Linear-Algebra-Done-Right · Unit 18 · Video 1 · Interactive Practice

The Adjoint T∗T^*: Moving TT into the Second Slot of ⟨Tv,w⟩\langle Tv, w \rangle

IKey Formulas

FormulaNameWhat it says
⟨Tv,w⟩=⟨v,T∗w⟩\langle Tv, w \rangle = \langle v, T^*w \rangle for all v∈Vv \in V, w∈Ww \in WDefinition of T∗:W→VT^*: W \to VFor T∈L(V,W)T \in \mathcal{L}(V, W) with VV, WW finite-dimensional inner product spaces over F\mathbf{F}: T∗wT^*w is the Riesz vector of v↦⟨Tv,w⟩v \mapsto \langle Tv, w \rangle, so it exists and is unique
T∗(w1+w2)=T∗w1+T∗w2T^*(w_1 + w_2) = T^*w_1 + T^*w_2,   T∗(λw)=λ T∗w\;T^*(\lambda w) = \lambda\, T^*wT∗∈L(W,V)T^* \in \mathcal{L}(W, V)The two conjugations in the homogeneity proof cancel
(S+T)∗=S∗+T∗(S + T)^* = S^* + T^*,   (λT)∗=λ‾ T∗\;(\lambda T)^* = \overline{\lambda}\, T^*,   (T∗)∗=T\;(T^*)^* = TRules (a)–(c)λ\lambda enters through the first slot, so it is conjugated once
(ST)∗=T∗S∗(ST)^* = T^*S^*,   I∗=I\;I^* = I,   (T∗)−1=(T−1)∗\;(T^*)^{-1} = (T^{-1})^*Rules (d)–(f)Products reverse order; S∈L(W,U)S \in \mathcal{L}(W, U) in (d); in (f), TT is invertible, and then so is T∗T^*

Key Insight: To compute T∗wT^*w, start from ⟨Tv,w⟩\langle Tv, w \rangle, rearrange until vv stands alone in the first slot, and read off whatever sits in the second slot.

IIReading Off T∗wT^*w from ⟨Tv,w⟩\langle Tv, w \rangle

For T(x,y)=(x−y,2x,y)T(x, y) = (x - y, 2x, y), the vector T∗wT^*w reproduces ⟨Tv,w⟩\langle Tv, w \rangle as ⟨v,T∗w⟩\langle v, T^*w \rangle for every vv.

IIIA Scalar Through the First Slot Comes Out Conjugated

λ\lambda leaves the first slot unchanged but enters the second slot conjugated, so (λT)∗=λ‾ T∗(\lambda T)^* = \overline{\lambda}\, T^*, not λT∗\lambda T^*.

IVWhy (ST)∗=T∗S∗(ST)^* = T^*S^*: The Adjoints Run Back in Reverse

With T(x,y)=(x−y,2x,y)T(x, y) = (x - y, 2x, y) as above and S(a,b,c)=a+b−cS(a, b, c) = a + b - c from R3\mathbf{R}^3 to R\mathbf{R}, moving STST across the inner product strips off SS first, then TT.

VQuiz Questions

Problem 1 · Compute an Adjoint by Regrouping

Given: T∈L(R2,R3)T \in \mathcal{L}(\mathbf{R}^2, \mathbf{R}^3), T(x,y)=(x+2y, 3y, −x)T(x, y) = (x + 2y,\ 3y,\ -x), with the Euclidean inner products. Find T∗(w1,w2,w3)T^*(w_1, w_2, w_3).

✅ Correct! ⟨T(x,y),w⟩=x(w1−w3)+y(2w1+3w2)=⟨(x,y),(w1−w3,2w1+3w2)⟩\langle T(x, y), w \rangle = x(w_1 - w_3) + y(2w_1 + 3w_2) = \langle (x, y), (w_1 - w_3, 2w_1 + 3w_2) \rangle.
❌ Not quite. These are the right coefficients in the wrong slots: 2w1+3w22w_1 + 3w_2 multiplies yy, so it is the second coordinate of T∗wT^*w, and w1−w3w_1 - w_3 multiplies xx, so it is the first.
❌ Not quite. The third coordinate of T(x,y)T(x, y) is −x-x, so it contributes −x w3-x\,w_3; the coefficient of xx is w1−w3w_1 - w_3, not w1+w3w_1 + w_3.
❌ Not quite. TT goes from R2\mathbf{R}^2 to R3\mathbf{R}^3, so T∗T^* comes back from R3\mathbf{R}^3 to R2\mathbf{R}^2: T∗wT^*w must have exactly two coordinates, one for each of xx and yy.
Show solution

Start from ⟨Tv,w⟩\langle Tv, w \rangle with v=(x,y)v = (x, y) and w=(w1,w2,w3)w = (w_1, w_2, w_3):

⟨T(x,y),w⟩=(x+2y)w1+3y w2−x w3.\langle T(x, y), w \rangle = (x + 2y)w_1 + 3y\,w_2 - x\,w_3.

Regroup around xx and yy so that (x,y)(x, y) stands alone in the first slot:

=x(w1−w3)+y(2w1+3w2)=⟨(x,y), (w1−w3, 2w1+3w2)⟩.= x(w_1 - w_3) + y(2w_1 + 3w_2) = \big\langle (x, y),\ (w_1 - w_3,\ 2w_1 + 3w_2) \big\rangle.

The second slot is T∗wT^*w, so T∗(w1,w2,w3)=(w1−w3, 2w1+3w2)T^*(w_1, w_2, w_3) = (w_1 - w_3,\ 2w_1 + 3w_2).

Check: T(1,1)=(3,3,−1)T(1, 1) = (3, 3, -1) and ⟨(3,3,−1),(1,1,1)⟩=5\langle (3, 3, -1), (1, 1, 1) \rangle = 5; T∗(1,1,1)=(0,5)T^*(1, 1, 1) = (0, 5) and ⟨(1,1),(0,5)⟩=5\langle (1, 1), (0, 5) \rangle = 5.

Problem 2 · Conjugates in the Second Slot

Given: T∈L(C2,C3)T \in \mathcal{L}(\mathbf{C}^2, \mathbf{C}^3), T(z1,z2)=(iz1, z1+z2, 2z2)T(z_1, z_2) = (i z_1,\ z_1 + z_2,\ 2z_2), with the Euclidean inner products ⟨a,b⟩=∑kakbk‾\langle a, b \rangle = \sum_k a_k \overline{b_k}. Find T∗(w1,w2,w3)T^*(w_1, w_2, w_3).

✅ Correct! The coefficient of z1z_1 is iw1‾+w2‾i\overline{w_1} + \overline{w_2}, and moving it into the second slot conjugates it to −iw1+w2-i w_1 + w_2.
❌ Not quite. You conjugated the constant ii but left the wk‾\overline{w_k} alone; the whole coefficient iw1‾+w2‾i\overline{w_1} + \overline{w_2} must be conjugated, which turns each wk‾\overline{w_k} back into wkw_k. As written, this map is not even linear in ww.
❌ Not quite. You undid the bars on the wkw_k but forgot that the constant ii is conjugated too: iw1‾‾=−iw1\overline{i\overline{w_1}} = -i w_1. With +i+i, ⟨z,T∗w⟩\langle z, T^*w \rangle would contain −iz1w1‾-i z_1 \overline{w_1} instead of iz1w1‾i z_1 \overline{w_1}.
❌ Not quite. These are the coefficients of z1z_1 and z2z_2 read off directly, but the second slot conjugates whatever sits in it: ⟨z,u⟩=z1u1‾+z2u2‾\langle z, u \rangle = z_1 \overline{u_1} + z_2 \overline{u_2}, so u1u_1 must be the conjugate of iw1‾+w2‾i\overline{w_1} + \overline{w_2}.
Show solution

Start from ⟨Tz,w⟩\langle Tz, w \rangle with z=(z1,z2)z = (z_1, z_2):

⟨T(z1,z2),w⟩=iz1w1‾+(z1+z2)w2‾+2z2w3‾=z1(iw1‾+w2‾)+z2(w2‾+2w3‾).\langle T(z_1, z_2), w \rangle = i z_1 \overline{w_1} + (z_1 + z_2)\overline{w_2} + 2z_2 \overline{w_3} = z_1\big(i\overline{w_1} + \overline{w_2}\big) + z_2\big(\overline{w_2} + 2\overline{w_3}\big).

To write z1c1z_1 c_1 as z1u1‾z_1 \overline{u_1} we need u1=c1‾u_1 = \overline{c_1}; a coefficient entering the second slot is conjugated:

iw1‾+w2‾‾=−iw1+w2,w2‾+2w3‾‾=w2+2w3.\overline{i\overline{w_1} + \overline{w_2}} = -i w_1 + w_2, \qquad \overline{\overline{w_2} + 2\overline{w_3}} = w_2 + 2w_3.

So ⟨Tz,w⟩=⟨(z1,z2), (−iw1+w2, w2+2w3)⟩\langle Tz, w \rangle = \big\langle (z_1, z_2),\ (-i w_1 + w_2,\ w_2 + 2w_3) \big\rangle and T∗(w1,w2,w3)=(−iw1+w2, w2+2w3)T^*(w_1, w_2, w_3) = (-i w_1 + w_2,\ w_2 + 2w_3), which is linear in ww, as the theorem promises.

Check: z=(1,0)z = (1, 0), w=(1,0,0)w = (1, 0, 0): ⟨Tz,w⟩=⟨(i,1,0),(1,0,0)⟩=i\langle Tz, w \rangle = \langle (i, 1, 0), (1, 0, 0) \rangle = i, and ⟨(1,0),T∗w⟩=⟨(1,0),(−i,0)⟩=−i‾=i\langle (1, 0), T^*w \rangle = \langle (1, 0), (-i, 0) \rangle = \overline{-i} = i.

Problem 3 · Combining the Rules

Given: T∈L(V,W)T \in \mathcal{L}(V, W) and S∈L(W,U)S \in \mathcal{L}(W, U), where VV, WW, UU are finite-dimensional inner product spaces over C\mathbf{C}. Simplify each adjoint.

What is ((2i) ST)∗\big((2i)\,ST\big)^*?

What is (((3−i) T)∗)∗\big(((3 - i)\,T)^*\big)^*?

✅ Correct! Rule (b) conjugates the scalar and rule (d) reverses the product; applying the adjoint twice undoes everything, scalar included.
❌ Not quite. The order T∗S∗T^*S^* is right, but by rule (b) the scalar comes out conjugated: 2i‾=−2i\overline{2i} = -2i.
❌ Not quite. The scalar −2i-2i is right, but (ST)∗=T∗S∗(ST)^* = T^*S^*: (ST)∗(ST)^* maps UU back to VV, so S∗:U→WS^*: U \to W must act first and T∗:W→VT^*: W \to V second. S∗T∗S^*T^* would apply T∗T^* to a vector of UU.
❌ Not quite. Two errors: the scalar must be conjugated, 2i‾=−2i\overline{2i} = -2i, and the product must reverse, (ST)∗=T∗S∗(ST)^* = T^*S^*.
❌ Not quite. You conjugated only once. The first adjoint gives (3+i) T∗(3 + i)\,T^*, and the second adjoint conjugates the scalar again: 3+i‾=3−i\overline{3 + i} = 3 - i.
❌ Not quite. (3+i) T∗(3 + i)\,T^* is ((3−i) T)∗((3 - i)\,T)^*, the result after ONE adjoint; the question takes the adjoint twice, and (T∗)∗=T(T^*)^* = T.
❌ Not quite. The scalar is right after two conjugations, but the map is not: (T∗)∗=T(T^*)^* = T, and T∗T^* goes from WW to VV, the wrong direction for a map built from T∈L(V,W)T \in \mathcal{L}(V, W).
Show solution

First part. By rule (b) with λ=2i\lambda = 2i, then rule (d):

((2i) ST)∗=2i‾ (ST)∗=−2i (ST)∗=−2i T∗S∗.\big((2i)\,ST\big)^* = \overline{2i}\,(ST)^* = -2i\,(ST)^* = -2i\,T^*S^*.

Second part. Apply rule (b) twice and rule (c):

((3−i) T)∗=3−i‾ T∗=(3+i) T∗,((3+i) T∗)∗=3+i‾ (T∗)∗=(3−i) T.\big((3 - i)\,T\big)^* = \overline{3 - i}\,T^* = (3 + i)\,T^*, \qquad \big((3 + i)\,T^*\big)^* = \overline{3 + i}\,(T^*)^* = (3 - i)\,T.

This is rule (c) applied to the single map (3−i) T(3 - i)\,T: the adjoint of an adjoint returns the original map.

Problem 4 · The Adjoint of an Inverse

Given: With the Euclidean inner products, T(x,y)=(x+2y, 3y, −x)T(x, y) = (x + 2y,\ 3y,\ -x) from R2\mathbf{R}^2 to R3\mathbf{R}^3 and S(a,b,c)=(a, b+c)S(a, b, c) = (a,\ b + c) from R3\mathbf{R}^3 to R2\mathbf{R}^2, so ST(x,y)=(x+2y, 3y−x)ST(x, y) = (x + 2y,\ 3y - x) is invertible on R2\mathbf{R}^2. Find ((ST)−1)∗(5,0)\big((ST)^{-1}\big)^*(5, 0).

✅ Correct! (ST)∗(u1,u2)=(u1−u2, 2u1+3u2)(ST)^*(u_1, u_2) = (u_1 - u_2,\ 2u_1 + 3u_2), and by rule (f) the answer solves (ST)∗u=(5,0)(ST)^*u = (5, 0), giving u=(3,−2)u = (3, -2).
❌ Not quite. (3,1)=(ST)−1(5,0)(3, 1) = (ST)^{-1}(5, 0): you inverted STST but never took the adjoint. STST is not its own adjoint here, since (ST)∗(u1,u2)=(u1−u2, 2u1+3u2)≠ST(u1,u2)(ST)^*(u_1, u_2) = (u_1 - u_2,\ 2u_1 + 3u_2) \neq ST(u_1, u_2).
❌ Not quite. (5,10)=(ST)∗(5,0)(5, 10) = (ST)^*(5, 0): you took the adjoint but not the inverse. By rule (f), ((ST)−1)∗=((ST)∗)−1\big((ST)^{-1}\big)^* = \big((ST)^*\big)^{-1}, so solve (ST)∗u=(5,0)(ST)^*u = (5, 0) instead.
❌ Not quite. Only the operator ST∈L(R2)ST \in \mathcal{L}(\mathbf{R}^2) needs to be invertible, and it is; (ST)−1(ST)^{-1} is never split into T−1S−1T^{-1}S^{-1}, and the adjoint of any linear map exists.
Show solution

Step 1: (ST)∗(ST)^* by regrouping. For v=(x,y)v = (x, y) and u=(u1,u2)u = (u_1, u_2):

⟨ST(x,y),u⟩=(x+2y)u1+(3y−x)u2=x(u1−u2)+y(2u1+3u2),\langle ST(x, y), u \rangle = (x + 2y)u_1 + (3y - x)u_2 = x(u_1 - u_2) + y(2u_1 + 3u_2),

so (ST)∗(u1,u2)=(u1−u2, 2u1+3u2)(ST)^*(u_1, u_2) = (u_1 - u_2,\ 2u_1 + 3u_2). Rule (d) agrees: S∗(u1,u2)=(u1,u2,u2)S^*(u_1, u_2) = (u_1, u_2, u_2) and T∗(w1,w2,w3)=(w1−w3, 2w1+3w2)T^*(w_1, w_2, w_3) = (w_1 - w_3,\ 2w_1 + 3w_2), so T∗S∗(u1,u2)=(u1−u2, 2u1+3u2)T^*S^*(u_1, u_2) = (u_1 - u_2,\ 2u_1 + 3u_2).

Step 2: rule (f). ((ST)−1)∗=((ST)∗)−1\big((ST)^{-1}\big)^* = \big((ST)^*\big)^{-1}, so the answer is the uu with (ST)∗u=(5,0)(ST)^*u = (5, 0):

u1−u2=5,2u1+3u2=0⟹u2=−2, u1=3.u_1 - u_2 = 5, \qquad 2u_1 + 3u_2 = 0 \quad\Longrightarrow\quad u_2 = -2,\ u_1 = 3.

Check against the definition: (ST)−1(1,0)=(35,15)(ST)^{-1}(1, 0) = \big(\tfrac{3}{5}, \tfrac{1}{5}\big), and ⟨(35,15),(5,0)⟩=3=⟨(1,0),(3,−2)⟩\big\langle (\tfrac{3}{5}, \tfrac{1}{5}), (5, 0) \big\rangle = 3 = \langle (1, 0), (3, -2) \rangle.

Solved: 0 / 4