back to WRITINGS
Rational in Retrospect - Reflections on the Logic of Rationality in Recent Design
Patrik Schumacher 1999
Published In: AA files 38, Annals of the Architectural Association School of Architecture
One of the most striking aspects of Un Studio's Arnhem Central master plan is its ability to maintain its
radical spatial concept in the face of the constraints and pressures that come with such a big investment.
The project formulates an infra-structural knot that comprises two bus terminals to be linked with a train
The key to the political and economic robustness of this spectacular public space might not only lie in the
scheme and the design method from which it emerges, but rather in the seamless presentation of the project
and its design process as eminently rational. The process itself has a veritable complexity as it brackets
ongoing formal experiments with a rigorous form finding process based on the geometric and quantitative
analysis of all traffic functions. The claim is made that the project evolves through the articulation of
the various required transfer flows. However, the real unfolding of this process is not reducible to the logic
of linear functional determination. Neither is the design imposing a preconceived architectural figure. It
rather operates via the experimental groping into design options not yet at hand. Any determination is
relative to a presupposed ontology, i.e. a given 'design worldą that opens and delimits a space of
possibilities in which a certain optimisation is definable. The jump into unexpected spaces of possibility
will undercut any previous determination and set up another optimisation cycle according to new rules. This
happened in the case of the Arnhem project when the investigation moved from the two-dimensional plane into a
three-dimensional single surface, i.e. a layered space that multiplies the available transfer surface while
avoiding the usual bottlenecks and the dichotomous segmentation of stacked space. This solution could not
have been derived within a linear engineering process that operates within a given, well-formalised ontology.
The a posteriori documentation and rationalisation of this process becomes the medium to convert an explorative
experiment - with all its leaps, bounds and loops - into a reproducible professional repertoire. Through such
reflection the project is able to legitimize itself as an effective material contribution to a given site as
well as an innovative contribution to the culture of architectural research.
This condensed sketch and appraisal of a project, its method of development and its final
self-legitimization is pointing towards a complex new logic of legitimate design speculation that requires to
be elaborated in more detail in order to be fully appreciated.
This is thus the aim of this paper: to utilise the project and design process of the Arnhem master-plan to
clarify questions concerning the methods, purposes and final rationality of current research and practise in
In their contribution to Any Magazine 23 Ben van Berkel and Caroline Bos express the uncertainty and
vacillating mood of current avant-garde practises with respect to the question of rational principles that
might guide and legitimize the work. They are among those struggeling to define a practise that, on the one
hand assumes it to be "natural and right that architects strive to be reasonable, responsible partners" (2)
in a cooperative process with clients, authorities and users, so that "large investments can be safely
entrusted to them"(3), while on the other hand becoming increasingly sceptical with respect to the pressure
of rationality and "the demand to present the "right" solution, even when the contents of that concept have
become very uncertain".(4) Van Berkel and Bos are referring to the objectivity and rationality demanded by
the client (and delivered by the architects) as a "retrospective justification" or "after-theory" that "blocks
the view of what went on behind it".(5) They bemoan the lack of "real" (vs post-rationalising) architectural
theory. They speak of the way their strategies react to their dependence on being selected for work and of
the resultant fear to critically analyse their internal discourse. This frank admission of being somewhat
alienated from their own discourse marks a courageous step, but yet begs the question of how a free and
self-critical practise would proceed and present itself.
The method and rhetoric of functional determination
Prima facie Un Studio's (Van Berkel & Bos) Arnheim Project participates in the recent re-orientation of Dutch
avant-garde architecture towards a parameter driven elaboration of spatial form. As a complex infra-structural
project, i.e. a transport interchange required to integrate train, bus, taxi, car, bicycle, and pedestrian
movement, it could well become a paradigmatic testbed for the scope and validity of the method of parameter
based derivation of form. Such a method - the so called 'data-scape' method - has been most explicitly
proclaimed and practised by MVRDV. (6) In its most ruthless guise, this approach considers the design-process
as an explicit optimisation process guided by some (narrow) set of performance criteria. Viewed as such it
would follow in the tracks of Hannes Meyer and the 60s and 70s efforts to resolve architectural design into a
formalisable science.(7) Although important differences in "sensibility" need to be registered, the
continuities with historical functionalism should not be denied. The data-scape method is to be understood as
the relativist heir to a realist functionalism. In my opinion this is a tradition that deserves to be claimed.
Attempts to pose and formalise systematic accounts of design method and result remain indispensable, even if
such accounts will become ever more complex, transitory and relative.
The various presentations of the Arnhem project I have witnessed ( by Ben van Berkel himself and by the project
architects Tobias Wallisser and Peter Trummer) tell a fairly systematic and straightforward story of
functional determination: All required transfer routes are plotted out as required. The lines swell into
volumes according to the respective circulation quantities etc. The account comes close to be thoroughly
convincing, although from the outside real and rhetorical rigor are hard to tell apart.
In order to trace the uncertainties and gaps in the design process that give its traditional format of
presentation the sense of unease that van Berkel & Bos express in the Any 23 article, it is required here not
only to rehearse the design process, as it has been presented, but moreover to attempt a reconstruction
according to the ideal schema of formalised decision analysis. On the basis of such a logical reconstruction
it will be possible to appreciate the deviations and complications of the non-traditional forms of rational
procedure that actually operate here.
The justification of the Arnhem scheme usually takes the form of tracing its rational design process. In the
outllook of traditional decision analysis this is perfectly respectable. In the ideal schema of rationality
one would expect that the hierarchy of arguments - from the overall system level to successive subsystem levels
- that justifies the resultant scheme, would re-present the string of successive decisions - from major to
minor decisions - by which the actual design or decision process did proceed. In this scenario the occurrence
of post-rationalisation would throw doubt on the rigor and certainty of both process and result.
(Behind the fact of post-rationalisation lurks the accidental and therefore precarious nature of the supposed
However, the direct translation between design process and legitimation does no longer hold relative to
recent practise. At least in education post-rationalisation has become a (legitimate) common place. Today's
design process gropes, stumbles, backtracks and only then succeeds. The emergent rationality of this process,
i.e. a structured, reproducible method, can only be reconstructed retrospectively by cutting the dead ends,
short-circuiting the loops etc. Innovative theory is always "after-theory". Therefore van Berkel & Bos'
frustration with "a posteriori rationalisation" and "after-theory"(8) must be interpreted as concerning the
pretence of such after-theory to describe the actual process rather than the fact of post-rationalisation as
such. (Through the fact of post-rationalisation shines the open-ended nature of research.)
The ideal schema of rationality
Before we embark on the concrete reconstruction (of some key aspects) of the Arnhem project we should
visualise the most basic formal structure of a rational design or decision process as given in fig.A.
The design process is structured as a hierarchical decision tree. The process follows the ramifications of an
initial decision in a linear fashion. It is analogous to the navigation through a menu-structure. Each choice
opens up a further finite series of dependent sub-choices. A decision, in order to qualify as rational, needs
to be reconstructed as such a successive selection from measurable sets of alternatives. The minimal
requirement for any act to be worthy of being called a decision at all, would thus be that it proceeded by
comparison, i.e. evaluation against at least one specified alternative. The ideal case of a rational process,
combining the strongest claim with the simplest structure, would be defined through the following conditions
whereby decisions are
1. Hierarchical: The rational decision process is fully resolvable into a linear chain of discrete and
self-sufficient decisions. This means that there is a clear hierarchy of decisions whereby a later ramification
can never put previous decisions into doubt. (No loops and iterations are required.)
2. Ranked: The hierarchy of decisions presupposes a finite and stable list and ranking order of all objectives or performance criteria to be addressed.
3. Comprehensive: On each level or branching point of the decision tree the menu or space of options is finite (or at least computationally exhaustible) and known to be comprehensive.
4. Decidable: At each level the space of solutions is measurable and unambiguously decidable relative to given performance criteria.
5. Coherent: A primary objective that was giving reason to a primary decision can not be sacrificed for another reason at a later stage without subverting the whole process.
6. Decomposable: In as much as there are parallel objectives (as well as objectives in an order of subsumption) those parallel objectives will be adressed in parallel decision trees. This requires the independence of those objectives and their solutions from each other.(9)
Rationalising the Arnhem design process
On the basis of this schema we can know investigate the deviations and complications encountered in the Arnhem
The given situation and the first decision may seem simple enough: The major access road to the site runs
parallel to the train tracks at about 120-meter distance. The project is wedged between those two edges.
The part of the project to be considered here - the most significant and formally richest part - is
the pedestrian zone of interchange that mediates between the major means of transport that meet on the
site. This zone is framed by the train station (north); the pedestrian access from the city (east and south);
the regional bus-station (west); and the trolley bus station (south-east). To separate the latter two was the
first major project decision that framed the further study of possible configurations. This way a central
space was established in which the necessary transfer flows could be configured. This concentric figure has
been established by comparison with (and elimination of) a figure according to which regional and city busses
would have been butted against each other, and where the interchange/waiting space would have been the bulk
head instead of the centre of the system (fig.1). This decision might seem self-evident but in order to be
formalised and made transparent the selection criterion would have to be stated, i.e. the primary criterion
of this primary decision. Also:The quality and credibility of the decision would be enhanced if the selection
would proceed from a greater number of configurational options. Ideally it would have to be shown that the
offered alternative configurations exhaust the space of configurational possibility. Only then could we reach
"the triumphant conclusion that the particular design under discussion is the only objectively justifiable
one."(10) The simplest way to formally assure exhaustion of possibilities at any stage is to proceed by means
of successive dichotomies.
Concerning the missing explicit objective or selection criterion for the chosen centralising configuration,
any of the following might be proposed:
- the veto on any visual obstruction between any two means of transport
to be linked.
- the provision of a point of total visual orientation
- the maximisation of such points
- the veto on any indirect (or chained) link between any two means of transport
- volumetric compactness ( ratio of circumference to surface ) of the transfer space for the sake of
- the minimization of total length of transfer paths to be constructed
- the minimisation of the average detour-factor imposed upon any desired transfer link
Any of these criteria (or certain combinations) would privilege the selected alternative. We might even
feel inclined to cite them all as so many good reasons to choose and start with the centralising figure.
But as, e.g. the last two criteria are going to be placed into direct contradiction when it comes to
delineate the paths that run across the central space, the mere enumeration of good reasons at any point
does not by itself qualify for a rational design process. Not all criteria will be maintainable throughout
and it could very well make a difference for the further elaboration of the scheme which criterion was
implied as primary. In order to maintain coherence the performance according to the respective postulated
criterion (e.g. orientation) would have to be protected from cancellation in the further development of the
The next series of diagrammes that Un Studio offers (fig.2-3) to describe the further elaboration of the
scheme share the ubiquitous notation of network analysis.
Fig.2 determines the necessary connections that have to be made between the various means of transport.
Each means of transport is represented as node in the network, whereby the relative quantities of passengers
that each means of transport brings to the system are represented by the relative size of the respective node.
The links are differentiated by line thickness thus classifying and ranking (rather than quantifying) the
respective binary transfer quantities. Spatial relations are represented topologically.
Fig.3 somehow translates fig.2 into a scaled plan representation. The density of originating pedestrian traffic
as well as pedestrian transfer traffic density is now annotated by absolute quantities (cardinal numbers).
Fig. 3 therefore is in certain respects more concrete, in other respects more abstract than fig.2.
Design Worlds and Deleuzian diagrammes
Both representations (fig.2 & fig.3) think of the problem - traffic interchange - in terms of nodes and
links between nodes. This seems natural and straightforward enough. But in order to capture a key aspect
of the new rationality that is evolving here - the proliferation of modes of representation - , we have to
reflect upon the categorical imposition that any type of diagramme or notational language effects. Any
analysis or design operation that proceeds via network graphs presupposes that the world consist (or should
consist) of nothing but nodes and links, as well as higher order entities like full networks, rings, chains,
trees, stars etc. Instead of speaking of the notational language we might speak of the formal a priori, the
graphic universe, or the "design world."(11) Each design world, i.e. each diagramming technique, drawing type,
or software package imposes is own brand of primitives, rules of association and manipulation and thus opens
and delimits a universe of speculation in which both problems and solutions are lodged. Each design world
implies a quasi-ontology and quasi-laws of nature. We might start to critisise them in terms of what they
leave out. We can do this by reference to another tool that in turn has its own limits. The double-bind of
revelation and blindness of any language is definable only relative to other languages. There is nothing
self-evident, objective or compelling in the imposition of any language or design world. It remains an
arbitrary imposition until it is rationalised by means of comparative evaluation against alternatives.
But such are not at hand at will. Below we will sketch out what it takes to create a language or design world,
i.e. a recurrent social practise. Also: Different languages or design worlds might well be incommensurable
and can be compared only on the basis of an arbitrating meta-language. We lack an ultimate meta-position. All
we can do is to experiment practically within various given languages and hope that viable repertoires
crystallise through competition in practise.
The standard scheme of rationality implicitly assumes that such a competition has already resulted in a stable
and ranked selection of tools: An appropriate tool for each specifiable task. This rationality thus
assumes that progress and history has come to a standstill, at least as far as the evolution of design
worlds are concerned.(13) What we do today is thus no longer (fully) covered by that schema of rationality.
Only in the last 20 years has the architectural avant-garde experimented with new types of diagrammes,
drawings and recently new digital tools. Since the refoundation of the discipline in the early 1920s the
architect's design world has been a singular and stable system of hierarchically scaled line drawings. From
the scale-less (topological) sketch to the working drawings this world distributes nothing but outlines and
boundaries. Everything is about the distribution of horizontal and vertical planes. The meaning of each drawing
resides in its position and role in the chain of translation from one drawing to the next (more detailed)
drawing and from there to the construction process and the building itself. Within this routinised practise
of translation, from the abstract to the concrete, it is habitually known how each drawing constraints the
next set of decisions, until the detailed lines finally translate into physical edges. (And we all have
learned to perceive and inhabit space along those edges.) Only within such an order of repetition can one
speak of a well defined notational system. The concept of a drawing - here termed "representation" - that is
firmly lodged in such routine practise, is the model against which the Deleuzian "diagramme" is defined. The
difference does not reside within the object, but in the patterns of its use. The question here is whether
or not (yet) it functions within a stable social practise. The diagramme does not yet know its place in a
routine operation. It is creatively engaged in the formation of such a (potentially reproducible) practise.
It therefore is worked upon without stable interpretation, without predetermined consequences. Work is
assimilated to play.(14) At least since the mid-eighties virtually all design efforts at the AA, and soon
after at most other schools in London, were conducted through "diagrammatic" processes. Only recently these
not-yet-methods (or drifting non-methods) have been moving into real life practise.
Leaps and loops
The reflection on the dependency of any design effort upon the design world it operates in, allows
the discussion of the next crucial "steps" of the Arnhem design process.
"Movement studies" are set to be "the cornerstone of the proposal". (15) Such movement studies could
have taken many forms. Plotting a network graph is only one (rather economical) form such a study could
take. Although the graph does constrain the further design moves, and thus is certainly not meaningless, a
strict rule of translation from such a graph into a floor plan does not exist. The claim that the "station
emerges from these motion studies" (16) is therefore as yet rather anecdotal than rational. Without rule there
is no determination. What can be rightfully claimed here, at least, is the negative implication that the main
space of the station was not based on a preconceived (platonic) figure or any known (classical plaza)
typology. Indeed it seems to be the amorphous irregularity of the scheme (fig.4-7) which inclines us to
grant credibility to the cited claim that the figure emerges from movement studies.
The move from fig.3 to fig.4, in terms of network- or graph-theory, implies an abrupt shift from an all-
line graph to a branching graph (see fig). Another leap is the emergence of irregular curves in fig.4.
(Figure 5 is just the 3D version of this preliminary diagramme.)
The leap from fig.3 to fig.4 is certainly not (yet) motivated by any known routine practise. (Only in
the recent avant-garde canon are such irregular figures at all.) That does not mean that one could not,
in the end, legitimately post-rationalise such shifts and leaps by constructing a rule that would define
the one figure as the regular translation of the other. Principally any anecdotal connection might be found
to be productively reproducible and thus can be retrospectively elevated from a haphazard leap into a
rational, methodical move.
What would motivate or justify the branching graph (fig.4/5) here? The centralisation of control? The
economy of paths? It depends. What we do know is that (according to formalised decision analysis)
any motivation or justification would have to take the form of a criterion-based selection from a
field of possible configurations as I have tried to specify in figureB. But any such rationality is
vulnerable to subversion when yet unthought of possibilities emerge. Van Berkel and Bos are talking
about providing a hybrid between a centralised and an all line system (fig.6). My list (fig.B) does not
know of "hybrids". But once things called "hybrids" have been admitted into this world, any analysis,
decision or motivation based on my list is obsolete.
Figure 9 suggests a hyberdization of figures 6&7. It also seems to be inspired by the so
called Klein-bottle, assuming transformational qualities in analogy to the spatial logic of the bottle
(fig.8) These moves are far from being well defined. One might be inclined to dismiss the introduction of
the Klein-bottle as an arbitrary graft. But then again, to refuse the bottle and rest with the usual a
priori ontology of platonic primitives etc. would even be more arbitrary.
In Fig.6 the project suddenly appears in the guise of a 3D computer model operating with smoothly bent cones.
As these cones traverse the space they intersect, merge and branch. Yet another new design world has
been opened, irreducible to any network graph. This world does not know of nodes and links between nodes
whereas in turn the graph did not know of interpenetrations and fusions. Are these really cones? Are we
looking at one, two, three, four or five pieces here? In figure 6 it is, in fact, impossible to
unambiguously count the number of elements to be distinguished. There are no discrete elements here and
no clearly nameable spots and places. The project leapt into a different world, a yet uncharted world.
However, rules of translation back into a network graph might very well be definable. One could even
construct such translation-rules that would allow any graph like fig.2 or fig.3 (plus some additional
information e.g. about peak time flows ) to be transformed into an image like fig.6. But those rules do
not yet exist, at least they have not been made explicit. And no performance criteria have been specified
yet. But again, it is never too late. We are never principally unprincipled. We might learn to trust our
intuitions and expect that at least some of our creative action will assume their principle retrospectively,
against the arrow of time.(17)
Figure 7 announces a further transfiguration of the design process into yet another irreducible ontology and system of spatial manipulation: a single surface that peels and splinters smoothly into differential levels. This is the crucial moment referred to in the opening paragraph. This way the design process exploits and responds to existing differential levels on the site. This solution could not have been derived within a linear engineering process that would always be presumed to operate within a given, well-formalised ontology. No such response was available to the discipline until fairly recently.(18) Once discovered this possibility offers surprising advantages: a space is created that multiplies the available transfer surface while avoiding the usual bottlenecks (lifts or stair-cases) and the disorienting dichotomous segmentation of orthogonal, stacked space.
The logic of design innovation is analogous to the logic of biological evolution. It goes through the
necessary moments of mutation, selection and reproduction.
Methodologically it is important to notice that the last reflections were inspired by the formal universe of
the single surface with its subtle distribution of gradients, moulds, ridges and sweeping cuts. Only on the
back of the original leap into this world could the ambitions of a certain type of choreography be envisaged.
A system of tunnels, lifts and travelators, which would have been a much more straightforward, linear
elaboration of the initial network graphs, could never have brought forward this discourse. Here we would
have remained in the realm of mechanical traffic management.
Throughout the design process new good reasons have been (and will continue to be) introduced. Those reasons
will post-validate explorations that were not necessarily and at all times covered by the strictures of the
ideal schema of rationality. But even the result of a strict process might reveal unexpected qualities. In
any such case, in order to recuperate the option of principled and rational conduct, one would have to admit
the subversion of one's course of action, precisely because one has found more reasons than one was looking
for. Armed with those new values and criteria one would have to loop back and reinvestigate the design path
traveled, and give systematic account of any "final" result , on the basis of both well established and newly
The attempt to reconstruct the Arnhem design process makes it abundantly obvious that the overall dynamic
pattern of current design and design-research can no longer, at all times and in its totality, be cast into
the mould of the ideal schema of rational conduct presented above. Nevertheless it deserves to be noticed that
we are not strictly talking about the abandonment of the schema. Rather we are witnessing its dynamisation
and complication. As a backbone of definition from which the new practises deviate but ultimately around which
they oscillated and gravitate, and to which they recursively return , if only in a relative and temporarily
mode, the ideal schema of a formalised decision process remains indispensable.
When we design today we will, overall, still be climbing up a decision tree, however much we will temporarily
spread out laterally. We still will have to segment our process: Pick up this tool first and address another
task later. We will therefore try to identify a hierarchy and start with the most important decisions. Should
we find that the order of importance turned out to be otherwise, there still remains, until further notice,
an order of importance to be considered. But we will have to give up that we could know or take account of
all options at each junction. New branches will sprout as we move up the tree. They also grow besides us from
below, may be precisely because we moved up the other branch. We will have to loop back or jump branches. No
time to backtrack. A previously abandoned sidetrack might afford help. But for the next time we know the
straight line. A whole new tree might cast our whole path into the shadow. But it is too risky to jump and
too late to go back to the root. But then the latest branches of the new tree might fall back onto the old
tree. Now several path seem to converge rather than ramify further. The pursuits of parallel objectives get
entangled. We will initially seek synergies but then cut as much as possible to keep the process manageable.
The tree keeps mutating into a rhizome. We keep trimming it into a tree-shape. This is not fundamentalism but
a matter of economy. We are moving from the illusion of absolute rationality to "bounded rationality"(19)
and "good enough reason". Reason and research don't always come cheap and easy. How can one assess the balance
of investment in time
and resources between research and execution? There is no answer outside experience. The competitive economy
is the final court of appeal. It passes verdict on any course of action but without locating the moments of
failure or triumph within it. We know we have won or lost but we are never told exactly where and why.(20)
Research, more than ever, means taking risks.
back to WRITINGS