Let X = (xₙ) be a bounded sequence in ℝ and let x* ∈ ℝ. Write vₘ = sup {xₙ : n ≥ m}, and let L be the set of all v ∈ ℝ such that some subsequence of X converges to v. The following statements are equivalent. (a) x* = lim sup (xₙ). (b) For each ε > 0 there are at most finitely many n ∈ ℕ with x* + ε < xₙ, but infinitely many n ∈ ℕ with x* − ε < xₙ. (c) x* = inf {vₘ : m ∈ ℕ}. (d) x* = lim (vₘ). (e) x* = sup L.
Proof. Assume (a), so x* = inf V, and let ε > 0. Since x* + ε exceeds the greatest lower bound of V, it is not a lower bound, so some v ∈ V satisfies v < x* + ε. Because V contains everything to the right of any of its members, x* + ε ∈ V; that is, at most finitely many n satisfy x* + ε < xₙ.
這一步的所求:把 x* + ε 塞進 V,因為 V 的入場條件講的就是「超過它的項只有有限多個」。用的是 inf 的定義(6.2:inf V 是最大的下界)——既然 x* + ε 比最大的下界還大,它就不可能是下界,所以 V 裡必定有成員比它小。找到這個成員之後,射線的形狀直接把 x* + ε 也拉進 V。舉例對照:X = ((−1)ⁿ) 時 V = [1, ∞)、x* = 1,取 ε = 0.3,1.3 確實在 V 裡,而超過 1.3 的項一個也沒有。
On the other hand x* − ε < x* = inf V, and every member of V is at least inf V. Hence x* − ε ∉ V, which by the definition of V says that infinitely many n satisfy x* − ε < xₙ. Therefore (a) implies (b).
這一步的所求:把 x* − ε 擋在 V 外面。理由比上一步短——inf V 是下界,所以 V 的每個成員都不小於它,而 x* − ε 嚴格小於它。關鍵在「不屬於 V」這句話怎麼翻譯:V 的入場條件是「至多有限多個 n 滿足 v < xₙ」,它的否定就是「有無限多個 n 滿足」——「有限」的否定是「無限」,中間沒有第三種可能,這一步用掉的正是這件事。
Now assume (b) and let ε > 0. The set F = {n ∈ ℕ : x* + ε < xₙ} is finite, so there is m₀ ∈ ℕ larger than every member of F. For n ≥ m₀ we then have xₙ ≤ x* + ε, so x* + ε is an upper bound of {xₙ : n ≥ m₀} and therefore inf {vₘ : m ∈ ℕ} ≤ v_{m₀} ≤ x* + ε.
Conversely, fix m ∈ ℕ. Since infinitely many n satisfy x* − ε < xₙ, at least one such n is ≥ m; for it, x* − ε < xₙ ≤ vₘ. As m was arbitrary, x* − ε is a lower bound of {vₘ : m ∈ ℕ}, so x* − ε ≤ inf {vₘ : m ∈ ℕ}.
Combining, x* − ε ≤ inf {vₘ : m ∈ ℕ} ≤ x* + ε for every ε > 0. If the two numbers differed, taking ε smaller than half the difference would be a contradiction; hence inf {vₘ : m ∈ ℕ} = x* and (c) holds.
Proof. Assume (c). Since {xₙ : n ≥ m + 1} ⊆ {xₙ : n ≥ m}, the supremum over the smaller set is no larger, so v_{m+1} ≤ vₘ and the sequence (vₘ) is monotone decreasing. It is bounded below by inf {xₙ : n ∈ ℕ}. By 16.2 it converges, and its limit is inf {vₘ : m ∈ ℕ} = x*. Hence (c) implies (d).
這一步只是把 (c) 的 inf 換個名字。遞減的理由是集合越縮越小:第 m+1 段是第 m 段去掉一項,能選的東西變少,最小上界只會往下或不動。有界則來自 X 本身有界。兩個條件湊齊,§16-1 的 16.2(單調遞減且有界的數列收斂,極限恰是它的 inf)就把「取 inf」與「取極限」畫上等號。上一張圖的紅色階梯正是這條數列。
Assume (d) and let X′ = (x_{nₖ}) be any convergent subsequence, with limit v. Since nₖ ≥ k, the term x_{nₖ} belongs to {xₙ : n ≥ k} and hence x_{nₖ} ≤ v_k. Passing to the limit with 15.8 gives v ≤ x*. Thus x* is an upper bound of L.
這一步的所求:沒有任何子數列的極限超得過 x*。兩條數列逐項比大小,比完之後一起過極限——不等號會保留下來,這是 §15-4 的 15.8(非嚴格不等式過得了極限)。要能逐項比,得先知道 x_{nₖ} 落在第 k 段裡,也就是 nₖ ≥ k;這件事在 §15-1 的 15.2 證明裡已經立過(子數列的編號嚴格遞增,第 k 個編號至少是 k)。
For the reverse, construct a subsequence converging to x*. Put n₀ = 0. Given n_{k−1}, set mₖ = n_{k−1} + 1; since v_{mₖ} − 1/k is not an upper bound of {xₙ : n ≥ mₖ}, choose nₖ ≥ mₖ with v_{mₖ} − 1/k < x_{nₖ} ≤ v_{mₖ}.
The indices mₖ are strictly increasing, so (v_{mₖ}) is a subsequence of (vₘ) and converges to x* by 15.2. Since |x_{nₖ} − v_{mₖ}| < 1/k, the difference tends to zero, and therefore x_{nₖ} → x*. So x* ∈ L, and together with the previous paragraph x* = sup L. Hence (d) implies (e).
Finally assume (e) and put w = sup L; note L ≠ ∅ because X is bounded and 16.4 supplies a convergent subsequence. Let ε > 0. If infinitely many n satisfied w + ε < xₙ, those terms would form a bounded subsequence, which by 16.4 has a convergent subsequence; its limit lies in L and is at least w + ε by 15.8, contradicting w = sup L. Hence w + ε ∈ V and lim sup X ≤ w + ε.
這一步的所求:把 w + ε 送進 V,走反證。假設超過 w + ε 的項有無限多個,把它們照編號順序排成一條子數列——它有界,§16-3 的 16.4(Bolzano-Weierstrass:有界數列必有收斂子數列)再從中挑出收斂的一條,極限不會小於 w + ε。而子數列的子數列仍是子數列(兩層嚴格遞增的編號合成起來仍嚴格遞增),所以那個極限屬於 L,卻比 L 的 sup 還大,矛盾。
Since w − ε < w = sup L, some ℓ ∈ L satisfies w − ε < ℓ. A subsequence converging to ℓ has all but finitely many terms above w − ε, so infinitely many n satisfy w − ε < xₙ and w − ε ∉ V. As V lies to the right of w − ε, we get lim sup X ≥ w − ε. Since ε > 0 was arbitrary, lim sup X = w, which is (a).
最後一步的所求:把 w − ε 擋在 V 外。先用 6.3 左測在 L 裡挑一個超過 w − ε 的成員 ℓ,再看收斂到 ℓ 的那條子數列:它的項最終會擠進 ℓ 附近,而 ℓ 嚴格大於 w − ε,所以除了有限多項之外全都超過 w − ε——這就是無限多項。至此圈閉合,五句話兩兩等價。
這張圖在說 (e) 的內容:藍點在三處聚集,每一處都有子數列收斂過去,所以 L 有三個成員。上極限挑的是最右邊那一個。要注意 L 未必是有限集,也未必只由「看得出來」的聚集位置構成——(e) 保證的是不論 L 長什麼樣,它的 sup 一定被 L 自己取到(證明裡那條遞迴挑出的子數列就是見證)。
COROLLARY
Let X = (xₙ) be a bounded sequence in ℝ, write wₘ = inf {xₙ : n ≥ m}, and let L be as in 18.2. For x* ∈ ℝ the following are equivalent. (a′) x* = lim inf (xₙ). (b′) For each ε > 0 there are at most finitely many n with xₙ < x* − ε, but infinitely many with xₙ < x* + ε. (c′) x* = sup {wₘ : m ∈ ℕ}. (d′) x* = lim (wₘ). (e′) x* = inf L.
Proof. Write Y = −X. The mirror lemma gives lim sup Y = −lim inf X. Moreover sup {yₙ : n ≥ m} = −inf {xₙ : n ≥ m} = −wₘ, and a subsequence of Y converges to v exactly when the corresponding subsequence of X converges to −v, so L(Y) = {−ℓ : ℓ ∈ L}.
這一步的所求:一張翻譯表。三筆對應都只用到取負號會把 sup 與 inf 互換這件事(6.5):Y 第 m 段的 sup 是 X 第 m 段 inf 的相反數;Y 的子數列極限集是 L 的相反數集(由 15.6(b) 的純量倍,係數取 −1)。
Applying 18.2 to Y with the number −x* and rewriting each of the five statements through this dictionary turns (a)–(e) into (a′)–(e′) respectively; for instance −x* = inf {−wₘ : m ∈ ℕ} becomes x* = sup {wₘ : m ∈ ℕ}. Since the five statements about Y are equivalent, so are the five about X.