The Geometry of the Dot and Cross Products

Tevian Dray
Department of Mathematics
Oregon State University
Corvallis, OR 97331
[email protected]
Corinne A. Manogue
Department of Physics
Oregon State University
Corvallis, OR 97331
[email protected]
January 15, 2008
Abstract
We argue for pedagogical reasons that the dot and cross products
should be dened by their geometric properties, from which algebraic
representations can be derived, rather than the other way around.
1 Introduction
Most students rst learn the algebraic formula for the dot and cross prod-
ucts in rectangular coordinates, and only then are shown their geometric
interpretations. We believe this should be done in the other order. Students
tend to remember best the rst denition they use; this should not be an
algebraic formula devoid of context. The geometric denition is coordinate
independent, and therefore conveys invariant properties of these products,
not just a formula for calculating them. Furthermore, it is easier to derive
the algebraic formula from the geometric one than the other way around, as
we demonstrate below.
1

v
v
w
| w|
w
Figure 1: The dot product is fundamentally a projection.
2 Dot Product
The dot product is fundamentally a projection. As shown in Figure 1, the
dot product of a vector with a unit vector is the projection of that vector in
the direction given by the unit vector. This leads to the geometric formula
v w = |v|| w| cos (1)
for the dot product of any two vectors v and w.
An immediate consequence of (1) is that the dot product of a vector with
itself gives the square of the length, that is
v v = |v|
2
(2)
In particular, taking the square of any unit vector yields 1, for example
= 1 (3)
where as usual denotes the unit vector in the x direction.
1
Furthermore,
it follows immediately from the geometric denition that two vectors are
orthogonal if and only if their dot product vanishes, that is
v w v w = 0 (4)
For instance, if denotes the unit vector in the y direction, then
= 0 (5)
1
We follow standard usage among scientists and engineers by putting hats on unit
vectors.
2

B

C

A
Figure 2: The Law of Cosines is just the denition of the dot product!
The geometry of an orthonormal basis is fully captured by these properties;
each basis vector is normalized, which is (3), and each pair of vectors is
orthogonal, which is (5).
The components of a vector v in an orthonormal basis are just the dot
products of v with each basis vector. For instance, in two dimensions, setting
v
x
= v
v
y
= v (6)
implies v = v
x
+ v
y
. The component form of the dot product now follows
from its properties given above. For example, if w = w
x
+ w
y
, then
v w = (v
x
+ v
y
) (w
x
+ w
y
)
= v
x
w
x
+ v
y
w
y
+ v
x
w
y
+ v
y
w
x
(7)
= v
x
w
x
+ v
y
w
y
This computation clearly works for any orthonormal basis. A special case
is the dot product of a vector with itself, which reduces to the Pythagorean
theorem, for example
v v = |v|
2
= v
2
x
+ v
2
y
(8)
What happens if you dont use an orthonormal basis? Consider Figure 2,
in which

A+

C =

B, or equivalently

C =

B

A. Then

C

C = (

A+

B) (

A+

B)
=

A

A+

B

B 2

A

B (9)
3
u
v
w
Figure 3: A geometric proof of the linearity of the dot product.
or equivalently
|

C|
2
= |

A|
2
+ |

B|
2
2|

A||

B| cos (10)
which is just the Law of Cosines! The Law of Cosines is usually used to
derive the geometric form of the dot product (1) from the algebraic form (7),
which is taken as the denition. Instead, by starting with geometry, the Law
of Cosines follows immediately.
Not so fast! Did you spot the aw in the above argument? In the com-
putation (7) of the algebraic formula for the dot product in terms of com-
ponents, it was assumed without comment that the dot product distributes
over addition, or in other words that the dot product is linear. If one starts
with the geometric denition (1), this must be proved.
However, the proof is straightforward, as shown in Figure 3.
2
We must
show that
(v + u) w = v w + u w (11)
But this is equivalent to showing that the projection of v + u along w is
the sum of the projections of v and u, which is immediately obvious from
Figure 3.
3 Examples
What is the bonding angle of carbon tetrachloride? Take a tetrahedron, and
connect each vertex (a chlorine atom) to the center (a carbon atom). What
is the angle between the lines that meet at the center?
2
Active versions of this gure are available online at [1] in both Java and Maple formats.
4
Figure 4: The geometry of the bonding angle of carbon tetrachloride.
This problem can be done by brute force using high school geometry try
it. A simpler approach is to represent the tetrahedron using vectors. It helps
to realize that a tetrahedron is formed by connecting alternating vertices of
a cube, as shown in Figure 4, and that the center of the tetrahedron is at the
center of the cube. It is now straightforward to write down the coordinates
of the vertices, thus obtaining the components of the vectors from the center
(shown in red in the gure). Compute the dot product (algebraically), divide
by the magnitudes, and use the geometric expression for the dot product to
read o the cosine of the angle. The answer? A bit more than 109

.
A simpler version of this problem would be to determine the angle between
the diagonal of a cube and one of its edges, that is, between the green (or
red) lines and the adjacent black lines in Figure 4. Or between the diagonals
of adjacent faces the blue lines in the gure. Again, both the geometric
and algebraic expressions for the dot product are involved in the solution.
4 Gram-Schmidt Orthogonalization
The most common use of the dot product in applications in physics and
engineering is to decompose vectors into their components parallel and per-
pendicular to a given vector, for which an understanding of the geometric
denition (1) is essential.
The Gram-Schmidt orthogonalization process uses this idea to construct
an orthonormal basis from a given set of (linearly independent) vectors. This
can be beautifully illustrated in three dimensions, as shown in Figure 5 and
described below. Stick out three ngers of one hand in arbitrary directions.
5
Figure 5: The geometry of the Gram-Schmidt orthogonalization process.
Pick one of these vectors (black) as the rst of the new orthogonal basis
vectors. Then pick a second (thick red), and subtract from this vector its
projection parallel to the rst, resulting in a vector perpendicular to the
rst (thin red). Now subtract from the remaining vector (thick blue) its
projections parallel to both the rst and second vectors, resulting in a vector
perpendicular to both (thinnest blue). If your ngers are limber enough, you
have an orthogonal basis. These three vectors (black, thin red, thinnest blue)
can be normalized if desired, yielding an orthonormal basis. Students who
see this geometric idea behind the Gram-Schmidt process have a much easier
time sorting their way through the morass of formulas in textbooks.
In advanced courses, the fact that two vectors are perpendicular if their
dot product is zero may be used in more abstract settings, such as Fourier
analysis. A problem which asks students to nd the vector perpendicular to a
given vector, rst in two and then in three dimensions, provides an excellent
introduction to this idea.
5 Cross Product
The cross product is fundamentally a directed area. The magnitude of the
cross product is dened to be the area of the parallelogram shown in Figure 6.
This leads to the formula
|v w| = |v|| w| sin (12)
an immediate consequence of which is that
v w v w =

0 (13)
6

| w|
v |v| sin
w
Figure 6: The geometric denition of the cross product, whose magnitude is
dened to be the area of the parallelogram.
The direction of the cross product is given by the right-hand rule, so that in
the example shown v w points into the page. This implies that
v w = w v (14)
so that the cross product is not commutative.
3
Another important property
of the cross product is that
v v =

0 (15)
which also follows immediately from (12).
In terms of the standard orthonormal basis, the geometric formula quickly
yields
=

k


k = (16)

k =
with the remaining products being determined by (14) and (15). This cyclic
nature of the cross product can be emphasized by diagramming the multipli-
cation table as shown in Figure 7.
4
Products in the direction of the arrow
get a plus sign (e.g.

k = + ), while products against the arrow get a minus
sign (e.g.

k = ).
3
This may be the rst example some students have seen of a product which is not
commutative. It is worth pointing out that the cross product also fails to be associative,
7
k
j i
Figure 7: The cross product multiplication table.
Using an orthonormal basis such as {, ,

k}, the geometric formula re-
duces to the standard component form of the cross product. If v = v
x
+
v
y
+ v
z

k and and w = w
x
+ w
y
+ w
z

k, then
v w = (v
x
+ v
y
+ v
z

k) (w
x
+ w
y
+ w
z

k)
= v
x
w
x
+ v
x
w
y
+ ... (17)
= (v
y
w
z
v
z
w
y
) + (v
z
w
x
v
x
w
z
) + (v
x
w
y
v
y
w
x
)

k
which is often written as the symbolic determinant
v w =

k
v
x
v
y
v
z
w
x
w
y
w
z

(18)
As with the dot product, this derivation works in any (right-handed) or-
thonormal basis.
OK, this time you surely noticed; we must again check linearity, that
is, whether the cross product distributes over addition, or in other words
whether
w (v + u) = w v + w u (19)
since for example ( ) = but ( ) =

0.
4
This is really the multiplication table for the unit imaginary quaternions, a number
system which generalizes the familiar complex numbers. Quaternions predate vector anal-
ysis, which borrowed the quaternionic units i, j, k as labels for the rectangular basis
vectors , ,

k. A more logical name for the rectangular basis vectors would be x, y, z,
which is used by many physicists.
8
u
v
w
Figure 8: A geometric proof of the linearity of the cross product.
As we now show, this follows with a little thought from Figure 8.
2
Consider
in turn the vectors v, u, and v + u. The cross product of each of these
vectors with w is proportional to its projection perpendicular to w. These
projections are shown as solid lines in the gure. Since the projections lie in
the plane perpendicular to w, they can be combined into the triangle shown
in the middle of the gure. Two of the vectors making up the sides of a
triangle add up to the third; in this case, the sides are the projections of v,
u, and v+u, and the latter is clearly the sum of the rst two. But each cross
product is now just a rotation of one of the sides of this triangle, rescaled
by the length of w; these are the arrows perpendicular to the faces of the
prism. Two of these vectors therefore still add to the third, as indicated by
the vector triangle in front of the prism. This establishes (19).
6 Cross Product Pedagogy
We digress briey to mention some pedagogical issues which arise when teach-
ing students about the cross product.
First of all, we strongly discourage teaching, or even reviewing, the dot
and cross products at the same time students tend to get them mixed up!
As for the calculation of the cross product, we encourage students to
compute the determinant (18), rather than memorizing (17). But we also
encourage students to use the multiplication table directly for simple cross
products, such as ( + 3 )

k, rather than using determinants at all.
9
However, we discourage students from working out the determinant using
minors, which would result in the nal expression in (17) but with two sign
changes in the term involving . Instead, we encourage students to take 33
determinants in the form


k
v
x
v
y
v
z
v
x
v
y
w
x
w
y
w
z
w
x
w
y
@
@
@
@
@
@
@
@
@
@
@
@
@
@
@

(20)
where one multiplies the terms along each diagonal line, subtracting the
products obtained along lines going down to the left from those along lines
going down to the right. While this method works only for (2 2 and) 3 3
determinants, and is therefore usually omitted from a linear algebra course,
it has the denite advantage of emphasizing the cyclic nature of the cross
product whereas students who use minors often make sign errors.
7 Divergence and Curl
The geometric approach extends naturally to the divergence and curl of a
vector eld.
It is straightforward to dene divergence as ux per unit volume. This
amounts to taking an innitesimal version of the Divergence Theorem as
the denition of divergence; the computation traditionally done to prove the
theorem become instead the derivation of the component representation of
the divergence, expressed in rectangular coordinates.
This approach is beautifully summarized in [2], and is included in some
physics textbooks, such as [3, 4, 5], but has not been adopted in many math-
ematics textbooks (a notable exception being [6]). It not only emphasizes
the geometric relationship between divergence and ux, but makes clear how
to derive the component representation in other coordinate systems.
It then seems natural to dene curl as (oriented) circulation per unit area,
turning an innitesimal version of Stokes Theorem into a denition, which
is then used to derive the component representation of curl in rectangular
coordinates. This is indeed the approach taken in [2] (and [6]).
However, there is an issue of principle which we have not seen addressed
in textbooks: One must argue that this procedure denes a vector eld! It
is not enough to derive the separate formulas for the circulation per unit
10
z
x y
Figure 9: The decomposition of a triangle into component triangles in the
coordinate planes.
area in each (rectangular) coordinate plane, one must also show that the
circulation in any plane is equal to an appropriate sum of these circulations.
But this follows immediately from the diagram in Figure 9, which decom-
poses an arbitrary triangle into three triangles in the coordinate planes; it
is straightforward to check that the circulation around the original triangle
is equal to the sum of the (appropriately oriented!) circulations around the
component triangles. The contributions of the circulations along all the
extra sides cancel pairwise.
8 Discussion
We have shown how to prove geometrically that the dot and cross products
are linear (i.e. that they distribute over addition), using Figures 3 and 8,
respectively. These proofs are important for faculty wishing to emphasize
the geometric denitions in their classroom. However, it is not at all clear
that these proofs are relevant for students, especially in introductory courses.
Most students will accept, say, the argument presented in (7) without ques-
tion, failing to notice that linearity is an issue. While faculty should of course
be prepared to justify this if asked, we see nothing wrong with not raising
the issue of linearity otherwise. Similarly, the proof that the curl is a vector
eld can be omitted.
As shown above, the Law of Cosines follows immediately from the ge-
ometric denition of the dot product, in direct contrast to the traditional
treatment, in which the order is reversed. Indeed, many students who have
11
v
u

Figure 10: Deriving the addition formula for cosine using the dot product.
memorized the Law of Cosines are delighted with this derivation. This sug-
gests to us that it might be benecial to include the dot product in trigonom-
etry courses, even in high school, rather than saving it for later. A strong
argument in favor of such an approach is the ease with which the addition for-
mulas could then be derived: From the unit circle denition of the trig func-
tions, any vector on the unit circle can be written in the form cos +sin ,
where is the angle to the positive x-axis. But the cosine of the angle be-
tween any two such vectors is just their dot product, since both are unit
vectors, so that
cos( ) = u v = (cos + sin ) (cos + sin )
= cos cos + sin sin (21)
as illustrated in Figure 10.
Why arent geometric proofs such as those presented here more common?
We believe this lies in part with the diculty in translating them into words
on the printed page. Consider for example the Gram-Schmidt orthogonal-
ization process. As discussed in Section 4, the geometry of this construction
can be beautifully illustrated by (literally) waving ones hands. It is easy to
write down the formulas after seeing this demonstration, but the formulas
by themselves are much less enlightening. You may have had some trouble
sorting through the arguments in this paper, but try them in the classroom,
where you can point at the pictures. They work like a charm.
12
Acknowledgments
This work was partially supported by NSF grants DUE0088901 and DUE
0231032. Early versions of the gures used to prove linearity were constructed
while we were Noyce Visiting Professors at Grinnell College, an opportunity
for which we are grateful. The print version of this manuscript was pre-
pared using L
A
T
E
X, then converted to MathML using T
T
M [7] for online use;
the color gures were originally drawn using Maple, then converted to Java
applets using JavaView [8].
References
[1] Tevian Dray and Corinne A. Manogue, Dot and Cross Products webpage.
Retrieved June 20, 2005 from
http://www.math.oregonstate.edu/bridge/ideas/dotcross.
[2] H. M. Schey, Div, Grad, Curl and all that, 4th edition Norton, New
York, 2005.
[3] David J. Griths, Introduction to Electrodynamics, 3rd edition,
Prentice-Hall, New York, 1999.
[4] Paul Lorrain, Dale P. Corson, and Fran cois Lorrain, Electromagnetic
Fields and Waves, 3rd edition, Addison-Wesley, New York, 1988.
[5] John R. Reitz, Frederick J. Milford, Robert W. Christy, Foundations
of Electromagnetic Theory, 4th edition, W. H. Freeman, New York,
1993.
[6] William McCallum, Deborah Hughes-Hallett, Andrew Gleason, et al.,
Multivariable Calculus, 3rd edition Wiley, New York, 2001.
[7] Ian Hutchinson, T
T
M website. Retrieved January 28, 2006 from
http://hutchinson.belmont.ma.us/ttm.
[8] Konrad Polthier, JavaView website. Retrieved June 19, 2005 from
http://www.javaview.de.
13