vector space 的純量乘法是「實數乘向量、得向量」。很多空間還配得起另一種乘法——「向量乘向量、得實數」。這個數日後要一手包辦長度與角度,所以規矩得先立好。
8.3 DEFINITION
An inner product (dot product) on a vector space V is a function from V × V to ℝ, written (x, y) ↦ x · y, satisfying: (i) x · x ≥ 0 for all x ∈ V; (ii) x · x = 0 if and only if x = 0; (iii) x · y = y · x; (iv) x · (y + z) = x · y + x · z and (x + y) · z = x · z + y · z; (v) (ax) · y = a(x · y) = x · (ay) for all a ∈ ℝ. A vector space with an inner product is an inner product space.
A norm on a vector space V is a function x ↦ ‖x‖ from V to ℝ satisfying: (i) ‖x‖ ≥ 0 for all x ∈ V; (ii) ‖x‖ = 0 if and only if x = 0; (iii) ‖ax‖ = |a| ‖x‖ for all a ∈ ℝ, x ∈ V; (iv) ‖x + y‖ ≤ ‖x‖ + ‖y‖ for all x, y ∈ V (the Triangle Inequality). A vector space with a norm is a normed space.
Let V be an inner product space and set ‖x‖ = (x · x)^{1/2} for x ∈ V. Then x ↦ ‖x‖ is a norm on V, and (∗) x · y ≤ ‖x‖ ‖y‖ for all x, y ∈ V. Moreover, for non-zero x, y equality holds in (∗) exactly when x = cy for some strictly positive real c.
Proof. Properties 8.5(i)–(iii) come straight from the inner product. The square root is positive or zero, giving (i). If ‖x‖ = 0, then x · x = 0, and 8.3(ii) forces x = 0 — with the converse immediate — giving (ii). And (ax) · (ax) = a²(x · x) by two uses of 8.3(v), so taking square roots — unique non-negative roots multiply — yields ‖ax‖ = |a| ‖x‖, giving (iii).
For (∗), let a, b ∈ ℝ and form z = ax − by. Expanding with 8.3(i), (iii), (iv), (v), 0 ≤ z · z = a²(x · x) − 2ab(x · y) + b²(y · y). Now choose a = ‖y‖ and b = ‖x‖: 0 ≤ ‖y‖²‖x‖² − 2‖x‖‖y‖(x · y) + ‖x‖²‖y‖² = 2‖x‖‖y‖(‖x‖‖y‖ − x · y). If x and y are both non-zero, then ‖x‖‖y‖ > 0, and dividing gives x · y ≤ ‖x‖‖y‖; if either vector is 0, both sides of (∗) vanish and the inequality is trivial.
這一步的所求:一條對任何 x、y 都成立的內積上限。整場證明只有一個主意:自乘必非負(8.3(i))是免費的不等式產生器——挑一個聰明的組合向量 z = ax − by,把 z · z ≥ 0 攤開,不等式就自己長出來。攤開用掉對稱與雙線性(交叉項 −2ab(x · y) 合併自兩筆相同的帳)。係數怎麼挑是唯一的巧思:取 a = ‖y‖、b = ‖x‖,前後兩項疊成同一個 ‖x‖²‖y‖²,整式收攏成 2‖x‖‖y‖(‖x‖‖y‖ − x · y) ≥ 0——括號外的因子非負,於是括號內非負(兩人皆非零時因子嚴格正,同除即得;有一方為零時 (∗) 兩邊都是 0,setup 的小帳收尾)。數字對照:x = (1, 0)、y = (1, 1)——x · y = 1 ≤ 1 × √2 ✓。
這張圖是 Schwarz 的幾何直觀(證明本身是 z · z ≥ 0 的代數帳):內積可讀成「x 投在 y 方向的帶號投影」乘上 ‖y‖——帶號的意思是它會是負的,x 與 y 指向相反的一側時(例如 x = (−1, 0)、y = (1, 0),內積 −1)投影朝後。無論朝前朝後,投影的大小都不超過 ‖x‖——這正是 8.8 絕對值版 |x · y| ≤ ‖x‖‖y‖ 的畫面;(∗) 只管朝前那一側,等號發生在 x 整個躺平在 y 的正方向時,即取等條件「x = cy、c > 0」。
For the equality case, suppose first x = cy with c > 0. Then x · y = c(y · y) = c‖y‖² while ‖x‖ = c‖y‖, so both sides of (∗) equal c‖y‖². Conversely, if x · y = ‖x‖‖y‖ with x, y non-zero, run the display above with z = ‖y‖x − ‖x‖y: it gives z · z = 0, so z = 0 by 8.3(ii), that is ‖y‖x = ‖x‖y. Dividing by ‖y‖ > 0 exhibits x = c y with c = ‖x‖/‖y‖ > 0.
取等的雙向各走一步。順向是驗算:x = cy 代進去,兩邊都算出 c‖y‖²(c > 0 讓 |c| = c,例行)。逆向的機制漂亮:等號成立時,主帳 2‖x‖‖y‖(‖x‖‖y‖ − x · y) 歸零,也就是那個組合向量自乘歸零——8.3(ii) 說自乘歸零的只有零向量,於是 ‖y‖x = ‖x‖y:兩向量被迫共線,倍率 c = ‖x‖/‖y‖ 嚴格正。數字對照回 inst:(3, 4) 與 (6, 8)——c = 1/2⋯⋯從 y 看 x 是一半、從 x 看 y 是兩倍,方向一致即可。
Proof. Apply (∗) twice: to the pair x, y it gives x · y ≤ ‖x‖‖y‖; to the pair −x, y it gives −(x · y) = (−x) · y ≤ ‖−x‖‖y‖ = ‖x‖‖y‖, using 8.3(v) and 8.5(iii). The two bounds read −‖x‖‖y‖ ≤ x · y ≤ ‖x‖‖y‖, and Theorem 5.11(d) converts this band into |x · y| ≤ ‖x‖‖y‖. For the equality case with y ≠ 0: if x = cy, then |x · y| = |c|‖y‖² = ‖x‖‖y‖. Conversely |x · y| = ‖x‖‖y‖ means x · y = ±‖x‖‖y‖; the + case is 8.7 (giving c > 0, or x = 0 = 0·y), and the − case applied to the pair −x, y gives −x = c′y with c′ > 0, that is x = (−c′)y.
機制與 5.12 的下半場同款:把 x 換成 −x 再引用一次原不等式,正負兩側各得一道牆,帶寬翻譯 5.11(d) 一開,絕對值收工。取等的討論按符號分兩案,各自退回 8.7 的取等條件(x = 0 的邊界寫成 0 · y 收進「任意實倍數」裡)——例行分案。