Finding the coordinates of a vector in a basis usually means setting up a system of equations and row-reducing a matrix, and we want to know when we can find each coordinate on its own without that algebra.
1
Why 90∘ angles stop basis vectors from interfering
To write a target vector v=[79] as a combination of two tilted basis vectors b1=[12] and b2=[31], we look for the coordinates c1 and c2 in
c1[12]+c2[31]=[79]
In the standard basis [10] and [01], we can read off the coordinates of [34] right away because the axes sit at 90∘ to each other and each basis vector has length 1.
Why 90-degree angles stop cross-talk
Try tilting b2 below to see when c1b1 matches the perpendicular shadow of v on b1.
Tilting the second basis vector changes how much of the first one you need to reach v, and only at a right angle does the first piece match v's perpendicular shadow.
2
Orthogonal sets and orthogonal bases
An orthogonal set is a group of vectors in which every vector is at a 90∘ angle to every other vector in the set.
Two vectors meet at 90∘ when their dot product is 0. To test whether u1=110,u2=−110, and u3=005 form an orthogonal set in R3, we check the dot product of every pair.
Orthogonal sets and orthogonal bases
u1⋅u2u1⋅u3u2⋅u3=?=?=?
Text solution
Dot the first pair (u1 and u2)Multiply matching entries and add them up to check that u1 and u2 meet at 90∘.
u1⋅u2=(1)(−1)+(1)(1)+(0)(0)=−1+1+0=0
Dot the remaining two pairs (u1,u3 and u2,u3)Check every other pair to make sure those meet at 90∘ too.
Count non-zero orthogonal vectors in R3Because each new vector is at 90∘ to the previous ones, it points into a new dimension and cannot lie in their span, making the set linearly independent.
orthogonal set of 3 non-zero vectors in R3⎩⎨⎧110,−110,005⎭⎬⎫⟹orthogonal basis for R3
So {u1,u2,u3} is an orthogonal basis for R3, which guarantees that any vector v in R3 can be written as a unique combination v=c1u1+c2u2+c3u3.
For three vectors, make sure to check all three pairs (u1⋅u2,u1⋅u3, and u2⋅u3). Checking only adjacent pairs like u1⋅u2 and u2⋅u3 is a common mistake, because u1 and u3 still might not be at 90∘ to each other.
3
Finding coordinates in an orthogonal basis
Given our orthogonal basis u1=110,u2=−110,u3=005 and a target vector v=2610, we want to find the coordinates c1,c2,c3 in
v=c1u1+c2u2+c3u3
without row reduction.
Finding coordinates in an orthogonal basis
x
y
z
u1
u2
u3
v=2610
vc1,c2,c3=c1u1+c2u2+c3u3=?
Taking the dot product of both sides with u1 zeroes out the terms with the other basis vectors because their dot products with u1 are zero:
Add the projected vector piecesMultiply each coordinate by its basis vector to build v from its perpendicular projections along the three axes.
v=4110+2−110+2005=projection on u1440+projection on u2−220+projection on u30010=2610
So the coordinates of v in {u1,u2,u3} are c1=4,c2=2, and c3=2, and each piece ciui is the perpendicular projection of v onto that basis line.
Try dragging v parallel to u2 below to see that changing c2 leaves the projection c1u1 and the coordinate c1 completely alone.
Because the two axes meet at a right angle, your vector splits into two perpendicular pieces along the axes that add up to the vector, and sliding parallel to one axis changes only that axis's coordinate.
4
Normalizing to an orthonormal basis
An orthonormal set is an orthogonal set whose vectors all have length 1 (unit vectors), and an orthonormal basis is a basis that is an orthonormal set.
In ci=ui⋅uiv⋅ui, we had to divide by ui⋅ui only because our basis vectors were not 1 unit long. Since a vector's length is ∥ui∥=ui⋅ui, dividing each vector in {u1,u2,u3} by its own length shrinks or stretches it to length 1 without changing its direction.
Normalizing to an orthonormal basis
x
y
z
u3
v
10
c3=u3⋅u3v⋅u3=2550why 25?
x
y
z
u1
u2
u3
∥u1∥=∥u2∥=2,∥u3∥=5
x
y
z
q1
q2
q3
∥q1∥=∥q2∥=∥q3∥=1
Text solution
Divide u1 by its length ∥u1∥Compute ∥u1∥=12+12+02=2 and divide each entry of u1 by 2.
q1=∥u1∥u1=21110=21210
Divide u2 by its length ∥u2∥Compute ∥u2∥=(−1)2+12+02=2 and divide each entry of u2 by 2.
q2=∥u2∥u2=21−110=−21210
Divide u3 by its length ∥u3∥Compute ∥u3∥=02+02+52=5 and divide each entry of u3 by 5.
q3=∥u3∥u3=51005=001
So {q1,q2,q3} is an orthonormal basis for R3, keeping the exact same 90∘ directions as {u1,u2,u3} with every vector now of length 1.
5
Finding coordinates in an orthonormal basis
Now we want to find the coordinates c1,c2,c3 of the same target vector v=2610 in our orthonormal basis {q1,q2,q3}:
v=c1q1+c2q2+c3q3
Finding coordinates in an orthonormal basis
c1q1⋅q1=q1⋅q1v⋅q1=?
Because every qi is a unit vector of length 1, its dot product with itself is qi⋅qi=12=1, and the denominator in the coordinate formula disappears:
ci=1qi⋅qiv⋅qi=v⋅qi
Every vector v is built directly from these dot products:
v=(v⋅q1)q1+(v⋅q2)q2+(v⋅q3)q3
Text solution
Dot v with q1 to get c1Multiply matching entries of v=2610 and q1=21210.
c1=v⋅q1=(2)(21)+(6)(21)+(10)(0)=28=42
Dot v with q2 to get c2Multiply matching entries of v and q2=−21210.
c2=v⋅q2=(2)(−21)+(6)(21)+(10)(0)=24=22
Dot v with q3 to get c3Multiply matching entries of v and q3=001.
c3=v⋅q3=(2)(0)+(6)(0)+(10)(1)=10
Multiply the new coordinates by the unit vectors to check the piecesThe 2 factors cancel out, giving the exact same projected vectors along each axis as before.
So the coordinates of v in {q1,q2,q3} are 42,22, and 10: dividing the length of each basis vector ui by ∥ui∥ means we need ∥ui∥ times as many copies of qi to get the same projected piece.
Try changing the length of u1 below to see how the coordinate c1 adjusts so the vector piece c1u1 stays the same, and what happens right when ∥u1∥=1.
Changing the basis vector's length changes how many copies it takes to reach the projection, while the projected piece itself never moves.
6
Matrices with orthonormal columns
When we put an orthonormal set of vectors q1,q2,q3 into the columns of a matrix Q, we want to see what happens when we multiply Q on the left by its transpose QT.
Taking the transpose turns the columns of Q into rows, so multiplying rows by columns in QTQ takes the dot product of every pair of vectors.
Matrices with orthonormal columns
Q=
q1
q2
q3
QTQ=?
Down the main diagonal, each unit vector dots with itself (qi⋅qi=1), and off the diagonal, orthogonal vectors dot with each other (qi⋅qj=0):
When Q is a square matrix (meaning its columns form an orthonormal basis), its inverse is simply its transpose:
Q−1=QT
Key concepts
Everything here starts from checking whether vectors sit at 90∘ to each other. A set of vectors is orthogonal when every pair has dot product 0 (ui⋅uj=0 for i=j); for three vectors, make sure to check all three pairs u1⋅u2,u1⋅u3, and u2⋅u3. Because each perpendicular vector points into a brand-new dimension, non-zero orthogonal vectors are automatically linearly independent, so n of them in Rn automatically form an orthogonal basis (such as u1=110,u2=−110,u3=005 in R3).
When finding coordinates in a basis, first check whether the basis is orthogonal, and then whether its vectors have length 1:
Key concept
If the basis vectors are not orthogonal (like b1=[12] and b2=[31]), moving along one changes your position along the other; the dot-product coordinate formulas do not apply, and you must solve the system by row reduction.
Key concept
In an orthogonal basis{u1,…,un} where lengths are not 1, find any coordinate on its own by dividing by ui⋅ui:
ciprojected piece=ui⋅uiv⋅ui,=ciui
For v=2610 along u3=005,c3=2550=2 and the vector piece is 2005=0010. If a problem only asks for coordinates in an orthogonal basis {u1,u2,u3}, use ui⋅uiv⋅ui directly, and don't normalize the basis into square roots first.
Key concept
In an orthonormal basis{q1,…,qn} where each vector has already been normalized to length 1 (qi=∥ui∥ui, so qi⋅qi=1), the denominator disappears:
ci=v⋅qi
For v=2610 along q3=001,c3=v⋅q3=10. Dividing the length of ui by ∥ui∥ multiplies its coordinate by ∥ui∥, while the vector piece ciui=(v⋅qi)qi stays the same.
Key concept
When orthonormal vectors are put into the columns of a matrix Q, the dot products of rows and columns give QTQ=I. If Q is square, Q−1=QT and Qx=b has the immediate solution x=QTb.
7
Practice
1
Let u1=12−1,u2=2−10, and u3=012. Notice that u1⋅u2=0 and u1⋅u3=0. Is {u1,u2,u3} an orthogonal set?
How many pairs do three vectors make in total, and which pair hasn't been tested yet?
Stopping after the two given dot products u1⋅u2=0 and u1⋅u3=0 gives the wrong answer "Yes"; in an orthogonal set, every pair must have a dot product of 0, so we must also check u2⋅u3:
u2⋅u3=(2)(0)+(−1)(1)+(0)(2)=−1=0
No
2
Let u1=120,u2=−215, and u3=2ab. Find the values of a and b that make {u1,u2,u3} an orthogonal set, and explain why {u1,u2,u3} is then guaranteed to be a basis for R3.
Watch the sign when moving 2 across the equals sign: writing a=1 here carries into the second equation and gives the wrong value b=53.
Between u1⋅u3=0 and u2⋅u3=0, which equation contains only one unknown so you can solve it first?
First note that u1⋅u2=(1)(−2)+(2)(1)+(0)(5)=0. Setting u1⋅u3=0 gives
u1⋅u3=(1)(2)+(2)(a)+(0)(b)=2+2a=0⟹a=−1
(Watch the sign when moving 2 across the equals sign: writing a=1 here carries into the second equation and gives the wrong value b=53.) Substituting a=−1 into u2⋅u3=0 gives
u2⋅u3=(−2)(2)+(1)(−1)+(5)(b)=−4−1+5b=0⟹b=1
Because any orthogonal set of non-zero vectors is automatically linearly independent, 3 non-zero orthogonal vectors in R3 automatically form a basis for R3, with a=−1 and b=1.
3
The vectors u1=1−10,u2=111, and u3=11−2 form an orthogonal basis for R3. Find the coordinates c1,c2,c3 of v=4−24 in this basis.
Do not leave off the denominator ui⋅ui as if the basis were orthonormal (which gives 6,6,−6), and divide by ui⋅ui rather than the square-root length ∥ui∥ (which gives 32,23,−6).
Check the lengths of u1,u2,u3: are they unit vectors, or do you need the denominator ui⋅ui for each coordinate?
Use ci=ui⋅uiv⋅ui. Do not leave off the denominator ui⋅ui as if the basis were orthonormal (which gives 6,6,−6), and divide by ui⋅ui rather than the square-root length ∥ui∥ (which gives 32,23,−6):
Suppose {u1,u2,u3} is an orthogonal basis for R3 with u1=2−12, and u2 and u3 are not given. If v=7−21 is written as v=c1u1+c2u2+c3u3, find the coordinate c1 and the individual vector piece c1u1 along u1.
Make sure to divide v⋅u1=18 by u1⋅u1=9 rather than by the length ∥u1∥=3 (which gives the wrong coordinate c1=6 and the wrong vector piece 12−612).
What happens to the c2u2 and c3u3 terms when you dot both sides of v=c1u1+c2u2+c3u3 with u1?
Dotting both sides with u1 zeroes out the u2 and u3 terms. Make sure to divide v⋅u1=18 by u1⋅u1=9 rather than by the length ∥u1∥=3 (which gives the wrong coordinate c1=6 and the wrong vector piece 12−612):
Multiply c1=2 by the basis vector u1 (not by v) to get the projected vector piece:
c1u1=22−12=4−24
5
Turn the orthogonal basis u1=340,u2=−430,u3=002 into an orthonormal basis {q1,q2,q3} for R3, and find the coordinates of v=105−6 in {q1,q2,q3}.
Once you divide each ui by its length ∥ui∥ to get qi, what is the denominator qi⋅qi when finding the coordinates in {q1,q2,q3}?
Divide each vector by its length (∥u1∥=32+42=5,∥u2∥=(−4)2+32=5,∥u3∥=2):
q1=53540,q2=−54530,q3=001
Because {q1,q2,q3} is now orthonormal, each coordinate is simply ci=v⋅qi (do not divide by ∥ui∥ a second time, which gives the old {u1,u2,u3} coordinates 2,−1,−3):
c1=v⋅q1=(10)(53)+(5)(54)+(−6)(0)=10
c2=v⋅q2=(10)(−54)+(5)(53)+(−6)(0)=−5
c3=v⋅q3=(10)(0)+(5)(0)+(−6)(1)=−6
6
A vector v in R3 has coordinates (3,−2,5) in an orthogonal basis {u1,u2,u3} where ∥u1∥=2,∥u2∥=6, and ∥u3∥=3. When {u1,u2,u3} is normalized into the orthonormal basis {q1,q2,q3} with qi=∥ui∥ui, what are the coordinates of v in {q1,q2,q3}?
Because each unit vector qi=∥ui∥ui is ∥ui∥1 times as long as ui, reaching the exact same vector pieces takes ∥ui∥ times as many copies (do not divide the coordinates by the lengths ∥ui∥, which gives (23,−31,35)).
If a basis vector u1 of length 2 is replaced by a unit vector q1 of length 1, do you need twice as many copies or half as many copies to build the same vector piece?
Because each unit vector qi=∥ui∥ui is ∥ui∥1 times as long as ui, reaching the exact same vector pieces takes ∥ui∥ times as many copies (do not divide the coordinates by the lengths ∥ui∥, which gives (23,−31,35)):
c1c2c3=3×2=6,=−2×6=−12,=5×3=15
(6,−12,15)
7
The columns of Q=31313121−2106161−62 form an orthonormal basis for R3. Without row reduction, find QTQ,Q−1, and the solution x to Qx=600.
For a square matrix Q with orthonormal columns, how is Q−1 related to QT, and what happens when you multiply Qx=600 on the left by QT?
Because the columns of Q are orthonormal and Q is square:
To solve Qx=600, multiply on the left by QT (make sure to multiply QT600 rather than Q600, which picks the first column of Q instead of the first column of QT and gives 232323):