When a transformation moves every point on a grid to a new spot all at once, we want a simple way to figure out where every point lands without tracking an infinite number of points one by one.
1
Keeping the Grid Straight
A transformation (or map) takes each point on the grid and moves it to a new spot. In linear algebra, we focus on transformations that keep the grid lines straight, parallel, and evenly spaced, with the center (0,0) pinned in place.
Keeping the Grid Straight
Because the grid never bends or bunches up unevenly, the point (3,2) is reached by the exact same steps before and after the move:
Before: ⟶After: 3 green arrows+2 pink arrows3 new green arrows+2 new pink arrows
2
Where Every Point Lands
Now we want to find the exact coordinates where the vector [32] lands when the green arrow [10] lands at [21] and the pink arrow [01] lands at [−12].
Where Every Point Lands
Text solution
Break the starting vector into arrowsWrite [32] as 3 green arrows plus 2 pink arrows:
[32]=3[10]+2[01]
Replace with the landed arrows and addTake 3 of the new green arrow plus 2 of the new pink arrow to find where [32] lands:
3[21]+2[−12]=[63]+[−24]=[47]
Pack the landed arrows into a matrixPut the new green arrow into the first column and the new pink arrow into the second column to write the move as a matrix multiplying [32]:
landed arrows[21−12][32]=[47]
So the point (3,2) lands at (4,7), and the two columns of the matrix record where the green and pink arrows land.
Try dragging the green and pink arrows around (0,0) to see how (3,2) and the matrix columns follow them, even if you line the two arrows up and squish the whole grid onto a single line.
Where the green and pink arrows land sets the two columns of the matrix and fixes where every point on the grid lands, always at the same combination of those two arrows.
3
The Two Rules of Linearity
When we aren't given a picture of a grid, or when we work with vectors that don't live on a 2D grid, we need a way to tell from a formula whether a transformation T follows these straight-grid rules. A linear transformation is a transformation T where adding vectors and scaling vectors work the exact same way before and after the move.
The Two Rules of Linearity
To be a linear transformation, T must obey two rules for all vectors u,v and any scaling number c:
Because the grid lines stay straight, parallel, and evenly spaced, the parallelogram made by u and v tilts and stretches without coming apart, so T(u+v) still lands at the tip of T(u)+T(v).
Because the grid lines stay straight, parallel, and evenly spaced, the parallelogram tilts and stretches without coming apart, so adding two vectors before transforming always lands on the exact same spot as transforming them first and adding them afterward.
4
Testing Formulas for Linearity
Given a formula for T([xy]), we want to test whether it is a linear transformation. Setting c=0 in the scaling rule T(cv)=cT(v) gives a quick Zero Check to try before anything else:
Test a shift using the Zero Check: T([xy])=[x+5y]Plug in [00] to see if the origin stays pinned:
T([00])=[0+50]=[50]=[00]
So T([xy])=[x+5y] is not linear because it moves the origin to (5,0).
Test a squared coordinate using a counterexample: T([xy])=[x2y]Even though T([00])=[00], stretching v=[10] by c=3 before applying T does not match applying T first and stretching by 3 afterward:
Check adding on two general vectors and write the matrixPlug in u+v=[x1+x2y1+y2], group the x1,y1 terms and the x2,y2 terms into separate vectors, and match the formula to a matrix product:
So T([xy])=[2x−yx+2y] is a linear transformation. In fact, because matrix multiplication obeys A(u+v)=Au+Av and A(cv)=c(Av), you can spot right away that any formula that can be written as Av (with no added constants like +5 or powers like x2) is linear. Still, when you need to prove linearity from the rules, write out the steps with [cxcy] and [x1+x2y1+y2].
Try moving v under T([xy])=[x2y] to see how the uneven horizontal spacing pulls T(2v) away from 2T(v) whenever v has a non-zero x-coordinate, even though (0,0) stays pinned.
Even when the origin stays put, unevenly spaced grid lines mean doubling the input no longer just doubles the output.
5
Linear Transformations on Polynomials
Polynomials like 3x2+2x+5 are vectors too because we can add them together and scale them by numbers, even though there is no 2D grid to look at. We want to find what the derivative transformation D does to 3x2+2x+5 using only where D sends the simple building-block polynomials 1,x, and x2:
D(1)=0,D(x)=1,D(x2)=2x
Remember from calculus that the derivative of a sum is the sum of the derivatives, and constant multiples pull out front:
D(f+g)andD(cf)=D(f)+D(g)=cD(f)
Because D obeys both rules of linearity, breaking an input into building blocks works the exact same way on polynomials.
Linear Transformations on Polynomials
Text solution
Break the polynomial into building blocksWrite 3x2+2x+5 as a combination of the building blocks x2,x, and 1:
3x2+2x+5=3(x2)+2(x)+5(1)
Split addition and pull out the scaling numbersUse Rules 1 and 2 so D acts on each building block separately:
Plug in the transformed building blocksSubstitute D(x2)=2x,D(x)=1, and D(1)=0:
3(2x)+2(1)+5(0)=6x+2
So D sends 3x2+2x+5 to 6x+2, using the same building-block method we used for arrows on a grid.
Key concepts
A linear transformation can be recognized in two ways that mean the exact same thing: visually on a grid, it keeps grid lines straight, parallel, and evenly spaced (even if squished flat onto a line) with (0,0) pinned; algebraically for any vectors anywhere, adding and scaling work the same before and after:
T(u+v)andT(cv)=T(u)+T(v)=cT(v)
When you already know T is linear and want to find where an input lands
Break the input into simple building blocks ([10] and [01] on a 2D grid, or 1,x, and x2 for polynomials), pull the scaling numbers out front, and plug in where the building blocks land:
On a 2D grid, placing the landed green and pink arrows into the columns packs the transformation into a matrix:
T([xy])=[21−12][xy]
When you are given a formula and asked to test if it is linear
Start with the Zero CheckT(0)=0. If the origin moves, as in T([xy])=[x+5y] where T([00])=[50], stop right away; you do not need to test u+v or cv. If (0,0) stays pinned but the formula has exponents like T([xy])=[x2y], do not write out general letters x1,x2. Just pick one specific vector and number (such as v=[10] and c=3, giving T(3v)=[90]=3T(v)=[30]) to show a rule breaks. Save the full algebra with general vectors [cxcy] and [x1+x2y1+y2] for proving that a matrix-style formula like T([xy])=[2x−yx+2y] obeys both rules for every vector.
6
Practice
1
A linear transformation T sends the green basis arrow [10] to [3−1] and the pink basis arrow [01] to [14]. Where does T send the vector [25], and what matrix represents T?
[25] is built from how many green arrows and how many pink arrows? When you build the matrix, do the landed arrows go in the rows or the columns?
Write [25]=2[10]+5[01] and replace each basis arrow with where it lands:
T([25])=2[3−1]+5[14]=[6−2]+[520]=[1118]
(Watch the negative sign in 2(−1)+5(4)=−2+20=18; dropping the minus sign gives [1122].)
To write the matrix, put the landed arrows into the columns. Putting them into the rows is a common slip that gives the transposed matrix [31−14] (which would send [25] to [122]):
[3−114]
2
Test whether each transformation is linear: (a) T([xy])=[xy−3] and (b) T([xy])=[xy3]. For each one that is not linear, show a specific rule it breaks.
Try the Zero Check on [00] first; if [00] stays pinned, test the scaling rule T(cv)=cT(v) on a simple vector whose y-coordinate isn't zero.
Neither transformation is linear.
For (a), don't be fooled by the fact that x and y have no exponents. Check where the origin lands:
T([00])=[00−3]=[0−3]=[00]
It fails the Zero Check.
For (b), T([00])=[00] passes the Zero Check, so don't stop there. Test the scaling rule T(cv)=cT(v) using v=[01] and c=2(avoid picking v=[10], because y=0 hides the y3 term and makes both sides [20]):
Use the two rules of linearity, T(cv)=cT(v) and T(u+v)=T(u)+T(v), to prove that T([xy])=[3x+y−x+4y] is a linear transformation, and write T([xy]) as a matrix multiplying [xy].
Plug a general scaled vector [cxcy] and a general sum [x1+x2y1+y2] into the formula, then factor or regroup to match cT(v) and T(u)+T(v).
For scaling, plug in cv=[cxcy] and factor c out of both rows:
For adding u=[x1y1] and v=[x2y2], plug in u+v=[x1+x2y1+y2](be careful to distribute the minus sign to both x1 and x2 in −(x1+x2)=−x1−x2, or the second vector won't match T(v)):
Both rules hold, and in matrix form (keeping the −1 coefficient on x in the bottom-left entry):
T([xy])=[3−114][xy]
4
A linear transformation T on polynomials sends the building blocks 1,x, and x2 to T(1)=2,T(x)=x+1, and T(x2)=x2−3x. What does T do to the polynomial 4x2−2x+5?
Write the constant term 5 as 5(1) so you can pull the numbers out front: how does T act on x2,x, and 1?
Split up the addition and pull out the scaling numbers:
T(4x2−2x+5)=4T(x2)−2T(x)+5T(1)
Substitute the given outputs (watch out for two common slips: distribute −2 to both terms in T(x)=x+1, and multiply 5 by T(1)=2 rather than treating T like a derivative where T(5)=0, which would give 4x2−14x−2, or leaving 5 unchanged, which would give 4x2−14x+3):
4(x2−3x)−2(x+1)+5(2)=4x2−12x−2x−2+10=4x2−14x+8
5
A linear transformation T sends [10] to [23] and [11] to [1−1]. Where does T send the pink basis arrow [01], and where does it send [3−2]?
Don't use [1−1] directly as T([01]), and don't subtract in reverse order [23]−[1−1]=[14].
Write [11] as a sum of [10] and [01], then apply Rule 1 to solve for T([01]) first.
Because [11]=[10]+[01], Rule 1 gives T([11])=T([10])+T([01]). Subtract T([10]) from T([11])(don't use [1−1] directly as T([01]), and don't subtract in reverse order [23]−[1−1]=[14]):
T([01])=[1−1]−[23]=[−1−4]
Now break [3−2] into 3[10]−2[01](watch the double negatives when multiplying −2 by [−1−4]; subtracting instead of adding gives [41]):