Skip to content

Latest commit

 

History

History
455 lines (308 loc) · 27.2 KB

Linear algebra.md

File metadata and controls

455 lines (308 loc) · 27.2 KB

Lecture material

Part 1. Vector spaces

Lecture 1

Definition. A field $F$ is a set with addition and multiplication, both of which are associate, commutative, have identities and inverses, and are distributive.

Definition. A vector space $V$ over a field $F$ is a set with the two operations: addition ($+$) and multiplication ($\cdot$) by elements in $F$ (scalar) that satisfies the following axioms:

  • Associativity ($+$);
  • Existence of 0;
  • Existence of inverse;
  • Commutativity; (this is an Abelian group properties)
  • Associativity ($\cdot$);
  • Distributivity for scalars;
  • Distributivity for vectors;
  • Existence of 1.

Definition. A subset $W \in V$ is called a subspace if $W$ is a vector space with respect to the induced operations (i.e. the same addition and scalar multiplication as in $V$).

Lecture 2

Statement. Non-empty subset $W \subset V$ is a subspace $\Leftrightarrow$ $W$ is closed w.r.t the induced operations.

Statement. Neutral element $0_V$ of $V$ equals neutral element $0_W$ of $W$.

Definition. Let $V$ be a vector space over some field $F$. Then $$c_1 v_1 + \ldots + c_k v_k = \sum_{i = 1}^k c_i v_i$$ for $c_i \in F$, $v_i \in V$, $i \in {1, \ldots k}$ is called linear combination.

Definition. For any subset $S \subset V$ we define $$\text{span} (S) = {\text{all linear combinations } \sum_{i = 1}^k c_i s_i \text{ with }c_i \in F, s_i \in S, i \in {1, \ldots, k}, k \in \mathbb{N}}$$ Remark.

  • For any subset $S \subset V$, $\text{span}(S)$ is a subspace;
  • *$\text{span}(S)$ is the smallest subspace that includes $S$;
  • if $W \subset V$ is a subspace, then $\text{span}(W) = W$.

Definition. A subset $S \subset V$, s.t. $\text{span}(S) = V$ is called generating set (or spanning set).

Definition. A minimal generating set is called a basis of $V$ (minimal means no element can be removed).

Definition. A subset $S \subset V$ is called linearly independent if $\sum_{i = 1}^k c_i s_i = 0$ for $c_i \in F$, $s_i \in S$ always implies $c_i = 0$ $\forall i \in {1, \ldots, k}$. Otherwise, it's called linearly dependent.

Theorem. Let $V$ be a vector space over a field $F$, and $E \subset V$ a subset. Then the following are equivalent:

  • $E$ is a basis of $V$;
  • $E$ is a maximal linearly independent set;
  • every $v \in V$ can uniquely be written as a linear combination of $e_i \in E$.

Lecture 3

Definition. $V$ is called a finite dimensional if it has finite basis. If not, it is called infinite dimensional.

Theorem. If $V$ is finite dimensional, then the number of elements in a basis, does not depend on the basis.

Definition. If $V$ has a basis with $n$ elements, then $n$ is called the dimension of $V$, i.e. $\text{dim} V = n$.

Lecture 4

Theorem (Zorn's lemma). Let $S$ be a partially ordered set, s.t. every linearly ordered subset has an upper bound. Then $S$ has a maximal element.

Theorem. Every vector space has a basis.

Part 2. Maps

Definition. $m: S_1 \to S_2$ is called a map or function, if $V$, $W$ are vector spaces over a field $F$, $f: V \to W$ is called linear if $f(c_1 v_1 + c_2 v_2) = c_1 f(v_1) + c_2 f(v_2)$ $\forall c_1, c_2 \in F$, $v_1, v_2 \in V$. We define $\mathfrak{L}(V, W) = {\text{all linear maps from } V \text{ to } W}$ and $\mathfrak{L}(V) = {\text{all linear maps from }V \text{ to } V}$.

Lemma. $\mathfrak{L}(V, W)$ is a vector space.

Lemma. Let ${v_1, \ldots, v_n}$ be a basis in $V$ (in particular, $\text{dim} V = n < \infty$) and ${w_1, \ldots, w_n} \subset W$. Then there exists a unique linear map $f: V \to W$ with $f(v_i) = w_i$ $\forall i \in {1, \ldots, n}$.

Definition. If $V$ is a vector space over $F$, then elements of $\mathfrak{L}(V, F)$ are called functionals of $V$. And $\mathfrak{L}(V, F) =: V^*$ is called a dual space if $V$.

Definition. Let ${v_1, \ldots, v_n}$ be a basis of $V$ (finite dimension). Then ${v_1^, \ldots, v_n^} \subset V^$, with $v_i^(\sum_{j = 1}^n c_j v_j) = c_i$ ($v_i^*(v_j) = \delta_{ij}$) is called a dual basis.

Lemma. The dual basis is indeed a basis.

Remark. Wrong for infinite dimensions (not generating).

Lecture 5

Definition. Bijective $f \in \mathfrak{L}(V, W)$ are called isomorphisms. If there exists an isomorphism between $V$ and $W$, then $V$ and $W$ are called isomorphic, and we write $V \cong W$.

Theorem. Let $V$ and $W$ be finite dimensional vector spaces. Then they are isomorphic if and only if $\text{dim} V = \text{dim} W$.

Definition. Isomorphisms that do not depend on "arbitrary choices" (e.g., basis) are called canonical or natural isomorphisms. Otherwise, they are called non-canonical or accidental.

Example (canonical isomorphism). Consider $\text{dim} V = n$ and isomorphism $V \to (V^)^ = V^{}$ (double dual). Let's define isomorphism $\varepsilon: V \to V^{}$, $v \mapsto v^{} = [f \mapsto f(v)]$ or $\varepsilon(v) = v^{}$ with $v^{**}(f):= f(v)$ ($f \in V^*$).

Definition. Let $f: V \to W$ be linear. Then dual map $f^: W^ \to V^$ is defined by $$f^(w^)(v) = w^(f(v)) \quad \forall v \in V$$ Lecture 6

Theorem. Dual map linear and unique.

Definition. Let $f: V \to W$ be linear. Then

  • kernel or null space is $\text{ker} f = {v \in V: f(v) = 0} \subset V$;
  • image or range is $\text{im} f = {w \in W: \exists v \in V \text{ with } f(v) = w} \subset W$.

Remark. Kernel and image are actually subspaces.

Lemma. Let $f \in \mathfrak{L}(V, W)$. Then $f$ injective $\Leftrightarrow$ $\text{ker} f = {0}$.

Theorem (Fundamental theorem of linear maps). Let $f \in \mathfrak{L}(V, W)$ with $V$ fin-dim. Then $$\text{dim} V = \text{dim}(\text{im} f) + \text{dim}(\text{ker} f)$$ Corollary. Let $f \in \mathfrak{L}(V, W)$, $\text{dim} V < \infty$. Then $$f \text{ injective} \Leftrightarrow \text{dim} V = \text{dim} (\text{im} f)$$ Corollary. Let $f \in \mathfrak{L}(V, W)$, $\text{dim} V = \text{dim} W < \infty$. Then $$f \text{ isomorphism} \Leftrightarrow \text{ker} f = {0} \Leftrightarrow \text{im} f = W$$

Part 3. Matrices

Lecture 7

Definition. Let $f \in \mathfrak{L}(V, W)$, where $V$ and $W$ are finite dimensional. We choose basis $B_V = {v_1, \ldots, v_n}$ of $V$ and $B_W = {w_1, \ldots, w_m}$ of $W$. $f$ defined by action on $v_k$, can be written as linear combination with $B_W$: $$f(v_k) = \sum_{i = 1}^m a_{ik} w_i$$ which can be associated to matrix $A_{B_V, B_W} = (a_{ik})_{i \in {1, \ldots, m}, k \in {1, \ldots, n}}$.

Remark. Composition of linear maps corresponds to matrix multiplication.

Definition. Basis change matrix from $B_V$ to $B_V'$ is $T_{B_V B_V'}$.

Statement. Let $f \in \mathfrak{L}(V, W)$, choose bases $B_V$ and $B_V'$ of $V$ and bases $B_W$ and $B_W'$ of $W$. Let $T_{B_V B_V'}$ be matrix of basis change $B_V$ to $B_V'$, $T_{B_W B_W'}$ of $B_W$ to $B_W'$. Let $A_{B_V B_W}$ be a matrix of $f$ in bases $B_V$, $B_W$, and $A_{B_V' B_W'}$ in bases $B_V'$, $B_W'$. Then $$A_{B_V' B_W'} = T_{B_W B_W'}^{-1} A_{B_V B_W} T_{B_V B_V'}$$ If $V = W$, i.e. $B_V = B_W$, $B_V' = B_W'$, we have $T_{B_V B_V'} = T_{B_W B_W'} = T$, hence $$A_{B_V' B_V'} = T^{-1} A_{B_V B_V} T,$$ this is called conjugation of by $T$.

Definition. We can now define functionals $\varphi$, which are defined via $A_{B_V}$, but are actually independent of basis choice, i.e. $\varphi(f) = \varphi(A_{B_V}^f) = \varphi(A_{B_V'}^f)$. Such $\varphi$ are called invariants.

Examples. Two important examples for $f \in \mathfrak{L}(V, V)$ are:

  • Trace. We define $\text{tr} f := \text{tr} A_{B_V}^f := \sum_i (A_{B_V}^f)_{ii}$ (sum of diagonal entries for some basis $B_V$);
  • Determinant. We define $\text{det} f := \text{det} A_{B_V}^f$.
Part 4. Sums

Definition. Let $V_1, \ldots, V_1$ be subsets of vector space $V$. Then their sum is $$\sum_{i = 1}^n V_i = \left{\sum_{i = 1}^n v_i \in V, \text{ s.t. } v_i \in V_i, i \in {1, \ldots, n}\right}$$ Remark. If $V_1, \ldots, V_n$ are subspaces of $V$, then their sum is their $\text{span}$.

Definition. Let $V_1, \ldots, V_n$ be vector spaces. Then external direct sum $$V = \bigoplus_{i = 1}^n V_i$$ is defined by

  • $(v_1, \ldots, v_n) \in V$, $v_i \in V_i$;
  • $c(v_1, \ldots, v_n) + c'(v_1', \ldots, v_n') = (cv_1 + c'v_1', \ldots, cv_n + c'v_n')$.

Theorem. Let $V_1$, $V_2$ be subspaces of $V$, $\text{dim} V < \infty$. Then $$\text{dim}(V_1 \cap V_2) + \text{dim}(V_1 + V_2) = \text{dim} V_1 + \text{dim} V_2$$ Lecture 8

Definition. Let $V_1, \ldots, V_n \subset V$ and $V_1, \ldots, V_n' \subset V$ be subspaces. We can say that ${V_i}$ and ${V_i'}$ are identically arranged, if there is a linear isomorphism $f: V \to V$, s.t. $f(V_i) = V_i'$.

Definition. We can say that $V_1$ and $V_2$ are in general position if $\text{dim} V_1 \cap V_2$ is minimal.

Theorem. Let $V_1, \ldots, V_n \subset V$ be subspaces with $\sum_{i = 1}^n V_i = V$. Then $$V = \bigoplus_{i = 1}^n V_i \Leftrightarrow V_{i_0} \cap \left( \sum_{i = 1, i \neq i_0}^n V_i\right) = {0} \quad 1 \leq i_0 \leq n \Leftrightarrow \sum_{i = 1}^n \text{dim} V_i = \text{dim} V$$ Definition. $p \in \mathfrak{L}(V)$ is a projector if $p^2 = p$ ($p^2 = p \circ p$).

Remark. Let $V = \bigoplus_{i = 1}^n V_i$ be given, i.e. any $v$ can be uniquely written as $v = \sum_{i = 1}^n$ with $v_i \in V_i$. We can define $p_j(\sum_{i = 1}^n v_i) = v_j$, then $p_j$ clearly is a projector, $p_i p_j = 0$ for $i \neq j$, $\sum_{i = 1}^n p_i = \text{id}$, and $V_i = \text{im} p_i$.

Theorem. Let $p_1, \ldots, p_n \in \mathfrak{L}(V)$ be projectors with $\sum_{i = 1}^n p_i = \text{id}$ and $p_i p_j = 0$ for all $i \neq j$, then $$V = \bigoplus_{i = 1}^n \text{im} p_i$$ Remark. Mapping between direct sums is a direct sum of mappings.

Part 5. Quotient spaces

Definition. Let $M \subset L$ be a subspace, $e \in L$, translation of $M$ by $e$ is $e + M = {e + m: m \in M}$ is an "affine subset".

Lemma. Let $M_1, M_2 \subset L$ be subspaces, $e_1, e_1 \in L$. Then $$e_1 + M_1 = e_2 + M_2 \Leftrightarrow M_1 = M_2 = M \text{ and } e_1 - e_2 \in M$$ Lecture 9

Definition. The quotient space (factor space) $L/M$ is defined as $$L/M = {l + M : l \in L}$$ with

  • addition $(l_1 + M) + (l_2 + M) = (l_1 + l_2) + M$;
  • scalar multiplication $c(l + M) = cl + M$.

Lemma. $L/M$ is a vector space.

Remark. Equivalence relation.

Remark. There is a canonical map $q: L \to L/M$, $l \mapsto q(l) = l + M$, called the quotient map. It is surjective, linear, linear image (fiber) of $\widetilde{l} + M$ is $\widetilde{l} + M$ (subset of $L$) and its kernel is $M$.

Theorem. For $\text{dim} L < \infty$, we have $\text{dim}L/M = \text{dim}L - \text{dim}M$ (called codimension of $M$ in L).

Definition. Given $f: L \to M$ (linear), we can define $\text{coim} f := L/\text{ker} f$ (coimage of $f$), $\text{coker} f := M / \text{im} f$ (cokernel of $f$).

Remark. Some important diagrams are in lecture notes.

Lemma (Universal property). Let $f: L \to M$, $g: L \to N$, then $h$ so that $hf = g$ exists iff $\text{ker} f \subset \text{ker} g$. If additionally $\text{im} f = M$, then $h$ is unique.

Lecture 10

Definition. The four fundamental spaces of a linear map $f$ are: $\text{ker} f$, $\text{coker} f$, $\text{ker} f^$, $\text{coker} f^$. $$\text{ker} f \subset L \xrightarrow{f} M \supset \text{im} f \quad \quad \text{im} f^* \subset L^* \xleftarrow{f^} M^ \supset \text{ker} f^$$ Definition. Let $M \subset L$ be a subspace. Then $M^{\perp} \subset L^$, the orthogonal complement of $M$, is defined as ${m^* \in L^* : m^*(m) = 0 \forall m \in M}$.

Remark. $M^{\perp}$ is a subspace.

Remark. Important diagram is in lecture notes.

Lemma.

  • For $M$ a subspace of $L$, $L^/ M^{\perp} \cong M^$ (canonical isomorphism);
  • For $M$ a subspace of $L$, $(L/M)^* \cong M^{\perp}$ (canonical isomorphism).
Part 6. Linear operators

Lecture 11

Remark: In this chapter: $L$, $M$ are finite dimension vector spaces.

Theorem. Let $f \in \mathfrak{L}(L, M)$. Then there are $\widetilde{L} \subset L$ and $M \subset M'$ such that $L = \text{ker} f \oplus \widetilde{L}$, $M = \text{im} f \oplus M'$ and $\widetilde{f}: \widetilde{L} \to \text{im} f$, $\widetilde{f} = f|{\widetilde{L}}$ is an isomorphism. Also, there are bases of $L$ and $M$ such that the matrix $A_f$ of $f$ is a matrix with $A{ii} = 1$ for $i \in {1, \ldots, r}$ with $r =: \text{rank} f$ and zeros everywhere else.

Remark. This means that any $m \times n$ matrix can be brought into that form by basis change.

Definition. Let $\widetilde{L} \subset L$, $f \in \mathfrak{L}(L)$

  • $\widetilde{L}$ is called invariant if $f(\widetilde{L}) \subset \widetilde{L}$;
  • If $\widetilde{L}$ is invariant and $\text{dim} \widetilde{L} = 1$, then $\widetilde{L}$ is called a proper subspace (for $f$). Then $f|{\widetilde{L}}(l) = \lambda l$ for some $\lambda \in F$, or $f|{\widetilde{L}} = \lambda \text{id}{\widetilde{L}}$, i.e. $f|{\widetilde{L}}$ is multiplication by a constant. $\lambda$ is called eigenvalue of $f$ and any $\widetilde{L} \ni \widetilde{l} \neq 0$ is called eigenvector;
  • If $\widetilde{L}$ is invariant and $f|_{\widetilde{L}}$ is multiplication by a constant, then $\widetilde{L}$ is eigenspace.

Definition. $f \in \mathfrak{L}(L)$ is called diagonalizable, if $L = \bigoplus_{i = 1}^n L_i$ with proper subspaces $L_i$ of $f$ (i.e. $\text{dim} L_i = 1$ for all $i$).

Remark. This is equivalent to existence of a basis such that the matrix of $f$ is diagonalizable.

Definition. For $f \in \mathfrak{L}$, we call $P_f(t) = \text{det}(t \cdot \text{id} - f)$ the characteristic polynomial of $f$.

Remark. The characteristic polynomial can be computed with matrix of $f$, and it is independent of basis choice.

Theorem. $\lambda$ is an eigenvalue of $f$ iff $\lambda$ is a root of $P_f(t)$ in $F$.

Lecture 12

Remark. In the lecture notes, there are some examples.

Definition. A polynomial $Q$ annihilates $f$ if $Q(f) = 0$. $Q(t) = \sum_{i = 0}^n c_i t^i$ with $c_n = 1$ and minimal $n$ such that still $Q(t) = 0$ is called minimal polynomial.

Remark. There is a polynomial with degree $\leq n^2$ which annihilates $f$. Minimal polynomial is unique.

Theorem (Cayley-Hamilton). $P_f(f) = 0$.

Definition.

  • $l \in L$ is called a root vector of $f$ if $(f - \lambda)^r(l) = 0$ for some $r > 0$;
  • If $\lambda$ is additionally an eigenvalue, then such $l$ which are non-zero are called generalized eigenvector;
  • Generalized eigenspace $G(\lambda) = {\text{generalized eigenvectors to } \lambda} \cup {0}$.
Part 7. Jordan form

Lecture 13

Definition. $f$ is called nilpotent if $f^j = 0$ for some $j$.

Theorem (abstract Jordan decomposition). Let $f \in \mathfrak{L}(L)$ with $\text{dim} L = n < \infty$, and $L$ a vector space over the algebraically closed field $F$. Then $L = \bigoplus_i G(\lambda_i)$ and $f = \bigoplus_i f_i$ with $f_i: G(\lambda_i) \to G(\lambda_i)$, where the $\lambda_i$ are the distinct eigenvalues.

Lemma. Let $f \in \mathfrak{L}(L)$ and $\text{dim} L = n$. Then $L = \text{ker} f^n \oplus \text{im} f^n$.

Lemma. $G(\lambda) = \text{ker}(f - \lambda)^n$.

Lemma. Let $\lambda_1, \ldots, \lambda_n$ be the distinct eigenvalues with corresponding generalized eigenvectors $l_1, \ldots, l_m$. Then $l_1, \ldots, l_m$ are linearly independent.

Lemma. $\text{ker} p(f)$ and $\text{im} p(f)$ are invariant for any polynomial $p$.

Corollary. $L$ has a basis of generalized eigenvectors.

Corollary. $f$ has simple spectrum $\Rightarrow$ $f$ is diagonalizable.

Remark.

  • Multiplicity of eigenvalue $\lambda = \text{dim} G(\lambda)$, so $\text{dim} L$ is a sum of multiplicities of all eigenvalues.
  • An $r \times r$ matrix of the form $$J_r(\lambda) = \begin{pmatrix} \lambda & 1 & 0 & \cdots & 0 \ 0 & \lambda & 1 & \cdots & 0 \ 0 & 0 & \lambda & \cdots & 0 \ \vdots & \vdots & \vdots & \ddots & \vdots \ 0 & 0 & 0 & \cdots & \lambda \end{pmatrix} $$ is called a Jordan block.
  • $J$ with Jordan blocks on a diagonal is called a Jordan matrix.
  • A Jordan basis of $f$ is a basis such that $f$ is represented by Jordan matrix, i.e. has Jordan normal form: there is non-singular $X$ s.t. $X^{-1} A_f X = J$.

Lecture 14

Theorem (Jordan normal form). Every $f \in \mathfrak{L}(L)$ has a Jordan basis.

Lemma. Let $g \in \mathfrak{L}(L)$ be nilpotent. Then there are $l_1, \ldots, l_n \in L$ and integers $m_1, \ldots, m_n > 0$ such that $g^{m_n}(l_1), g^{m_1 - 1}(l_1), \ldots, g(l_1), l_1, \ldots, g^{m_n}(l_n), \ldots, g(l_n), l_n$ is a basis of $L$ and $g^{m_i + 1}(l_i) = 0$ for all $i \in {1, \ldots, n}$.

Remark.

  • Nonzero vectors $l, g(l), \ldots, g^k(l)$ such that $g^{k + 1}(l) = 0$ are called a sting of $g$;
  • Every nilpotent $g$ has a Jordan matrix.

Lecture 15

Remark. Some useless shit about decomplexification is in lecture notes.

Part 8. Bilinear forms

Lecture 16

Definition. Let $L_1$, $L_2$, $M$ be vector spaced over the field $F$. Then $f: L_1 \times L_2 \to M$, $(l_1, l_2) \mapsto f(l_1, l_2)$ is called bilinear map if it is linear with respect to both variables.

Definition. If $M = F$, then such $f$ are called bilinear forms.

Remark. We study bilinear forms $L \times \overline{L} \to F$ or $L \times \overline{L} \to \mathbb{C}$; these are called inner products. We write $L \times \overline{L} \to \mathbb{C}$ as a sesquilinear form $g: L \times L \to \mathbb{C}$, i.e, $g(al_1, bl_2) = a\overline{b} g(l_1, l_2)$ (linear in first, autilinear in second argument).

Definition. Gram matrix is $G = (g(e_i, e_j))_{i, j = 1, \ldots, n}$.

Remark.

  • Bilinear: $g^T(l, m) := g(m, l)$; $G$ changes to $G^T$;
  • Sesquilinear: $\overline{g}^T(l, m) := \overline{g(m, l)}$ (still linear in first argument); $G$ changes to $\overline{G}^T$.

Definition.

  • Symmetric: $g^T = g$ ($G$ symmetric), orthogonal geometry;
  • Antisymmetric or symplectic: $g^T = -g$ ($G$ antisymmetric);
  • Hermitian: $\overline{g}^T = g$ ($G$ Hermitian) (note: $\overline{g}^T(l, l) = \overline{g}(l, l) = \overline{g(l, l)} = g(l, l)$ is real).

Definition. Let $(L, g)$ be an inner product space. Then $l_1$, $l_2$ are called orthogonal if $g(l_1, l_2) = 0$. $L_1, L_2 \subset L$ are orthogonal if $g(l_1, l_2) = 0$ for all $l_1 \in L_1$, $l_2 \in L_2$.

Remark. If $g = \pm g^T$, then $g(l_1, l_2) = 0$ iff $\pm g^T(l_1, l_2) = 0$ iff $g(l_2, l_2) = 0$.

Definition.

  • $\text{ker} g = {m \in L: g(m, l) = 0 \forall l \in L}$;
  • If $\text{ker} g = {0}$, then $g$ is called non-degenerate;

Remark.

  • $\text{ker} g = \text{ker} \widetilde{g}$, where $\widetilde{g}: L \to \overline{L^*}$ by rule $\widetilde{g}(l)(m) := g(l, m)$;
  • $g$ is non-degenerate iff $G$ is non-singular;
  • $\text{rank} g := \text{dim }\text{im} \widetilde{g} = \text{rank} G$.

Lecture 17

Definition. Let $(L_1, g_1)$ and $(L_2, g_2)$ be inner product spaces. A linear isomorphism $f: L_1 \to L_2$ is called isometry if $g_1(l, l') = g_2(f(l), f(l'))$ $\forall l, l' \in L_1$. $(L_1, g_1)$ and $(L_2, g_2)$ are called isometric if there is an isometry for them.

Remark. Some classifications are in lecture notes.

Definition. A subspace $L_0 \subset L$ is called

  • non-degenerate if $g|_{L_0}$ is non-degenerate;
  • isotropic if $g|_{L_0} = 0$.

Definition. The orthogonal complement $L_0^{\perp}$ of $L_0 \subset L$ is $$L_0^{\perp} := {l \in L : g(l_0, l) = 0 \text{ for all } l_0 \in L_0} \subset L$$

Lecture 18

Lemma. Let $(L, g)$ be an inner product with space with $\text{dim} L < \infty$. Then

  • $L_0 \subset L$ non-degenerate $\Rightarrow$ $L = L_0 \oplus L_0^{\perp}$;
  • $L_0 \subset L$ and $L_0^{\perp}$ non degenerate $\Rightarrow$ $(L_0^{\perp})^{\perp} = L_0$.

Theorem. Let $(L, g)$ be an inner product space with $\text{dim} L < \infty$. Then $L = \bigoplus_{i = 1}^m L_i$, where $L_i$'s'are pairwise orthogonal and

  • 1-dimensional for symmetric and Hermitian forms;
  • 1-dimensional degenerate or 2-dimensional non-degenerate for symplectic forms.

Definition. $(r_0, r_+, r_-)$ is called signature of $(L, g)$, where

  • $r_0 = \text{dim} \text{ ker} g$;
  • $r_+$ is a number of positive $L_i$'s;
  • $r_-$ is a number of negative $L_i$'s.

Lecture 19

Theorem.

  • Symplectic over $\forall$ field $F$, symmetric over $\mathbb{C}$ up to isometry determined by $n$, $r_0$;
  • (Inertia theorem or Sylvester's law of inertia) Symmetric over $\mathbb{R}$, Hermitian over $\mathbb{C}$: up to isometry determined by signature, independent of choice of orthogonal decomposition.

Corollary.

  • Orthogonal and Hermitian spaces have an orthogonal basis $g(e_i, e_j) = 0$ for $i \neq j$, Gram matrix looks like $$ \begin{pmatrix} E_{r_+} & 0 & 0 \ 0 & -E_{r_-} & 0 \ 0 & 0 & 0_{r_0} \end{pmatrix} $$
  • Symplectic spaces hace a symplectic basis ${e_1, \ldots, e_r, \widetilde{e_1}, \ldots, \widetilde{e_r}, e_1', \ldots, e_{n - 2r}'}$ with $g(e_1, \widetilde{e_i}) = -g(\widetilde{e_i}, e_i) = 1$ and all other combinations equal to 0. Gram matrix looks like $$ \begin{pmatrix} 0 & E_{r} & 0 \ -E_{r} & 0 & 0 \ 0 & 0 & 0 \end{pmatrix} $$
Part 9. Orthogonalization

Lecture 20

Definition. Let $h: L \times L \to F$ be a bilinear form, then $q: L \to F$, $q(l) = h(l, l)$ is called a quadratic form. A symmetric bilinear form $g$ such that $q(l) = g(l, l)$ is called polarization of $q$.

Algorithm (Gram-Schmidt orthogonalization).

  • Start with $e_1 = e_1'$;
  • Say $e_{1}, \ldots, e_{i-1}$ already given. Then look for $e_i=e_i' - \sum_{j=1}^{i-1} Y_j e_j \quad\left(Y_j \in F\right)$
  • Need $g\left(e_{i}, e_j\right)=0$ $\forall j=l_1, \ldots,-1 \Rightarrow 0=g\left(e_i^{\prime}, e_j\right)-Y_j g\left(e_{j, e_j}\right)$ $$ \begin{aligned} & \Rightarrow Y_j=\frac{g\left(e_i^{\prime}, e_j\right)}{g\left(e_{j, e_j}\right)} \ & \Rightarrow e_i=e_i^{\prime}-\sum_{j=1}^{i-1} \frac{g\left(e_{i}, e_j'\right)}{g\left(e_j, e_j\right)} e_j \end{aligned} $$ Remark. Some examples of polynomials and orthogonalizations are in lecture notes.

Lecture 21

Definition. A Euclidean space $(L, g)$ is a real vector finite dimension space $L$, with symmetric and positive defined inner product $g$ ($g(l, l) > 0$ for $l \neq 0$, or $r_0 = r_- = 0$). A unitary space $(L, g)$ is a complex vector space $L$ with Hermitian and positive defined inner product $g$.

Remark. Write as follows: $g(l, m) =: \langle l, m \rangle$, $\sqrt{\langle l, l \rangle} =: ||l||$ (lenght), and such $g$ is called *scalar product.

Statement (Cauchy-Schwarts-Bunyakovskii). $|\langle l_1, l_2\rangle| \leq ||l_1|| \cdot ||l_2||$ with equality iff $l_1$ and $l_2$ linearly independent.

Remark. Triangle inequality for lenght holds, hence $||\cdot||$ is a norm and $d(l_1, l_2) = ||l_1 - l_2||$ is a metric. Angles in Euclideans space are defined as follows: $$\cos \varphi = \frac{\langle l_1, l_2\rangle}{||l_1||\cdot ||l_2||}$$ Distance between objects is defined as a minimum between their points.

Lecture 22

Definition. Isometries on Euclidean (unitary) spaces are called orthogonal (unitary) operators.

Lemma. $f$ is isometry iff

  • $||f(l)|| = ||l|| \quad \forall l \in L$;
  • ${e_j}_{j = 1, \ldots, n}$ basis of $l$, $G$ Gram matrix of scalar product, $U$ matrix of $f$, then $U^TG U = G$;
  • $f$ maps any ONB into another ONB;
  • matrix $U$ of $f$ in any ONB satisfies $U^T U = E_n$.

Theorem.

  • $f$ unitary iff $f$ diagonizable in some ONB with $|\lambda_j = 1|$ ($\lambda_j$ eigenvalue, $j \in {1, \ldots, n}$);
  • $f$ orthogonal iff in some ONB the matrix of $f$ is $$ \begin{pmatrix} U(\varphi_1) & \cdots & 0 & 0 & \cdots & 0 & 0 & \cdots & 0 \ \vdots & \ddots & \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots \ 0 & \cdots & U(\varphi_n) & 0 & \cdots & 0 & 0 & \cdots & 0 \ 0 & \cdots & 0 & 1 & \cdots & 0 & 0 & \cdots & 0 \ \vdots & \ddots & \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots \ 0 & \cdots & 0 & 0 & \cdots & 1 & 0 & \cdots & 0 \ 0 & \cdots & 0 & 0 & \cdots & 0 & -1 & \cdots & 0 \ \vdots & \ddots & \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots \ 0 & \cdots & 0 & 0 & \cdots & 0 & 0 & \cdots & -1 \end{pmatrix} $$ with $$ U(\varphi) = \begin{pmatrix} \cos \varphi & -\sin \varphi \ \sin \varphi & \cos \varphi \end{pmatrix} $$ Lecture 23

Definition. $f: L \to L$ with $\langle f(l_1), l_2 \rangle = \langle l_1, f(l_2) \rangle$ $\forall l_1, l_2 \in L$ is called self-adjoint.

Lemma. Let $f: L \to L$ be diagonizable in some basis with real eigenvectors. Then $f$ is self-adjoint.

Theorem.

  • (Spectral theorem) $f: L \to L$ is self-adjoint iff $f$ diagonizeable in some ONB with real eigenvectors;
  • If $f: L \to L$ is self-adjoint, then eigenvectors for different eigenvalues are orthogonal.

Definition. Normal operators are ${f : ff^* = f^*f}$.