RAISING STUDENTS’ UNDERSTANDING: LINEAR ALGEBRA

Marianna Bogomolny
Southern Oregon University
This study is a contribution to the ongoing research in undergraduate mathematics
education, focusing on linear algebra. It is guided by the belief that better
understanding of students’ difficulties leads to improved instructional methods. The
questions posed in this study are: What is students’ understanding of linear
(in)dependence? What can example-generation tasks reveal about students’
understanding of linear algebra? This study identifies some of the difficulties
experienced by students with learning the concepts of linear dependence and
independence, and also isolates some possible obstacles to such learning. In addition,
this study introduces learner-generated examples as a pedagogical tool that helps
learners partly overcome these obstacles.

BACKGROUND AND THEORETICAL PERSPECTIVE

There is a common concern expressed in the literature that students leaving a linear
algebra course have very little understanding of the basic concepts, mostly knowing
how to manipulate different algorithms. Carlson (1993) stated that solving systems of
linear equations and calculating products of matrices is easy for the students. However,
when they get to subspaces, spanning, and linear independence students become
confused and disoriented: “it is as if the heavy fog has rolled in over them.” Carlson
(1993) further identified the reasons why certain topics in linear algebra are so difficult
for students. Presently linear algebra is taught far earlier and to less sophisticated
students than before. The topics that create difficulties for students are concepts, not
computational algorithms. Also, different algorithms are required to work with these
ideas in different settings.
Dubinsky (1997) pointed out slightly different sources of students’ difficulties in
learning linear algebra. First, the overall pedagogical approach in linear algebra is that
of telling students about mathematics and showing how it works. The strength, and at
the same time the pedagogical weakness, of linear situations is that the algorithms and
procedures work even if their meaning is not understood. Thus, students just learn to
apply certain well-used algorithms on a large number of exercises, for example,
computing echelon forms of matrices using the Gaussian row elimination method.
Secondly, students lack the understanding of background concepts that are not part of
linear algebra but important to learning it. Dorier, Robert, Robinet, and Rogalski
(2000) identified students’ lack of knowledge of set theory, logic needed for proofs,
and interpretation of formal mathematical language as being obstacles to their learning
of linear algebra. Thirdly, there is a lack of pedagogical strategies that give students a
chance to construct their own ideas about concepts in linear algebra.
Examples play an important role in mathematics education. Students are usually
provided with examples by teachers, but are very rarely faced with example-generation
2007. In Woo, J. H., Lew, H. C., Park, K. S. & Seo, D. Y. (Eds.). Proceedings of the 31st Conference of
the International Group for the Psychology of Mathematics Education, Vol. 2, pp. 65-72. Seoul: PME. 2-65
Bogomolny

tasks, especially as undergraduates. As research shows (Hazzan & Zazkis, 1999;

Watson & Mason, 2004), the construction of examples by students contributes to the
development of understanding of the mathematical concepts. Simultaneously,
learner-generated examples may highlight difficulties that students experience. This
study examines how and in what way example-generation tasks can inform about and
influence students’ understanding of linear algebra. The APOS theoretical framework
was adopted in this study to interpret and analyse students’ responses (Asiala et al,
1996).
METHOD
The participants of the study were students enrolled in Elementary Linear Algebra
course at a Canadian University. The course is a standard one-semester introductory
linear algebra course. It is a required course not only for mathematics majors but also
for students majoring in computing science, physics, statistics, etc. 113 students
participated in the study. Later in the course the students were asked to participate in
individual, clinical interviews. A total of six students volunteered to participate in the
interviews. These students represented different levels of achievement and
sophistication.
The data for this study comes from the following sources: students’ written responses
to the questions designed for this study, and clinical interviews. To follow the
example-generation process, Task: Linear (in)dependence was included in the
interview questions as well. Having students generate examples and justify their
choices through written responses and in an interview setting provided an opportunity
not only to observe the final product of a student’s thinking process but also to follow it
through interaction with a student during his/her example-generation.
Task: Linear (in)dependence
a). (1). Give an example of a 3x3 matrix A with real nonzero entries whose columns a1,
a2, a3 are linearly dependent.
(2). Now change as few entries of A as possible to produce a matrix B whose columns
b1, b2, b3 are linearly independent, explaining your reasoning.
(3). Interpret the span of the columns of A geometrically
b). Repeat part a (involving A and B), but this time choose your example so that the
number of changed entries in going from A to B takes a different value from before.
The prerequisite knowledge for many concepts in linear algebra is the linear
dependence relation between vectors. The purpose of the task was to investigate
students’ understanding of the concept of linear dependence and linear independence
of vectors, in particular, in R3. Many concepts of linear algebra are connected, and
students should be able to use all these terms freely and with understanding. On one
hand, this task connects the number of linearly independent columns in a matrix A, the
number of pivots in an echelon form of A, and the dimension of the vector space

2-66 PME31―2007
 Bogomolny

spanned by the column vectors of A. On the other hand, it connects the minimum
number of entries required to be changed in A to make its columns linearly independent,
and the number of free variables in the matrix equation Ax = 0. This task also explores
the possible proper subspaces of a vector space R3 (excluding the subspace spanned by
the zero vector, Span{0}). It can be further extended to a 4x4 case, and then to the
general case of nxn matrices.
This is an open-ended task with no learnt procedures to accomplish it. The routine
tasks ask students to determine if a set of vectors is linearly dependent or independent
by applying the definition or theorems presented in the course. In part (a1) of the task,
the given and the question are reversed. Zazkis and Hazzan claim that ‘such
“inversion” usually presents a greater challenge for students than a standard situation’
(1999, p.433). To complete Task (a1) students have to adjust their prior experiences in
order to construct a set of three linearly dependent vectors in R3, viewed as columns of
a 3x3 matrix A.

RESULTS: LINEAR DEPENDENCE AND LINEAR INDEPENDENCE

The two concepts of linear dependence and independence are closely connected. To
have a solid understanding of one of them involves having understanding of the other.
I first present the summary of students’ responses for constructing matrix A with
linearly dependent columns, and then use APOS theoretical framework to analyze
students’ understanding of linear (in)dependence.
Constructing matrix A with linearly dependent columns
In Task (a1) the students were required to give an example of three linearly dependent
vectors represented as a 3x3 matrix with nonzero real entries. Table 1 presents the
summary of different approaches used to complete this part of the task. The total
frequencies exceed the number of participants as some students provided two
examples for the task. Although all but 6% of the students constructed correct
examples, their methods indicate different levels of understanding.
The responses to the remaining parts of the task depended on the construction of a
matrix A. The results and analysis of these remaining parts are presented with
examples of students’ work below.
Linear dependence as action
For students using a guess-and-check strategy the linear dependence was concluded as
an outcome of an action performed on a chosen 3x3 matrix A. To complete Task (a1),
these students had to pick 9 numbers to perform a set of operations on these numbers
getting a certain result, in this case, at least one zero row in the modified form of A.
This is a consequence of the condition for the columns of a matrix to be linearly
dependent. These students had to go through calculations explicitly to verify that their
example satisfied the requirement of the task.

PME31―2007 2-67
Bogomolny

Table 1: Constructing 3x3 matrix A with linearly dependent columns

Method Examples Frequency of

occurrence

Guess-and-Check 1 4 2  1 4 2  1 4 2  17%
method A = 2 5 1 → 2 5 1 → 2 5 1
3 6 7 3 6 3 3 6 0

Rows method: same 1 2 3 4 6 9  23%

rows, one row multiple A = 2 5 1 ; or A = 8 − 7 − 5 
  
   
of another row 2 5 1 2 3 9 2

Echelon method: start 1 0 1 1 0 1 5%

with echelon form U of U = 0 1 1    → 0 1 1
  R 3→ R 3+ R 2 + R1
   
A 0 0 0 1 1 2

 2 1 3
R1→ R1+ R 3

→ 1 2 3 = A
R 2→ R 2 + R 3
  
1 1 2

Identical columns 1 1 1  1 1 1 8%
method: A= 2 2 2 ; or A= 1 1 1
 
3 3 3 1 1 1
[a1 a1 a1] where a1 has
nonzero real entries

Multiple columns 1 3 9 19%

method: two columns A = [a1 3a1 9a1] = 1 3 9 ;

are multiples of the 1 3 9
first one - [a1 ca1 da1]
where a1 has nonzero 3 6 9
real entries and c and A = [a1 2a1 3a1] = 2 4 6 

d are both nonzero 4 8 12
real numbers

2-68 PME31―2007
 Bogomolny

Two multiple columns 1 1 1  2 4 5 21%

method: two identical A = 2 2 2 ; or A = [a1 2a1 a3] = 3 6 6
 
   
columns or two 3 3 4 4 8 1
columns multiples of
each other and the
third column having
any nonzero real
entries: [a1 ca1 a3]

Linear combination 1 3 4  18%

method: any two A = [a1 a2 a1+a2] = 2 4 6 ;
columns, a1 and a2, 3 5 8
having nonzero real
entries and a3 = ca1 + 1 2 1 
da2 A = [a1 a2 2a1+ (− 1 2) a2] =  1 − 2 3 

1 2 − 1 3 2

Linear dependence as process

Applying the APOS theoretical framework, students are operating with the process
conception of linear dependence when they construct a matrix A emphasizing the
relations between the rows. They may know that in order for the columns of A to be
linearly dependent an echelon form of a matrix has to have a zero row. The row
reduction process is an intended action in this case. It is performed mentally, and then
reversed to generate a required matrix.
There is an intermediate step that links the linear dependence of columns of a matrix
and its echelon form having a zero row. The definition of linear dependence of a set of
vectors is given in terms of a solution to a vector equation. That is, a set of vectors {v1,
…, vn} is linearly dependent if the vector equation, c1v1 + … + cnvn = 0 has a nontrivial
solution. The solution set of this vector equation corresponds to the solution set of a
matrix equation Ax = 0 having the vi’s as columns which in turn corresponds to the
solution set of the system of linear equations whose augmented matrix is [A 0]. In the
prior instruction it was shown that the linear system Ax = 0 has a nontrivial solution if it
has free variables, and this can be inferred from an echelon form of A. Thus, some
students formed the connection: linear dependence ↔ free variables ↔ zero row in
echelon form. As a result some examples were justified with the following statements:
‘a linearly dependent matrix is a matrix with free variables’, ‘columns of A are linearly
dependent since x3 is free variable which implies Ax = 0 has not only trivial solution’,
or ‘when the forms are reduced into reduced echelon form, the linearly dependent
matrix has a free variable x3; however, the linearly independent doesn’t – it has a
unique solution’.

PME31―2007 2-69
Bogomolny

Only 50% of the students completed both parts of the task, with 63% of incorrect
responses to Task (b). In the majority of incorrect responses students ignored the
different structures of linear dependence relations between vectors. They either used
the same matrix A in both parts of the task or a matrix A having the same linear
dependence relations between columns. Then if the students changed the correct
number of entries in Task (a), their response to Task (b) was incorrect. For instance,
one student constructed matrix A for both parts with the same dependence relation
between columns, a3 in Span{a1, a2}, and changed 1 entry in the first part but 3 entries
in the second part.
Linear dependence as object
The row reduction process is central to linear algebra. It is an essential tool, an
algorithm that allows students to compute concrete solutions to elementary linear
algebra problems. However, encapsulation of linear dependence as an object requires a
movement beyond the outcome of actual or intended procedures of row reduction
toward a conceptual understanding of the structure of linear dependence relations in a
set of vectors.
An indication of the construction of linear dependence as an object is demonstrated
when students emphasize the relation between vectors, when they use the linear
combination method to construct their examples of three linearly dependent vectors. In
the linear combination method, there could be recognized different levels of generality
for constructing an example. Either students gave a specific example of a matrix with a
linear dependence relations between columns easily identified, as can be seen in Table
1, or they identified a general strategy for constructing a class of 3x3 matrices with
linearly dependent columns. For example, Amy wrote: ‘to be linearly dependent, at
least one of the columns of a matrix A has to be a linear combination of the others …
x1a1 + x2a2 + x3a3 = 0 with weights not all zero. Pick a1 and a2. Then for a1, a2, a3 to be
linearly independent, a3 has to be a linear combination of a1 and a2. So, let a3 = a1 + a2’.
In the latter case, students applied the property that if u and v are linearly independent
vectors in Rn, then the set of three vectors {u, v, w} is linearly dependent if and only if
w is in Span{u, v} (i.e. w is a linear combination of u and v).
During clinical interviews, students were able to move from the process understanding
of linear dependence to the object level. Initially, both Anna and Leon used matrix A
with the same linear dependence relation between columns to complete Task (b). They
were changing different number of entries but knew that some changes were
unnecessary.
Working through this task helped students understand the connection between the
linear dependence relations, the geometric interpretation, and the minimum number of
entries needed to change:
Leon: …actually, I think to make two changes is minimum, because all three vectors are
linearly dependent to one another. Changing one will not change the relationship
overall. You will still have at least two linearly dependent vectors. So can I draw

2-70 PME31―2007
 Bogomolny

from that with three linearly dependent vectors you need two changes and with
only two linearly dependent vectors you only need one change.
Anna even attempted to generalize her strategy for an nxn case:
Anna: First, if my vectors are the same it’s going to take more than one step to make them
linearly independent. But if two of the vectors are different and the last one is the
same as one of the other ones, I just need to change the leading entry number in that
matrix; so when I row reduce it, I have an identity matrix…If I have an n by n
matrix, and I have {v1, …, vn-1} and then I have 2v1. This one vector is twice v1, or
three times v1, just to keep it general, as my vn. So if I make {v1, …, vn-1} linearly
independent, and the very last one is cv1 then I need to change only one entry.
Object conception of linear dependence relation includes mastery of all possible
characterizations of a linearly dependent set of vectors, in particular, the ability to
recognize the possible ways to alter a set in order to obtain a linearly independent set.
In Task: Linear (in)dependence, encapsulation of linear dependence as an object
includes viewing a matrix as a set of column vectors, not as discrete entries that have
certain values after performing algebraic manipulations. The latter perspective inhibits
students’ geometric interpretation of the span of columns, because the structure of
linear dependence relations is not visible. Thus, the students that correctly completed
both parts of the task might be operating with the object conception of linear
dependence.

CONCLUSION
In general, example-generation tasks provide a view of an individual’s schema of basic
linear algebra concepts. Through the construction process and students’ examples we
see the relationships between the different concepts. Task: Linear (in)dependence
revealed that the connections linear dependence ↔ free variables / pivot positions /
zero row in echelon form, and linear independence ↔ no free variables / vectors not
multiples of each other are strong in students’ schema.
Learners’ responses to Task: Linear (in)dependence showed that many students treat
linear dependence as a process. They think of linear dependence in reference to the row
reduction procedure. Some students connected linear dependence to the homogeneous
linear system Ax = 0 having free variables that in turn corresponds to the nxn matrix A
having a zero row in an echelon form. Other students linked the linear independence of
vectors to a homogeneous linear system having only basic variables and therefore n
pivot positions. However, few students considered the different structures of the linear
dependence relations.
Even though geometric representation helps in visualizing the concepts, for some
students geometric and algebraic representations seem completely detached. This can
be seen in students’ attempts to provide a geometric interpretation of the span of the
columns of a matrix. There was a common confusion of the span of the columns of A
with the solution set of Ax = 0. Instead of providing a geometric interpretation of the
span of the columns of A, some students gave a geometric interpretation of the solution
set of the homogeneous system Ax = 0.

PME31―2007 2-71
Bogomolny

The tasks soliciting learner-generated examples were developed in this research for the
purpose of data collection. However, these tasks are also effective pedagogical tools
for assessment and construction of mathematical knowledge, and can contribute to the
learning process. Part of the power of Task: Linear (in)dependence is that it anticipates
the concept of rank, long before students are exposed to it. In playing with the
examples (assigned after only two weeks of classes), students develop their intuition
about what linear (in)dependence “really means”. The students may not be able to
articulate why the second example works differently from the first, but they are starting
to develop a “feel” for the difference. This task can be further extended to higher
dimensional vector spaces.
It is hoped that by examining students’ learning, the data collected can lead to teaching
strategies, which will help students expand their example spaces of mathematical
concepts and broaden their concept images/schemas. It is proposed that further studies
could discuss the design and implementation of example-generation tasks intended
specifically as instructional strategies, and evaluate their effectiveness.
References
Asiala, M., Brown, A., DeVries, D., Dubinsky, E, Mathews, D., and Thomas, K. (1996). A
Framework for Research and Curriculum Development in Undergraduate Mathematics
Education. Research in Collegiate Mathematics Education II, CBMS Issues in
Mathematics Education, 6, 1-32.
Carlson, D. (1993). Teaching linear algebra: must the fog always roll in? College
Mathematics Journal, 12(1), 29-40.
Dorier, J.-L., Robert, A., Robinet, J., and Rogalski, M. (2000). The obstacle of formalism in
linear algebra. In Dorier, J.-L. (ed), On the teaching of linear algebra. 85-124. Dordrecht,
the Netherlands: Kluwer Academic Publishers.
Dubinsky, E. (1997). Some thoughts on a first course in linear algebra on the college level. In
Carlson D., Johnson, C, Lay, D., Porter, D., Watkins, A, & Watkins, W. (eds.). Resources
for Teaching Linear Algebra, pp.107-126. MAA Notes, Vol. 42.
Hazzan, O. & Zazkis, R. (1999). A perspective on “give an example” tasks as opportunities to
construct links among mathematical concepts. FOCUS on Learning Problems in
Mathematics, 21(4), 1-13.
Watson, A. & Mason, J. (2004). Mathematics as a constructive activity: learners generating
examples. Mahwah, NJ: Lawrence Erlbaum Associates Inc.
Zazkis, R. and Hazzan, O. (1999). Interviewing in mathematics education research: Choosing
the questions. Journal of Mathematical Behavior, 17(4), 429-239.

2-72 PME31―2007