Matrix Algebra

01 / UNDERSTAND & APPLY

1. Shape is part of the mathematical object

A matrix is a rectangular array with a specified number of rows and columns. An m×n matrix has m rows and n columns, and its entry aᵢⱼ lies in row i, column j. Two matrices are equal only when their shapes match and every corresponding entry is equal. An ordered arrangement matters: swapping entries usually produces a different matrix.

Matrices can store data or represent linear transformations. A matrix A of shape m×n maps a column vector with n coordinates to a column vector with m coordinates by multiplication. This input-output interpretation explains the dimension rule before any arithmetic: the input must have the number of entries the transformation expects.

Addition and subtraction are entrywise and require equal shapes. Scalar multiplication multiplies every entry by the same scalar. The zero matrix has every entry zero and a shape appropriate to the context. The identity matrix Iₙ is square, with ones on its main diagonal and zeros elsewhere; it leaves n-coordinate vectors unchanged.

We use real matrices unless a different number system is stated. A binary matrix here means its entries are chosen from {0,1} but operations are ordinary real arithmetic. It does not automatically mean arithmetic modulo two. This distinction becomes important when determinant or invertibility conditions are imposed.

02 / UNDERSTAND & APPLY

2. Row by column is composition, not entrywise multiplication

If A is m×n and B is n×p, the product AB is m×p. Its entry in row i, column j is Σₖ₌₁ⁿaᵢₖbₖⱼ. Take row i of A, column j of B, multiply matching entries and add. The matching inner dimension n is the length of each such dot product.

If column vectors are used, AB means first apply B and then apply A: ABx = A(Bx). This explains why order matters. Even when AB and BA both exist and have the same shape, they need not be equal. A horizontal shear followed by a vertical shear generally differs from doing them in reverse order.

Multiplication is associative where the dimensions allow it: (AB)C = A(BC). It distributes over addition. However, the usual scalar expansion (A + B)² = A² + 2AB + B² needs AB = BA. Without commutation, the middle terms are AB + BA. Likewise, (A − B)(A + B) need not equal A² − B².

A product can be zero even when neither factor is zero. This is a major difference from real scalar multiplication. Matrix cancellation from AB = AC requires additional information such as invertibility of A; merely knowing A ≠ 0 is not enough.

03 / UNDERSTAND & APPLY

3. Transposition swaps rows and columns and reverses products

The transpose Aᵀ has entry (Aᵀ)ᵢⱼ = aⱼᵢ. An m×n matrix becomes n×m. Transposition respects sums and scalar multiples, but a product reverses order: (AB)ᵀ = BᵀAᵀ. The dimensions themselves suggest this reversal, and the row-column entry formula proves it.

A square real matrix is symmetric if Aᵀ = A and skew-symmetric if Aᵀ = −A. For a skew-symmetric real matrix, aᵢᵢ = −aᵢᵢ forces every diagonal entry to be zero. Its lower-triangular entries are the negatives of the corresponding upper-triangular entries. This is why skew-symmetric matrices have fewer independent choices.

Every real square matrix has a unique decomposition A = S + K with S symmetric and K skew-symmetric: S = (A + Aᵀ)/2 and K = (A − Aᵀ)/2. Adding gives A, while transposing checks the two required properties. Uniqueness follows because a real matrix that is both symmetric and skew-symmetric must be zero.

A product of symmetric matrices need not be symmetric. If A and B are symmetric, then (AB)ᵀ = BA, so AB is symmetric exactly when AB = BA. In contrast, AᵀA is always symmetric and has xᵀAᵀAx = |Ax|² ≥ 0 for every real column x. The latter identity often provides a useful structural check.

04 / UNDERSTAND & APPLY

4. An inverse undoes a transformation on the correct side

A square matrix A is invertible if there is a matrix A⁻¹ satisfying AA⁻¹ = A⁻¹A = I. For A = [[a,b],[c,d]], the inverse exists exactly when ad − bc ≠ 0, and equals (1/(ad−bc))[[d,−b],[−c,a]]. The denominator is the determinant, not the sum of diagonal entries.

To solve AX = B with invertible A, multiply on the left: X = A⁻¹B. To solve XA = B, multiply on the right: X = BA⁻¹. The side is determined by where A appears. A casual instruction to “divide by A” is ambiguous and should be replaced by a correctly ordered multiplication.

For invertible square matrices, (AB)⁻¹ = B⁻¹A⁻¹. The reversal is necessary because AB B⁻¹A⁻¹ = I. Also (Aᵀ)⁻¹ = (A⁻¹)ᵀ. An inverse of a sum generally has no simple sum formula: (A + B)⁻¹ is not A⁻¹ + B⁻¹, even when all three inverses exist.

A singular matrix may still allow solutions of AX = B for some B, but the inverse shortcut is unavailable and uniqueness can fail. Solving the associated linear system is then the appropriate method. Singularity does not mean every equation involving A is inconsistent.

05 / UNDERSTAND & APPLY

5. Polynomial identities can simplify high powers

A matrix is idempotent when P² = P, as happens for many projection matrices. A matrix is nilpotent when Nᵏ = 0 for some positive integer k. Neither condition permits ordinary scalar factor cancellation. From P(P − I) = 0 one cannot conclude P = 0 or P = I; for example diag(1,0) is a third possibility.

For N² = 0, powers of I + N simplify because I commutes with N and the ordinary binomial expansion is valid for this pair. Every term containing N² vanishes, leaving (I + N)ᵐ = I + mN for nonnegative integer m. Its inverse is I − N because (I + N)(I − N) = I − N² = I.

More generally, a polynomial relation satisfied by A can reduce high powers to lower powers. If A² = αA + βI, repeated substitution expresses every higher power as a linear combination of A and I. The scalar coefficients follow a recurrence. Verify the starting matrix identity directly before using it as a shortcut.

An orthogonal real matrix satisfies QᵀQ = I and preserves Euclidean lengths: |Qx|² = xᵀQᵀQx = |x|². Its inverse is Qᵀ. A symmetric matrix is not automatically orthogonal, and an orthogonal matrix need not be symmetric. Similar-looking names describe different equations.

06 / UNDERSTAND & APPLY

6. Count independent entries rather than visible boxes

If every entry of an m×n matrix can independently take one of q values, there are q^(mn) matrices. The word independently is doing the work. A condition such as symmetry ties pairs of entries together, a diagonal condition may fix entries, and an invertibility condition couples several choices through a determinant.

For an n×n symmetric matrix with entries in a fixed q-element set, choose its n diagonal entries and its n(n−1)/2 entries above the diagonal. The entries below are then forced by reflection. The count is q^[n(n+1)/2]. No ordering of the chosen values is forgotten: each chosen position remains distinct.

For a skew-symmetric real matrix, the diagonal is zero and each lower entry is the negative of an upper entry. If the allowed set S is closed under negation and contains zero, with |S| = q, the count is q^[n(n−1)/2]. If S lacks some negatives, not every upper choice is allowed. For S = {0,1}, skew-symmetry forces every off-diagonal entry to be zero too, so only the zero matrix remains.

An upper-triangular matrix has n(n+1)/2 potentially free entries, including the diagonal. With a common q-choice alphabet containing zero, its count is q^[n(n+1)/2]. If the diagonal is fixed in advance, only n(n−1)/2 positions remain free. A diagonal matrix has only n independent positions, and a scalar matrix λI has only one.

07 / UNDERSTAND & APPLY

7. Structural restrictions can change a count completely

A triangular matrix is invertible exactly when every diagonal entry is nonzero, since its determinant is their product. For upper-triangular binary matrices under real arithmetic, every diagonal must therefore be 1 and the entries above the diagonal remain arbitrary. The count is 2^[n(n−1)/2], not the count of all triangular matrices.

A permutation matrix has exactly one 1 in each row and each column, with other entries zero. Choose the column for the first row in n ways, then a different column for the second in n − 1 ways, and continue. The count is n!, and all such matrices are invertible. Requiring exactly one 1 in each row alone would give nⁿ because repeated columns would be allowed.

For small matrices, case splits on a strategic entry or row are often safer than an unsupported general formula. A 2×2 binary matrix is invertible over the reals if its two rows are nonzero and not scalar multiples. Among the three nonzero binary row vectors, no two distinct ones are scalar multiples, so there are 3·2 = 6 invertible matrices.

The field of arithmetic must be stated. A real binary 3×3 matrix can have determinant −2 and be invertible over the reals, while the same entries represent a singular matrix under arithmetic modulo two because −2 becomes zero. Formulas counting invertible matrices over finite fields cannot be transferred to real matrices merely because their entries are 0 and 1.

08 / UNDERSTAND & APPLY

8. Original practice: dimensions, order and structure

Before multiplying, write the shapes. Before cancelling, identify an inverse or another justified condition. These habits are more reliable than checking the final entries alone.

09 / UNDERSTAND & APPLY

9. Original practice: state the free choices explicitly

Every count should identify a set of independent choices and explain why each admissible matrix is counted exactly once. If the rule couples entries, classify that coupling before using a power.

10 / UNDERSTAND & APPLY

10. A matrix answer needs both algebra and a shape check

Dimensions decide whether an expression exists before arithmetic decides its value. Multiplication order records transformation order. Transposition and inversion reverse products. Singular matrices block cancellation, while special identities such as N² = 0 can make powers easy. These structural facts explain the rules rather than leaving them as isolated exceptions.

For counting, begin with unconstrained positions and then identify which entries are fixed, mirrored, negated or coupled by invertibility. Positive counts such as q^(mn) rely on independence; they are not universal formulas for every matrix condition. For small cases, a direct case split or enumeration is a valuable check on a proposed general argument.

The most common errors are treating multiplication as entrywise, assuming AB = BA, putting the inverse on the wrong side, counting symmetric off-diagonal partners independently, allowing nonzero skew-symmetric diagonal entries over the reals, and importing a finite-field count into real arithmetic.

Verify a matrix equation by multiplication, a symmetry claim by transposition, an inverse by both-sided identity checks, and a count by explaining a one-to-one correspondence between choices and matrices. These are short checks that protect against different kinds of error and make a solution independently readable.