A Recursive Theory of Mathematical

Understanding
SUSAN PIRIE, TOM KIEREN

"Everything said is said by an observer" ematical procedures ( e g. counting) to make mathematical

[Maturana, 1980) abstractions reflected through the use of a notational
system
"The experiencing organism now turns into a
Both of the above models of understanding involve
builder of cognitive structures intended to solve
levels or components which appear to have predicate
such problems as the organism perceives or
quality- they define complexes of components in unique
conceives. among which is the never ending
categorical terms In that sense they give a picture of the
problem of consistent organization that we call
components which might be involved in the process of
understanding." [Von G!asersfeld, 1987]
understanding Von Glasersfe1d [1987), however, sees
Over the past 20 years or so there has been a continuing understanding as a continuing problem-solving process of
c:onsistent~y organizing one's mathematical structures
dialogue, much of it through Psychology of Mathematics
Education Conferences, on what it means for a person to let us consider the following example drawn from a
understand mathematics One of the features of this dia- study of 7-9 year olds working in groups doing fraction
logue has been the theoretical identification of different comparison tasks [Wales, 1984; Kieren & Pirie, forth-
kinds of understanding, principally instrumental and rela- coming) In the task children were asked to compare the
tional understanding but also concrete, procedural, sym- amount of pizza person "A" would get if sharing 3 pizzas
bolic and formal understanding. Pirie [1988] has suggested between 7 people with the amount person "B" would get
that thus describing different kinds of understanding is sharing 1 pizza between 3 people Here is a commentary
by Hanne working with 2 friends (all aged 7)
inadequate as a means of differentiating children's per-
formances which exhibit mathematical understanding Hanne: A is hard - let's skip it
She claims, and illustrates from extensive taped interac- Hanne: B is easy, you "Y" it
tions of children doing mathematics, that mathematical
understanding is a complex phenomenon for the child
doing it A single category does not well describe it, nor do
such categories capture understanding as a process rather
than as a single acquisition What is needed is an incisive (Draws "Y" and explains her process to her
way of viewing the whole process of gaining under- friends The numbers are ours to illustrate
standing her explanation.)
There have indeed been recent efforts to go beyond a Hanne: (I) let's use "Ys" on A
cataloguing of kinds of understanding or thinking of (Action I, draws:
mathematical understanding as a singular acquisition.
Ohlsson [1988) performed a detailed mathematical and
applicational analysis of fraction-related concepts From
this elaborated example, he suggested that mathematical
understanding entails three things: knowledge of the math- i . e , Hanne cuts each of the three pizzas into
ematical construct and related theory, knowing the class "fair shares" in order to give one third each.)
of situations to which this theory can be applied, and a Hanne: (II) (Action 2, draws:
referential mapping between the theory and the situations
He does not however, suggest how this mapping is devel-
oped or grows. He infers but does not give a process model.
Herscovics and Bergeron [1988] give a two-tiered
model of understanding and illustrate it using the under-
standing of number and pre-number in young children i.e , she cuts the remaining piece into seven
smaller pieces )
The first tier involves three levels of physical understand-
Hanne: (Ill) Oh, I see! A gets a third and a bite. A gets
ing: intuitive (perceptual awareness), procedural ( e g 1-1
more
correspondence) and logico-physical abstraction ( e g phy-
sical invariance) The second tier is non-hierarchical and What has happened here? Hanne starts by not under-
includes as components of understanding the use of math- standing how to divide 3 among 7 It is clear from the

For the Learning of Mathematic~ 9, 3 (November !989)

Fl M Publishing Association., Montreal, Quebec, Canada 7
complete tape that she can divide I among n for small n learner is self-referencing, a significant means of cognition
and in particular has formalised this act for I~ 3 ("Y it") is recursive in nature; growth is knowing, i e when effec-
In I and II we see her take this "form" from the particular tive action occurs through new thought-actions which
environment in which it was constructed and use it, firstly coordinate or distinguish prior thought-actions.
to divide each separate pizza into thirds ("Y them"), and In giving a more detailed analysis of this recursive
then to divide the remaining piece, now forming the growth, Iomm [1989] envisions it as occurring in a situa-
"whole 1," into seven pieces (I among n) At III she tion where one is in interaction with another person such
signifies her realisation that she has a successful new or- as a fellow student or teacher He described the first recur-
ganization of sharing, or division, based on the combina- sive level as the "consensual coordination" of action. The
tion of two previously unconnected actions This leads us recursion on this linguistic action is described as "languag-
to ask: what does Hanne's understanding entail? Indeed, ing," an action which makes something "a thing" for
what does mathematical understanding itself entail? How someone. He sees several levels of recursion beyond this.
can we depict this growing process? Our answer can be The model presented below has the levelled feature of
summarized as follows: IOmm's description, but it also envisions growth in know-
ing as possibly involving internal, as well as external,
Mathematical understanding can be characterized as
levelled but non-linear It is a recursive phenomenon interactions The terms used for the levels in the model
and recursion is seen to occur when thinking moves below reflect this vision and also relate to its use as a
model of mathematical understanding.
between levels of sophistication (as with Hanne
above) Indeed each level of understanding is con- The process of coming to know starts at a level we call
tained within succeeding levels Any particular level "primitive doing." Action at this level may involve physi-
is dependent on the forms and processes within and, cal objects, figures, graphics or symbols. "Primitive" here
further, is constrained by those without does not imply low-level mathematics, but rather a start-
ing place for the growth of any particular mathematical
While it is beyond the scope of this paper either to com- understanding
pletely delineate this theory of mathematical understand- The first recursion occurs when the learner begins to
ing, which we call transcendent recunion, or to fully fOrm images out ofthis "doing "The effective actions here
connect it to data on children's mathematical behaviour involve "image making." At the next level these action-
gathered in England and Canada, some major tenets of the tied images are replaced by a form for the images From
theory are highlighted below and then illustrated with the mathematical point of view it is this "image having"
reference to an understanding of quadratic functions Sub- which frees a person's mathematics from the need to take
sequently we use the model to analyse an episode of stu- particular actions as examples This is a first level of
dent behaviour with a view to comprehending the growth abstraction; but it is critical to note that it is the learner
of their understanding We see "consistent organisation" who makes this abstraction by recursively building on
in action images based in action For understanding to grow, these
In saying that mathematical understanding is levelled images cannot be imposed from outside
and recursive we are trying to observe it as a complex Because knowing has to be effective action, the recur-
levelled phenomenon, defined by Vitale [1988], which is sions do not stop here . lhe images can now be examined
recursive if each level is in some way defined in terms of for specific or relevant properties lhis may involve notic-
itself (self-referenced, self-similar), yet each level is not the ing distinctions, combinations, or connections between
same as the previous level (level-stepping} To this defini-
tion we have added an idea taken from Margenau's [1987]
notion of growth of scientific constructs New constructs
transcend but are compatible with old ones (they are not
simple extensions) What is in fact more critical is that the
"new" transcendent knowing frees one from the actions of
the prior knowing
The work of Maturana and Tomm [1986] and Iomm
[1989] allows us a more detailed view of recursion in
knowing. Maturana and Varela [1987] see knowing ex-
hibited by effective actions as these are determined by an
observer. That is, if one wishes to see if a person knows
how to play the piano, the person might actually be asked
to play Then the observer determines if this action is
effective So, as we will try to show below, it also is with
mathematics. Who, or rather, what is the knower? Again,
following Maturana and Varela [1987], we see the human
knower of mathematics as self-referencing and self-main-
taining in a particular niche of behavioural possibilities
For human beings, mental states and actions are crucial
elements of this niche, or sphere, of possibilities Since a

8
 images. This level of "property noticing" is the outermost quality If one were to ask such students, "What do you
level of unselfconscious knowing. (The word "outer" has know about quadratic functions?," their effective action
been carefully chosen to imply that the levels of under- would be related to the images of specific quadratics For
standing wrap round each other, as illustrated in Figure 1, them to say anything further would require their making
and contain, indeed require, the possibility of access to all more tables or doing more plotting; their comments about
previous levels. levels of understanding do not equate quadratics would pertain to their actions
with higher or lower levels of mathematics) The next stage is reached when they have generalized
The next level of transcendent recursion entails thinking these specific images. They are now able to say, "Quadratic
consciously about the noticed properties, abstracting com- functions? Well, they all form sort of U-shaped curves"
mon qualities and discarding the origins of one's mental They have now replaced their actions on quadratics with
action It is at this level that full mathematical definitions an image and can talk about it. There is an ability, but no
can occur as one becomes aware of classes of objects that necessity, to continue to plot individual graphs, and cer-
one has constructed from the formation of images and the tainly no need to make very extensive tables of values. This
abstraction of their properties stage of image-having is a recursion on the stage of image-
One is now in a position to observe one's own thought making
structures and organise them consistently. One is aware of Because they can now talk about quadratics as objects,
being aware, and can see the consequences of one's students can move to a level of noticing features or distinc-
thoughts It is clear at this point that while this outer level tions among them Quadratics have vertices, they open up
of understanding is transcendent, in other words funda- or down, they are not linear, they can have maxima or
mentally new in some way, it has to be consistent with all minima elsewhere than at the origin . Effective action is
previous levels of knowing not now related to a particular graph but to properties of
For fuller understanding one must now be able to answer quadratics Students here have a generalized image
why the consequences of thoughts must be true This calls The next recursion occurs when students realize that
for an awareness of associations and of sequence among despite these distinctions quadratic functions can be con-
one's previous thoughts, of their interdependence In math- sidered as a class of things: students can talk about the set
ematical terms it might be setting one's thinking within an of functions y = a.x2 + b.x + c and see this as a consciously-
axiomatic structure made mathematical definition. They are using self-con-
All of these levels of recursion are referenced in a direct scious thought to formalize their understanding They
way to previous levels Although new levels transcend or now know quadratic functions as mathematical objects
make one free of actions at an "inner" level, in some sense and not simply as graphical ones.
these actions on previous levels become initiating condi- The next level of recursion is signalled by the students'
tions which constrain one's knowing At the highest level interest in the question, "What is true for all quadratic
of recursion, as TOmm suggests, knowers act as free agents functions?" At this level the quadratic formula or the
We call this the level of inventing *Now one can choose to method of completing the square are pieces of a possible
initiate a sequence or structure of thought which is a theory and not simply techniques for computation as they
recursion on the previous one in the sense that it exists as a might earlier have been The arrival at this level, that of
base, but is freely, yet compatibly, created For example, observing the class of quadratic functions as a mathemati-
mathematicians such as lobachewski, Bolyai, and Gauss cal object, allows us to make another point critical in our
deliberately invented non-Euclidean geometries rather theory of understanding Up to now, the recursions may
than try to deduce or explain the parallel postulate as a have appeared to have a sequential or linear quality; but
necessary consequence of Euclidean geometry in the way the development of understanding does not have this qua-
that Saccheri did lity Students at this new level might decide to address the
To illustrate this growth through recursion in less gener- question "Can quadratic equations have just one real sol u-
al terms, we turn to an example of mathematical knowing, tion?" and be prompted to drop to the level of primitive
the knowing of quadratic functions. The particular exam- doing, to making tables and graphs of particular examples
ple is hypothetical but allows us to illustrate the entire However, the return to image-making is not to image-
chain of recursions from primitive doing to inventing. It making as it was previously perceived The students now
also allows illustrations of the fact that one should not know they are looking for an image which can be distin-
think of action in simplistic terms. Any action, and this guished by certain features but which can still be character-
one, plotting, in particular, may already be a recursion on ized as belonging to the class of quadratics In other words,
previous thought or actions while actions, images, and distinctions, formed initiating
We start with the first level of recursion, image-making conditions for the definition of a class of quadratics, the
The students are taking given, specific, quadratic functions, local class of quadratics now forms an environment in
making tables ofvalues, and then plotting them. The image- which a new kind of exploratory action takes place Theory
making can be seen in comments like "Does x 2 + 2x+ I go folds back to constrain one's actions In contrast to the
like this?" or "I get 1 for -1 in 2x'- x- 2 " What is critical original image-making, one is now acting on what may be
here is that these tables and graphs have an individual a specific but, one hopes, representative example. Thus
knowing and understanding quadratics is not a linear
*This use of· 'inventing"· is not intended to suggest that persons do not or process: knowing folds back on itself and has the quality of
could not develop essentially new (for them) ideas at other levels being circular without contradiction. Such a folding back

9
allows one to validate or to see as consistent all the levels Robert: Could I try another one?
of quadratic knowing up to this point, and permits the Robert: (III) One to each
recursive reconstruction of quadratic knowledge in the Robert: Now half to each
face of new challenges One and a half [pieces] each
Let us return now to the level of quadratic knowledge (Now Robert returns to the 3 for 2
we have termed "observing " As suggested, recursions at example and does the following.)
this level revolve around the question, "What is true for all
quadratic functions?" Not only does this lead to theoreti-
cal consequences with respect to maxima, minima and
roots, but it also provides a new focus for one's self-
conscious idea of quadratic functions For example, one
may now talk about quadratics as contrasted with linear
functions and generate comparative observations about
these
The next level of recursion is driven by the need to
logically situate the consequences observed at the previous
level: for example, to explain the "truth" that quadratics Robert: (IV) One and a half each too!
have two roots occurring in particular pairs One can think All three children (at (I) above) realize that the first
of this level as the level of axiomatic-deductive proof, or
answer can be obtained by sharing between two persons
the level where quadratics are situated as an exemplary and simply use their "forms" of halving It would appear
system within the system of polynomial functions
that they have an image for halving as an object and can
At the highest level, one would be deliberately thinking use it. They are then (at (II)) faced with a dividing up
of quadratic functions in a completely new way: for exam- situation which they cannot immediately comprehend
ple, the study of all complex functions of the form z = kz 2 + c, Like Hanne, earlier, who obtains thirds by "Y-ing," they
which while essentially quadratic might lead one into the simply call their halving action into play again- this time
field of recursive dynamic phenomena, strange attractors, unsuccessfully. At (III) Robert drops back to an inner-
fractals, and a whole new thought structure. action level of "one each" to try to cope with 9 shared
We now tum to using our model to analyze specific,
among 6 and is then able to use halving to reach a solution
actual student behaviours That is, while we have used At (IV) he quickly applies this same pattern of action to
transcendent recursion to typify the complex-levelled,
the "3 for 2" case
non-linear phenomena of understanding, these tenets also The non-linearity of effective action is visible in this
allow for a more informed study of such phenomena at the
example The children seem to have an image of halving,
micro level Consider another example drawn from the but using this proves ineffective for them in the 9-for-6
study of pupils working in groups on fraction-comparison situation, so Robert drops back to the level of image-
tasks In this task the children were asked to compare the making (sharing wholes, sharing halves) The fact that this
amount person "A" would get if 3 pizzas were shared
return to the inner level is not the same inner-level activity
between 2 people with the amount person "B" would get if
as earlier, is seen by the way that Robert then uses his
9 pizzas were being shared between 6 people Here is a patterns of action in the 3-for-2 situation He clearly has
commentary by Robert working with 2 friends
not simply gotten an "answer" for 9 from 6; he appears to
Robert: (I) (with nods of agreement from the have realized that he has a possible pattern for other
others) dividing-up situations. (It is interesting to note, however,
Oh, A is easy. A gets 3 pieces (He that Robert has not yet made connections among quotient
draws lines dividing each pizza in half, situations or moved to the level of self-conscious thinking
then draws lines connecting one piece which would allow him to identify "3 pieces" with "one
from each pizza to each of the 2 people) and a half ")
Dan: (II) B is hard though More generally one might say that the children's halving
(All three attempt to use the technique language blinded them to the dividing-up action from
of cutting the 9 pizzas in half and then which it came. Robert, however, removed these blinders:
distributing these pizzas pieces, but the for him the initiating conditions of sharing were still oper-
diagrams become hopelessly confused ative and led to a new coordination of actions within the
(see below) and all but Robert give up) environment of his knowledge of the existence of halving
as a pattern It was probably this knowledge of pattern
which enabled him to call forth the idea that there could be
a solution pattern here too, not simply a quantitative
solution. Certainly we can observe the complex, levelled,
but non-linear sequence of actions, as Robert tries to see
his knowledge structures as consistent, as he tries to
understand dividing up

10
summary and concluding remarks roles of language and thought both at any level and in the
The demonstration of the extended quadratic example and growth between levels
the analysis of two transcripts of children's mathematical
behaviour have been our attempt to observe the nature of
recursive levels and the inter-relationships among them
We have seen that image-making and image-having can References
function as initiating conditions for effective action: Herscovics N. and 1 Bergeron [1988] An Extended Model of Under-
children use inner-level knowledge to build more complex standing. In C. l acompagne and M Behr ( eds) Proaedings of PME-
knowing However, the actual process of knowing and NA X De Kalb: PME/NA:NIU Pp. 15-22
Kieren, I' E and S E B. Pirie (in press) Recursion and I he Mathematical
understanding does not move in a linear fashion through
Experience In l Steffe (ed) Epistemologu.a! foundations of the
these levels, and when one folds back to a previous level of mathematual experien{f.'. New York: Springer Verlag
effective action, this is not merely doing the previous Margenau, H. [ 1987] The mira eft of uistence Boston & london. New
action again The outer-level knowing acts as an environ- Science library. Shambhalu
mental constraint within which forms or processes at the Maturana, H [1980] in H. Maturana and F Varela ( eds) A uropou sis and
wgnition Boston University Philosophy of Science Vol 42
inner levels are "called out," and these are then directly Dordrecht: D Riedel p xxi
constrained by the outer level concerns Maturana H and K Tomm [1986] Languaging and the Emotion Flow
The viewing of mathematical understanding as a dy- Paper delivered at a conference of the Department of Psychiatry
namic process allows us to see a person's current state as Calgary: University of Calgary
Maturana. H. and Varela F J [1987] The tree of knowledge Boston &
containing other levels which are different, but compa-
london New Science library, Shambhalu
tible, ways of understanding the mathematics, and which Ohlsson, S [1988] Mathematical Meaning and Applicational Meaning
allow the person to validate outer-level knowledge or in the Semantics of Fractions and Related Concepts In J Hiebert and
which provide a basis for facing unknown but related M Behr (eds) Numba wnapts and operations in the midd!t grade5
mathematics Reston: NCTM/l E A Pp. 53-91
Pirie, S E.B [ 1988] Understanding- Instrumental Relational, For mal
Considering mathematical understanding as a recursive Intuitive . How Can We Know? in For tht Learning of Mathe-
phenomenon is not meant to replace the contemporary matics Vol 8. No 3. Pp 2-6
views of understanding suggested by Ohlsson or by Hers- Tomm, K [1989] Consciousness and Intentionality in the work of
cavies and Bergeron It is meant to provide insight into Hamberto Maturana A presentation for the Faculty of Education
University of Alberta, Edmonton
how such understanding grows and how the elements
Vitale, B [1988] Elusive Recursion: A I rip in a Recursive land New
these authors describe can be integrated into a whole As ldws m P5ychology, November
such, the theory sketched above should allow for the Von Glasersfeld, E [1987] learning as a Constructive Activity InC
dynamic, levelled analysis of mathematical understanding Janvier (ed) Problems of rtpreswtation in the lwrning and /W{htng
In particular, it should allow one to see the self-similarity of mathematics Hillsdale: Lawrence Erlbaum Assoc
Wales, B [1984] A s·rudy of Chi/dun :s Language Ust When So!vmg
but transcendence in the levels, to see the process of vali- Partitioning Problems. Grades Two Through Four An unpublished
dation of personal knowledge and to comprehend transfer M Ed Thesis Edmonton: University of Alberta Department of Ele-
as recursive reconstruction It enables one to identify the mentary Education

II