math.wiki
Linear algebra › Abstract vector spaces › Sums and direct sums of subspaces
Sign in

Sums and direct sums of subspaces

When a space is built from two or more of its subspaces, which vectors are sums of pieces from each, and when is the split of every vector into pieces forced to be unique?

Where you are 5 in Abstract vector spaces

Why this exists

Mei and Andre run an embroidery machine. Two motors move the hoop that holds the fabric: one slides it along a rail in the direction (3, 1), the other along a rail in the direction (1, 2). Every stitch is a move \mathbf{v} \in \mathbb{R}^2, and the controller must turn it into one slide along each rail. Can every move be made? In only one way? And what changes if Andre bolts on a third motor?

The rails are lines L_1 and L_2 through \mathbf{0}, the moves the machine can make form their sum, and “only one way” makes the sum direct. Later lessons cut a space into eigenspaces the same way, using the tests and dimension counts built here.

Definition

Let U_1, \dots, U_m be subspaces of a vector space V over \mathbb{F} (Vector spaces from axioms). Their sum is the set of all vectors obtained by adding one vector from each:

U_1 + \dots + U_m = \{\mathbf{u}_1 + \dots + \mathbf{u}_m : \mathbf{u}_j \in U_j\}.

The sum is direct if each of its vectors equals \mathbf{u}_1 + \dots + \mathbf{u}_m with \mathbf{u}_j \in U_j for only one choice of \mathbf{u}_1, \dots, \mathbf{u}_m; then it is written U_1 \oplus \dots \oplus U_m, and \mathbf{u}_j is the U_j-piece of the vector. (\oplus named a made-up addition in one example of Vector spaces from axioms; here it only marks a sum as direct.)

If V = U \oplus W, then W is a complement of U in V. This is not the complement of a set: U and W always share \mathbf{0}, and U \cap W is their intersection.

Intuition

The union of two lines through \mathbf{0} is not a subspace: (3, 1) + (1, 2) lies on neither line. The sum is everything addition can reach. Directness asks that each vector have a well-defined piece in each summand, as a vector of \mathbb{F}^n has well-defined coordinates. Two splits of one vector differ by a split of \mathbf{0}, so only \mathbf{0} needs testing. For two summands that means sharing only \mathbf{0}; three lines in the plane share only \mathbf{0} in pairs, yet the third adds nothing new.

Results

Results 1 to 7 hold over any field. Throughout, U_1, \dots, U_m, U and W are subspaces of V.

Result 1 (the sum is the smallest subspace containing the summands). U_1 + \dots + U_m is a subspace of V containing every U_j, and every subspace of V containing every U_j contains it. If each U_j is the span of a list L_j, the sum is the span of the lists L_1, \dots, L_m joined in order.

Result 2 (the zero test). U_1 + \dots + U_m is direct if and only if the only way to write \mathbf{0} = \mathbf{u}_1 + \dots + \mathbf{u}_m with each \mathbf{u}_j \in U_j is \mathbf{u}_1 = \dots = \mathbf{u}_m = \mathbf{0}.

Result 3 (two summands). U + W is direct if and only if U \cap W = \{\mathbf{0}\}.

Result 4 (several summands). For m \ge 2, U_1 + \dots + U_m is direct if and only if

U_j \cap (U_1 + \dots + U_{j-1}) = \{\mathbf{0}\} \quad \text{for } j = 2, \dots, m.

Pairwise intersections equal to \{\mathbf{0}\} do not suffice (Examples 1 and 4).

Result 5 (bases join). Let each U_j be finite-dimensional with basis \mathcal{B}_j, and let \mathcal{B} be \mathcal{B}_1, \dots, \mathcal{B}_m joined in order. Then \mathcal{B} spans U_1 + \dots + U_m, and the sum is direct if and only if \mathcal{B} is linearly independent. In that case \mathcal{B} is a basis of the sum and

\dim(U_1 \oplus \dots \oplus U_m) = \dim U_1 + \dots + \dim U_m.

Conversely, if \dim(U_1 + \dots + U_m) = \dim U_1 + \dots + \dim U_m, the sum is direct.

Result 6 (dimension of a sum). If U and W are finite-dimensional, then

\dim(U + W) = \dim U + \dim W - \dim(U \cap W).

Result 7 (complements exist). If V is finite-dimensional, every subspace U has a complement, and every complement of U has dimension \dim V - \dim U.

Why it is true

“Result k there” means Result k of Bases and dimension, proved.

Result 1. \mathbf{0} = \mathbf{0} + \dots + \mathbf{0} lies in the sum. If \mathbf{x} = \sum_j \mathbf{u}_j and \mathbf{y} = \sum_j \mathbf{u}'_j are in it and c \in \mathbb{F}, the commutative, associative and distributive axioms give \mathbf{x} + c\mathbf{y} = \sum_j (\mathbf{u}_j + c\mathbf{u}'_j), and \mathbf{u}_j + c\mathbf{u}'_j \in U_j because U_j is a subspace. Each \mathbf{u} \in U_k is the sum with \mathbf{u} in place k and \mathbf{0} elsewhere. A subspace containing every U_j contains each \mathbf{u}_j, so their sum. For the lists, the span S of the joined list is a subspace containing each L_j, hence each U_j (Result 1 of Span and independence, proved), hence the sum; and the sum contains every entry of the joined list, hence S.

Result 2. If the sum is direct, \mathbf{0} + \dots + \mathbf{0} is the one way. Conversely, if \sum_j \mathbf{u}_j = \sum_j \mathbf{u}'_j with \mathbf{u}_j, \mathbf{u}'_j \in U_j, subtracting gives \mathbf{0} = \sum_j (\mathbf{u}_j - \mathbf{u}'_j) with \mathbf{u}_j - \mathbf{u}'_j \in U_j, so every \mathbf{u}_j = \mathbf{u}'_j.

Result 3. Let U + W be direct and \mathbf{x} \in U \cap W. Then \mathbf{0} = \mathbf{x} + (-\mathbf{x}) with \mathbf{x} \in U and -\mathbf{x} \in W, so \mathbf{x} = \mathbf{0} by Result 2. Conversely, let U \cap W = \{\mathbf{0}\} and \mathbf{0} = \mathbf{u} + \mathbf{w}. Then \mathbf{u} = -\mathbf{w} lies in W as well as U, so \mathbf{u} = \mathbf{0}, then \mathbf{w} = \mathbf{0}, and Result 2 applies.

Result 4. Let the sum be direct, j \ge 2, and \mathbf{x} \in U_j with \mathbf{x} = \mathbf{u}_1 + \dots + \mathbf{u}_{j-1}, each \mathbf{u}_i \in U_i. Then

\mathbf{0} = \mathbf{u}_1 + \dots + \mathbf{u}_{j-1} + (-\mathbf{x})

is a split of \mathbf{0} (pieces \mathbf{0} in the places after j), so \mathbf{x} = \mathbf{0} by Result 2. Conversely, assume the intersections are \{\mathbf{0}\} and \mathbf{0} = \mathbf{u}_1 + \dots + \mathbf{u}_m. If some \mathbf{u}_j \neq \mathbf{0}, let j be the largest such index. It is not 1, since then \mathbf{u}_1 = \mathbf{0}. So j \ge 2 and \mathbf{u}_j = -(\mathbf{u}_1 + \dots + \mathbf{u}_{j-1}) lies in U_j \cap (U_1 + \dots + U_{j-1}) = \{\mathbf{0}\}, which is false. So every \mathbf{u}_j = \mathbf{0}, and Result 2 applies.

Result 5. \mathcal{B} spans the sum by Result 1. Let the sum be direct, and take a combination of \mathcal{B} equal to \mathbf{0}. The terms from \mathcal{B}_j add to some \mathbf{w}_j \in U_j, and \mathbf{w}_1 + \dots + \mathbf{w}_m = \mathbf{0}, so every \mathbf{w}_j = \mathbf{0} (Result 2); as \mathcal{B}_j is independent, its coefficients are all 0. Conversely, let \mathcal{B} be independent and \mathbf{0} = \mathbf{u}_1 + \dots + \mathbf{u}_m. Writing each \mathbf{u}_j in the basis \mathcal{B}_j turns this into a combination of \mathcal{B} equal to \mathbf{0}, so all its coefficients are 0 and every \mathbf{u}_j = \mathbf{0}. Then \mathcal{B} is an independent spanning list of length \sum_j \dim U_j, a basis. Finally, if the sum has dimension N = \sum_j \dim U_j, then \mathcal{B} spans it and has length N, so it is a basis by Result 5 there, hence independent, and the sum is direct.

Result 6. U \cap W contains \mathbf{0} and is closed under sums and multiples since U and W are, so it is a subspace of U, finite-dimensional by Result 6 there, with a basis (\mathbf{x}_1, \dots, \mathbf{x}_k). Extend it to a basis (\mathbf{x}_1, \dots, \mathbf{x}_k, \mathbf{u}_1, \dots, \mathbf{u}_p) of U and to a basis (\mathbf{x}_1, \dots, \mathbf{x}_k, \mathbf{w}_1, \dots, \mathbf{w}_q) of W (Result 3 there). The list of all the \mathbf{x}_i, \mathbf{u}_r and \mathbf{w}_s spans U + W by Result 1. For independence, suppose

\sum_{i} a_i\mathbf{x}_i + \sum_{r} b_r\mathbf{u}_r + \sum_{s} c_s\mathbf{w}_s = \mathbf{0}.

Then \sum_s c_s\mathbf{w}_s = -\sum_i a_i\mathbf{x}_i - \sum_r b_r\mathbf{u}_r lies in W and in U, so it equals \sum_i d_i\mathbf{x}_i for some scalars d_i. That gives \sum_i d_i\mathbf{x}_i - \sum_s c_s\mathbf{w}_s = \mathbf{0}, a relation on the basis of W, so every c_s = 0. What remains is a relation on the basis of U, so every a_i and b_r is 0. Hence \dim(U + W) = k + p + q = (k + p) + (k + q) - k.

Result 7. Take a basis (\mathbf{u}_1, \dots, \mathbf{u}_k) of U (Result 6 there), extend it to a basis (\mathbf{u}_1, \dots, \mathbf{u}_k, \mathbf{w}_1, \dots, \mathbf{w}_l) of V (Result 3 there), and let W = \operatorname{span}\{\mathbf{w}_1, \dots, \mathbf{w}_l\}, which has basis (\mathbf{w}_1, \dots, \mathbf{w}_l). The joined list is independent, so U + W is direct by Result 5, and it spans V, so U + W = V by Result 1. Any complement W' is finite-dimensional (Result 6 there), so Result 5 gives \dim U + \dim W' = \dim V.

Worked examples

Example 1

Two rails and a third

(a) Let L_1 = \operatorname{span}\{(3, 1)\} and L_2 = \operatorname{span}\{(1, 2)\} be the two rails. Show that \mathbb{R}^2 = L_1 \oplus L_2, split the move \mathbf{v} = (7, 9), and give the split of a general (x, y). (b) Andre adds a rail L_3 = \operatorname{span}\{(1, -1)\}. Show that L_1 + L_2 + L_3 is not direct although each pair of rails meets only in \mathbf{0}, and give a second way to make \mathbf{v}.

Solution

(a) If a(3, 1) = b(1, 2), then 3a = b and a = 2b = 6a, so a = b = 0: L_1 \cap L_2 = \{\mathbf{0}\} and the sum is direct (Result 3). Its dimension is 1 + 1 = 2 (Result 5), so it is a subspace of \mathbb{R}^2 of full dimension, hence \mathbb{R}^2 (Result 6 of Bases and dimension, proved). To split (x, y) = a(3, 1) + b(1, 2), solve 3a + b = x, a + 2b = y:

a = \frac{2x - y}{5}, \qquad b = \frac{3y - x}{5}.

For \mathbf{v}, a = 1 and b = 4, so \mathbf{v} = (3, 1) + (4, 8).

\mathbf{u}
\mathbf{w}
\mathbf{v}
L_1
L_2
\mathbf{0}
Figure 1Since \mathbb{R}^2 = L_1 \oplus L_2, the move \mathbf{v} = (7, 9) splits in exactly one way, as \mathbf{u} = (3, 1) along L_1 plus \mathbf{w} = (4, 8) along L_2.

(b) No two of (3, 1), (1, 2), (1, -1) are multiples of each other, so each pair of rails meets only in \mathbf{0}. But L_1 + L_2 = \mathbb{R}^2 contains L_3, so L_3 \cap (L_1 + L_2) = L_3 and Result 4 fails. Splitting (1, -1) by part (a) gives \displaystyle a = \frac{3}{5}, \displaystyle b = -\frac{4}{5}, which cleared of fractions is

3(3, 1) - 4(1, 2) - 5(1, -1) = (0, 0).

Adding this split of \mathbf{0} to the old one gives \mathbf{v} = (12, 4) + (0, 0) + (-5, 5): three motors no longer fix the settings.

Example 2

A sum that is not direct

In \mathbb{R}^4 let U = \{\mathbf{x} : x_3 = x_1 + x_2,\ x_4 = 2x_1 - x_2\} and W = \operatorname{span}\{(1, 1, 2, 1), (1, -1, 0, 0)\}. Find U \cap W, \dim(U + W), and one equation that cuts out U + W.

Solution

In U, x_1 = a and x_2 = b are free, so U consists of the vectors (a, b, a + b, 2a - b), and U = \operatorname{span}\{(1, 0, 1, 2), (0, 1, 1, -1)\}. These two are independent (look at the first two entries), so \dim U = 2; the two vectors spanning W are not multiples, so \dim W = 2. A vector of W is

c(1, 1, 2, 1) + d(1, -1, 0, 0) = (c + d, c - d, 2c, c).

It lies in U when 2c = (c + d) + (c - d), which always holds, and c = 2(c + d) - (c - d) = c + 3d, which forces d = 0. So U \cap W = \operatorname{span}\{(1, 1, 2, 1)\}, the sum is not direct (Result 3), and Result 6 gives \dim(U + W) = 2 + 2 - 1 = 3.

By Result 1, U + W is spanned by (1, 0, 1, 2), (0, 1, 1, -1), (1, -1, 0, 0), each satisfying x_3 = x_1 + x_2. That equation cuts out a subspace H of dimension 3 (x_1, x_2, x_4 free) containing U + W, and equal dimensions give U + W = \{\mathbf{x} : x_3 = x_1 + x_2\} (Result 6 of Bases and dimension, proved).

Example 3

The tangent line as a piece

Let W = \{p \in \mathcal{P}_3(\mathbb{R}) : p(1) = 0 \text{ and } p'(1) = 0\}. Show that \mathcal{P}_3(\mathbb{R}) = \mathcal{P}_1(\mathbb{R}) \oplus W, describe the \mathcal{P}_1(\mathbb{R})-piece of any p, and split z^3.

Solution

W is a subspace, since (p + cq)(1) = p(1) + cq(1) and (p + cq)' = p' + cq' (Definition of the derivative). By the The remainder and factor theorems, p(1) = 0 gives p = (z - 1)q with q \in \mathcal{P}_2(\mathbb{R}); then p'(1) = q(1), and q(1) = 0 gives q = (z - 1)r with r \in \mathcal{P}_1(\mathbb{R}). So W = \{(z - 1)^2 r : r \in \mathcal{P}_1(\mathbb{R})\}, with basis ((z - 1)^2, z(z - 1)^2), independent because the degrees differ: \dim W = 2.

A nonzero (z - 1)^2 r has degree at least 2, so \mathcal{P}_1(\mathbb{R}) \cap W = \{0\}, the sum is direct (Result 3), and its dimension 2 + 2 = 4 (Result 5) makes it \mathcal{P}_3(\mathbb{R}).

If p = \ell + w with \ell \in \mathcal{P}_1(\mathbb{R}) and w \in W, then \ell(1) = p(1) and \ell'(1) = p'(1): the piece \ell is the tangent line to the graph of p at 1. For p = z^3, p(1) = 1 and p'(1) = 3, so \ell = 1 + 3(z - 1) = 3z - 2 and

z^3 = (3z - 2) + (z - 1)^2(z + 2).
Example 4

Three summands and a parameter

In \mathbb{F}^4 let U_1 = \operatorname{span}\{(1, 1, 0, 0)\}, U_2 = \operatorname{span}\{(0, 1, 1, 0), (0, 0, 1, 1)\} and U_3 = \operatorname{span}\{(1, 0, 0, c)\} with c \in \mathbb{F}. For which c is U_1 + U_2 + U_3 direct? For the other values, show that the pairwise intersections are still \{\mathbf{0}\}.

Solution

Use Result 4. A vector of U_2 is (0, a, a + b, b), with first entry 0, so t(1, 1, 0, 0) \in U_2 only for t = 0: U_1 \cap U_2 = \{\mathbf{0}\}, and U_1 + U_2 has dimension 3 (Result 5). The three spanning vectors satisfy x_1 - x_2 + x_3 - x_4 = 0, which cuts out a subspace H of dimension 3, so U_1 + U_2 = H as in Example 2.

Now t(1, 0, 0, c) \in H exactly when t(1 - c) = 0. If c \neq 1, only t = 0 works, the sum is direct, and its dimension 4 makes it \mathbb{F}^4. If c = 1, U_3 \subseteq H and the sum is not direct; indeed

\begin{aligned} \mathbf{0} &= (1, 1, 0, 0) + (0, -1, 0, 1) \\ &\quad + (-1, 0, 0, -1) \end{aligned}

has nonzero pieces, the middle one being -(0, 1, 1, 0) + (0, 0, 1, 1) \in U_2. Yet for c = 1, t(1, 1, 0, 0) = s(1, 0, 0, 1) forces t = s = 0, and nonzero vectors of U_3 have first entry s \neq 0, so all three pairwise intersections are \{\mathbf{0}\}.

Example 5

Even and odd functions

Call f \in \mathbb{R}^{\mathbb{R}} (Polynomials and functions as vectors) even if f(-x) = f(x) for all x and odd if f(-x) = -f(x) for all x, and let E and O be the sets of even and odd functions. Prove that \mathbb{R}^{\mathbb{R}} = E \oplus O, and split f(x) = e^x.

Solution

The zero function is even and odd, and (f + cg)(-x) = f(-x) + cg(-x) shows that E and O are closed under sums and multiples, so they are subspaces. If f \in E \cap O, then f(x) = f(-x) = -f(x), so 2f(x) = 0 and f(x) = 0 for every x: E \cap O = \{0\}. For any f put

f_e(x) = \frac{f(x) + f(-x)}{2}, \qquad f_o(x) = \frac{f(x) - f(-x)}{2}.

Replacing x by -x swaps the terms in each numerator, so f_e \in E, f_o \in O, and f_e + f_o = f. Hence \mathbb{R}^{\mathbb{R}} = E + O, direct by Result 3.

For f(x) = e^x the pieces are \displaystyle \frac{e^x + e^{-x}}{2} and \displaystyle \frac{e^x - e^{-x}}{2}, the functions \cosh x and \sinh x. In this infinite-dimensional space no dimension count is available; the proof wrote the pieces down.

Boundaries

Pairwise tests stop at two summands. For three or more, pairwise intersections \{\mathbf{0}\} do not give directness (Examples 1 and 4); Result 4 is the test.

No inclusion-exclusion for three. Result 6 has no three-term version. For the three rails of Example 1 every intersection is \{\mathbf{0}\}, so 1 + 1 + 1 - 0 - 0 - 0 + 0 = 3, yet the sum \mathbb{R}^2 has dimension 2.

Halving needs 2 \neq 0. Example 5 and Exercise 10 divide by 2, which is allowed in \mathbb{R} and \mathbb{C}. Abstract algebra meets fields in which 1 + 1 = 0; there -A = A, so the symmetric and skew-symmetric matrices coincide and their sum is not direct.

Complements are not unique. Every line through \mathbf{0} other than L_1 is a complement of L_1 in \mathbb{R}^2. Result 7 extends a finite basis; in infinite dimensions a complement may be written down, as O for E, but in general its existence rests on the axiom of choice (Bases and dimension, proved).

Where this goes next

When an operator is diagonalizable shows that an operator is diagonalizable exactly when V is the direct sum of its eigenspaces, and Generalized eigenvectors splits every finite-dimensional complex space into generalized eigenspaces; Result 5 counts dimensions in both. Orthonormal bases and Gram-Schmidt, proved builds the complement U^{\perp}. Back down, Subspaces and Span are the \mathbb{R}^n versions.

Exercises

Sums and intersections

  1. In \mathbb{R}^3 let U = \operatorname{span}\{(1, 2, 0), (0, 1, 1)\} and W = \operatorname{span}\{(1, 0, 1), (2, 1, 3)\}. Give each by one equation, then find U \cap W and U + W. Is the sum direct?
  2. In \mathcal{P}_2(\mathbb{R}) let U = \operatorname{span}\{1 + z, z^2\} and W = \operatorname{span}\{z, 1 + z^2\}. Find U \cap W and U + W.
  3. Let U and W be 4-dimensional subspaces of a 6-dimensional space. List the possible values of \dim(U \cap W) and of \dim(U + W), and realize each in \mathbb{R}^6.

Direct sums

  1. Another hoop has rails along (2, 1) and (-1, 3). Show that \mathbb{R}^2 is the direct sum of the two rails, give the split of a general (x, y), and split (1, 11).
  2. In \mathbb{F}^4 let U_1 = \operatorname{span}\{(1, 0, 1, 0)\}, U_2 = \operatorname{span}\{(0, 1, 0, 1)\} and U_3 = \operatorname{span}\{(1, 1, 1, 1)\}. Show that each pair meets only in \mathbf{0} but U_1 + U_2 + U_3 is not direct, and write \mathbf{0} with nonzero pieces.
  3. Let U = \{p \in \mathcal{P}_3(\mathbb{R}) : p(0) = p(1) = 0\}. Show that \mathcal{P}_3(\mathbb{R}) = U \oplus \mathcal{P}_1(\mathbb{R}) and split z^3 + 2z.
  4. Show that \mathbb{R}^{2 \times 2} = \operatorname{span}\{I\} \oplus \{A : \operatorname{tr} A = 0\}, where \operatorname{tr} A is the sum of the diagonal entries, and split \begin{bmatrix} 5 & 2 \\ -1 & 3 \end{bmatrix}.
  5. Show that \mathbb{C}^2 = \operatorname{span}\{(1, i)\} \oplus \operatorname{span}\{(1, -i)\} and split (0, 2) (Complex numbers).
  6. Let U = \operatorname{span}\{(1, 1, 1, 1), (1, 2, 0, 3)\} in \mathbb{R}^4. Follow the proof of Result 7, appending \mathbf{e}_1, \mathbf{e}_2, \mathbf{e}_3, \mathbf{e}_4 in order, to find a complement.

Proofs

  1. Let \mathbb{F} be \mathbb{R} or \mathbb{C}. Prove that \mathbb{F}^{n \times n} is the direct sum of the symmetric matrices (A^{T} = A) and the skew-symmetric matrices (A^{T} = -A) (The transpose and symmetric matrices).
  2. Prove that U + U = U, and that U + W = W if and only if U \subseteq W.
  3. Prove that if U_1 + \dots + U_m is direct, then U_j \cap U_k = \{\mathbf{0}\} whenever j \neq k.
  4. Give a Counterexample to: if V = U \oplus W_1 and V = U \oplus W_2, then W_1 = W_2. Then prove that \dim W_1 = \dim W_2 when V is finite-dimensional.
  5. Let (\mathbf{v}_1, \dots, \mathbf{v}_n) be a basis of V. Prove that V = \operatorname{span}\{\mathbf{v}_1\} \oplus \dots \oplus \operatorname{span}\{\mathbf{v}_n\}.
  6. Let V be finite-dimensional and U a subspace with \{\mathbf{0}\} \neq U \neq V. Prove that U has at least two complements.
Answers
  1. U = \{2x - y + z = 0\} and W = \{z = x + y\}: each spanning vector satisfies its equation, and each equation cuts out a plane. On both, z = x + y turns 2x - y + z = 0 into 3x = 0, so x = 0 and z = y: U \cap W = \operatorname{span}\{(0, 1, 1)\}. By Result 6, \dim(U + W) = 2 + 2 - 1 = 3, so U + W = \mathbb{R}^3. Not direct.
  2. U \cap W = \operatorname{span}\{1 + z + z^2\} and U + W = \mathcal{P}_2(\mathbb{R}). If a(1 + z) + bz^2 = cz + d(1 + z^2), the constant, z and z^2 terms give a = d, a = c, b = d, so all four are equal. Then \dim(U + W) = 2 + 2 - 1 = 3.
  3. \dim(U + W) \in \{4, 5, 6\} and \dim(U \cap W) \in \{4, 3, 2\}, matched in that order: the sum contains U and lies in the space, and Result 6 gives \dim(U \cap W) = 8 - \dim(U + W). With U = \operatorname{span}\{\mathbf{e}_1, \dots, \mathbf{e}_4\}, take W = U, then W = \operatorname{span}\{\mathbf{e}_2, \dots, \mathbf{e}_5\}, then W = \operatorname{span}\{\mathbf{e}_3, \dots, \mathbf{e}_6\}.
  4. The directions are not multiples, so the rails meet only in \mathbf{0}, and the direct sum has dimension 2, so it is \mathbb{R}^2. Solving 2a - b = x, a + 3b = y gives \displaystyle a = \frac{3x + y}{7}, \displaystyle b = \frac{2y - x}{7}. For (1, 11), a = 2, b = 3: (1, 11) = (4, 2) + (-3, 9).
  5. No two of the vectors are multiples, so each pair of lines meets only in \mathbf{0}. But (1, 1, 1, 1) = (1, 0, 1, 0) + (0, 1, 0, 1), so U_3 \subseteq U_1 + U_2 and Result 4 fails: \mathbf{0} = (1, 0, 1, 0) + (0, 1, 0, 1) + (-1, -1, -1, -1).
  6. z^3 + 2z = (z^3 - z) + 3z. By the factor theorem U = \{z(z - 1)r : r \in \mathcal{P}_1(\mathbb{R})\}, of dimension 2, and its nonzero elements have degree at least 2, so U \cap \mathcal{P}_1(\mathbb{R}) = \{0\} and the direct sum, of dimension 4, is \mathcal{P}_3(\mathbb{R}). The \mathcal{P}_1-piece is the line through (0, p(0)) and (1, p(1)), here (0, 0) and (1, 3); and z^3 - z = z(z - 1)(z + 1) \in U.
  7. 4I + \begin{bmatrix} 1 & 2 \\ -1 & -1 \end{bmatrix}. The trace-zero matrices \begin{bmatrix} a & b \\ c & -a \end{bmatrix} form a subspace of dimension 3. \operatorname{tr}(cI) = 2c = 0 only for c = 0, and dimension 1 + 3 = 4 gives \mathbb{R}^{2 \times 2}. The piece along I is \displaystyle \frac{\operatorname{tr} A}{2}I, here 4I.
  8. (0, 2) = (-i, 1) + (i, 1), that is -i(1, i) + i(1, -i). If a(1, i) = b(1, -i), then a = b and ia = -ib, so a = b = 0; the direct sum has dimension 2, so it is \mathbb{C}^2. Solving a + b = 0, i(a - b) = 2 gives a = -i, b = i.
  9. W = \operatorname{span}\{\mathbf{e}_1, \mathbf{e}_2\}. A combination a(1, 1, 1, 1) + b(1, 2, 0, 3) has third entry a and fourth a + 3b, so it is \mathbf{e}_1 only if a = b = 0, which fails: \mathbf{e}_1 is kept. Adding c\mathbf{e}_1, those entries still force a = b = 0, and the second entry is then 0, not 1: \mathbf{e}_2 is kept. Four independent vectors form a basis of \mathbb{R}^4, so \mathbf{e}_3, \mathbf{e}_4 are dropped.
  10. Call the two sets \mathcal{S} and \mathcal{K}. If A^{T} = \pm A, B^{T} = \pm B (same sign) and c \in \mathbb{F}, then (A + cB)^{T} = A^{T} + cB^{T} = \pm(A + cB), and 0^{T} = 0, so both are subspaces. If A \in \mathcal{S} \cap \mathcal{K}, then A = A^{T} = -A, so 2A = 0 and A = 0 since 2 \neq 0 in \mathbb{F}. For any A, (A + A^{T})^{T} = A + A^{T} and (A - A^{T})^{T} = -(A - A^{T}), so \displaystyle A = \frac{1}{2}(A + A^{T}) + \frac{1}{2}(A - A^{T}) splits A into a symmetric and a skew-symmetric piece. So \mathbb{F}^{n \times n} = \mathcal{S} + \mathcal{K}, direct by Result 3.
  11. U + U contains U (Result 1), and \mathbf{u} + \mathbf{u}' \in U for \mathbf{u}, \mathbf{u}' \in U, so U + U = U. If U \subseteq W, then W is a subspace containing U and W, so U + W \subseteq W by Result 1, and W \subseteq U + W. Conversely, if U + W = W, then U \subseteq U + W = W by Result 1.
  12. Let \mathbf{x} \in U_j \cap U_k with j \neq k. Put \mathbf{x} in place j, -\mathbf{x} in place k and \mathbf{0} elsewhere: each piece lies in its own summand and the pieces add to \mathbf{0}. By Result 2 every piece is \mathbf{0}, so \mathbf{x} = \mathbf{0}.
  13. In \mathbb{R}^2 take U = \operatorname{span}\{(1, 0)\}, W_1 = \operatorname{span}\{(0, 1)\} and W_2 = \operatorname{span}\{(1, 1)\}. Distinct lines meet only in \mathbf{0}, so each sum is direct of dimension 2, hence \mathbb{R}^2; but W_1 \neq W_2. If V is finite-dimensional, so are W_1 and W_2 (Result 6 of Bases and dimension, proved), and Result 5 gives \dim U + \dim W_1 = \dim V = \dim U + \dim W_2.
  14. Each \mathbf{v}_j \neq \mathbf{0} (Result 2 of Span and independence, proved), so (\mathbf{v}_j) is a basis of \operatorname{span}\{\mathbf{v}_j\}. Joined in order these one-entry bases give (\mathbf{v}_1, \dots, \mathbf{v}_n), which is independent, so the sum is direct by Result 5; it spans V, so the sum is V by Result 1.
  15. Let W be a complement (Result 7), with basis (\mathbf{w}_1, \dots, \mathbf{w}_l), where l = \dim V - \dim U \ge 1 as U \neq V. Let (\mathbf{u}_1, \dots, \mathbf{u}_k) be a basis of U, pick \mathbf{u} \neq \mathbf{0} in U, and let W' = \operatorname{span}\{\mathbf{w}_1 + \mathbf{u}, \mathbf{w}_2, \dots, \mathbf{w}_l\}. The list of the \mathbf{u}_i followed by \mathbf{w}_1 + \mathbf{u}, \mathbf{w}_2, \dots, \mathbf{w}_l has span containing U, hence \mathbf{w}_1 = (\mathbf{w}_1 + \mathbf{u}) - \mathbf{u} and every \mathbf{w}_j, hence U + W = V (Result 1). Its length is \dim V, so it is a basis (Result 5 of Bases and dimension, proved); its last l entries are a basis of W', and Result 5 here gives V = U \oplus W'. Finally W' \neq W: if \mathbf{w}_1 + \mathbf{u} \in W, then \mathbf{u} = (\mathbf{w}_1 + \mathbf{u}) - \mathbf{w}_1 \in U \cap W = \{\mathbf{0}\}, which is false.
math.wiki · CC-BY-SA-4.0