When we send an entire space of input vectors through a linear map, we want to know which outputs it can actually reach and whether different inputs ever land on the same spot.
1
Image and Onto
The Image is the set of all possible outputs a map can produce, and a map is Onto when its Image covers the entire target space. Let's find the Image of the matrix [1224] and see whether it is Onto R2.
Image and Onto
Text solution
Combine the columnsPlugging in an input vector like ⟨3,1⟩ combines the columns of the matrix to produce the output.
[1224]⟨3,1⟩=3⟨1,2⟩+1⟨2,4⟩=⟨5,10⟩
Keep the independent directionsBecause the second column ⟨2,4⟩ is just double the first column ⟨1,2⟩, every combination of the columns stays on the single line through ⟨1,2⟩.
Column SpaceImage=Span{⟨1,2⟩}
So the map is not Onto, because its 1D Image only covers a line inside the 2D target space and misses points off that line like ⟨3,1⟩.
The Image fills the entire 2D target space to make the map Onto whenever the second column is off the line through ⟨1,2⟩, and squishes flat onto a 1D line only when one column is a multiple of the other.
Two columns fill the whole 2D Target Space whenever they don't lie on the same line, and collapse onto a single line only when one is a multiple of the other.
2
Kernel and One-to-One
When a 2D input grid squishes down onto a 1D line, different inputs are forced to land on top of each other. A map is One-to-One when every input gets its own unique output, and the Kernel is the set of all inputs that get sent to the zero vector. Let's see how the Kernel tells us right away that [1224] is not One-to-One.
Kernel and One-to-One
Text solution
Subtract two colliding inputsBoth u=⟨5,0⟩ and v=⟨3,1⟩ land on the same output ⟨5,10⟩, so their difference cancels out to the zero output.
u⟨5,0⟩−v⟨3,1⟩⟹2⟨1,2⟩−1⟨2,4⟩=⟨2,−1⟩=⟨0,0⟩
Check what lands on zeroBecause the difference between any two colliding inputs lands in the Kernel, a map has no collisions anywhere if and only if the zero vector is the only input in the Kernel.
One-to-One⟺Null SpaceKernel={⟨0,0⟩}
Since the nonzero vector ⟨2,−1⟩ gets sent to ⟨0,0⟩, inputs collide across the entire space and the map is not One-to-One.
For the matrix [1224], whose Kernel is the line Span{⟨2,−1⟩}, every line of inputs parallel to the Kernel lands on a single output point on the Image line, because the difference u−v between any two inputs on that line lies on the Kernel line and maps to ⟨0,0⟩.
Inputs that share an output form a line parallel to the Kernel, because the difference between any two of them lies in the Kernel and gets sent to zero.
3
Finding Kernel and Image by Row Reduction
For a larger matrix where we can't spot how the columns relate by eye, row reduction simplifies the matrix so we can read off the Kernel, the Image, One-to-One, and Onto all at once. Now let's find all four for the 3×3 matrix 101213415.
Finding Kernel and Image by Row Reduction
101213415
Kernel=?One-to-One?Image=?Onto?
Text solution
Row-reduce to find the pivotsRow reduction keeps the exact same relationships between the columns, showing pivots in Columns 1 and 2 and no pivot in Column 3.
101213415row-reduce100010210
Cancel the free column for the KernelThe reduced third column shows that Col 3=2(Col 1)+1(Col 2), so taking −2 of Column 1, −1 of Column 2, and 1 of Column 3 cancels to zero.
Kernel=Span{⟨−2,−1,1⟩}
Take the original pivot columns for the ImageBecause Columns 1 and 2 have the pivots, the first two columns of the original matrix span all reachable outputs.
Image=Span{⟨1,0,1⟩,⟨2,1,3⟩}
So the map is not One-to-One because Column 3 has no pivot (sending Span{⟨−2,−1,1⟩} to zero), and it is not Onto because Row 3 has no pivot (leaving the 2D Image unable to fill R3).
4
One-to-One vs. Onto: Tall and Wide Matrices
In both square matrices above, the map failed One-to-One and Onto at the same time, which can make them look like the same test. Removing the third column leaves the tall 3×2 matrix 101213 mapping R2→R3, where the column check and the row check come apart.
One-to-One vs. Onto: Tall and Wide Matrices
x
y
z
R3
One-to-One?Onto?
Text solution
Check the columns for the Kernel and One-to-OneRow reduction puts a pivot in every column, so the only way to combine the columns to get ⟨0,0,0⟩ is 0⟨1,0,1⟩+0⟨2,1,3⟩.
101213row-reduce100010⟹Kernel={⟨0,0⟩}
Check the rows for the Image and OntoBoth columns have pivots, but the third row is all zeros because two columns can only span a 2D plane inside 3D target space.
Image=Span{⟨1,0,1⟩,⟨2,1,3⟩}
So this 3×2 map is One-to-One because every column has a pivot, and not Onto because the third row has no pivot.
Every input dimension (column) either stays in the Image as a pivot column or gets sent to zero in the Kernel as a free column:
In R3, the two columns of a 3×2 matrix can at most span a flat 2D plane, so the map is never Onto R3, though it stays One-to-One for every column direction except when one column lines up with the other.
Two columns in 3D space can at most make a flat plane, so the map stays one-to-one whenever the columns don't line up, and is never onto.
5
Kernel and Image of Polynomial Maps
Linear maps also act on vectors like polynomials and functions without starting as a matrix, which is why we use the general names Kernel and Image instead of Null Space and Column Space. Let's find the Kernel and Image of the derivative map D:P2→P1 given by D(a+bx+cx2)=b+2cx, first directly from the formula and then by turning the map into a matrix.
Kernel and Image of Polynomial Maps
D(a+bx+cx2)=b+2cx
P2Input Space
P1Target Space
D
0
0
Kernel=?
Image=?
Text solution
Set the output to zero for the KernelSolving b+2cx=0 forces b=0 and c=0, leaving every constant polynomial p(x)=a.
D(a+bx+cx2)⟹Kernel=0=Span{1}
Collect all reachable outputs for the ImagePicking any numbers for b and c in b(1)+2c(x) builds every linear polynomial in the target space P1.
Image=Span{1,x}=P1
Or write the matrix in the standard basesPlugging the input basis {1,x,x2} into D and writing the outputs in the basis {1,x} gives D(1)=0→⟨0,0⟩,D(x)=1→⟨1,0⟩, and D(x2)=2x→⟨0,2⟩.
[001002]
So D is not One-to-One because Column 1 (for the constant 1) has no pivot, and D is Onto because both rows have pivots, spanning all of P1.
Key concepts
Before row-reducing a matrix, check whether you can answer the question directly. If you are given two different inputs that share the same output, like u=⟨5,0⟩ and v=⟨3,1⟩ both mapping to ⟨5,10⟩, don't set up a matrix; just subtract them to get a nonzero vector u−v=⟨2,−1⟩ directly in the Kernel, which immediately shows the map is not One-to-One. Similarly, if you only need to know what is possible for a tall or wide matrix, count the dimensions without row-reducing:
A wide matrix like 2×3 (R3→R2) sends 3 input dimensions into a 2D target space, forcing dim(Kernel)≥1 so it can never be One-to-One. A tall matrix like 3×2 (R2→R3) starts with only 2 columns, so its Image is at most 2D and can never be Onto.
When you do have a matrix (or build one from a polynomial map using input and output bases), row-reduce it and check the columns and rows separately:
Columns tell you the Kernel (Null Space) and One-to-One
The Kernel is every input that gets sent to 0. If every column has a pivot, as in 100010, then Kernel={⟨0,0⟩} and the map is One-to-One. If a column has no pivot, as in 100010210 where Col 3=2(Col 1)+1(Col 2), cancel it to get Kernel=Span{⟨−2,−1,1⟩} and the map is not One-to-One.
Rows tell you the Image (Column Space) and Onto
The Image is every output the map can build, found by taking the original columns that have pivots. If every row has a pivot, as in [001002] for D:P2→P1, the Image covers the whole target space (Span{1,x}=P1) and the map is Onto. If any row is all zeros, the Image cannot reach the whole target space and the map is not Onto. When a map is both One-to-One and Onto, no dimensions are squished to zero and no dimensions in the target space are missed, so the entire map can be reversed.
6
Practice
1
A linear map sends two different input vectors u=⟨4,1⟩ and v=⟨1,3⟩ to the exact same output vector ⟨7,−2⟩. Find a nonzero vector in the Kernel of this map, and state whether the map is One-to-One.
If u and v produce the exact same output, what output do you get when you subtract them?
Subtract the two colliding inputs so their shared output ⟨7,−2⟩ cancels to ⟨0,0⟩:
u−v=⟨4,1⟩−⟨1,3⟩=⟨3,−2⟩
(Watch out: subtract the two inputs from each other. Don't use the output ⟨7,−2⟩ or subtract the output from an input to get ⟨4,1⟩−⟨7,−2⟩=⟨−3,3⟩, because ⟨7,−2⟩ lives in the target space, not the input space.)
Since a nonzero vector gets sent to zero, a nonzero vector in the Kernel is ⟨3,−2⟩ (or any nonzero multiple), and the map is not One-to-One.
2
Row-reduce the matrix 1023172−13 to find its Kernel and Image, and determine whether the map is One-to-One and whether it is Onto.
After row-reducing, how does the third column combine the first two columns, and which matrix do you take the pivot columns from for the Image?
Row-reducing the matrix gives:
1023172−13row-reduce1000105−10
Columns 1 and 2 have pivots, and Col 3=5(Col 1)−1(Col 2), so −5(Col 1)+1(Col 2)+1(Col 3)=⟨0,0,0⟩.(Watch out: remember to flip the signs on the numbers multiplying the pivot columns when canceling Column 3, or you'll get ⟨5,−1,1⟩ instead of ⟨−5,1,1⟩; and take Columns 1 and 2 from the original matrix for the Image, not ⟨1,0,0⟩ and ⟨0,1,0⟩ from the reduced matrix.)
Find the Kernel and Image of the 3×2 matrix 12−125−1, and determine whether it is One-to-One and whether it is Onto.
Check the columns and rows separately in the reduced matrix: do both columns have pivots, and do all 3 rows have pivots?
Row-reducing the matrix gives:
12−125−1row-reduce100010
Both columns have pivots, while the third row has no pivot. (Watch out: don't mix up the column check and the row check. Having a pivot in both columns makes the map One-to-One, not Onto, and the Kernel lives in the 2D input space as {⟨0,0⟩}, not {⟨0,0,0⟩}.)
Without row-reducing, consider a linear map given by a 3×5 matrix (mapping R5 to R3). Can this map be One-to-One? Can it be Onto? What is the smallest possible dimension of its Kernel?
Watch out: don't swap rows and columns and treat 3×5 as R3→R5, or assume a well-chosen matrix can always have dim(Kernel)=0. At least 2 input dimensions must get sent to zero when 5 dimensions are mapped into R3.
In dim(Image)+dim(Kernel)=5, what is the largest dimension the Image can have inside a 3D target space?
The map has 5 input dimensions (columns) landing in a 3D target space (rows), so dim(Image)≤3. Since the dimensions of the Image and Kernel add up to 5:
dim(Image)+dim(Kernel)=5⟹dim(Kernel)≥5−3=2
(Watch out: don't swap rows and columns and treat 3×5 as R3→R5, or assume a well-chosen matrix can always have dim(Kernel)=0. At least 2 input dimensions must get sent to zero when 5 dimensions are mapped into R3.)
The map can never be One-to-One, it can be Onto (if all 3 rows have pivots), and the smallest possible dimension of its Kernel is 2.
5
Let T:P2→P1 be the linear map T(a+bx+cx2)=a+(b−c)x. Write the matrix of T using the standard input basis {1,x,x2} and output basis {1,x}, find the Kernel and Image of T, and determine whether T is One-to-One and whether it is Onto.
Plug each input basis polynomial 1,x, and x2 into T to build the three columns of the matrix, then turn the Kernel and Image back into polynomials.
Plugging in the input basis gives T(1)=1→⟨1,0⟩,T(x)=x→⟨0,1⟩, and T(x2)=−x→⟨0,−1⟩:
[10010−1]
Columns 1 and 2 have pivots while Col 3=−1(Col 2), so 1(Col 2)+1(Col 3)=⟨0,0⟩, giving the vector ⟨0,1,1⟩ in the Kernel; both rows have pivots. (Watch out: don't leave the answers as vectors Span{⟨0,1,1⟩} and Span{⟨1,0⟩,⟨0,1⟩}, and don't forget to flip the sign on −1(Col 2), which would give x−x2 instead of x+x2.)