Problem
I-1. Suppose that is a
matrix
with the property that
has the form
, no matter what the numbers
and
are.
Show that the entries of the third column of
must be
.
Problem
I-2. Let
,
,
.
(a) Find an affine transformation that moves the standard
triangle to the triangle (with
,
etc.).
(b) Find an affine transformation that moves to
the standard triangle.
(c) Find an affine transformation that takes to
respectively.
Problem
I-3. (a) Show that if is some sort of
transformation obeying the law
x
y
x
y
, then
0
0. (Therefore an affine
transformation does not obey this law unless it happens to
be a homogeneous linear transformation.)
(b) Show that if is an affine transformation, then
does obey the law
x
y
z
x
y
z
. (Start by writing
x
x
b.)
Problem
I-4. Let be a parallelogram, as pictured in Section
2 above. It is an easy fact that
can be expressed
algebraically in terms of
as
. Use
this fact and the result of Problem I-
to derive
algebraically the statement made in §
above that if
is an affine transformation taking three consecutive
vertices of a parallelogram to three consecutive vertices of
another parallelogram, then
takes the fourth vertex of the
first parallelogram to the fourth vertex of the second. (Here
vertices
are called ``consecutive'' if
and
are edges of the parallelogram.)
Problem
I-5. Let be three points in
R
,
and let
be the affine transformation that takes the
standard triangle to the triangle
.
(a) What is the area of the standard triangle in
R?
(b) An interesting quantity associated with a triangle
has several equivalent expressions:
Show that these expressions do in fact have the same value, so
that any one of them can be used to define . (Method:
Use facts about determinant expansion and row operations.)
(c) Show that the area of a triangle is given by
.
(Method: Think about the affine transformation taking the
standard triangle to
, as in Problem I-
. This
transformation is constructed as the composition of a homogeneous
linear transformation and a translation. By what factor does
each change areas?)
(d) Explain why the sign of tells whether you are
going counterclockwise or clockwise if you go from
to
to
and back to
. (Method: As in (c). Notice
that for the standard triangle the vertices are listed going
counterclockwise.)
(e) For the specific example ,
,
, find the area of the triangle
and say from
whether
are listed going counterclockwise or
clockwise.
(f) If
, what does that say about how the points
lie with respect to one another?
Problem
I-6.
As you know, the graph of a function
R
R is a set
of points in the plane, namely
R
.
More generally, the graph of any function between sets,
,
is the set
. In this
definition, recall that
just means the set of all
pairs
with
.
(a) Show that a function
R
R
is a homogeneous
linear transformation if and only if the graph of
is a subspace
of
R
R
, or equivalently, of
R
.
(This is a
problem; show each direction separately. Do
the
direction first. In the
direction,
notice that to start with you are not assuming that
is a homogeneous
linear transformation; it's just some function with a graph that is
a subspace, and you must verify that the defining properties of a homogeneous
linear transformation hold.)
(b) Show that a function
R
R
is an affine
transformation if and only if its graph is flat, meaning that its
graph is a subspace translated by some constant vector. (For the
constant vector you can use
0
.)
Problem
I-7.
Suppose that a graphics package uses device coordinates (i.e.,
screen coordinates) in which the upper left corner of the
screen is the origin and the lower right corner is
. Suppose that you want to plot points in a
window that in world coordinates is expressed by
and
. What affine transformation
transforms world-coordinate pairs to device-coordinate pairs?
Solve this problem by using a general method for finding an
affine transformation that takes one given parallelogram to
another.
Problem
I-8. An older laser printer has a graphics mode with
device coordinates in which the lower left corner of the
screen is the origin and the upper right corner is
. Suppose someones asks you to plot data
pairs from a window that is taller than it is wide, say with
and
. To accomplish
this, you want to transform the data pairs so that the user
origin is at the upper left corner of the screen and the
graph is on its side; of course, after the graph is plotted
on the Imagen you can turn the paper any way you like. By
what affine transformation should you transform the data
pairs before sending the data to the printer? (Be sure that
the graph does not come out non-uniformly scaled or
reflected. A window that is wider than it is tall is said
to have landscape shape; a window that is taller than it
is wide is said to have portrait shape.)
Problem
I-9.
PostScript printers assume an origin at the lower left
corner of the output paper, if you're holding at the paper with
the longer direction being up and down (``portrait position'').
The assumed unit length is one ``point'' (= 1/72 of an inch).
Suppose you want to print a picture that is wider than it is tall
(``landscape position''), with the origin also at the lower left,
using the whole sheet of 8.5 11-inch paper. One way
would be to transform your output points by an affine transformation
so that the picture is rotated by
and the origin is
moved to the lower right-hand corner. Give the extended matrix
of such a transformation. (Remember that your transformation is
not in inches. Actually, in PostScript you can simply declare a
new coordinate system in terms of the old one, but the same
matrix is needed to do this. In this problem, which position is
in user coordinates and which in device coordinates?)
Problem
I-10. Suppose you have a program that plots on a
screen in Tektronix coordinates (lower left , upper
right
). You then want to show four plots on
the screen at once, each half as large in each dimension, by
dividing the screen into four rectangles of the same shape
(four viewports). Give four affine transformations,
which, if applied to the output of your program, will put
the results in the four viewports. For example, the first
transformation could transform the whole screen to the upper
left quarter-screen.
Problem
I-11. In
R, by the standard tetrahedron
let us mean the tetrahedron with vertices
e
,
e
,
e
,
0
, listed in that order.
(a) Find the volume of the standard tetrahedron. (From calculus, the volume of a tetrahedron is 1/3 times the area of the base times the height.)
(b) Using the geometrical interpretation of the determinant and
the two-step construction of , analogously to
Problem I-
, explain why the volume of a tetrahedron
is given by
where
.
(c) By using facts about determinant expansion and and row operations,
show that can be written either as the determinant of
the extended matrix for
or even more symmetrically:
(d) Let
,
,
,
. Find the volume of the tetrahedron
.
(e) Can you find a geometrical interpretation of the sign of
?
Problem
I-12. (a) The standard square in
R is
the square with vertices
,
,
,
. Find an affine transformation that takes the
standard square to the square
, with
going
to
and
going to
, where
,
, and the vertices are listed counterclockwise.
(Method: You could try to find
and use triangle
methods, but it is simpler to combine a homogeneous linear
transformation of the form
with a translation.)
(b) A former student from this class, now in industry, has
asked how to find an affine transformation on
R
that preserves all shapes (but not size) and takes a given
line segment
to another given line segment
. Describe such a method.
Problem
I-13. In
R, consider the standard square
(mentioned in Problem I-
). Write down matrices for all
eight affine linear transformations that take this square to itself
rigidly, not necessarily leaving vertices fixed.
Problem
I-14. Suppose you are given a
matrix
. The homogeneous linear transformation
x
x
of course leaves the origin fixed. Suppose you want an
affine transformation
that is like
except that
it leaves a certain point
fixed instead of the
origin. It makes sense to let
be of the form
x
x
b, for the same
and for some suitable
vector
b. But what
b should you use to make
? (Your answer should express
b in terms
of
and
.)
Problem
I-15. Suppose and
are two
congruent triangles in
R
. Say explicitly how to
find an affine transformation that takes the first triangle
to the second with a rigid motion. (A detailed proof is not
required.)
(Suggested method: In
R, the analogue of the
triangle method in
R
maps tetrahedra, not
triangles. You could make tetrahedra just by choosing
arbitrarily a fourth vertex to go with each triangle, but
the resulting affine transformation is not necessarily
rigid. What you need is a way of choosing a fourth
vertex for each triangle in a regular way so that the two
resulting tetrahedra are congruent. Then the motion will be
rigid on the tetrahedra and so, you may assume, on all of
R
. One way: Find the fourth vertices
and
by taking the cross product of two sides of each
triangle and adding to a vertex. Your answer may be given
as a product
with entries of
and
clearly stated. Interestingly,
and
individually
may not give rigid motions, but
will.)
Problem I-16. This is a discussion of how to make the smoke in Picture #1 of a house in Handout C, by using an idea that occurs frequently in this course: Make drawings in strange positions by first making easy figures in easy positions and then applying affine transformations.
To make an easy wavy graph somewhat like the smoke, consider the
graph of
for
. Since
has values bouncing between
and
,
has values bouncing between the line
and
the line
. This same graph can be expressed
parametrically as
for
; parametric curves are more easily mapped.
The problem: Find an affine transformation that when applied to this graph gives the smoke in the correct position. If you wish, you may express your answer as a product of matrices and inverses of matrices, each with explicit entries.
One possible method: Think of the curve as drawn in the triangle
PQR, where
,
, and
is the origin. Transform the triangle several times to
get it to be the correct size and in the correct position.
Or another possible method: Using the triangle-to-triangle method
of Problem , transform the triangle just mentioned to a
triangle containing the smoke, whose base goes from
to
(the top of the
chimney) and whose top vertex is
.
Problem
I-17. Suppose you want to have a computer imitate
handwritten letters. One good approach is to design ``model
letters'' in a rectangle and then apply transformations to
reshape them a bit as desired. For example, your model
lower-case ``d'' might consist of a circle of radius
and
center
, together with the line
segment from
to
.
(a) Give a transformation that would turn the model ``d'' into a
thinner letter, in the rectangle
,
.
(b) Give a specific example of a transformation that would turn the model ``d'' into a version that is slanted and thinner but of the same height, somewhat like italic ``d'' but without the curl at the lower right. (You may choose where to position the image.)
Problem
I-18. In
R a line is like a
one-dimensional subspace, except that a subspace goes
through the origin and a line does not have to. You could
say that a line is just a one-dimensional subspace that has
been translated. The same is true of lines in
R
.
Planes in
R
could be described as two-dimensional
subspaces that have been translated. More generally, in
R
, a flat is any set that can be obtained by
translating a subspace. Thus lines and planes are examples
of flats. In
R
, for any
a
-simplex is a set of
points that does not lie
in a
-dimensional flat.
(a) In
R, what is a more common name for a
2-simplex? For a 3-simplex? For a 1-simplex?
(b) State without proof a fact about the possibility of
mapping one -simplex to another in
R
.
Problem
I-19. (a) Which homogeneous linear transformations
on
R R
take the line
to itself,
i.e., take points of the form
to points of the
same form
? (b) Which homogeneous linear
transformations on
R
R
take the plane
to itself? (In each case, give your reasoning; the
answer should describe the possible matrices precisely.)
Problem I-20. (a) Show that the formula for the matrix that maps one triangle onto another can be written more symmetrically:
.
(Method:
can be changed
into
by row operations,
and row operations can be achieved by pre-multiplying by a
suitable elementary matrix
, or in other words, a matrix
that is like the identity matrix except for some entries in one
column or one row. Find
explicitly. The right side of the
matrix equation then has the form
. Simplify.)
(b) Of course, all this only works for an expression of the
form
. We cannot use
in place of
generally. Show that even so,
.
(This fact was behind earlier problems for formulas for the area of a triangle and volume of a tetrahedron. Method: Use the fact that determinants are compatible with matrix multiplication.)
Problem
I-21.
In
R, suppose
are vertices of a square
obtained by rotating the standard square by
, so
that
,
that
, and
that
.
Find a matrix for an affine transformation taking
to the
parallelogram
. You may leave your answer
as a product of matrices with explicit entries.
Problem
I-22. (a) Using calculus, verify that the region under
the parabola
takes up
of the
area of the rectangle with
,
.
(b) A fact due to Archimedes is that a parabola inscribed in a
rectangle, from one vertex to an adjacent vertex and tangent to
the opposite side, bounds a region with of the area
of the rectangle. Use (a) and a transformation to prove this
statement. (Assume without proof the fact that each rectangle
with two specified adjacent vertices contains only one inscribed
parabola through them.)
Problem
I-23. A certain device has a window with lower left coordinates
and upper right coordinates
, and you need to write a routine to map
to it an arbitrarily specified user window with lower left
coordinates
and upper right
coordinates
. Explain how
to map the windows in each of these cases:
(a) The user window is to be mapped onto the device window
nonuniformly (i.e., the horizontal and vertical coordinates
work independently). Your answer will be a
matrix whose homogeneous part is diagonal and whose entries
are formulas involving the corner data given.
(b) The user window is to be mapped into the device window
uniformly (so the horizontal and vertical axes are scaled by the
same factor). If the two windows are not the same shape, then it
will not be possible to map one window onto the other.
Instead, transform the user window so that its image fills the
device window horizontally and is centered vertically, or
vice-versa. Your answer will be a
matrix
whose homogeneous part is a
scalar matrix and
whose entries depend on the corner data. Say how to compute
these entries. You will need either to break the solution
into cases or else to use
max
or min
functions.