Intuition The one-line ideaThe transpose flips a matrix across its main diagonal : rows become columns and columns become rows. Nothing is added or destroyed — you are just relabelling which index comes first. Everything else (the "reverse-order" rule, symmetry, dot products) falls out of this single flip.
For a matrix A A A m × n m \times n m × n A i j A_{ij} A ij i i i j j j transpose A ⊤ A^\top A ⊤ n × m n \times m n × m ( A ⊤ ) i j = A j i . (A^\top)_{ij} = A_{ji}. ( A ⊤ ) ij = A j i . i i i j j j A ⊤ A^\top A ⊤ j j j i i i A A A
WHY this definition? The main diagonal (entries A i i A_{ii} A ii ( j , i ) (j,i) ( j , i ) ( i , j ) (i,j) ( i , j ) m × n m\times n m × n n × m n\times m n × m
Worked example Basic flipA = [ 1 2 3 4 5 6 ] ( 2 × 3 ) ⟹ A ⊤ = [ 1 4 2 5 3 6 ] ( 3 × 2 ) . A = \begin{bmatrix} 1 & 2 & 3 \\ 4 & 5 & 6 \end{bmatrix} \quad (2\times 3) \implies A^\top = \begin{bmatrix} 1 & 4 \\ 2 & 5 \\ 3 & 6 \end{bmatrix} \quad (3\times 2). A = [ 1 4 2 5 3 6 ] ( 2 × 3 ) ⟹ A ⊤ = 1 2 3 4 5 6 ( 3 × 2 ) . Why this step? Row 1 of A A A ( 1 , 2 , 3 ) (1,2,3) ( 1 , 2 , 3 ) column 1 of A ⊤ A^\top A ⊤ ( A ⊤ ) 21 = A 12 = 2 (A^\top)_{21} = A_{12} = 2 ( A ⊤ ) 21 = A 12 = 2
Everything below is proved from the single rule ( A ⊤ ) i j = A j i (A^\top)_{ij} = A_{ji} ( A ⊤ ) ij = A j i
( ( A ⊤ ) ⊤ ) i j = ( A ⊤ ) j i = A i j ⟹ ( A ⊤ ) ⊤ = A . \big((A^\top)^\top\big)_{ij} = (A^\top)_{ji} = A_{ij} \implies (A^\top)^\top = A. ( ( A ⊤ ) ⊤ ) ij = ( A ⊤ ) j i = A ij ⟹ ( A ⊤ ) ⊤ = A . Why? Flipping twice returns every entry to its home. So ⊤ \top ⊤ involution .
( ( A + B ) ⊤ ) i j = ( A + B ) j i = A j i + B j i = ( A ⊤ ) i j + ( B ⊤ ) i j . ((A+B)^\top)_{ij} = (A+B)_{ji} = A_{ji}+B_{ji} = (A^\top)_{ij}+(B^\top)_{ij}. (( A + B ) ⊤ ) ij = ( A + B ) j i = A j i + B j i = ( A ⊤ ) ij + ( B ⊤ ) ij . Why? Transpose only moves entries around; addition is done entry-by-entry, so the two operations commute.
Start from the matrix-product definition ( A B ) i j = ∑ k A i k B k j (AB)_{ij} = \sum_k A_{ik}B_{kj} ( A B ) ij = ∑ k A ik B k j ( ( A B ) ⊤ ) i j = ( A B ) j i = ∑ k A j k B k i . \big((AB)^\top\big)_{ij} = (AB)_{ji} = \sum_k A_{jk}B_{ki}. ( ( A B ) ⊤ ) ij = ( A B ) j i = ∑ k A j k B k i . A j k = ( A ⊤ ) k j A_{jk}=(A^\top)_{kj} A j k = ( A ⊤ ) k j B k i = ( B ⊤ ) i k B_{ki}=(B^\top)_{ik} B k i = ( B ⊤ ) ik = ∑ k ( B ⊤ ) i k ( A ⊤ ) k j = ( B ⊤ A ⊤ ) i j . = \sum_k (B^\top)_{ik}(A^\top)_{kj} = (B^\top A^\top)_{ij}. = ∑ k ( B ⊤ ) ik ( A ⊤ ) k j = ( B ⊤ A ⊤ ) ij . ( A B ) ⊤ = B ⊤ A ⊤ (AB)^\top = B^\top A^\top ( A B ) ⊤ = B ⊤ A ⊤
reverses
Think of dimensions: if A A A m × n m\times n m × n B B B n × p n\times p n × p A B AB A B m × p m\times p m × p ( A B ) ⊤ (AB)^\top ( A B ) ⊤ p × m p\times m p × m p × m p\times m p × m B ⊤ ( p × n ) B^\top (p\times n) B ⊤ ( p × n ) A ⊤ ( n × m ) A^\top (n\times m) A ⊤ ( n × m ) forced by shape-matching — it can't be A ⊤ B ⊤ A^\top B^\top A ⊤ B ⊤
Transpose both sides of A A − 1 = I A A^{-1} = I A A − 1 = I I ⊤ = I I^\top = I I ⊤ = I ( A − 1 ) ⊤ A ⊤ = I ⟹ ( A − 1 ) ⊤ = ( A ⊤ ) − 1 . (A^{-1})^\top A^\top = I \implies (A^{-1})^\top = (A^\top)^{-1}. ( A − 1 ) ⊤ A ⊤ = I ⟹ ( A − 1 ) ⊤ = ( A ⊤ ) − 1 . A − ⊤ A^{-\top} A − ⊤
The Leibniz formula sums over permutations σ \sigma σ ∏ i A i , σ ( i ) \prod_i A_{i,\sigma(i)} ∏ i A i , σ ( i ) ∏ i A σ ( i ) , i \prod_i A_{\sigma(i),i} ∏ i A σ ( i ) , i σ − 1 \sigma^{-1} σ − 1
Definition Symmetric / skew-symmetricA A A symmetric if A ⊤ = A A^\top = A A ⊤ = A A i j = A j i A_{ij}=A_{ji} A ij = A j i A A A skew-symmetric (antisymmetric) if A ⊤ = − A A^\top = -A A ⊤ = − A A i j = − A j i A_{ij}=-A_{ji} A ij = − A j i A i i = 0 A_{ii}=0 A ii = 0
A ⊤ A A^\top A A ⊤ A ( A ⊤ A ) ⊤ = A ⊤ ( A ⊤ ) ⊤ = A ⊤ A (A^\top A)^\top = A^\top (A^\top)^\top = A^\top A ( A ⊤ A ) ⊤ = A ⊤ ( A ⊤ ) ⊤ = A ⊤ A Gram matrices , covariance matrices , and the normal equations A ⊤ A x = A ⊤ b A^\top A x = A^\top b A ⊤ A x = A ⊤ b
Also: for a vector x x x x ⊤ x = ∑ i x i 2 = ∥ x ∥ 2 x^\top x = \sum_i x_i^2 = \|x\|^2 x ⊤ x = ∑ i x i 2 = ∥ x ∥ 2
Worked example Verify the reverse-order ruleA = [ 1 2 0 1 ] , B = [ 1 0 3 1 ] A=\begin{bmatrix}1&2\\0&1\end{bmatrix},\; B=\begin{bmatrix}1&0\\3&1\end{bmatrix} A = [ 1 0 2 1 ] , B = [ 1 3 0 1 ] A B = [ 7 2 3 1 ] ⇒ ( A B ) ⊤ = [ 7 3 2 1 ] AB = \begin{bmatrix}7&2\\3&1\end{bmatrix}\Rightarrow (AB)^\top=\begin{bmatrix}7&3\\2&1\end{bmatrix} A B = [ 7 3 2 1 ] ⇒ ( A B ) ⊤ = [ 7 2 3 1 ] B ⊤ A ⊤ = [ 1 3 0 1 ] [ 1 0 2 1 ] = [ 7 3 2 1 ] B^\top A^\top = \begin{bmatrix}1&3\\0&1\end{bmatrix}\begin{bmatrix}1&0\\2&1\end{bmatrix} = \begin{bmatrix}7&3\\2&1\end{bmatrix} B ⊤ A ⊤ = [ 1 0 3 1 ] [ 1 2 0 1 ] = [ 7 2 3 1 ] Why this step? We compute both sides independently and match — confirming the derivation, not assuming it.
Worked example Split a matrix into symmetric + skewA = [ 2 5 1 4 ] A=\begin{bmatrix}2&5\\1&4\end{bmatrix} A = [ 2 1 5 4 ] A ⊤ = [ 2 1 5 4 ] A^\top=\begin{bmatrix}2&1\\5&4\end{bmatrix} A ⊤ = [ 2 5 1 4 ] S = 1 2 ( A + A ⊤ ) = [ 2 3 3 4 ] S=\tfrac12(A+A^\top)=\begin{bmatrix}2&3\\3&4\end{bmatrix} S = 2 1 ( A + A ⊤ ) = [ 2 3 3 4 ] K = 1 2 ( A − A ⊤ ) = [ 0 2 − 2 0 ] K=\tfrac12(A-A^\top)=\begin{bmatrix}0&2\\-2&0\end{bmatrix} K = 2 1 ( A − A ⊤ ) = [ 0 − 2 2 0 ] S + K = [ 2 5 1 4 ] = A S+K = \begin{bmatrix}2&5\\1&4\end{bmatrix}=A S + K = [ 2 1 5 4 ] = A Why this step? Averaging picks the part symmetric under the flip; half the difference picks the anti-part.
( A B ) ⊤ = A ⊤ B ⊤ (AB)^\top = A^\top B^\top ( A B ) ⊤ = A ⊤ B ⊤ Why it feels right: transpose is linear, so students expect it to "pass through" a product term by term, like it does for a sum.
The fix: a product is not entrywise — it mixes indices via ∑ k A i k B k j \sum_k A_{ik}B_{kj} ∑ k A ik B k j forces the reversal: ( A B ) ⊤ = B ⊤ A ⊤ (AB)^\top = B^\top A^\top ( A B ) ⊤ = B ⊤ A ⊤ A ⊤ B ⊤ A^\top B^\top A ⊤ B ⊤
Common mistake "Transpose changes the determinant / eigenvalues."Why it feels right: A ⊤ A^\top A ⊤ The fix: det ( A ⊤ ) = det ( A ) \det(A^\top)=\det(A) det ( A ⊤ ) = det ( A ) det ( A − λ I ) \det(A-\lambda I) det ( A − λ I ) A A A A ⊤ A^\top A ⊤ same eigenvalues (eigenvectors can differ).
Common mistake "A skew-symmetric matrix can have nonzero diagonal."Why it feels right: you focus only on off-diagonal sign flips.
The fix: A i i = − A i i ⇒ 2 A i i = 0 ⇒ A i i = 0 A_{ii} = -A_{ii} \Rightarrow 2A_{ii}=0 \Rightarrow A_{ii}=0 A ii = − A ii ⇒ 2 A ii = 0 ⇒ A ii = 0 always zero.
Recall Feynman: explain to a 12-year-oldImagine a class seating chart drawn as a grid. Each number says "this kid in row i i i j j j transpose is what you get if you turn the grid on its side so rows become columns — like tilting a photo 90° along its diagonal mirror. The kid who was in "row 3, seat 1" is now in "row 1, seat 3." You didn't add or remove any kids, you just changed how you read the grid. And a cool rule: if you first do task A A A B B B B B B A A A
"Flip about the diagonal" — diagonal entries never move.Reverse rule: "Socks-and-shoes" — ( A B ) ⊤ = B ⊤ A ⊤ (AB)^\top = B^\top A^\top ( A B ) ⊤ = B ⊤ A ⊤ Symmetric factory: A ⊤ A A^\top A A ⊤ A always symmetric ("A-transpose-A is A-OK / symmetric").
What is the defining entry rule for the transpose? ( A ⊤ ) i j = A j i (A^\top)_{ij} = A_{ji} ( A ⊤ ) ij = A j i — swap the two indices.
How does transpose change the shape of an m × n m\times n m × n It becomes
n × m n\times m n × m .
State the reverse-order (socks-and-shoes) rule. ( A B ) ⊤ = B ⊤ A ⊤ (AB)^\top = B^\top A^\top ( A B ) ⊤ = B ⊤ A ⊤ .
Derive why ( A B ) ⊤ = B ⊤ A ⊤ (AB)^\top=B^\top A^\top ( A B ) ⊤ = B ⊤ A ⊤ A ⊤ B ⊤ A^\top B^\top A ⊤ B ⊤ ( A B ) i j ⊤ = ∑ k A j k B k i = ∑ k ( B ⊤ ) i k ( A ⊤ ) k j (AB)^\top_{ij}=\sum_k A_{jk}B_{ki}=\sum_k (B^\top)_{ik}(A^\top)_{kj} ( A B ) ij ⊤ = ∑ k A j k B k i = ∑ k ( B ⊤ ) ik ( A ⊤ ) k j ; also shapes force the reversal.
What is ( A ⊤ ) ⊤ (A^\top)^\top ( A ⊤ ) ⊤ A A A — transpose is an involution (self-inverse).
When is A A A Symmetric:
A ⊤ = A A^\top=A A ⊤ = A . Skew:
A ⊤ = − A A^\top=-A A ⊤ = − A (zero diagonal).
Why is A ⊤ A A^\top A A ⊤ A ( A ⊤ A ) ⊤ = A ⊤ ( A ⊤ ) ⊤ = A ⊤ A (A^\top A)^\top = A^\top(A^\top)^\top = A^\top A ( A ⊤ A ) ⊤ = A ⊤ ( A ⊤ ) ⊤ = A ⊤ A .
How do you split any square A A A A = 1 2 ( A + A ⊤ ) + 1 2 ( A − A ⊤ ) A=\tfrac12(A+A^\top)+\tfrac12(A-A^\top) A = 2 1 ( A + A ⊤ ) + 2 1 ( A − A ⊤ ) .
Relationship between det ( A ) \det(A) det ( A ) det ( A ⊤ ) \det(A^\top) det ( A ⊤ ) Equal; hence same characteristic polynomial and eigenvalues.
What is ( A − 1 ) ⊤ (A^{-1})^\top ( A − 1 ) ⊤ ( A ⊤ ) − 1 (A^\top)^{-1} ( A ⊤ ) − 1 (proved by transposing
A A − 1 = I AA^{-1}=I A A − 1 = I ).
Inner-product / adjoint property of transpose? ⟨ A x , y ⟩ = ⟨ x , A ⊤ y ⟩ \langle Ax,y\rangle=\langle x,A^\top y\rangle ⟨ A x , y ⟩ = ⟨ x , A ⊤ y ⟩ .
Why must a skew-symmetric matrix have zero diagonal? A i i = − A i i ⇒ A i i = 0 A_{ii}=-A_{ii}\Rightarrow A_{ii}=0 A ii = − A ii ⇒ A ii = 0 .
Reverse-order: (AB)^T = B^T A^T
Inverse: (A^-1)^T = (A^T)^-1
Symmetry and dot products
Intuition Hinglish mein samjhoDekho, transpose ka funda bilkul simple hai: matrix ko uski main diagonal ke around palat do. Jo rows the woh columns ban jaate hain, aur jo columns the woh rows. Rule sirf ek hai — ( A ⊤ ) i j = A j i (A^\top)_{ij} = A_{ji} ( A ⊤ ) ij = A j i i i i j j j m × n m\times n m × n n × m n\times m n × m
Sabse important property jise students galat karte hain: ( A B ) ⊤ = B ⊤ A ⊤ (AB)^\top = B^\top A^\top ( A B ) ⊤ = B ⊤ A ⊤ A ⊤ B ⊤ A^\top B^\top A ⊤ B ⊤ A A A m × n m\times n m × n B B B n × p n\times p n × p ( A B ) ⊤ (AB)^\top ( A B ) ⊤ p × m p\times m p × m B ⊤ B^\top B ⊤
ML mein transpose kyun matter karta hai? Kyunki A ⊤ A A^\top A A ⊤ A hamesha symmetric hota hai — proof: ( A ⊤ A ) ⊤ = A ⊤ A (A^\top A)^\top = A^\top A ( A ⊤ A ) ⊤ = A ⊤ A A ⊤ A x = A ⊤ b A^\top A x = A^\top b A ⊤ A x = A ⊤ b
Ek aur deep baat: transpose actually dot product ko "doosri taraf" le jaata hai — ⟨ A x , y ⟩ = ⟨ x , A ⊤ y ⟩ \langle Ax, y\rangle = \langle x, A^\top y\rangle ⟨ A x , y ⟩ = ⟨ x , A ⊤ y ⟩
