represented by GF is a projection on
y2 y2 Ir Ir  
( I
  S represented by R along the nullspace
Ir JK
The columns of represent a basis for S, the
  JK   N …q† ˆ N …f † represented by N …F†.
0 I
columns of a basis for N …f †, the columns of r a We will now take a look into the special types of
Id 0 generalized inverses resulting by adding to the property
94
 
0
(G1) a combination from the properties (G2), (G3), all be orthogonal to the columns of spanning
Id
(G4).
N …f †. This can be expressed in the form
Lemma. For a re¯exive generalized inverse G of F (g of " #T  
f ) it holds that K ˆ JKH, i.e., Kÿ1 0
ˆ 0 , KÿT T T
r 0 ‡ J Id ˆ J ˆ 0
 ÿ1  J Id
K H
Gˆ …62†
J JKH ,Jˆ0 (

Proof. It follows directly from the property (G2): Remark. Upon looking into the explicit form of the
GFG ˆ G. ( matrices    
In the linear case where X and Y are inner-product I KH Ir 0
FG ˆ r and GF ˆ , it is easy to
linear spaces with distances determined by the norm, the 0 0 JK 0
minimum-distance generalized inverse is called the least- see that H ˆ 0 , …FG†T ˆ FG, while J ˆ 0 ,
squares generalized inverse and choosing x0 ˆ 0, the …GF†T ˆ GF. Thus we obtain the more familiar form of
0-nearest generalized inverse is called minimum- norm the properties (G3) and (G4) for least-squares and
generalized inverse. minimum-norm generalized inverses of matrices.

Lemma. For a least-squares generalized inverse it holds Corollary. A re¯exive least-squares inverse is represent-
that H ˆ 0, i.e., ed by
 ÿ1   ÿ1 
K 0 K 0
Gˆ …63† Gˆ …65†
J K J 0
a re¯exive minimum-norm inverse by
Proof. If G represents a least-squares inverse g then FG  ÿ1 
must represent an orthogonal projection p on R…f †. In K H
Gˆ …66†
other words, for any y 2 Y ; y^ ˆ p…y† represented by 0 0
^
y ˆ FGy is the closest element to y from R…f † if a least squares-minimum norm inverse by
y^ ÿ y ? R…f †, i.e., y^ ÿ y 2 R…f †? , where R…f †? is the  ÿ1 
orthogonal complement of R…f † in X . Since y^ ÿ y 2 C K 0
and C is a subspace complementary to R…f † it must hold Gˆ …67†
  0 K
that C ˆ R…f †? . ÿKH
Since C is represented by R it must hold and ®nally a re¯exive least squares-minimum norm
If
that inverse (pseudo inverse) by
 ÿ1 
 T K 0
ÿKH Gˆ …68†
Fx ˆ 0 for every x 0 0
If
 T   T (
ÿKH K 0 x1
, ˆ0 A generalized inverse G of the matrix F is speci®ed if the
If 0 0 x2 subspaces C  Y and S  X are known by means of
, ÿHT Kx1 ˆ 0 for every x1 , H ˆ 0 ( speci®c representations and in addition the action of g
on C is speci®ed. Alternatively, G is uniquely determined
once (in addition to C and S) we know the images under
Lemma. For a minimum-norm generalized inverse it g of the elements of a basis of C, which images span and
holds that J ˆ 0, i.e., thus determine both the action of g on C and the
 ÿ1  subspace D ˆ g…C†  N …f †  X .
K H Let C; S and D be represented in the original bases by
Gˆ …64†
0 K R…C0 †; R…S0 † and R…D0 † with D0 ˆ G0 C0 .
In the special bases introduced by the singular value
decomposition
  C is represented by the range R…C†
Proof. If G represents a minimum-norm inverse g then q C1
represented by GF must project every x0 2 Fy to an ˆR with jC2 j 6ˆ 0 (since dimC ˆ f ; C is n  f ,
C2
element x^ 2 Fy which has the minimum norm among all
elements of the same ®ber Fy . Since Fy ˆ x0 ‡ N …f † this so that C1 is r  f and C2 is f  f †, which means that
happens only when x^ ? N …f †. We have seen that q maps the f columns of C represent the elements of a basis for
elements of X on the section S, and in view of the same C.
dimension of these subspaces In the same special bases S is represented by the
 ÿ1it must hold that   
K S1
S ˆ N …f †? . The columns of span S and must range R…S† ˆ R with jS1 j 6ˆ 0 (since dimS ˆ r; S
J S2
 95

is m  r, so that S1 is r  r and S2 is d  r), which means D1 ˆ UT1 D0 ˆ 0 …78†

that the r columns of S represent the elements of a basis
for S. D2  QC2 ˆ …K ÿ JKH†C2 …79†
The subspace D ˆ g…C† is represented by the matrix
and
D ˆ GC, which means that the f columns of D represent
a spanning set of D, which, however, is not a basis of D K ˆ Q ‡ JKH ˆ D2 Cÿ1
2 ‡ JKH
because dim D ˆ Dr < f in general.
In order to relate the two types of representations we ˆ UT2 D0 …VT2 C0 †ÿ1 …80†
may apply Eq. (48) to the columns of S; C, and D to ÿ1
ÿ UT2 S0 …UT1 S0 † Kÿ1 VT1 C0 …VT2 C0 † ÿ1
obtain
The upper zero submatrix D1 in D is a direct conse-
S0 ˆ US; C0 ˆ VC; D0 ˆ UD …69†
quence of the fact the D  N …f † where N …f † is
  
with inverse relations 0
represented by R . The same fact is expressed
Id
S ˆ U T S0 ; C ˆ VT C0 ; D ˆ UT D0 …70†
in the original bases by the condition UT1 D0 ˆ 0.
Setting
Remark. The matrix D is more than a representation of
U ˆ ‰U1 U2 Š; V ˆ ‰V1 V2 Š …71† the subspace D. This would be the case if the matrix T
was chosen arbitrarily. The choice T ˆ C2 makes D
where U1 is m  r, U2 is m  d, V1 is n  r and V2 is equal to GC and thus the columns of D are images under
n  f , we obtain g of the columns of C, i.e., of the speci®c given
description for C. Thus D not only describes the
S1 ˆ UT1 S0 ; S2 ˆ UT2 S0 …72† subspace D but it also describes the action of g on the
subspace C. Due to the fact that D  N …f † the relation
D1 ˆ UT1 D0 ; D2 ˆ UT2 D0 …73† D ˆ GC degenerates into D2 ˆ QC2 . Every column of C
is uniquely determined by the corresponding column of
C2 . Indeed if ei the i-th column of If then the i-th
C1 ˆ VT1 C0 ; C2 ˆ VT2 C0 …74† column of C is
 
The relations can be used to obtain G0 (the represen- ÿKH
Cei ˆ C2 ei and is determined by C2 ei , the i-th
tation of g in the original bases) in terms of the matrices If
S0 ; C0 ; D0 which represent the subspaces S; C; D; respec- column of C2 . Therefore C2 represents in a certain sense
tively, in the original bases. C and thus the subspace C. The matrix Q, which is
    uniquely related to K once H and J have been ®xed by
S1 I the choice of C and S, de®nes through the relation
Since R ˆR ; there exists an r  r non-
S2 JK D2 ˆ QC2 the action of g on C. (
     
S1 I I
singular matrix L such that ˆ Lˆ Thus the submatrices J; H and K which determine G
S2 JK JKL have now been related to the matrices C0 ; S0 , and D0
which implies that L ˆ S1 and which determine the subspaces C; S, and D, respectively,
J ˆ S2 Sÿ1 ÿ1
ˆ UT2 S0 …UT1 S0 †ÿ1 Kÿ1 as well as the action of g on C. The matrix G0 ˆ UGVT
1 K …75†
can now be expressed in terms of the matrices C0 ; S0 ,
    and D0 and the change of bases matrices U and V, which
C1 ÿKH also assume geometric representations.
Since R ˆR there exists an f  f
C2 If  
C1
non-singular matrix T such that Lemma. In the original basis R…f † is represented by
    C
ÿKH ÿKHT
2
R…V1 †; R…f †? by R…V2 †; N …f † by R…U2 † and N …f †? by
ˆ Tˆ which implies that T ˆ C2 R…U1 †.
If T
and Proof. Since V and U are orthogonal we obtain
ÿ1
H ˆ ÿKÿ1 C1 Cÿ1 ÿ1 T T
2 ˆ ÿK V1 C0 …V2 C0 † …76† I ˆ UT U )
In correspondence to the speci®c representation of C by UT1 U1 ˆ Ir ; UT2 U2 ˆ Id ; UT1 U2 ˆ 0; UT2 U1 ˆ 0
R…C† the subspace D ˆ g…C† is represented by R…D†
where the m  f matrix D is now given by Eq. (61) with I ˆ VT V )
T ˆ C2
      VT1 V1 ˆ Ir ; VT2 V2 ˆ Id ; VT1 V2 ˆ 0; VT2 V1 ˆ 0
0 0 0   
Dˆ C ˆ ˆ …77† Ir
K ÿ JKH 2 QC2 D2 R…f † is represented in the special basis by R , so
0
where that in the original basis it will be represented according
96
 
Ir
to Eq. (48) by the column range of U ˆ U1 . From The representation of a general generalized inverse
0
given by Eq. (81), despite its algebraic form and origin,
UT2 U1 ˆ 0, it follows that R…f †? is representedby R…U
 2 †. has a purely geometric character. All the submatrices
0 appearing in it have clear geometric meanings. A similar
N …f † is represented in the special basis by R , so
Id representation has been presented by Teunissen (1985,
that in the original basis it will be represented
  according Eq. 2.8), which in our notation can be written as
0
to Eq. (48) by the column range of V ˆ V2 . From G0 ˆ S0 ‰…C? T ÿ1 ? T
Id 0 † F0 S0 Š …C0 † …82†
T ?
V1 V2 ˆ 0, it follows that N …f † is represented by ? T ÿ1
‡ D0 ‰……F0 S0 † † C0 Š ……F0 S0 † † ? T

R…V1 †. ( where for a p  q matrix M with q < m, M? denotes a

p  …p ÿ q† matrix such that rank‰MjM? Š ˆ p and
Remark. Since the column space of a matrix remains the MT M? ˆ 0 (The column span of M? is an orthogonal
same when the matrix is multiplied from the left with a complement of the column span of M in Ep ). We prove
nonsingular matrix, we have the equally valid represen- the identity of the two representations of Eqs. (81) and
tations of R…f † by R…V1 K† and of N …f †? by R…U1 K†. ( (82) in Appendix B.
In the representation of Eq. (82) the matrices C? 0 and
Lemma. …F0 S0 †? are not uniquely de®ned, although G0 is. Our
(admittedly less elegant) representation by Eq. (81) gives
G0 ˆ S0 ‰…U1 K†T S0 Šÿ1 ‰VT1 ÿ VT1 C0 …VT2 C0 †ÿ1 VT2 Š a unique expression for G0 in terms of the chosen spe-
…81†
‡ D0 ‰VT2 C0 Šÿ1 VT2 ci®c descriptions of the subspaces C; S; D and the action
of g on C, as well as, in terms of the speci®c descriptions
of R…f †, N …f †, R…f †? and N …f †? associated with the
Proof. nonzero eigenvalues and the eigenvectors of the matrices
" #" # FT0 F0 and F0 FT0 . Of course the choice of descriptions for
T Kÿ1 H VT1
G0 ˆ UGV ˆ ‰U1 U2 Š the above subspaces is not unique.
J K VT2 It is easy to derive corresponding representation for
ˆ U1 K ÿ1
VT1 ‡ U1 HVT2 ‡ U2 JVT1 ‡ U2 KVT2 special types of generalized inverses. We can transform
the relevant representations Eqs. (56), (62)±(68) in terms
ˆ U1 Kÿ1 VT1 ÿ U1 Kÿ1 VT1 C0 …VT2 C0 †ÿ1 VT2 of the submatrices H, J, K to the original bases using
Eqs. (75), (76), (80). An alternative approach is to
‡ U2 UT2 S0 …UT1 S0 †ÿ1 Kÿ1 VT1 transform the restricting properties K ˆ JKH, H ˆ 0,
J ˆ 0, to the original bases using Eqs. (75), (76), (80) and
‡ U2 UT2 D0 …VT2 C0 †ÿ1 VT2
then apply the resulting conditions D0 ˆ 0, VT1 C0 ˆ 0,
ÿ U2 UT2 S0 …UT1 S0 †ÿ1 Kÿ1 VT1 C0 …VT2 C0 †ÿ1 VT2 UT2 S0 ˆ 0, respectively, on the general representation
Eq. (81). The results are:
ˆ U1 UT1 S0 …UT1 S0 †ÿ1 Kÿ1 VT1
Re¯exive generalized inverse:
ÿ U1 UT1 S0 …UT1 S0 †ÿ1 Kÿ1 VT1 C0 …VT2 C0 †ÿ1 VT2
G0 ˆ S0 ‰…U1 K†T S0 Šÿ1 VT1 ‰I ÿ C0 …VT2 C0 †ÿ1 VT2 Š …83†
‡ U2 UT2 S0 …UT1 S0 †ÿ1 Kÿ1 VT1
‡ U2 UT2 D0 …VT2 C0 †ÿ1 VT2 Least-squares generalized inverse:

ÿ U2 UT2 S0 …UT1 S0 †ÿ1 Kÿ1 VT1 C0 …VT2 C0 †ÿ1 VT2 G0 ˆ S0 ‰…U1 K†T S0 Šÿ1 VT1 ‡ D0 ‰VT2 C0 Šÿ1 VT2 Š …84†
 T
T U Minimum-norm generalized inverse:
which in view of I ˆ UU ˆ ‰U1 U2 Š 1T ˆ
U2
U1 UT1 ‡ U2 UT2 and the condition UT1 D0 ˆ 0 becomes G0 ˆ U1 …V1 K†ÿ1 ÿ S0 ‰…U1 K†T S0 Šÿ1 VT1 C0 …VT2 C0 †ÿ1 VT2
‡ D0 ‰VT2 C0 Šÿ1 VT2 …85†
G0 ˆ S0 …UT1 S0 †ÿ1 Kÿ1 VT1
ÿ S0 …UT1 S0 †ÿ1 Kÿ1 VT1 C0 …VT2 C0 †ÿ1 VT2 Re¯exive least-squares inverse:

‡ …I ÿ U1 UT1 †D0 …VT2 C0 †ÿ1 VT2 G0 ˆ S0 ‰…U1 K†T S0 Šÿ1 VT1 …86†

ˆ S0 …UT1 S0 †ÿ1 Kÿ1 VT1 Re¯exive-minimum norm inverse:

ÿ S0 …UT1 S0 †ÿ1 Kÿ1 VT1 C0 …VT2 C0 †ÿ1 VT2 G0 ˆ S0 ‰…U1 K†T S0 Šÿ1 ‰VT1 ÿ VT1 C0 …VT2 C0 †ÿ1 VT2 Š
‡ D0 …VT2 C0 †ÿ1 VT2 ‡ D0 ‰VT2 C0 Šÿ1 VT2 …87†

ˆ S0 ‰KUT1 S0 Šÿ1 ‰VT1 ÿ VT1 C0 …VT2 C0 †ÿ1 VT2 Š Least squares-minimum norm inverse:
‡ D0 ‰VT2 C0 Šÿ1 VT2 ( G0 ˆ U1 …V1 K†ÿ1 ‰I ÿ C0 …VT2 C0 †ÿ1 VT2 Š …88†
 97

Pseudo-inverse: X oxi
…xi ÿ x0i †T
oh
G0 ˆ U1 Kÿ1 VT1 ˆ …U1 K†Kÿ1 …V1 K†ÿ1 …89† i
X …A7†
ˆ …kRzi ‡ t ÿ x0i †T …ÿkR‰zi ŠRT X† ˆ 0
All these representations involve the submatrices C0 , S0 , i
V1 , V2 , (V1 K) and (U1 K) which have a clear geometrical X
meaning. oxi X
…xi ÿ x0i †T ˆ …kRzi ‡ t ÿ x0i †T I ˆ 0 …A8†
i
ot i
Appendix A X oxi X
…xi ÿ x0i †T ˆ …kRzi ‡ t ÿ x0i †T …Rzi † ˆ 0
Derivation of the nonlinear Baarda transformations i
ok i
to the inner solution …A9†
We shall derive here the solution to the nonlinear which assuming that k 6ˆ 0 and jXj 6ˆ 0, can be rewritten
Baarda transformations, which are coordinate transfor- as
mations x ˆ S…z† from any minimum constraints solu- X X X
tion z to the x0 -nearest element x of the f -induced ®ber k ‰zi Šzi ‡ ‰zi ŠRT t ˆ ‰zi ŠRT x0i …A10†
i i i
Ff …z† . Here X X X
    kR zi ‡ tˆ x0i …A11†
xT ˆ xT1 xT2 . . . xTN ; xT0 ˆ xT01 xT02 . . . xT0N …A1† i i i
X X X
and in the most general case the three-dimensional k zTi zi ‡ zTi RT t ˆ zTi RT x0i …A12†
transformation sought can be expressed pointwise by i i i
the similarity transformation
  Taking into account that ‰zi Šzi ˆ 0 and setting
xi ˆ kR…h†zi ‡ t ˆ xi …zi ; p†; pT ˆ hT tT k …A2† 1X 1X
z zi ; x0  x0i …A13†
where the transformation parameters p consist of three N i N i
rotational parameters h ˆ ‰h1 h2 h3 ŠT de®ning the orthog- Eqs. (A10)±(A12) take the form
onal matrix R ˆ R…h†, three parallel displacement X
parameters t ˆ ‰t1 t2 t3 ŠT , and a scale parameter k. N ‰zŠRT t ˆ ‰zi ŠRT x0i …A14†
The problem is to ®nd the optimal values of h; t; k i
which minimize f ˆ f …p† ˆ …x ÿ x0 †T …x ÿ x0 †. We ®rst kRz ‡ t ˆ x0 …A15†
note that for X X
k zTi zi ‡ N zT RT t ˆ zTi RT x0i …A16†
o
R_ ˆ R : i i
oa Solving Eq. (A15) for t
T _ T ‡ RR_ T ˆ 0
RR ˆ I ) RR t ˆ x0 ÿ kRz …A17†
) RR _ T †T  ‰xŠ ) R_ ˆ ‰xŠR
_ T ˆ ÿ…RR and replacing in Eq. (A16) gives
P T T
This can be applied for the derivatives with respect to hk zi R …x0i ÿ x0 †
by setting oho k R ˆ ‰xk ŠR. i
kˆ P T …A18†
zi zi ÿ N zT z
oxi o i
ˆk Rzi ˆ k‰xk ŠRzi ˆ ÿk‰…Rzi †Šxk
ohk ohk …A3† which can be also written in the form
ˆ ÿkR‰zi ŠRT xk P
  …x0i ÿ x0 †T R…zi ÿ z†
oxi oxi oxi oxi i
kˆ P …A19†
ˆ …A4† …zi ÿ z†T …zi ÿ z†
oh oh1 oh2 oh3
i
ˆ ÿkR‰zi ŠRT ‰x1 x2 x3 Š  ÿkR‰zi ŠRT X
To determine R (in fact h) we replace t in Eq. (A14)
where X
N ‰zŠRT x0 ÿ kN ‰zŠz ˆ ‰zi ŠRT x0i …A20†
oxi oxi i
ˆ I; ˆ Rzi …A5†
ot ok which in view of ‰zŠz ˆ 0 can be written as
To minimize f we simply set 1X
  ‰zi ŠRT x0i ÿ ‰zŠRT x0
of of ox T ox ox ox
N i
ˆ ˆ 2…x ÿ x0 † ˆ0 …A6† …A21†
op ox op oh ot ok 1X
ˆ ‰x0i ŠRzi ÿ ‰x0 ŠRz ˆ 0
N i
or explicitly
98

or in the equivalent form With this particular choice Eq. (A28) takes the form
1X a cos # ˆ b sin # …A32†
‰…x0i ÿ x0 †ŠR…zi ÿ z† ˆ 0 …A22†
N i
where
Equation (A22) is a nonlinear equation which can be 1X
solved to obtain the values of the parameters h in any a …x0i ÿ x0 †T W…zi ÿ z†
N i
particular parametrization of the rotation matrix R…h†. …A33†
The obtained values should be next substituted in 1X
b …x0i ÿ x0 †T …zi ÿ z†
Eq. (A19) in order to determine the value of the scale N i
parameter k. The similarity transformation can be re-
alized using these values, once t and k are replaced from There are two solutions: cos # ˆ  p 6
; sin # ˆ
Eq. (A17) and (A19), respectively, into Eq. (A2) to ob- a2 ‡b2
tain  p
6
2
a ‡b 2
(both positive or both negative). Replacing in
P Eq. (A29) we obtain
…x0i ÿ x0 †T R…zi ÿ z†
i p
xi ˆ x0 ‡ P R…h†…zi ÿ z† …A23† b cos # ‡ a sin # a2 ‡ b2
…zi ÿ z†T …zi ÿ z† kˆ ˆ  …A34†
i s2z s2z
In the case of the rigid transformation xi ˆ R…h†zi ‡ t we where
obtain only Eds. (A7), (A8) with k ˆ 1 and following the 1X
same procedure as before we arrive at the solution s2z  …zi ÿ z†T …zi ÿ z† …A35†
N i
1X
‰…x0i ÿ x0 †ŠR…zi ÿ z† ˆ 0 …A24† With the obtained values of sin #; cos # and k, the planar
N i
similarity transformation of Eq. (A30) becomes
xi ˆ x0 ‡ R…h†…zi ÿ z† …A25†  
1 b a
xi ˆ x0 ‡ 2 …zi ÿ z† …A36†
The planar (two-dimensional) case is somewhat dif- sz ÿa b
ferent since the 2  2 rotation matrix R…#† depends only
on a single parameter #. Setting In the case of the planar rigid transformation, the
solution is given by
o _ T
W RRT ˆ RR …A26† 1X
o# …x0i ÿ x†T WR…zi ÿ z† ˆ 0 …A37†
N i
where W is a 2  2 antisymmetric matrix, the only
di€erence from the three-dimensional similarity trans-
formation case (apart from the fact that the vectors xi ˆ x0 ‡ R…#†…zi ÿ z† …A38†
xi ; x0i ; zi ; x0 ; z involved are now two-dimensional) is that
Eq. (A7) is replaced by and for the above speci®c parametrization of R…#† we
obtain two solutions
X oxi X  
…xi ÿ x0i †T ˆ …kRzi ‡ t ÿ x0i †T …kWRzi † ˆ 0 1 b a
i
o# i xi ˆ x0  p …zi ÿ z† …A39†
a2 ‡ b2 ÿa b
…A27†
which correspond to two rotation angles # and # ‡ p. A
and the ®nal solution for the planar similarity transfor- further investigation is needed in each particular appli-
mation takes the form cation, in order to determine which of the two solutions
1X minimizes in fact the target function f ˆ
…x0i ÿ x0 †T WR…zi ÿ z† ˆ 0 …A28† …x ÿ x0 †T …x ÿ x0 †:
N i
P
…x0i ÿ x0 †T R…zi ÿ z†
i Appendix B
kˆ P …A29†
…zi ÿ z†T …zi ÿ z†
i Proof of the equivalence of the representations
of Eqs. (81) and (82)
xi ˆ x0 ‡ kR…#†…zi ÿ z† …A30†
Our starting point is Eq. (81)
We can proceed further with the solution by choosing
the usual parametrization
    G0 ˆ S0 ‰KUT1 S0 Šÿ1 ‰VT1 ÿ VT1 C0 …VT2 C0 †ÿ1 VT2 Š
cos # sin # 0 1 (B1)
R…#† ˆ )Wˆ …A31†
ÿ sin # cos # ÿ1 0 ‡ D0 ‰VT2 C0 Šÿ1 VT2
 99

We need an interpretation of this formula which will be With the obvious simplest choice B ˆ Id we obtain
independent of U1 ; U2 ; V1 ; V2 and depends only on  
S0 ; C0 ; D0 , and related matrices. ? 0
…FS† ˆ : (
Let us consider in addition to the matrix C with Id
column span R…C† a matrix C? with column span
R…C†? ˆ R…C†? . This means that C? must be such that Returning to the original bases we have

rank‰CjC? Š ˆ n and CT C? ˆ 0 0 ˆ …FS†T …FS†? ˆ …VT F0 UUT S0 †T …FS†?

ˆ …F0 S0 †T V…FS†?
?
Lemma. A possible choice for C is
  implying that a possible choice for …F0 S0 †? is
I  
C? ˆ (B2)
ÿCÿT
2 C1
T
…F0 S0 †? ˆ V…FS†? ˆ ‰V1 V2 Š
0
ˆ V2
Id


? A
Proof. Setting C ˆ we have Using the derived relations
 B
  A ÿT T
CT C? ˆ CT1 ; CT2 ˆ CT1 A ‡ CT2 B ˆ 0 C?
0 ˆ V1 ÿ V2 …V2 C0 † C0 V1 …B4†
B
yielding B ˆ ÿCÿT T
2 C1 A while A must be chosen so that …F0 S0 †? ˆ V2 …B5†
 
? C1 A and
rank‰CjC Š ˆ rank
C2 ÿCÿT T
2 C1 A
  …F0 S0 † ˆ VFUT US ˆ V…FS† …B6†
A C1
ˆ rank ˆn we obtain
ÿCÿT T
2 C1 A C2  
T KS1
With the obvious simplest choice A ˆ Ir we obtain …C?
0 † F0 S0 ˆ ‰VT1 ÿ VT1 C0 …VT2 C0 †ÿ1 VT2 Š‰V1 V2 Š
  0
I ˆ ‰VT1 ÿ VT1 C0 …VT2 C0 †ÿ1 VT2 ŠV1 KS1
C? ˆ T : (
ÿCÿT
2 C1
ˆ VT1 V1 KS1
Returning to the original bases we have 0 ˆ CT C?
ÿ VT1 C0 …VT2 C0 †ÿ1 VT2 V1 KS1
ˆ CT0 VC? ; implying that
ÿT
ˆ KS1 ˆ KUT1 S0 (B7)
C?
0
?
ˆ VC ˆ V1 ÿ V2 CÿT T
2 C1 ˆ V1 ÿ V2 …V2 C0 † CT0 V1
where we have used the fact that VT1 V1 ˆ I and
The range R…f † is represented by R…F† ˆ R…FS† since VT2 V1 ˆ 0 as a consequence of VT V ˆ I. Utilizing the
f …S† ˆ R…f † and derived relation KUT1 S0 ˆ …C? T
0 † F0 S0 and the one for
     ? T
…C0 † in Eq. (B4) the generalized inverse g is represent-
K 0 S1 KS1
FS ˆ ˆ ed in the original bases by the matrix
0 0 S2 0
T ÿ1 ? T
G0 ˆ S0 ‰…C?
0 † F0 S0 Š …C0 †
Lemma. A possible choice for …FS† is ? (B8)
‡ D0 ‰……F0 S0 †? †T C0 Šÿ1 ……F0 S0 †? †T
 
? 0
…FS† ˆ …B3† This equation is identical with the representation of
Id Eq. (82) derived by Teunissen (1985, Eq. 2.7).


? A
Proof. Setting …FS† ˆ for any matrix with
B References
?
column-span R…FS† we have
  Baarda W (1973) S-Transformations and criterion matrices.
  A Netherlands Geodetic Commission, Publications on Geodesy,
0 ˆ …FS†T …FS†? ˆ ST1 K 0 ˆ ST1 K A ˆ 0
B New Series, Vol. 5, No. 1
Bjerhammar A (1951) Rectangular reciprocal matrices with special
which yields A ˆ 0, while B must be chosen so that references to geodetic calculations. Bull Geod 52: 188±220
  Blaha G (1971) Inner adjustment constraints with emphasis on
h i KS1 0 range observations. Dept Geodetic Science Report No. 148, The
rank …FS†j…FS†? ˆ rank Ohio State University
A B Choquet-Bruhat Y, DeWitt-Morette C, Dillard-Bleick M (1977)
 
KS1 0 Analysis manifolds and physics. North-Holland, Amsterdam
ˆ rank ˆ m: Dermanis A (1991) A uni®ed approach to linear estimation and
A B prediction. Presented at the IUGG XXth General Assembly,
100
International Association of Geodesy, Vienna, August 11±24,
1991
Dermanis A, Grafarend E (1981) Estimability analysis of geodetic,
astrometric and geodynamical quantities in Very Long Baseline
Interferometry. Geophys J R Astronom Soc 64: 31±56
Dermanis A, Sanso F (1995) Nonlinear estimation problems for
nonlinear models. Manuscr Geod 20: 110±122
Grafarend E (1974) Optimization of geodetic networks. Boll Geod
Sci Ani 33: 351±406
Grafarend E, Kampmann G (1996) C10 …3†: the ten-parameter
conformal group as a datum transformation in three-dimen-
sional Euclidean space. Z Vermessungswesen 121: 68±77
Grafarend E, Scha€rin B (1974) Unbiased free net adjustment.
Surv Rev XXII, 171: 200±218
Grafarend E, Scha€rin B (1976) Equivalence of estimable quanti-
ties and invariants in geodetic networks. Z Vermessungswesen
101: 485±491
Grafarend E, Scha€rin B (1991) The planar trisection problem and
the impact of curvature on non-linear least squares estimation.
Comput Stat Data Anal 12: 187±199
Halmos PR (1974) Finite-dimensional vector spaces. Springer,
Berlin Heidelberg New York
Koch K-R (1982) S-transformations and projections for obtaining
estimable parameters. In: Blotwijk, M.J., F.J.J. Brouwer, D.T.
van Daalen, H.M. de Heus, J.T. Gravesteijn, H.C. van der
Hoek, J.J. Kok, P.J.G. Teunissen (eds) 40 Years of Thought,
Anniversary volume for Prof. Baarda's 65th Birthday Vol. 1,
Technische Hogeschool Delft, Delft, pp 136±144
Lohse P (1994) Ausgleichungsrechnung in nichtlinearen modellen.
Deutsche Geod Komm C 429
Loomis LH, Sternberg S (1980) Advanced calculus. Addison-
Wesley, Reading Mass
Meissl P (1965) UÈber die Innere Genauigkeit dreidimensionaler
Punkthaufen. Z Vermessungswesen 90: 109±118
Meissl P (1969) Zusammengfassung und Ausbau der inneren
Fehlertheorie eines Punkthaufens. Deutsche Geod Komm A
61: 8±21
Mierlo, J van (1980) Free network adjustment and S-transforma-
tions. Deutsche Geod Komm B 252: 41±54
Moore EH (1920) On the reciprocal of the general algebraic matrix
(abstract). Bull Amer Math Soc 26: 394±395
Penrose R (1955): A generalized inverse for matrices. Proc Cam-
bridge Philos Soc 52: 17±19
Rao CR, Mitra SK (1971) Generalized inverse of matrices and its
applications. Wiley, New York
Scha€rin B, Heidenreich E, Grafarend E (1977) A representation
of the standard gravity ®eld. Manuscr Geod 2: 135±174
Sternberg S (1964) Lectures on di€erential geometry. Prentice-Hall,
Englewood Cli€s, N.J.
Takos I (1976) Generalized inverse of matrices and its applications
(in Greek). Bull Hellenic Army Geogr Service 1976: 77±133
Teunissen PJG (1985) The geometry of geodetic inverse linear
mapping and non-linear adjustment. Netherlands Geodetic
Commission, Publications on Geodesy, New Series, Vol. 8,
No. 1
Teunissen PJG (1989a) Nonlinear inversion of geodetic and geo-
physical data: diagnosing nonlinearity. In: Brunner FK, Rizos
C (eds) Developments in four-dimensional geodesy. Lecture
Notes in Earth Sciences, No. 29. Springer, Berlin Heidelberg
New York pp 241±264
Teunissen PJG (1989b) First and second moments of non-linear
least-squares estimators. Bull Geod 63: 253±262
Teunissen PJG (1990) Non-linear least squares. Manuscr Geod 15:
137±150
Xu P (1995) Testing the hypotheses of non-estimable functions in
free net adjustment models. Manuscr Geod 20: 73±81