Visual walkthrough — Types of matrices — row, column, square, diagonal, identity, zero, symmetric, skew-symmetric

The parent note gave us the cast of characters — row, column, square, diagonal, identity, zero, symmetric, skew-symmetric. This page proves the single most beautiful fact that ties two of them together:

Any square matrix can be broken into a symmetric part plus a skew-symmetric part — and only one way.

We build it from nothing. If you have never seen a matrix, a transpose, or the word "symmetric", start at Step 1 and never look anything up.

---

Step 1 — What a matrix is, and what "flipping" it means

WHAT. A matrix is just a rectangular grid of numbers. We label each number by which row and which column it sits in. The number in row $i$, column $j$ is written $a_{ij}$.

WHY. Before we can flip or add grids, we need a name for each cell. The address $a_{ij}$ is that name — the first index is the row (counting down), the second is the column (counting across).

PICTURE. Look at the $2\times 2$ board below. The white cell in row 1, column 2 is $a_{12}$; its mirror across the diagonal (row 2, column 1) is $a_{21}$. These two "partner" cells are the heart of everything that follows.

---

Step 2 — The transpose: reflecting the grid across its diagonal

WHAT. The transpose of $A$, written $A^T$, is the grid you get by swapping every cell $a_{ij}$ with $a_{ji}$. In words: rows become columns.

$(A^T{)}_{ij}=\underset{\text{cell from the swapped address}}{\underbrace{a_{ji}}}$

Read the equation term by term: the cell at address $(i,j)$ inside $A^T$ is fetched from address $(j,i)$ inside the original $A$. The two indices trade places.

WHY. We need a machine that "mirrors" a matrix so we can ask "does the mirror equal the original?" That single question defines symmetry. The mirror line is the principal diagonal — diagonal cells ($i=j$) never move, because $a_{ii}\to a_{ii}$.

PICTURE. The blue arrow shows $a_{12}$ leaping across the diagonal to become the new $a_{21}$, and the pink arrow shows the reverse. Diagonal cells stay put (yellow).

---

Step 3 — Two special mirrors: symmetric and skew-symmetric

WHAT. Ask the mirror question and there are exactly two "perfect" answers:

• $A^T=A$ → the mirror is identical. Call $A$ symmetric. Cell-wise: $a_{ij}=a_{ji}$.
• $A^T=-A$ → the mirror is the exact negative. Call $A$ skew-symmetric. Cell-wise: $a_{ij}=-a_{ji}$.

WHY. These are the two building blocks we will split every matrix into. Symmetric = partners are equal; skew = partners are opposite.

There is a forced consequence for skew. Put $i=j$ (a diagonal cell): $a_{ii}=-a_{ii}⟹2a_{ii}=0⟹\underset{\text{every diagonal cell is zero}}{\underbrace{a_{ii}=0}}$ Term by term: a diagonal cell must equal its own negative; the only number equal to its own negative is $0$. So a skew-symmetric matrix always has a zero diagonal.

PICTURE. Left board: equal partners (both $+3$). Right board: opposite partners ($+3$ and $-3$) with a forced-zero diagonal.

---

Step 4 — The trick: build a symmetric matrix out of any matrix

WHAT. Take any square $A$. Form $S=\frac{1}{2}(\underset{\text{original}}{\underbrace{A}}+\underset{\text{its mirror}}{\underbrace{A^T}}).$

Claim: $S$ is symmetric. Check by transposing it, using two facts — transpose is linear ($(X+Y{)}^T=X^T+Y^T$) and self-undoing ($(A^T{)}^T=A$): $S^T=\frac{1}{2}(A^T+(A^T{)}^T)=\frac{1}{2}(A^T+A)=S.$ Every symbol: we transposed the sum, split it, undid the double-transpose on the second term, and landed back on $S$. Since $S^T=S$, it is symmetric. ✓

WHY. Adding a matrix to its own mirror forces partners to become equal — each off-diagonal pair becomes $\frac{1}{2}(a_{ij}+a_{ji})$ on both sides. The $\frac{1}{2}$ is bookkeeping so nothing gets doubled.

PICTURE. Watch the partners $a_{12}$ and $a_{21}$ average into the same value on both sides of the diagonal.

---

Step 5 — Build a skew-symmetric matrix out of the same $A$

WHAT. Now subtract the mirror: $K=\frac{1}{2}(\underset{\text{original}}{\underbrace{A}}-\underset{\text{its mirror}}{\underbrace{A^T}}).$

Claim: $K$ is skew-symmetric. Transpose it: $K^T=\frac{1}{2}(A^T-(A^T{)}^T)=\frac{1}{2}(A^T-A)=-\frac{1}{2}(A-A^T)=-K.$ Term by term: transpose the difference, undo the double-transpose, factor out the minus sign, recognise $K$ inside. Since $K^T=-K$, it is skew. ✓ And its diagonal is automatically zero, because $\frac{1}{2}(a_{ii}-a_{ii})=0$.

WHY. Subtracting a matrix from its mirror forces partners to become opposites: each pair becomes $\frac{1}{2}(a_{ij}-a_{ji})$ on one side and $\frac{1}{2}(a_{ji}-a_{ij})$ (its negative) on the other.

PICTURE. The partners split into $+d$ and $-d$; the diagonal collapses to $0$.

---

Step 6 — They add back to $A$ (the decomposition)

WHAT. Add the two pieces: $S+K=\frac{1}{2}(A+A^T)+\frac{1}{2}(A-A^T)=\frac{1}{2}(A+A^T+A-A^T)=\frac{2A}{2}=A.$ Every symbol earns its place: the $+A^T$ from $S$ cancels the $-A^T$ from $K$, leaving $A+A$, which the halves turn into a single $A$.

WHY. This is the whole point: any square matrix is a symmetric part plus a skew-symmetric part. $\boxed{A=\underset{\text{symmetric }S}{\underbrace{\frac{1}{2}(A+A^T)}}+\underset{\text{skew }K}{\underbrace{\frac{1}{2}(A-A^T)}}}$

PICTURE. The $S$ board and the $K$ board stack back into the original $A$ board.

---

Step 7 — Edge cases: does it ever break?

We must never leave a scenario unshown.

• $A$ already symmetric ($A^T=A$): then $K=\frac{1}{2}(A-A)=O$, the zero matrix. The skew part vanishes — as it should, there is nothing to fix.
• $A$ already skew ($A^T=-A$): then $S=\frac{1}{2}(A+(-A))=O$. The symmetric part vanishes.
• $A$ diagonal (a special symmetric matrix): again $K=O$; diagonal matrices are their own symmetric part.
• $A=O$ (zero matrix): both $S=O$ and $K=O$. Trivially true.
• $A$ not square: $A^T$ has a different shape, so $A+A^T$ can't even be formed. The decomposition needs a square matrix — this is why Step 1 insisted on square.

PICTURE. The four collapse-cases side by side, plus the "shape mismatch" that forbids non-square inputs.

---

The one-picture summary

One diagram, the whole derivation: mirror $A$ to get $A^T$; average them for the symmetric half $S$; take half their difference for the skew half $K$; the two halves add back to $A$.

Recall Feynman retelling — say it in plain words

Picture a grid of numbers and its mirror image reflected across the top-left-to-bottom-right diagonal. If you average the grid with its mirror, matching partners become equal — that's a symmetric grid $S$. If instead you take half the difference, partners become exact opposites and the diagonal is forced to zero — that's a skew-symmetric grid $K$. The clever bit: when you add $S$ and $K$ the mirror-copies cancel and you get your original grid back, perfectly. So every square grid is secretly a symmetric part plus a skew part, and there's only one way to do it. If the grid was already symmetric, the skew part is all zeros; if it was already skew, the symmetric part is all zeros; and if the grid isn't square, you can't even line it up with its mirror, so the trick doesn't apply.

Recall

Formula for the symmetric part of $A$ ::: $S=\frac{1}{2}(A+A^T)$ Formula for the skew-symmetric part ::: $K=\frac{1}{2}(A-A^T)$ Why is a skew matrix's diagonal always zero ::: $a_{ii}=-a_{ii}\Rightarrow a_{ii}=0$ What does the transpose do to cell $a_{ij}$ ::: sends it to address $(j,i)$, i.e. $(A^T{)}_{ij}=a_{ji}$ When does the skew part vanish ::: when $A$ is already symmetric ($A^T=A$)