Wzór Eulera-Maclaurina
Wielomiany, liczby i funkcje okresowe Bernoulliego
Definicja E1
Wielomiany [math]\displaystyle{ B_n(x) }[/math] spełniające warunki
● [math]\displaystyle{ B_0(x) = 1 }[/math] ● [math]\displaystyle{ {\small\frac{d}{d x}}B_n(x) = n B_{n - 1}(x) }[/math] ● [math]\displaystyle{ \int_0^1 B_n(t) d t = 0 \qquad \text{dla} \;\; n \geqslant 1 }[/math]
będziemy nazywali wielomianami Bernoulliego[1][2][3][4].
Zadanie E2
Korzystając z definicji E1 znaleźć jawną postać wielomianów [math]\displaystyle{ B_1 (x) }[/math], [math]\displaystyle{ B_2 (x) }[/math] i [math]\displaystyle{ B_3 (x) }[/math].
Twierdzenie E3*
Wielomiany Bernoulliego [math]\displaystyle{ B_n(x) }[/math] określone są następującym wzorem ogólnym
- [math]\displaystyle{ B_n(x) = \sum_{k = 0}^{n} {\small\frac{1}{k + 1}} \sum_{j = 0}^{k} (- 1)^j {\small\binom{k}{j}} (x + j)^n }[/math]
Przykład E4
W tabeli wypisaliśmy początkowe wielomiany Bernoulliego.
[math]\displaystyle{ \quad \;\: n \quad }[/math] [math]\displaystyle{ B_n(x) }[/math] [math]\displaystyle{ \quad \;\: 0 \quad }[/math] [math]\displaystyle{ 1 }[/math] [math]\displaystyle{ \quad \;\: 1 \quad }[/math] [math]\displaystyle{ x - {\small\frac{1}{2}} }[/math] [math]\displaystyle{ \quad \;\: 2 \quad }[/math] [math]\displaystyle{ x^2 - x + {\small\frac{1}{6}} }[/math] [math]\displaystyle{ \quad \;\: 3 \quad }[/math] [math]\displaystyle{ x^3 - {\small\frac{3}{2}} x^2 + {\small\frac{1}{2}} x }[/math] [math]\displaystyle{ \quad \;\: 4 \quad }[/math] [math]\displaystyle{ x^4 - 2 x^3 + x^2 - \small\frac{1}{30} }[/math] [math]\displaystyle{ \quad \;\: 5 \quad }[/math] [math]\displaystyle{ x^5 - {\small\frac{5}{2}} x^4 + \small\frac{5}{3} x^3 - \small\frac{1}{6} x }[/math] [math]\displaystyle{ \quad \;\: 6 \quad }[/math] [math]\displaystyle{ x^6 - 3 x^5 + \small\frac{5}{2} x^4 - \small\frac{1}{2} x^2 + \small\frac{1}{42} }[/math] [math]\displaystyle{ \quad \;\: 7 \quad }[/math] [math]\displaystyle{ x^7 - {\small\frac{7}{2}} x^6 + {\small\frac{7}{2}} x^5 - {\small\frac{7}{6}} x^3 + {\small\frac{1}{6}} x }[/math] [math]\displaystyle{ \quad \;\: 8 \quad }[/math] [math]\displaystyle{ x^8 - 4 x^7 + \small\frac{14}{3} x^6 - \small\frac{7}{3} x^4 + \small\frac{2}{3} x^2 - \small\frac{1}{30} }[/math] [math]\displaystyle{ \quad \;\: 9 \quad }[/math] [math]\displaystyle{ x^9 - \small\frac{9}{2} x^8 + 6 x^7 - \small\frac{21}{5} x^5 + 2 x^3 - \small\frac{3}{10} x }[/math] [math]\displaystyle{ \quad 10 \quad }[/math] [math]\displaystyle{ x^{10} - 5 x^9 + \small\frac{15}{2} x^8 - 7 x^6 + 5 x^4 - \small\frac{3}{2} x^2 + \small\frac{5}{66} }[/math] [math]\displaystyle{ \quad 11 \quad }[/math] [math]\displaystyle{ x^{11} - \small\frac{11}{2} x^{10} + \small\frac{55}{6} x^9 - 11 x^7 + 11 x^5 - \small\frac{11}{2} x^3 + \small\frac{5}{6} x }[/math] [math]\displaystyle{ \quad 12 \quad }[/math] [math]\displaystyle{ x^{12} - 6 x^{11} + 11 x^{10} - {\small\frac{33}{2}} x^8 + 22 x^6 - {\small\frac{33}{2}} x^4 + 5 x^2 - {\small\frac{691}{2730}} }[/math]
Przykład E5
Przedstawiamy wykresy wielomianów Bernoulliego [math]\displaystyle{ B_n(x) }[/math] dla [math]\displaystyle{ x \in [0, 1] }[/math]
Definicja E6
Liczbami Bernoulliego [math]\displaystyle{ B_n }[/math] będziemy nazywali wartości wielomianów Bernoulliego [math]\displaystyle{ B_n(x) }[/math] dla [math]\displaystyle{ x = 0 }[/math], czyli [math]\displaystyle{ B_n = B_n (0) }[/math].
Uwaga E7
Ze wzoru podanego w twierdzeniu E3 wynika natychmiast wzór ogólny dla liczb Bernoulliego.
- [math]\displaystyle{ B_n = B_n (0) = \sum_{k = 0}^{n} {\small\frac{1}{k + 1}} \sum_{j = 0}^{k} (- 1)^j {\small\binom{k}{j}} j^n }[/math]
Twierdzenie E8
Niech [math]\displaystyle{ B_n (x) }[/math] i [math]\displaystyle{ B_n }[/math] oznaczają odpowiednio wielomiany i liczby Bernoulliego. Prawdziwe są następujące wzory
[math]\displaystyle{ \quad 1. \quad }[/math] [math]\displaystyle{ B_n (1) = B_n (0) }[/math] [math]\displaystyle{ n \geqslant 2 }[/math] [math]\displaystyle{ \quad 2. \quad }[/math] [math]\displaystyle{ B_n (1 - x) = (- 1)^n B_n (x) }[/math] [math]\displaystyle{ n \geqslant 0 }[/math] [math]\displaystyle{ \quad 3. \quad }[/math] [math]\displaystyle{ B_{2 k + 1} (1) = B_{2 k + 1} (0) = B_{2 k + 1} \left( \tfrac{1}{2} \right) = 0 }[/math] [math]\displaystyle{ k \geqslant 1 }[/math] [math]\displaystyle{ \quad 4. \quad }[/math] [math]\displaystyle{ B_n \left( a x \right) = a^{n - 1} \sum_{k = 0}^{a - 1} B_n \left( x + \small\frac{k}{a} \right) }[/math] [math]\displaystyle{ n \geqslant 0 \quad \text{i} \quad a \in \mathbb{Z}_+ }[/math] [math]\displaystyle{ \quad 5. \quad }[/math] [math]\displaystyle{ \sum_{k = 1}^{a - 1} B_n \left( \small\frac{k}{a} \right) = (a^{1 - n} - 1) B_n }[/math] [math]\displaystyle{ n \geqslant 0 \quad \text{i} \quad a \in \mathbb{Z}_+ }[/math] [math]\displaystyle{ \quad 6. \quad }[/math] [math]\displaystyle{ B_n \left( \tfrac{1}{2} \right) = (2^{1 - n} - 1) B_n }[/math] [math]\displaystyle{ n \geqslant 0 }[/math] [math]\displaystyle{ \quad 7. \quad }[/math] [math]\displaystyle{ B_{2 k} \left( \tfrac{1}{3} \right) = \tfrac{1}{2} (3^{1 - 2 k} - 1) B_{2 k} }[/math] [math]\displaystyle{ k \geqslant 0 }[/math] [math]\displaystyle{ \quad 8. \quad }[/math] [math]\displaystyle{ B_{2 k} \left( \tfrac{1}{4} \right) = 2^{- 2 k} (2^{1 - 2 k} - 1) B_{2 k} }[/math] [math]\displaystyle{ k \geqslant 0 }[/math] [math]\displaystyle{ \quad 9. \quad }[/math] [math]\displaystyle{ B_{2 k} \left( \tfrac{1}{6} \right) = \tfrac{1}{2} (2^{1 - 2 k} - 1) (3^{1 - 2 k} - 1) B_{2 k} }[/math] [math]\displaystyle{ k \geqslant 0 }[/math] [math]\displaystyle{ \quad 10. \quad }[/math] [math]\displaystyle{ B_n (x + 1) - B_n (x) = n x^{n - 1} }[/math] [math]\displaystyle{ n \geqslant 0 }[/math]
Zadanie E9
Niech [math]\displaystyle{ k \in \mathbb{N}_0 }[/math]. Pokazać, że wykres funkcji [math]\displaystyle{ B_{2 k} (x) }[/math] jest symetryczny, a funkcji [math]\displaystyle{ B_{2 k + 1} (x) }[/math] jest antysymetryczny względem prostej [math]\displaystyle{ x = {\small\frac{1}{2}} }[/math].
Zadanie E10
Niech [math]\displaystyle{ k \in \mathbb{N}_0 }[/math]. Pokazać, że
- [math]\displaystyle{ \int_{0}^{1 / 2} B_{2 k + 1} (x) d x = - \int^1_{1 / 2} B_{2 k + 1} (t) d t }[/math]
- [math]\displaystyle{ \int_{0}^{1 / 2} B_{2 k + 2} (x) d x = 0 }[/math]
Zadanie E11
Niech [math]\displaystyle{ n \in \mathbb{Z}_+ }[/math]. Pokazać, że [math]\displaystyle{ B_n = - {\small\frac{n}{2 (1 - 2^{- n})}} \int_{0}^{1 / 2} B_{n - 1} (t) d t }[/math]
Twierdzenie E12
Niech [math]\displaystyle{ f(x) }[/math] i [math]\displaystyle{ f' (x) }[/math] będą ciągłymi funkcjami rzeczywistymi określonymi w przedziale [math]\displaystyle{ [a, b] }[/math] i różniczkowalnymi w [math]\displaystyle{ (a, b) }[/math]. Jeżeli dla pewnego punktu [math]\displaystyle{ r \in (a, b) }[/math] spełnione są warunki [math]\displaystyle{ f(a) = f (b) = f (r) = 0 }[/math], to istnieje taki punkt [math]\displaystyle{ t \in (a, b) }[/math], że [math]\displaystyle{ f'' (t) = 0 }[/math].
Twierdzenie E13
Niech [math]\displaystyle{ k \in \mathbb{Z}_+ }[/math]. Wielomian [math]\displaystyle{ B_{2 k + 1} (x) }[/math] ma dokładnie trzy pierwiastki w przedziale [math]\displaystyle{ [0, 1] }[/math]. Są to liczby [math]\displaystyle{ x = 0 }[/math], [math]\displaystyle{ x = {\small\frac{1}{2}} \, }[/math] i [math]\displaystyle{ \, x = 1 }[/math].
Twierdzenie E14
Niech [math]\displaystyle{ k \in \mathbb{N}_0 }[/math]. Liczby Bernoulliego [math]\displaystyle{ B_{2 k} }[/math] są różne od zera.
Twierdzenie E15
Niech [math]\displaystyle{ k \in \mathbb{N}_0 }[/math]. Dla wielomianów Bernoulliego [math]\displaystyle{ B_{2 k} (x) \, }[/math] i [math]\displaystyle{ \, x \in [0, 1] }[/math] prawdziwe jest następujące oszacowanie
- [math]\displaystyle{ | B_{2 k} (x) | \leqslant | B_{2 k} | }[/math]
Przykład E16
W tabeli przedstawiamy liczby Bernoulliego [math]\displaystyle{ B_n }[/math] oraz minimalne [math]\displaystyle{ m_n }[/math] i maksymalne [math]\displaystyle{ M_n }[/math] wartości wielomianów [math]\displaystyle{ B_n(x) }[/math] dla [math]\displaystyle{ x \in [0, 1] }[/math]
[math]\displaystyle{ \quad n \quad }[/math] [math]\displaystyle{ B_n(x) }[/math] [math]\displaystyle{ B_n }[/math] [math]\displaystyle{ m_n }[/math] [math]\displaystyle{ M_n }[/math] [math]\displaystyle{ \quad 0 \quad }[/math] [math]\displaystyle{ 1 }[/math] [math]\displaystyle{ 1 }[/math] [math]\displaystyle{ 1 }[/math] [math]\displaystyle{ 1 }[/math] [math]\displaystyle{ \quad 1 \quad }[/math] [math]\displaystyle{ x - {\small\frac{1}{2}} }[/math] [math]\displaystyle{ - {\small\frac{1}{2}} }[/math] [math]\displaystyle{ - {\small\frac{1}{2}} }[/math] [math]\displaystyle{ {\small\frac{1}{2}} }[/math] [math]\displaystyle{ \quad 2 \quad }[/math] [math]\displaystyle{ x^2 - x + {\small\frac{1}{6}} }[/math] [math]\displaystyle{ {\small\frac{1}{6}} }[/math] [math]\displaystyle{ - {\small\frac{1}{12}} }[/math] [math]\displaystyle{ {\small\frac{1}{6}} }[/math] [math]\displaystyle{ \quad 3 \quad }[/math] [math]\displaystyle{ x^3 - {\small\frac{3}{2}} x^2 + {\small\frac{1}{2}} x }[/math] [math]\displaystyle{ 0 }[/math] [math]\displaystyle{ - {\small\frac{\sqrt{3}}{36}} }[/math] [math]\displaystyle{ {\small\frac{\sqrt{3}}{36}} }[/math] [math]\displaystyle{ \quad 4 \quad }[/math] [math]\displaystyle{ x^4 - 2 x^3 + x^2 - {\small\frac{1}{30}} }[/math] [math]\displaystyle{ - {\small\frac{1}{30}} }[/math] [math]\displaystyle{ - {\small\frac{1}{30}} }[/math] [math]\displaystyle{ {\small\frac{7}{240}} }[/math] [math]\displaystyle{ \quad 5 \quad }[/math] [math]\displaystyle{ x^5 - {\small\frac{5}{2}} x^4 + {\small\frac{5}{3}} x^3 - {\small\frac{1}{6}} x }[/math] [math]\displaystyle{ 0 }[/math] [math]\displaystyle{ - {\small\frac{1}{6}} \sqrt{{\small\frac{1}{60}} + {\small\frac{\sqrt{30}}{1125}}} }[/math] [math]\displaystyle{ {\small\frac{1}{6}} \sqrt{{\small\frac{1}{60}} + {\small\frac{\sqrt{30}}{1125}}} }[/math] [math]\displaystyle{ \quad 6 \quad }[/math] [math]\displaystyle{ x^6 - 3 x^5 + {\small\frac{5}{2}} x^4 - {\small\frac{1}{2}} x^2 + {\small\frac{1}{42}} }[/math] [math]\displaystyle{ {\small\frac{1}{42}} }[/math] [math]\displaystyle{ - {\small\frac{31}{1344}} }[/math] [math]\displaystyle{ {\small\frac{1}{42}} }[/math]
Zauważmy, że [math]\displaystyle{ M_3 = {\small\frac{\sqrt{3}}{36}} \lt {\small\frac{3}{62}} }[/math], [math]\displaystyle{ \quad M_5 \lt {\small\frac{1}{40}} }[/math], [math]\displaystyle{ \quad M_7 \lt {\small\frac{1}{38}} \quad }[/math] oraz [math]\displaystyle{ \quad M_9 \lt {\small\frac{1}{21}} }[/math]
Przykład E17
Minima [math]\displaystyle{ m_n }[/math] i maksima [math]\displaystyle{ M_n }[/math] wielomianów Bernoulliego [math]\displaystyle{ B_n(x) }[/math] dla [math]\displaystyle{ x \in [0, 1] }[/math] są równe[6]
[math]\displaystyle{ n }[/math] [math]\displaystyle{ m_n }[/math] [math]\displaystyle{ M_n }[/math] [math]\displaystyle{ \text{uwagi} }[/math] [math]\displaystyle{ 2 k + 1 }[/math] [math]\displaystyle{ - \bigl| B_{2 k + 1} (x_{2 k}) \bigr| }[/math] [math]\displaystyle{ \bigl| B_{2 k + 1} (x_{2 k}) \bigr| }[/math] [math]\displaystyle{ B_{2 k} (x_{2 k}) = 0 \;\;\; \text{dla} \;\; x \in \left( 0, \tfrac{1}{2} \right) }[/math] [math]\displaystyle{ 4 k }[/math] [math]\displaystyle{ B_{4 k} (0) }[/math] [math]\displaystyle{ B_{4 k} \left( \tfrac{1}{2} \right) }[/math] [math]\displaystyle{ \text{dla} \;\; k \geqslant 1 }[/math] [math]\displaystyle{ 4 k + 2 }[/math] [math]\displaystyle{ B_{4 k + 2} \left( \tfrac{1}{2} \right) }[/math] [math]\displaystyle{ B_{4 k + 2} (0) }[/math] [math]\displaystyle{ }[/math]
W zamieszczonej niżej tabeli przedstawiamy liczby Bernoulliego [math]\displaystyle{ B_n }[/math] oraz minimalne i maksymalne wartości wielomianów [math]\displaystyle{ B_n(x) }[/math] dla [math]\displaystyle{ x \in [0, 1] }[/math] w zapisie dziesiętnym.
Definicja E18
Funkcje okresowe Bernoulliego [math]\displaystyle{ P_n(x) }[/math] definiujemy następująco
- [math]\displaystyle{ P_n(x) = B_n(x - \lfloor x \rfloor) }[/math]
Uwaga E19
Inaczej mówiąc funkcja okresowa Bernoulliego [math]\displaystyle{ P_n(x) }[/math] na odcinku [math]\displaystyle{ [0, 1] }[/math], przyjmuje te same wartości, co wielomian Bernoulliego [math]\displaystyle{ B_n(x) }[/math]. Wartości te powtarzają się dla kolejnych odcinków [math]\displaystyle{ [k, k + 1] }[/math], gdzie [math]\displaystyle{ k \in \mathbb{Z} }[/math].
Uwaga E20
Wprost z definicji funkcji okresowych Bernoulliego wynika, że dla [math]\displaystyle{ k \in \mathbb{Z} }[/math] jest
- [math]\displaystyle{ P_n (k) = B_n (k - \lfloor k \rfloor) = B_n (0) = B_n }[/math]
Twierdzenie E21
Własności funkcji okresowych Bernoulliego
● funkcja [math]\displaystyle{ P_0 (x) }[/math] jest ciągła i różniczkowalna ● funkcja [math]\displaystyle{ P_1 (x) }[/math] nie jest ciągła w punktach [math]\displaystyle{ x \in \mathbb{Z} }[/math] ● funkcja [math]\displaystyle{ P_2 (x) }[/math] jest ciągła, ale nie jest różniczkowalna w punktach [math]\displaystyle{ x \in \mathbb{Z} }[/math] ● dla [math]\displaystyle{ n \geqslant 3 }[/math] funkcje [math]\displaystyle{ P_n (x) }[/math] są ciągłe i różniczkowalne ● [math]\displaystyle{ {\small\frac{d}{d x}} P_n (x) = n P_{n - 1} (x) \qquad }[/math] o ile [math]\displaystyle{ n \neq 1, 2 }[/math] lub [math]\displaystyle{ n = 1, 2 }[/math] oraz [math]\displaystyle{ x \notin \mathbb{Z} }[/math] ● [math]\displaystyle{ \int^x_0 P_n (t) d t = {\small\frac{P_{n + 1} (x)}{n + 1}} - {\small\frac{B_{n + 1}}{n + 1}} }[/math]
Przykład E22
Przedstawiamy przykładowe wykresy funkcji okresowych Bernoulliego [math]\displaystyle{ P_n (x) }[/math]. Stanowią one bardzo dobrą ilustrację do twierdzenia E21.
Twierdzenie E23*
Niech [math]\displaystyle{ n \in \mathbb{Z}_+ }[/math]. Dla liczb Bernoulliego [math]\displaystyle{ B_{2 n} = (- 1)^{n + 1} | B_{2 n} | }[/math] prawdziwe są następujące oszacowania [7][8][9]
- [math]\displaystyle{ {\small\frac{2 (2 n) !}{(2 \pi)^{2 n}}} \cdot {\small\frac{1}{1 - 2^{- 2 n}}} \lt | B_{2 n} | \lt {\small\frac{2 (2 n) !}{(2 \pi)^{2 n}}} \cdot {\small\frac{1}{1 - 2^{1 - 2 n}}} }[/math]
i asymptotyki
- [math]\displaystyle{ B_{2 n} \sim (- 1)^{n + 1} \cdot {\small\frac{2 (2 n) !}{(2 \pi)^{2 n}}} }[/math]
- [math]\displaystyle{ B_{2 n} \sim (- 1)^{n + 1} \cdot 4 \sqrt{\pi n} \cdot \left( {\small\frac{n}{\pi e}} \right)^{2 n} }[/math]
Twierdzenie E24*
Niech [math]\displaystyle{ n \in \mathbb{Z}_+ }[/math]. Dla ilorazu kolejnych liczb Bernoulliego [math]\displaystyle{ B_{2 n} }[/math] prawdziwe są następujące oszacowania[10]
- [math]\displaystyle{ {\small\frac{2^{2 n - 1} - 1}{2^{2 n + 1} - 1}} \cdot {\small\frac{(2 n + 1) (2 n + 2)}{\pi^2}} \lt \left| {\small\frac{B_{2 n + 2}}{B_{2 n}}} \right| \lt {\small\frac{2^{2 n} - 1}{2^{2 n + 2} - 1}} \cdot {\small\frac{(2 n + 1) (2 n + 2)}{\pi^2}} }[/math]
i asymptotyka
- [math]\displaystyle{ {\small\frac{B_{2 n + 2}}{B_{2 n}}} \sim - {\small\frac{n^2}{\pi^2}} }[/math]
Wzór sumacyjny Eulera-Maclaurina
Uwaga E25
Często w twierdzeniu musimy założyć, że rozważana funkcja [math]\displaystyle{ f(x) }[/math] jest określona w pewnym zbiorze liczb rzeczywistych i jest funkcją ciągłą oraz wszystkie jej pochodne od [math]\displaystyle{ f' (x) }[/math] do [math]\displaystyle{ f^{(n)} (x) }[/math] istnieją i są ciągłe w tym zbiorze. Przekazanie tego prostego założenia wymaga użycia wielu słów, a samo twierdzenie staje się mało czytelne. Ze względów czysto praktycznych wprowadzamy pojęcie klasy funkcji.
Definicja E26
Funkcję [math]\displaystyle{ f(x) }[/math] określoną i ciągłą w zbiorze [math]\displaystyle{ A \subset \mathbb{R} }[/math] i mającą kolejno [math]\displaystyle{ n }[/math] ciągłych pochodnych w tym zbiorze będziemy nazywali funkcją klasy [math]\displaystyle{ C^n }[/math]. Jeżeli funkcja [math]\displaystyle{ f(x) }[/math] jest ciągła w [math]\displaystyle{ A }[/math], to powiemy, że jest klasy [math]\displaystyle{ C^0 }[/math]. Jeżeli funkcja [math]\displaystyle{ f(x) }[/math] jest klasy [math]\displaystyle{ C^n }[/math] dla dowolnego [math]\displaystyle{ n \in \mathbb{Z}_+ }[/math], to powiemy, że funkcja [math]\displaystyle{ f(x) }[/math] jest klasy [math]\displaystyle{ C^{\infty} }[/math]. W
przypadku, gdy chcemy jednocześnie zaznaczyć dziedzinę funkcji, to stosujemy zapis [math]\displaystyle{ C^0 (A) }[/math], [math]\displaystyle{ C^n (A) }[/math] i [math]\displaystyle{ C^{\infty} (A) }[/math].
Przykład E27
Tylko dla potrzeb tego przykładu funkcję [math]\displaystyle{ f(x) }[/math] określoną następująco
- [math]\displaystyle{ f(x) = \left\{ \begin{array}{lll} g (x) & & x \lt 0\\ h (x) & & x \geqslant 0 \end{array} \right. }[/math]
będziemy zapisywali jako [math]\displaystyle{ f(x) = \left \{ g (x) \big\rvert h (x) \right \} }[/math].
Przykłady funkcji klasy [math]\displaystyle{ C^0 (\mathbb{R}) }[/math]
- [math]\displaystyle{ \left \{ - x \big\rvert x \right \} \;\; \text{czyli} \;\; | x | , \quad \left \{ 0 \big\rvert x \right \} , \quad \left \{ 1 \big\rvert e^x \right \} , \quad \left \{ 1 + x \big\rvert \cos (x) \right \} }[/math]
Przykłady funkcji klasy [math]\displaystyle{ C^1 (\mathbb{R}) }[/math]
- [math]\displaystyle{ \left \{ 0 \big\rvert x^2 \right \} , \quad \left \{ 1 + x \big\rvert e^x \right \} , \quad \left \{ 1 \big\rvert \cos (x) \right \} }[/math]
Przykłady funkcji klasy [math]\displaystyle{ C^2 (\mathbb{R}) }[/math]
- [math]\displaystyle{ x^2 \sqrt{x^2} , \quad \left \{ 0 \big\rvert x^3 \right \} , \quad \left \{ 1 + x + \tfrac{1}{2} x^2 \big\rvert e^x \right \} , \quad \left \{ x \big\rvert \sin (x) \right \} }[/math]
Przykłady funkcji klasy [math]\displaystyle{ C^3 (\mathbb{R}) }[/math]
- [math]\displaystyle{ \left \{ 0 \big\rvert x^4 \right \} , \quad \left \{ 1 + x + \tfrac{1}{2} x^2 + \tfrac{1}{6} x^3 \big\rvert e^x \right \} , \quad \left \{ 1 - \tfrac{1}{2} x^2 \big\rvert \cos (x) \right \} }[/math]
Przykłady funkcji klasy [math]\displaystyle{ C^n (\mathbb{R}) }[/math]
- [math]\displaystyle{ P_{n + 2} (x) , \quad x^n \sqrt{x^2} , \quad \left \{ 0 \big\rvert x^{n + 1} \right \} , \quad \left\{ \sum_{k = 0}^{n} {\small\frac{x^k}{k!}} \biggr\rvert e^x \right\} }[/math]
Przykłady funkcji klasy [math]\displaystyle{ C^{\infty} (\mathbb{R}) }[/math]
- [math]\displaystyle{ x^k \;\; \text{dla} \;\; k \in \mathbb{N}_0 , \quad e^x , \quad \sin (x) , \quad \cos (x) }[/math]
Przykłady funkcji klasy [math]\displaystyle{ C^{\infty} (\mathbb{R}_+) }[/math]
- [math]\displaystyle{ {\small\frac{1}{x}} }[/math], [math]\displaystyle{ \sqrt{x} }[/math], [math]\displaystyle{ \log x }[/math]
Twierdzenie E28
Niech [math]\displaystyle{ f(x) }[/math] będzie funkcją rzeczywistą klasy [math]\displaystyle{ C^1 ( [k, k + 1] ) }[/math], gdzie [math]\displaystyle{ k \in \mathbb{Z} }[/math]. Jeżeli zastąpimy na jednostkowym odcinku pole prostokąta całką, to błąd, jaki popełnimy, jest równy
- [math]\displaystyle{ f(k) - \int_{k}^{k + 1} f(t) d t = \int_k^{k + 1} (t - \lfloor t \rfloor - 1) f'(t) d t }[/math]
Zadanie E29
Pokazać, że dla [math]\displaystyle{ x \gt 0 }[/math] całka [math]\displaystyle{ \int^x_0 (t - \lfloor t \rfloor)^n d t }[/math] jest równa
- [math]\displaystyle{ \int^x_0 (t - \lfloor t \rfloor)^n d t = {\small\frac{\lfloor x \rfloor + (x - \lfloor x \rfloor)^{n + 1}}{n + 1}} }[/math]
Twierdzenie E30
Niech [math]\displaystyle{ f(x) }[/math] będzie funkcją rzeczywistą klasy [math]\displaystyle{ C^1 ( [a, b] ) }[/math], gdzie [math]\displaystyle{ a, b \in \mathbb{Z} }[/math]. Możemy zastąpić sumowanie całkowaniem, stosując wzór
- [math]\displaystyle{ \sum_{k = a}^{b} f(k) = \int_a^b f(t) d t + {\small\frac{1}{2}} [f(b) + f(a)] + \int_a^b \left( t - \lfloor t \rfloor - {\small\frac{1}{2}} \right) f'(t) d t }[/math]
Powyższy wzór można zapisać w postaci
[math]\displaystyle{ \sum_{k = a}^{b} f(k) = \int_a^b f(t) d t + {\small\frac{1}{2}} [f(b) + f(a)] + \int_a^b P_1(t) f'(t) d t }[/math]
gdzie [math]\displaystyle{ P_1(t) }[/math] jest funkcją okresową Bernoulliego.
Uwaga E31
Czytelnik zapewne już domyśla się, w jakim kierunku zmierzamy. Całkując przez części i korzystając z własności funkcji okresowych Bernoulliego, przekształcimy całkę [math]\displaystyle{ \int_a^b P_1 (t) f' (t) d t }[/math] do postaci [math]\displaystyle{ \int_a^b P_2 (t) f'' (t) d t }[/math], a następnie do postaci [math]\displaystyle{ \int_a^b P_3 (t) f^{(3)} (t) d t }[/math] itd.
Twierdzenie E32
Niech [math]\displaystyle{ a, b \in \mathbb{Z} }[/math], a funkcje [math]\displaystyle{ P_n(t) }[/math], gdzie [math]\displaystyle{ n \geqslant 1 }[/math], będą funkcjami okresowymi Bernoulliego. Jeżeli funkcja rzeczywista [math]\displaystyle{ g(t) }[/math] jest klasy [math]\displaystyle{ C^1 ( [a, b] ) }[/math], to
- [math]\displaystyle{ \int_a^b P_n(t) g(t) d t = {\small\frac{B_{n + 1}}{n + 1}} [g(b) - g(a)] - {\small\frac{1}{n + 1}} \int_a^b P_{n + 1}(t) g'(t) d t }[/math]
Twierdzenie E33
Niech [math]\displaystyle{ a, b \in \mathbb{Z} }[/math], a funkcje [math]\displaystyle{ P_n (t) }[/math], gdzie [math]\displaystyle{ n \geqslant 1 }[/math], będą funkcjami okresowymi Bernoulliego. Jeżeli funkcja rzeczywista [math]\displaystyle{ g(t) }[/math] jest klasy [math]\displaystyle{ C^k ( [a, b] ) }[/math], to
- [math]\displaystyle{ \int_a^b P_n (t) g (t) d t = \sum_{j = 1}^k \frac{(- 1)^{j + 1} n! \cdot B_{n + j}}{(n + j) !} [g^{(j - 1)} (b) - g^{(j - 1)} (a)] + {\normalsize\frac{(- 1)^k n!}{(n + k) !}} \int_a^b P_{n + k} (t) g^{(k)} (t) d t }[/math]
Twierdzenie E34 (wzór sumacyjny Eulera-Maclaurina, [math]\displaystyle{ \sim }[/math]1735)
Niech [math]\displaystyle{ a, b \in \mathbb{Z} }[/math], a funkcje [math]\displaystyle{ P_r (t) }[/math], gdzie [math]\displaystyle{ r \geqslant 1 }[/math], będą funkcjami okresowymi Bernoulliego. Jeżeli funkcja rzeczywista [math]\displaystyle{ f(t) }[/math] jest klasy [math]\displaystyle{ C^r ( [a, b] ) }[/math], to
[math]\displaystyle{ \sum_{k = a}^b f(k) = \int_a^b f(t) d t + {\small\frac{1}{2}} [f(b) + f(a)] + \sum_{k = 2}^r {\small\frac{B_k}{k!}} [f^{(k - 1)}(b) - f^{(k - 1)}(a)] - {\small\frac{(- 1)^r}{r!}} \int_a^b P_r(t) f^{(r)}(t) d t }[/math]
Uwaga E35
Uwzględniając, że dla nieparzystych liczb [math]\displaystyle{ k \geqslant 2 }[/math] jest [math]\displaystyle{ B_k = 0 }[/math], możemy dla parzystego [math]\displaystyle{ r = 2 s }[/math] napisać
[math]\displaystyle{ \sum_{k = a}^b f(k) = \int_a^b f(t) d t + {\small\frac{1}{2}} [f(b) + f(a)] + \sum_{k = 1}^s {\small\frac{B_{2 k}}{(2 k) !}} [f^{(2 k - 1)}(b) - f^{(2 k - 1)}(a)] - {\small\frac{1}{(2 s) !}} \int_a^b P_{2 s}(t) f^{(2 s)}(t) d t }[/math]
W przypadku, gdy [math]\displaystyle{ r = 2 s + 1 }[/math] mamy [math]\displaystyle{ B_{2 s + 1} = 0 }[/math], zatem nie pojawi się nowy składnik sumy, ale ostatni wyraz ulegnie zmianie
[math]\displaystyle{ \sum_{k = a}^b f(k) = \int_a^b f(t) d t + {\small\frac{1}{2}} [f(b) + f(a)] + \sum_{k = 1}^s {\small\frac{B_{2 k}}{(2 k) !}} [f^{(2 k - 1)}(b) - f^{(2 k - 1)}(a)] + {\small\frac{1}{(2 s + 1) !}} \int_a^b P_{2 s + 1}(t) f^{(2 s + 1)}(t) d t }[/math]
Oczywiście
- [math]\displaystyle{ - {\small\frac{1}{(2 s) !}} \int_a^b P_{2 s} (t) f^{(2 s)} (t) d t = {\small\frac{1}{(2 s + 1) !}} \int_a^b P_{2 s + 1} (t) f^{(2 s + 1)} (t) d t }[/math]
(zobacz twierdzenie E32).
Uwaga E36
Poniżej wypisaliśmy gotowe wzory Eulera-Maclaurina dla [math]\displaystyle{ r = 1, \ldots, 9 }[/math]
- [math]\displaystyle{ \sum_{k = a}^b f(k) = \int_a^b f(t) d t + {\small\frac{1}{2}} [f(b) + f(a)] + Q_r }[/math]
gdzie
[math]\displaystyle{ \quad \;\: r \quad }[/math] [math]\displaystyle{ Q_r }[/math] [math]\displaystyle{ \quad \;\: 1. \quad }[/math] [math]\displaystyle{ \int_a^b P_1(t) f'(t) d t }[/math] [math]\displaystyle{ \quad \;\: 2. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] - {\small\frac{1}{2}} \int_a^b P_2(t) f''(t) d t }[/math] [math]\displaystyle{ \quad \;\: 3. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] + {\small\frac{1}{6}} \int_a^b P_3(t) f^{(3)}(t) d t }[/math] [math]\displaystyle{ \quad \;\: 4. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] - {\small\frac{1}{720}} [f^{(3)}(b) - f^{(3)}(a)] - {\small\frac{1}{24}} \int_a^b P_4(t) f^{(4)}(t) d t }[/math] [math]\displaystyle{ \quad \;\: 5. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] - {\small\frac{1}{720}} [f^{(3)}(b) - f^{(3)}(a)] + {\small\frac{1}{120}} \int_a^b P_5(t) f^{(5)}(t) d t }[/math] [math]\displaystyle{ \quad \;\: 6. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] - {\small\frac{1}{720}} [f^{(3)}(b) - f^{(3)}(a)] + {\small\frac{1}{30240}} [f^{(5)}(b) - f^{(5)}(a)] - {\small\frac{1}{720}} \int_a^b P_6(t) f^{(6)}(t) d t }[/math] [math]\displaystyle{ \quad \;\: 7. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] - {\small\frac{1}{720}} [f^{(3)}(b) - f^{(3)}(a)] + {\small\frac{1}{30240}} [f^{(5)}(b) - f^{(5)}(a)] + {\small\frac{1}{5040}} \int_a^b P_7(t) f^{(7)}(t) d t }[/math] [math]\displaystyle{ \quad \;\: 8. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] - {\small\frac{1}{720}} [f^{(3)}(b) - f^{(3)}(a)] + {\small\frac{1}{30240}} [f^{(5)}(b) - f^{(5)}(a)] - {\small\frac{1}{1209600}} [f^{(7)}(b) - f^{(7)}(a)] - {\small\frac{1}{40320}} \int_a^b P_8(t) f^{(8)}(t) d t }[/math] [math]\displaystyle{ \quad \;\: 9. \quad }[/math] [math]\displaystyle{ {\small\frac{1}{12}} [f'(b) - f'(a)] - {\small\frac{1}{720}} [f^{(3)}(b) - f^{(3)}(a)] + {\small\frac{1}{30240}} [f^{(5)}(b) - f^{(5)}(a)] - {\small\frac{1}{1209600}} [f^{(7)}(b) - f^{(7)}(a)] + {\small\frac{1}{362880}} \int_a^b P_9(t) f^{(9)}(t) d t }[/math]
Nim przejdziemy do przykładów, przypomnimy kilka podstawowych definicji i twierdzeń dotyczących całek niewłaściwych.
Całki niewłaściwe – zbieżność i kryteria zbieżności
Definicja E37
Niech funkcja [math]\displaystyle{ f(x) }[/math] będzie określona w przedziale [math]\displaystyle{ [a, + \infty) }[/math] i całkowalna w każdym podprzedziale [math]\displaystyle{ [a, b] }[/math] tego przedziału. Granicę
- [math]\displaystyle{ \lim_{b \to + \infty} \int^b_a f(x) d x }[/math]
będziemy nazywali całką niewłaściwą funkcji [math]\displaystyle{ f(x) }[/math] w granicach od [math]\displaystyle{ a }[/math] do [math]\displaystyle{ + \infty }[/math] i zapisywali symbolicznie jako
- [math]\displaystyle{ \int_{a}^{\infty} f(x) d x }[/math]
Jeżeli powyższa granica jest skończona, to powiemy, że całka [math]\displaystyle{ \int_{a}^{\infty} f(x) d x }[/math] jest zbieżna. Jeżeli granica jest nieskończona lub nie istnieje, to powiemy, że całka jest rozbieżna.
Twierdzenie E38 (kryterium porównawcze)
Jeżeli dla [math]\displaystyle{ x \geqslant a }[/math] funkcje [math]\displaystyle{ f(x) }[/math] i [math]\displaystyle{ g(x) }[/math] spełniają nierówności
- [math]\displaystyle{ 0 \leqslant f(x) \leqslant g(x) }[/math]
to
● ze zbieżności całki [math]\displaystyle{ \int_{a}^{\infty} g(x) d x }[/math] wynika zbieżność całki [math]\displaystyle{ \int_{a}^{\infty} f(x) d x }[/math] ● z rozbieżności całki [math]\displaystyle{ \int_{a}^{\infty} f(x) d x }[/math] wynika rozbieżność całki [math]\displaystyle{ \int_{a}^{\infty} g(x) d x }[/math]
Twierdzenie E39
Jeżeli funkcja [math]\displaystyle{ f(x) }[/math] jest całkowalna w każdym podprzedziale [math]\displaystyle{ [a, b] }[/math] przedziału [math]\displaystyle{ [a, + \infty) }[/math] i całka [math]\displaystyle{ \int_{a}^{\infty} | f(x) | d x }[/math] jest zbieżna, to zbieżna jest też całka [math]\displaystyle{ \int_{a}^{\infty} f(x) d x }[/math]. O całce [math]\displaystyle{ \int_{a}^{\infty} f (x) d x }[/math] powiemy wtedy, że jest bezwzględnie zbieżna.
Twierdzenie E40
Jeżeli całka [math]\displaystyle{ \int_{a}^{\infty} | f(x) | d x }[/math] jest zbieżna, a funkcja [math]\displaystyle{ g(x) }[/math] jest ograniczona, to zbieżna jest też całka [math]\displaystyle{ \int_{a}^{\infty} | f(x) g(x) | d x }[/math].
Twierdzenie E41
Niech [math]\displaystyle{ F(x) }[/math] oznacza funkcję pierwotną funkcji [math]\displaystyle{ f(x) }[/math]. Całka [math]\displaystyle{ \int_{a}^{\infty} f(x) d x }[/math] jest zbieżna wtedy i tylko wtedy, gdy granica [math]\displaystyle{ \lim_{x \to \infty} F(x) }[/math] jest skończona.
Twierdzenie E42
Jeżeli
● funkcja [math]\displaystyle{ f(x) }[/math] jest funkcją ciągłą i ma stały znak w przedziale [math]\displaystyle{ [a, + \infty) }[/math] ● całka [math]\displaystyle{ \int_{a}^{\infty} f (x) d x }[/math] jest zbieżna ● funkcja [math]\displaystyle{ g(x) }[/math] jest ograniczona w przedziale [math]\displaystyle{ [a, + \infty) }[/math], czyli dla [math]\displaystyle{ x \geqslant a }[/math] jest
- 1. [math]\displaystyle{ \qquad m \leqslant g (x) \leqslant M }[/math]
- 1. [math]\displaystyle{ \qquad m \leqslant g (x) \leqslant M }[/math]
lub
- 2. [math]\displaystyle{ \qquad | g (x) | \leqslant L }[/math]
● całka [math]\displaystyle{ \int^b_a g (x) d x }[/math] istnieje dla każdego [math]\displaystyle{ b \gt a }[/math]
to całki [math]\displaystyle{ \int_{a}^{\infty} | f (x) g (x) | d x }[/math] oraz [math]\displaystyle{ \int_{a}^{\infty} f (x) g (x) d x }[/math] są zbieżne i prawdziwe są następujące oszacowania
- 1. [math]\displaystyle{ \qquad s \:\! m \int_{a}^{\infty} f (x) d x \leqslant s \int_{a}^{\infty} f (x) g (x) d x \leqslant s M \int_{a}^{\infty} f (x) d x }[/math]
lub
- 2. [math]\displaystyle{ \qquad \int_{a}^{\infty} | f (x) g (x) | d x \leqslant L \left| \int_{a}^{\infty} f (x) d x \right| }[/math]
gdzie [math]\displaystyle{ s }[/math] jest znakiem funkcji [math]\displaystyle{ f(x) }[/math] w przedziale [math]\displaystyle{ [a, + \infty) }[/math].
Twierdzenie E43
Niech [math]\displaystyle{ P_n(t) }[/math], gdzie [math]\displaystyle{ n \geqslant 1 }[/math], będzie funkcją okresową Bernoulliego. Całka
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_n(t)}{t^{\alpha}}} d t }[/math]
gdzie [math]\displaystyle{ \alpha \gt 1 }[/math], jest zbieżna.
Twierdzenie E44
Niech [math]\displaystyle{ P_n (t) }[/math], gdzie [math]\displaystyle{ n \geqslant 1 }[/math], będzie funkcją okresową Bernoulliego. Całka
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_n(t)}{t^{\varepsilon}}} d t }[/math]
gdzie [math]\displaystyle{ \varepsilon \gt 0 }[/math], jest zbieżna.
Zadanie E45
Niech [math]\displaystyle{ P_n (t) }[/math], gdzie [math]\displaystyle{ n \geqslant 1 }[/math], będzie funkcją okresową Bernoulliego. Pokazać, że całka
- [math]\displaystyle{ \int_1^{\infty} P_n (t) t^{\varepsilon} d t }[/math]
gdzie [math]\displaystyle{ 0 \lt \varepsilon \lt 1 }[/math], jest rozbieżna.
Zadanie E46
Niech [math]\displaystyle{ P_n (t) }[/math], gdzie [math]\displaystyle{ n \geqslant 1 }[/math], będzie funkcją okresową Bernoulliego. Pokazać, że całka
- [math]\displaystyle{ \int_2^{\infty} {\small\frac{P_n (t)}{\log t}} d t }[/math]
jest zbieżna.
Zadanie E47
Niech [math]\displaystyle{ P_r (t) }[/math], gdzie [math]\displaystyle{ r \geqslant 1 }[/math], będzie funkcją okresową Bernoulliego oraz prawdziwe będzie następujące oszacowanie funkcji [math]\displaystyle{ P_r (t) }[/math]
- [math]\displaystyle{ m_r \leqslant P_r (t) \leqslant M_r }[/math]
Pokazać, że dla [math]\displaystyle{ \alpha \gt 1 }[/math] i [math]\displaystyle{ n \in \mathbb{Z}_+ }[/math] jest
- [math]\displaystyle{ {\small\frac{m_r}{\alpha - 1}} \cdot {\small\frac{1}{n^{\alpha - 1}}} \leqslant \int_n^{\infty} {\small\frac{P_r(t)}{t^{\alpha}}} d t \leqslant {\small\frac{M_r}{\alpha - 1}} \cdot {\small\frac{1}{n^{\alpha - 1}}} }[/math]
Podamy teraz kryterium Dirichleta, dzięki któremu moglibyśmy natychmiast uzyskać dowody twierdzeń E43 i E44 oraz rozwiązanie zadania E46. Celowo nie stosowaliśmy tego kryterium, aby Czytelnik mógł zapoznać się z ciekawym zastosowaniem twierdzenia E32.
Twierdzenie E48* (kryterium Dirichleta)
Jeżeli funkcje [math]\displaystyle{ f(x) }[/math] i [math]\displaystyle{ g(x) }[/math] są całkowalne w każdym podprzedziale [math]\displaystyle{ [a, b] }[/math] przedziału [math]\displaystyle{ [a, + \infty) }[/math] oraz spełniają warunki
● całka z funkcji [math]\displaystyle{ f(x) }[/math] jest ograniczona, czyli istnieje taka stała [math]\displaystyle{ M \gt 0 }[/math], że dla każdego [math]\displaystyle{ b \gt a }[/math] jest [math]\displaystyle{ \left| \int^b_a f(x) d x \right| \leqslant M }[/math] ● funkcja [math]\displaystyle{ g(x) }[/math] jest funkcją monotoniczną (czyli malejącą lub rosnącą) ● [math]\displaystyle{ \lim_{x \to \infty} g (x) = 0 }[/math]
to całka [math]\displaystyle{ \int_{a}^{\infty} f (x) g (x) d x }[/math] jest zbieżna.
Zadanie E49
Korzystając z kryterium Dirichleta, pokazać, że całki
- [math]\displaystyle{ \int_{0}^{\infty} {\small\frac{\sin x}{x}} d x = {\small\frac{1}{2}} \pi }[/math]
- [math]\displaystyle{ \int_{2}^{\infty} {\small\frac{P_1 (x)}{\log x}} d x = - 0.117923474371 \ldots }[/math]
są zbieżne.
Przykłady
Przykład E50
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^n k^2 }[/math]
Ponieważ [math]\displaystyle{ f(t) = t^2 }[/math], to [math]\displaystyle{ f'(t) = 2 t }[/math], [math]\displaystyle{ f''(t) = 2 }[/math], a dla [math]\displaystyle{ i \geqslant 3 }[/math] mamy [math]\displaystyle{ f^{(i)}(t) = 0 }[/math]. Zatem dla [math]\displaystyle{ r = 3 }[/math] wyraz [math]\displaystyle{ {\small\frac{1}{6}} \int_1^n P_3(t) f^{(3)}(t) d t }[/math] jest równy zero i otrzymujemy
- [math]\displaystyle{ \sum_{k = 1}^n k^2 = {\small\frac{1}{6}} n (n + 1) (2 n + 1) }[/math]
Przykład E51
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k^2}} }[/math]
Wiemy, że przypadku szeregu nieskończonego jest
- [math]\displaystyle{ \sum_{k = 1}^{\infty} {\small\frac{1}{k^2}} = {\small\frac{\pi^2}{6}} }[/math]
gdzie [math]\displaystyle{ {\small\frac{\pi^2}{6}} = 1.644934066848226436472415166646 \ldots }[/math]
Dla [math]\displaystyle{ r = 1 }[/math] mamy
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k^2}} = {\small\frac{3}{2}} - {\small\frac{1}{n}} + {\small\frac{1}{2 n^2}} - 2 \int_1^n {\small\frac{P_1 (t)}{t^3}} d t }[/math]
Przechodząc z [math]\displaystyle{ n }[/math] do nieskończoności, otrzymujemy
- [math]\displaystyle{ \sum_{k = 1}^{\infty} {\small\frac{1}{k^2}} = {\small\frac{3}{2}} - 2 \int_1^{\infty} {\small\frac{P_1 (t)}{t^3}} d t }[/math]
Rzeczywiście, całka po prawej stronie jest zbieżna i równa [math]\displaystyle{ \tfrac{1}{12} ( 9 - \pi^2 ) }[/math].
Jeżeli tak, to możemy sumę zapisać w postaci
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k^2}} = {\small\frac{3}{2}} - {\small\frac{1}{n}} + {\small\frac{1}{2 n^2}} - 2 \int_1^{\infty} {\small\frac{P_1 (t)}{t^3}} d t + 2 \int_n^{\infty} {\small\frac{P_1 (t)}{t^3}} d t }[/math]
- [math]\displaystyle{ \:\:\, = {\small\frac{\pi^2}{6}} - {\small\frac{1}{n}} + {\small\frac{1}{2 n^2}} + 2 \int_n^{\infty} {\small\frac{P_1 (t)}{t^3}} d t }[/math]
Ponieważ dla [math]\displaystyle{ P_1(t) = t - \lfloor t \rfloor - {\small\frac{1}{2}} }[/math] prawdziwe jest oszacowanie [math]\displaystyle{ - {\small\frac{1}{2}} \leqslant P_1(t) \leqslant {\small\frac{1}{2}} }[/math], to korzystając z pokazanego w zadaniu E47 wzoru, dostajemy
- [math]\displaystyle{ - {\small\frac{1}{4 n^2}} \leqslant \int_n^{\infty} {\small\frac{P_1 (t)}{t^3}} d t \leqslant {\small\frac{1}{4 n^2}} }[/math]
Teraz już łatwo otrzymujemy oszacowania
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k^2}} \leqslant {\small\frac{\pi^2}{6}} - {\small\frac{1}{n}} + {\small\frac{1}{n^2}} }[/math]
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k^2}} \geqslant {\small\frac{\pi^2}{6}} - {\small\frac{1}{n}} }[/math]
Jeśli we wzorze Eulera-Maclaurina uwzględnimy więcej wyrazów, to otrzymamy dokładniejsze oszacowania.
Przykład E52
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} }[/math]
Rozwinięcie asymptotyczne tej sumy jest dobrze znane[12]
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} = \log n + \gamma + {\small\frac{1}{2 n}} - \sum_{k = 1}^{\infty} {\small\frac{B_{2 k}}{2k \cdot n^{2 k}}} }[/math]
- [math]\displaystyle{ = \log n + \gamma + {\small\frac{1}{2 n}} - {\small\frac{1}{12 n^2}} + {\small\frac{1}{120 n^4}} - {\small\frac{1}{252 n^6}} + {\small\frac{1}{240 n^8}} - {\small\frac{1}{132 n^{10}}} + \cdots }[/math]
gdzie [math]\displaystyle{ \gamma = 0.57721566490153286060651209 \ldots }[/math] jest stałą Eulera.
Stosując wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 1 }[/math], mamy
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} = \log n + {\small\frac{1}{2}} + {\small\frac{1}{2 n}} - \int_1^n {\small\frac{P_1 (t)}{t^2}} d t }[/math]
W granicy, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, dostajemy
- [math]\displaystyle{ \lim_{n \to \infty} \left( \sum_{k = 1}^{n} {\small\frac{1}{k}} - \log n \right) = {\small\frac{1}{2}} - \int_1^{\infty} {\small\frac{P_1 (t)}{t^2}} d t }[/math]
Rzeczywiście, całka po prawej stronie jest zbieżna i równa [math]\displaystyle{ \tfrac{1}{2} - \gamma }[/math].
Zastosujemy teraz wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 3 }[/math] i znajdziemy oszacowanie
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} = \log n + {\small\frac{7}{12}} + {\small\frac{1}{2 n}} - {\small\frac{1}{12 n^2}} - \int_1^n {\small\frac{P_3 (t)}{t^4}} d t }[/math]
- [math]\displaystyle{ \: = \log n + {\small\frac{7}{12}} + {\small\frac{1}{2 n}} - {\small\frac{1}{12 n^2}} - \int_1^{\infty} {\small\frac{P_3 (t)}{t^4}} d t + \int_n^{\infty} {\small\frac{P_3 (t)}{t^4}} d t }[/math]
Oczywiście
- [math]\displaystyle{ {\small\frac{7}{12}} - \int_1^{\infty} {\small\frac{P_3 (t)}{t^4}} d t = \gamma }[/math]
Dostajemy
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} = \log n + \gamma + {\small\frac{1}{2 n}} - {\small\frac{1}{12 n^2}} + \int_n^{\infty} {\small\frac{P_3 (t)}{t^4}} d t }[/math]
Ponieważ prawdziwe są oszacowania (zobacz przykłady E16 i E17)
- [math]\displaystyle{ - {\small\frac{\sqrt{3}}{36}} \leqslant P_3 (t) \leqslant {\small\frac{\sqrt{3}}{36}} }[/math]
to korzystając z pokazanego w zadaniu E47 wzoru, dostajemy
- [math]\displaystyle{ - {\small\frac{\sqrt{3}}{108 n^3}} \leqslant \int_n^{\infty} {\small\frac{P_3 (t)}{t^4}} d t \leqslant {\small\frac{\sqrt{3}}{108 n^3}} }[/math]
Zatem
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} \leqslant \log n + \gamma + {\small\frac{1}{2 n}} - {\small\frac{1}{12 n^2}} + {\small\frac{\sqrt{3}}{108 n^3}} }[/math]
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} \geqslant \log n + \gamma + {\small\frac{1}{2 n}} - {\small\frac{1}{12 n^2}} - {\small\frac{\sqrt{3}}{108 n^3}} }[/math]
Powyższe nierówności mogą nam posłużyć do wyznaczenia stałej [math]\displaystyle{ \gamma }[/math]. Przykładowo dla [math]\displaystyle{ n = 10^6 }[/math], otrzymujemy
- [math]\displaystyle{ 0.5772156649015328605 \leqslant \gamma \leqslant 0.5772156649015328607 }[/math]
Przykład E53
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^{n} \log k }[/math]
Rozwinięcie asymptotyczne tej sumy jest dobrze znane[13]
- [math]\displaystyle{ \log n! = n \log n - n + {\small\frac{1}{2}} \log n + \tfrac{1}{2} \log \left( 2 \pi \right) + \sum_{k = 1}^{\infty} {\small\frac{B_{2 k}}{2 k (2 k - 1) \cdot n^{2 k - 1}}} }[/math]
- [math]\displaystyle{ \quad \; = n \log n - n + {\small\frac{1}{2}} \log n + \tfrac{1}{2} \log \left( 2 \pi \right) + {\small\frac{1}{12 n}} - {\small\frac{1}{360 n^3}} + {\small\frac{1}{1260 n^5}} - {\small\frac{1}{1680 n^7}} + {\small\frac{1}{1188 n^9}} - \cdots }[/math]
gdzie [math]\displaystyle{ \tfrac{1}{2} \log \left( 2 \pi \right) = 0.91893853320467274178 \ldots }[/math]
Stosując wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 1 }[/math], mamy
- [math]\displaystyle{ \sum_{k = 1}^{n} \log k = n \log n - n + {\small\frac{1}{2}} \log n + 1 + \int_1^n {\small\frac{P_1 (t)}{t}} d t }[/math]
W granicy, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, dostajemy
- [math]\displaystyle{ \lim_{n \to \infty} \left[ \sum_{k = 1}^{n} \log k - \left( n \log n - n + {\small\frac{1}{2}} \log n \right) \right] = 1 + \int_{1}^{\infty} {\small\frac{P_1(t)}{t}} d t }[/math]
Z twierdzenia E44 wiemy, że całka [math]\displaystyle{ \int_{1}^{\infty} {\small\frac{P_1 (t)}{t}} d t }[/math] jest zbieżna, a z rozwinięcia asymptotycznego wiemy, że granica po lewej stronie jest równa [math]\displaystyle{ \tfrac{1}{2} \log \left( 2 \pi \right) }[/math], zatem otrzymujemy
- [math]\displaystyle{ \int_{1}^{\infty} {\small\frac{P_1 (t)}{t}} d t = \tfrac{1}{2} \log (2 \pi) - 1 }[/math]
Stosując wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 4 }[/math], otrzymujemy
- [math]\displaystyle{ \sum_{k = 1}^{n} \log k = n \log n - n + {\small\frac{1}{2}} \log n + {\small\frac{331}{360}} + {\small\frac{1}{12 n}} - {\small\frac{1}{360 n^3}} + {\small\frac{1}{4}} \int_1^n {\small\frac{P_4 (t)}{t^4}} d t }[/math]
W granicy, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, dostajemy
- [math]\displaystyle{ \lim_{n \to \infty} \left[ \sum_{k = 1}^{n} \log k - \left( n \log n - n + {\small\frac{1}{2}} \log n \right) \right] = {\small\frac{331}{360}} + {\small\frac{1}{4}} \int_1^{\infty} {\small\frac{P_4 (t)}{t^4}} d t }[/math]
Ponieważ całka po prawej stronie jest zbieżna, to granica wypisana po lewej stronie istnieje i jest równa stałej – w tym przypadku [math]\displaystyle{ \tfrac{1}{2} \log \left( 2 \pi \right) }[/math]. Możemy teraz wzór Eulera-Maclaurina zapisać w postaci
- [math]\displaystyle{ \sum_{k = 1}^{n} \log k = n \log n - n + {\small\frac{1}{2}} \log n + {\small\frac{331}{360}} + {\small\frac{1}{12 n}} - {\small\frac{1}{360 n^3}} + {\small\frac{1}{4}} \int_1^{\infty} {\small\frac{P_4(t)}{t^4}} d t - {\small\frac{1}{4}} \int_n^{\infty} {\small\frac{P_4 (t)}{t^4}} d t }[/math]
Czyli
- [math]\displaystyle{ \sum_{k = 1}^{n} \log k = n \log n - n + {\small\frac{1}{2}} \log n + \tfrac{1}{2} \log \left( 2 \pi \right) + {\small\frac{1}{12 n}} - {\small\frac{1}{360 n^3}} - {\small\frac{1}{4}} \int_n^{\infty} {\small\frac{P_4 (t)}{t^4}} d t }[/math]
Z przykładów E16 i E17 wiemy, że prawdziwe są oszacowania
- [math]\displaystyle{ - {\small\frac{1}{30}} \leqslant P_4 (x) \leqslant {\small\frac{7}{240}} }[/math]
Zatem korzystając z pokazanego w zadaniu E47 wzoru, dostajemy
- [math]\displaystyle{ - {\small\frac{1}{90 n^3}} \leqslant \int_n^{\infty} {\small\frac{P_4 (t)}{t^4}} (t) d t \leqslant {\small\frac{7}{720 n^3}} }[/math]
Czyli
- [math]\displaystyle{ - {\small\frac{7}{2880 n^3}} \leqslant - {\small\frac{1}{4}} \int_n^{\infty} {\small\frac{P_4(t)}{t^4}} (t) d t \leqslant {\small\frac{1}{360 n^3}} }[/math]
Skąd natychmiast otrzymujemy oszacowania
- [math]\displaystyle{ \sum_{k = 1}^{n} \log k \leqslant n \log n - n + {\small\frac{1}{2}} \log n + \tfrac{1}{2} \log \left( 2 \pi \right) + {\small\frac{1}{12 n}} }[/math]
- [math]\displaystyle{ \sum_{k = 1}^{n} \log k \geqslant n \log n - n + {\small\frac{1}{2}} \log n + \tfrac{1}{2} \log \left( 2 \pi \right) + {\small\frac{1}{12 n}} - {\small\frac{1}{192 n^3}} }[/math]
Oczywiście, podobnie jak w poprzednim przykładzie, powyższe nierówności mogą służyć do wyznaczenia wartości stałej.
Przykład E54
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^{n} \sqrt{k} }[/math]
Stosując wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 4 }[/math], mamy
- [math]\displaystyle{ \sum_{k = 1}^{n} \sqrt{k} = {\small\frac{2}{3}} n^{3 / 2} + {\small\frac{1}{2}} n^{1 / 2} - {\small\frac{133}{640}} + {\small\frac{1}{24}} n^{- 1 / 2} - {\small\frac{1}{1920}} n^{- 5 / 2} + {\small\frac{5}{128}} \int_1^n {\small\frac{P_4 (t)}{t^{7 / 2}}} (t) d t }[/math]
W granicy, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, dostajemy
- [math]\displaystyle{ \lim_{n \to \infty} \left[ \sum_{k = 1}^{n} \sqrt{k} - \left( {\small\frac{2}{3}} n^{3 / 2} + {\small\frac{1}{2}} n^{1 / 2} \right) \right] = - {\small\frac{133}{640}} + {\small\frac{5}{128}} \int_1^{\infty} {\small\frac{P_4 (t)}{t^{7 / 2}}} (t) d t }[/math]
Ponieważ całka po prawej stronie jest zbieżna, to granica wypisana po lewej stronie istnieje i jest równa pewnej stałej [math]\displaystyle{ C }[/math]. Możemy teraz wzór Eulera-Maclaurina zapisać w postaci
- [math]\displaystyle{ \sum_{k = 1}^{n} \sqrt{k} = {\small\frac{2}{3}} n^{3 / 2} + {\small\frac{1}{2}} n^{1 / 2} - {\small\frac{133}{640}} + {\small\frac{1}{24}} n^{- 1 / 2} - {\small\frac{1}{1920}} n^{- 5 / 2} + {\small\frac{5}{128}} \int_1^{\infty} {\small\frac{P_4 (t)}{t^{7 / 2}}} (t) d t - {\small\frac{5}{128}} \int_n^{\infty} {\small\frac{P_4 (t)}{t^{7 / 2}}} (t) d t }[/math]
Zatem
- [math]\displaystyle{ \sum_{k = 1}^{n} \sqrt{k} = {\small\frac{2}{3}} n^{3 / 2} + {\small\frac{1}{2}} n^{1 / 2} + C + {\small\frac{1}{24}} n^{- 1 / 2} - {\small\frac{1}{1920}} n^{- 5 / 2} - {\small\frac{5}{128}} \int_n^{\infty} {\small\frac{P_4 (t)}{t^{7 / 2}}} (t) d t }[/math]
Z przykładów E16 i E17 wiemy, że prawdziwe są oszacowania
- [math]\displaystyle{ - {\small\frac{1}{30}} \leqslant P_4 (x) \leqslant {\small\frac{7}{240}} }[/math]
Zatem korzystając z pokazanego w zadaniu E47 wzoru, dostajemy
- [math]\displaystyle{ - {\small\frac{1}{75}} n^{- 5 / 2} \leqslant \int_n^{\infty} {\small\frac{P_4 (t)}{t^{7 / 2}}} (t) d t \leqslant {\small\frac{7}{600}} n^{- 5 / 2} }[/math]
Czyli
- [math]\displaystyle{ - {\small\frac{7}{15360}} n^{- 5 / 2} \leqslant - {\small\frac{5}{128}} \int_n^{\infty} {\small\frac{P_4 (t)}{t^{7 / 2}}} (t) d t \leqslant {\small\frac{1}{1920}} n^{- 5 / 2} }[/math]
I otrzymujemy oszacowania
- [math]\displaystyle{ \sum_{k = 1}^{n} \sqrt{k} \leqslant {\small\frac{2}{3}} n^{3 / 2} + {\small\frac{1}{2}} n^{1 / 2} + C + {\small\frac{1}{24}} n^{- 1 / 2} }[/math]
- [math]\displaystyle{ \sum_{k = 1}^{n} \sqrt{k} \geqslant {\small\frac{2}{3}} n^{3 / 2} + {\small\frac{1}{2}} n^{1 / 2} + C + {\small\frac{1}{24}} n^{- 1 / 2} - {\small\frac{1}{1024}} n^{- 5 / 2} }[/math]
Powyższe nierówności mogą nam posłużyć do wyznaczenia stałej [math]\displaystyle{ C }[/math]. Przykładowo dla [math]\displaystyle{ n = 10^6 }[/math], otrzymujemy
- [math]\displaystyle{ - 0.207886224977354567 \leqslant C \leqslant - 0.207886224977354565 }[/math]
Przykład E55
Pokażemy, dlaczego lepiej wybrać wartość [math]\displaystyle{ r }[/math] za dużą niż za małą i dlaczego należy sprawdzać zbieżność całki
- [math]\displaystyle{ \int_a^b P_r(t) f^{(r)}(t) d t }[/math]
korzystając z kryterium Dirichleta (twierdzenie E48) lub z twierdzenia E44. Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^{n} k^{3 / 2} }[/math]
Stosując wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 1 }[/math], mamy
- [math]\displaystyle{ \sum_{k = 1}^{n} k^{3 / 2} = {\small\frac{2}{5}} n^{5 / 2} + {\small\frac{1}{2}} n^{3 / 2} + {\small\frac{1}{10}} + {\small\frac{3}{2}} \int_1^n \sqrt{t} \cdot P_1 (t) d t }[/math]
W granicy, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, dostajemy
- [math]\displaystyle{ \lim_{n \to \infty} \left[ \sum_{k = 1}^{n} k^{3 / 2} - \left( {\small\frac{2}{5}} n^{5 / 2} + {\small\frac{1}{2}} n^{3 / 2} \right) \right] = {\small\frac{1}{10}} + {\small\frac{3}{2}} \int_1^{\infty} \sqrt{t} \cdot P_1(t) d t }[/math]
Jeszcze nie wszystkie wyrazy rozbieżne, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, zostały wyodrębnione – lewa strona jest rozbieżna. Podobnie rozbieżna jest całka po prawej stronie – rozbieżność tej całki informuje nas, że wybraliśmy za małą wartość [math]\displaystyle{ r }[/math].
Stosując wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 2 }[/math], mamy
- [math]\displaystyle{ \sum_{k = 1}^{n} k^{3 / 2} = {\small\frac{2}{5}} n^{5 / 2} + {\small\frac{1}{2}} n^{3 / 2} + {\small\frac{1}{8}} n^{1 / 2} - {\small\frac{1}{40}} - {\small\frac{3}{8}} \int_1^n {\small\frac{P_2(t)}{\sqrt{t}}} d t }[/math]
W granicy, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, otrzymujemy
- [math]\displaystyle{ \lim_{n \to \infty} \left[ \sum_{k = 1}^{n} k^{3 / 2} - \left( {\small\frac{2}{5}} n^{5 / 2} + {\small\frac{1}{2}} n^{3 / 2} + {\small\frac{1}{8}} n^{1 / 2} \right) \right] = - {\small\frac{1}{40}} - {\small\frac{3}{8}} \int_1^{\infty} {\small\frac{P_2(t)}{\sqrt{t}}} d t }[/math]
Całka po prawej stronie jest zbieżna, co wynika z kryterium Dirichleta. Zatem i lewa strona jest zbieżna, czyli wszystkie wyrazy rozbieżne, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, zostały wyodrębnione. Możemy to łatwo sprawdzić, obierając większą wartość [math]\displaystyle{ r }[/math].
Stosując wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 4 }[/math], mamy
- [math]\displaystyle{ \sum_{k = 1}^{n} k^{3 / 2} = {\small\frac{2}{5}} n^{5 / 2} + {\small\frac{1}{2}} n^{3 / 2} + {\small\frac{1}{8}} n^{1 / 2} - {\small\frac{49}{1920}} + {\small\frac{1}{1920}} n^{- 3 / 2} - {\small\frac{3}{128}} \int_1^n {\small\frac{P_4 (t)}{t^{5 / 2}}} d t }[/math]
W granicy, gdy [math]\displaystyle{ n }[/math] dąży do nieskończoności, dostajemy
- [math]\displaystyle{ \lim_{n \to \infty} \left[ \sum_{k = 1}^{n} k^{3 / 2} - \left( {\small\frac{2}{5}} n^{5 / 2} + {\small\frac{1}{2}} n^{3 / 2} + {\small\frac{1}{8}} n^{1 / 2} \right) \right] = - {\small\frac{49}{1920}} - {\small\frac{3}{128}} \int_1^{\infty} {\small\frac{P_4(t)}{t^{5 / 2}}} d t }[/math]
Uwaga E56
Rozwiązując przykłady znaleźliśmy wartości następujących całek oznaczonych
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_1 (t)}{t}} d t = {\small\frac{1}{2}} \log (2 \pi) - 1 }[/math]
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_1 (t)}{t^2}} d t = {\small\frac{1}{2}} - \gamma }[/math]
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_1 (t)}{t^3}} d t = {\small\frac{9 - \pi^2}{12}} }[/math]
Jeżeli funkcja rzeczywista [math]\displaystyle{ f(t) }[/math] ma ciągłą pochodną w [math]\displaystyle{ [a, b] \subset \mathbb{R} }[/math] oraz [math]\displaystyle{ \lim_{t \to \infty} f (t) = 0 }[/math], to dla [math]\displaystyle{ n \geqslant 1 }[/math] mamy
- [math]\displaystyle{ \int_a^{\infty} P_{n + 1} (t) f'(t) d t = - B_{n + 1} f(a) - (n + 1) \int_a^{\infty} P_n(t) f(t) d t }[/math]
(Jest to prosty wniosek z twierdzenia E32).
Powyższy rezultat możemy wykorzystać do wyliczenia kolejnych całek zawierających funkcje okresowe Bernoulliego. Przykładowo mamy
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_2 (t)}{t^2}} d t = \log (2 \pi) - {\small\frac{11}{6}} }[/math]
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_2 (t)}{t^3}} d t = {\small\frac{7}{12}} - \gamma }[/math]
- [math]\displaystyle{ \int_1^{\infty} {\small\frac{P_2 (t)}{t^4}} d t = {\small\frac{10 - \pi^2}{18}} }[/math]
Metody wyliczania stałej we wzorze Eulera-Maclaurina
Uwaga E57
W przedstawionych wyżej przykładach wyliczyliśmy wartość stałej we wzorze Eulera-Maclaurina (przykład E52 i E54) oraz pokazaliśmy, że wartość całki [math]\displaystyle{ \int_a^{\infty} P_r (t) f^{(r)} (t) d t }[/math] jest związana z wartością stałej (przykład E51, E52 i E53). Obecnie dokładnie omówimy ten problem.
Twierdzenie E58
Jeżeli założymy, że
● całka nieoznaczona [math]\displaystyle{ F(x) = \int f(x) d x }[/math] nie zawiera wyrazów, które nie zależą od [math]\displaystyle{ x }[/math] (takie wyrazy ulegają redukcji przy wyliczaniu całki [math]\displaystyle{ \int_a^b f(t) d t = F(b) - F(a) }[/math] i nie występują we wzorze Eulera-Maclaurina) ● dla pewnego [math]\displaystyle{ r \geqslant 1 }[/math] całka [math]\displaystyle{ \int_a^{\infty} P_r (t) f^{(r)} (t) d t }[/math] jest zbieżna
to wzór Eulera-Maclaurina może być zapisany w postaci
- [math]\displaystyle{ \sum_{k = a}^b f (k) = C (a) + E (b) }[/math]
gdzie
- [math]\displaystyle{ C(a) = - F (a) + {\small\frac{1}{2}} f (a) - \sum_{k = 2}^r {\small\frac{B_k}{k!}} f^{(k - 1)} (a) - {\small\frac{(- 1)^r}{r!}} \int_a^{\infty} P_r (t) f^{(r)} (t) d t }[/math]
- [math]\displaystyle{ E(b) = F (b) + {\small\frac{1}{2}} f (b) + \sum_{k = 2}^r {\small\frac{B_k}{k!}} f^{(k - 1)}(b) + {\small\frac{(- 1)^r}{r!}} \int_b^{\infty} P_r (t) f^{(r)} (t) d t }[/math]
Uwaga E59
We wzorze
- [math]\displaystyle{ \sum_{k = a}^b f (k) = C (a) + E (b) }[/math]
składnik [math]\displaystyle{ C(a) }[/math] jest wartością stałej [math]\displaystyle{ C }[/math] we wzorze Eulera-Maclaurina, a [math]\displaystyle{ E(b) }[/math] zwraca kolejne wyrazy rozwinięcia. Pokażemy, że wartość tej stałej możemy wyliczyć bezpośrednio ze wzoru
- [math]\displaystyle{ C = C (a) }[/math]
lub metodą pośrednią, wykorzystując związek
- [math]\displaystyle{ C(a) = \sum_{k = a}^b f (k) - E (b) }[/math]
W obydwu przypadkach obliczenia wykonamy dla znanej już Czytelnikowi sumy [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} }[/math] (przykład E48).
Przykład E60
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} }[/math]
Ponieważ
- [math]\displaystyle{ f(x) = {\small\frac{1}{x}} }[/math]
- [math]\displaystyle{ F(x) = \int {\small\frac{d x}{x}} = \log x }[/math]
- [math]\displaystyle{ f^{(r)} (x) = {\small\frac{d^r}{d x^r}} {\small\frac{1}{x}} = {\small\frac{(- 1)^r r!}{x^{r + 1}}} }[/math]
to wzór na wartość stałej z twierdzenia E57
- [math]\displaystyle{ C(a) = - F(a) + {\small\frac{1}{2}} f(a) - \sum_{k = 2}^r {\small\frac{B_k}{k!}} f^{(k - 1)}(a) - {\small\frac{(- 1)^r}{r!}} \int_a^{\infty} P_r(t) f^{(r)}(t) d t }[/math]
przyjmuje postać
- [math]\displaystyle{ C(1) = - \log (1) + {\small\frac{1}{2}} - \sum_{k = 2}^r {\small\frac{B_k}{k!}} \cdot {\small\frac{(- 1)^{k - 1} (k - 1) !}{x^k}} - {\small\frac{(- 1)^r}{r!}} \int_a^{\infty} P_r (t) \cdot {\small\frac{(- 1)^r r!}{x^{r + 1}}} d t }[/math]
- [math]\displaystyle{ C(1) = {\small\frac{1}{2}} + \sum_{k = 2}^r {\small\frac{B_k}{k}} - \int_1^{\infty} {\small\frac{P_r(t)}{t^{r + 1}}} d t }[/math]
Oznaczmy
- [math]\displaystyle{ C_r = {\small\frac{1}{2}} + \sum_{k = 2}^r {\small\frac{B_k}{k}} }[/math]
- [math]\displaystyle{ I_r = - \int_1^{\infty} {\small\frac{P_r (t)}{t^{r + 1}}} d t }[/math]
Wartość [math]\displaystyle{ I_r }[/math] obliczymy numerycznie w programie PARI/GP poleceniem
Int(r) = - intnum(t = 1,+oo, P(r, t)/t^(r+1), 12 )
gdzie
P(r, t) = B(r, t - floor(t))
jest funkcją okresową Bernoulliego [math]\displaystyle{ P_r (t) }[/math].
Ponieważ wyliczenie wartości [math]\displaystyle{ C_r }[/math] jest bardzo łatwe, to w tabeli przedstawiamy jedynie wartość całki [math]\displaystyle{ I_r }[/math] oraz wielkość błędu, z jakim wyliczyliśmy wartość stałej we wzorze Eulera-Maclaurina. Przy precyzji obliczeń w PARI/GP równej [math]\displaystyle{ 77 }[/math] cyfr znaczących (wyświetlanych jest tylko [math]\displaystyle{ 60 }[/math]) otrzymujemy
[math]\displaystyle{ \quad r \quad }[/math] [math]\displaystyle{ I_r }[/math] [math]\displaystyle{ C_r + I_r - \gamma }[/math] [math]\displaystyle{ \quad 2 \quad }[/math] [math]\displaystyle{ - 0.00611766843643217216316093584671186131649649607150165105785840 }[/math] [math]\displaystyle{ - 4.7 \cdot 10^{- 12} }[/math] [math]\displaystyle{ \quad 4 \quad }[/math] [math]\displaystyle{ 0.00221566490153286060651266099862945942063253146614696094725279 }[/math] [math]\displaystyle{ 5.8 \cdot 10^{- 25} }[/math] [math]\displaystyle{ \quad 6 \quad }[/math] [math]\displaystyle{ - 0.00175258906672110764745616388586252127113304104807585093607060 }[/math] [math]\displaystyle{ - 1.1 \cdot 10^{- 32} }[/math] [math]\displaystyle{ \quad 8 \quad }[/math] [math]\displaystyle{ 0.00241407759994555901921050278081512945505072420777227470125753 }[/math] [math]\displaystyle{ 2.0 \cdot 10^{- 40} }[/math] [math]\displaystyle{ \quad 10 \quad }[/math] [math]\displaystyle{ - 0.00516167997581201673836525479494244630271800893829613819256332 }[/math] [math]\displaystyle{ - 8.8 \cdot 10^{- 49} }[/math] [math]\displaystyle{ \quad 12 \quad }[/math] [math]\displaystyle{ 0.0159311161169840760577275412978536464933747871553748142613412 }[/math] [math]\displaystyle{ 4.4 \cdot 10^{- 57} }[/math] [math]\displaystyle{ \quad 14 \quad }[/math] [math]\displaystyle{ - 0.0674022172163492572756057920354796868399585461779585190763506 }[/math] [math]\displaystyle{ - 2.7 \cdot 10^{- 65} }[/math] [math]\displaystyle{ \quad 16 \quad }[/math] [math]\displaystyle{ 0.375857586705219370175374600121383058258080669508315990727571 }[/math] [math]\displaystyle{ 2.1 \cdot 10^{- 73} }[/math] [math]\displaystyle{ \quad 18 \quad }[/math] [math]\displaystyle{ - 2.67809674356490037362857973014873668554587366076180375307638 }[/math] [math]\displaystyle{ - 1.8 \cdot 10^{- 77} }[/math] [math]\displaystyle{ \quad 20 \quad }[/math] [math]\displaystyle{ 23.7781153776472208384926323910633845265753384604503174590448 }[/math] [math]\displaystyle{ - 2.6 \cdot 10^{- 76} }[/math] [math]\displaystyle{ \quad 22 \quad }[/math] [math]\displaystyle{ - 257.682029549889011045565338623429369096613067336651131816317 }[/math] [math]\displaystyle{ 3.6 \cdot 10^{- 74} }[/math] [math]\displaystyle{ \quad 24 \quad }[/math] [math]\displaystyle{ 3349.82851684815738700083270777461702894978497906139526623008 }[/math] [math]\displaystyle{ - 2.5 \cdot 10^{- 73} }[/math]
Zwróćmy uwagę, jak bardzo [math]\displaystyle{ C_r \approx \gamma - I_r }[/math] odbiega od wartości stałej [math]\displaystyle{ \gamma }[/math] dla dużych wartości [math]\displaystyle{ r }[/math] – dopiero suma [math]\displaystyle{ C_r + I_r }[/math] daje prawidłowy rezultat. Tylko po to, aby uwidocznić ten fakt, przedstawiliśmy dane dla tak wielu wartości [math]\displaystyle{ r }[/math].
Uwaga E61
W przykładzie E60 uzyskaliśmy zaskakująco dokładny wynik, ale wiemy o tym tylko dlatego, że znaliśmy wynik prawidłowy. Gdybyśmy nie znali wartości stałej [math]\displaystyle{ \gamma }[/math], to nie bylibyśmy w stanie określić, ile cyfr sumy [math]\displaystyle{ C_r + I_r }[/math] jest prawidłowych.
Nim przejdziemy do przedstawienia drugiego sposobu wyliczania stałej we wzorze Eulera-Maclaurina, udowodnimy twierdzenie, które pozwoli nam działać bardziej efektywnie.
Twierdzenie E62
Jeżeli założymy, że
● całka nieoznaczona [math]\displaystyle{ F(x) = \int f (x) d x }[/math] nie zawiera wyrazów, które nie zależą od [math]\displaystyle{ x }[/math] (takie wyrazy ulegają redukcji przy wyliczaniu całki [math]\displaystyle{ \int_a^b f (t) d t = F (b) - F (a) }[/math] i nie występują we wzorze Eulera-Maclaurina) ● funkcja [math]\displaystyle{ f^{(2 s)} (t) }[/math] jest funkcją ciągłą i ma stały znak w przedziale [math]\displaystyle{ [b, \infty) }[/math] ● [math]\displaystyle{ \lim_{t \to \infty} f^{(2 s - 1)} (t) = 0 }[/math]
to dla stałej [math]\displaystyle{ C(a) }[/math] we wzorze Eulera-Maclaurina prawdziwe jest oszacowanie
- [math]\displaystyle{ W - \Delta \leqslant C(a) \leqslant W + \Delta }[/math]
gdzie
- [math]\displaystyle{ W = W (s, a, b) = \sum_{k = a}^b f(k) - \left[ F (b) + {\small\frac{1}{2}} f(b) + \sum_{k = 1}^s {\small\frac{B_{2 k}}{(2 k) !}} f^{(2 k - 1)}(b) \right] }[/math]
- [math]\displaystyle{ \Delta = \Delta (s, b) = {\small\frac{| B_{2 s} |}{(2 s) !}} | f^{(2 s - 1)} (b) | }[/math]
Czyli błąd, jakim obarczony jest wynik [math]\displaystyle{ W }[/math], jest nie większy od wartości bezwzględnej ostatniego składnika sumy.
Przykład E63
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 1}^{n} {\small\frac{1}{k}} }[/math]
Ponieważ
- [math]\displaystyle{ f(x) = {\small\frac{1}{x}} }[/math]
- [math]\displaystyle{ F(x) = \int {\small\frac{d x}{x}} = \log x }[/math]
- [math]\displaystyle{ f^{(r)} (x) = {\small\frac{d^r}{d x^r}} {\small\frac{1}{x}} = {\small\frac{(- 1)^r r!}{x^{r + 1}}} }[/math]
to z twierdzenia E61 dostajemy
- [math]\displaystyle{ W = \sum_{k = 1}^n {\small\frac{1}{k}} - \left[ \log n + {\small\frac{1}{2 n}} - \sum_{k = 1}^s {\small\frac{B_{2 k}}{2 k \cdot n^{2 k}}} \right] }[/math]
- [math]\displaystyle{ \Delta = {\small\frac{| B_{2 s} |}{2 s \cdot n^{2 s}}} }[/math]
Dla [math]\displaystyle{ s = 4 }[/math] i [math]\displaystyle{ n = 10^8 }[/math] mamy
- [math]\displaystyle{ \Delta = 4.17 \cdot 10^{- 67} }[/math]
Uznając, że dokładność rzędu [math]\displaystyle{ 10^{- 65} }[/math] nas zadowala, otrzymujemy dla [math]\displaystyle{ s = 4 }[/math]
- [math]\displaystyle{ W = \sum_{k = 1}^n {\small\frac{1}{k}} - \left[ \log n + {\small\frac{1}{2 n}} - {\small\frac{1}{12 n^2}} + {\small\frac{1}{120 n^4}} - {\small\frac{1}{252 n^6}} + {\small\frac{1}{240 n^8}} \right] }[/math]
Wyliczając wartość prawej strony dla [math]\displaystyle{ n = 10^8 }[/math], dostajemy
- [math]\displaystyle{ W = 0.57721566490153286060651209008240243104215933593992359880576723488486772677766467 \ldots }[/math]
Ponieważ [math]\displaystyle{ \Delta = 4.17 \cdot 10^{- 67} }[/math], to ostatecznie możemy napisać
- [math]\displaystyle{ \gamma = 0.57721566490153286060651209008240243104215933593992359880576723488 \ldots }[/math]
Wyznaczyliśmy stałą [math]\displaystyle{ \gamma }[/math] z dokładnością [math]\displaystyle{ 65 }[/math] cyfr po przecinku. W rzeczywistości błąd jest mniejszy od [math]\displaystyle{ 10^{- 81} }[/math].
Uwaga E64
Zauważmy, że wyliczając wartość [math]\displaystyle{ \Delta }[/math], znamy wartość błędu jeszcze przed wykonaniem całości obliczeń. Dobierając odpowiednie wartości liczb [math]\displaystyle{ s }[/math] i [math]\displaystyle{ n }[/math] możemy sprawić, że błąd będzie odpowiednio mały. Unikamy numerycznego całkowania, które w przypadku bardziej skomplikowanych funkcji może być długie i obarczone znacznym i nieznanym błędem.
Przykład E65
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 2}^n \mathop{\text{li}}(k) }[/math]
W PARI/GP funkcję specjalną [math]\displaystyle{ \mathop{\text{li}}(x) = \int^x_0 {\small\frac{d t}{\log t}} }[/math] (logarytm całkowy[14][15]) możemy uzyskać następująco
li(x) = real( -eint1( -log(x) ) )
W powyższym wzorze wykorzystaliśmy zaimplementowaną pod nazwą [math]\displaystyle{ \text{eint1} (x) }[/math] inną funkcję specjalną [math]\displaystyle{ E_1 (x) = \int_{x}^{\infty} {\small\frac{e^{- t}}{t}} d t }[/math][16][17].
Mamy:
- [math]\displaystyle{ f(x) = \mathop{\text{li}}(x) }[/math]
- [math]\displaystyle{ F(x) = \int \mathop{\text{li}}(x) d x = x \mathop{\text{li}}(x) - \mathop{\text{li}}(x^2) }[/math]
- [math]\displaystyle{ f^{(1)} (x) = {\small\frac{1}{\log x}} }[/math]
dla [math]\displaystyle{ k \geqslant 2 }[/math] jest
- [math]\displaystyle{ f^{(k)} (x) = {\small\frac{d^{k - 1}}{d x^{k - 1}}} {\small\frac{1}{\log x}} = (- 1)^{k - 1} \sum_{j = 1}^{k - 1} {\small\frac{A^{k - 1}_j}{x^{k - 1} \log^{j + 1} x}} = \mathop{\text{DLog}}(k - 1, x) }[/math]
Oznaczenie [math]\displaystyle{ k }[/math]-tej pochodnej funkcji [math]\displaystyle{ {\small\frac{1}{\log x}} }[/math] jako [math]\displaystyle{ \mathop{\text{DLog}}(k, x) }[/math] znacząco ułatwi nam zapisywanie wzorów. Liczby naturalne [math]\displaystyle{ A^k_j }[/math] spełniają następujące równania rekurencyjne
- [math]\displaystyle{ A^k_1 = (k - 1) A^{k - 1}_1 }[/math]
- [math]\displaystyle{ A_j^k = j A^{k - 1}_{j - 1} + (k - 1) A^{k - 1}_j \qquad }[/math] dla [math]\displaystyle{ \quad j = 2, \ldots, k - 1 }[/math]
- [math]\displaystyle{ A^k_k = k A^{k - 1}_{k - 1} }[/math]
gdzie [math]\displaystyle{ A^1_1 = 1 }[/math] (zobacz twierdzenia E68 i E69).
Zauważmy, że dla [math]\displaystyle{ k \geqslant 2 }[/math] funkcje [math]\displaystyle{ f^{(k)} (x) = {\small\frac{d^{k - 1}}{d x^{k - 1}}} {\small\frac{1}{\log x}} }[/math] są funkcjami ciągłymi i mają stały znak dla [math]\displaystyle{ x \gt 1 }[/math] oraz [math]\displaystyle{ \lim_{x \to \infty} f^{(k - 1)} (x) = 0 }[/math]. Zatem dla dowolnego [math]\displaystyle{ k \geqslant 2 }[/math] spełnione są założenia twierdzenia E62. W przypadku rozpatrywanej przez nas sumy z twierdzenia E62 otrzymujemy
- [math]\displaystyle{ \Delta = \Delta (s, n) = {\small\frac{| B_{2 s} |}{(2 s) !}} | \mathop{\text{DLog}}(2 s - 2, n) | }[/math]
- [math]\displaystyle{ W = W (s, 2, n) = \sum_{k = 2}^n \mathop{\text{li}}(k) - \left[ n \mathop{\text{li}}(n) - \mathop{\text{li}}(n^2) + {\small\frac{1}{2}} \mathop{\text{li}}(n) + {\small\frac{B_2}{2 \log n}} + \sum_{k = 2}^s {\small\frac{B_{2 k}}{(2 k) !}} \mathop{\text{DLog}}(2 k - 2, n) \right] }[/math]
Obliczenia przeprowadziliśmy w programie PARI/GP. Wymagają one zwiększenia precyzji obliczeń do [math]\displaystyle{ 80 }[/math] miejsc znaczących i wcześniejszego przygotowania kilku funkcji omówionych szerzej w uwadze E70. Mamy
B(n, x) = sum(k = 0, n, 1/(k+1)*sum(j = 0, k, (-1)^j*binomial(k,j)*(x+j)^n)) A(n, k) = if( k == 1 || k == n, k*(n-1)!, k*A(n-1, k-1) + (n-1)*A(n-1, k) ) DLog(n, x) = (-1)^n * sum(k = 1, n, A(n,k)/( x^n * log(x)^(k+1) )) li(x) = real( -eint1( -log(x) ) ) delta(s, n) = B(2*s, 0)/(2*s)! * DLog(2*s-2, n) W(s, n) = sum(k = 2, n, li(k)) - n*li(n) + li(n^2) - 1/2*li(n) - B(2,0)/2! * 1/log(n) - sum(k = 2, s, B(2*k,0)/(2*k)! * DLog(2*k-2, n))
Dla [math]\displaystyle{ s = 5 }[/math] i [math]\displaystyle{ n = 10^7 }[/math] otrzymujemy (porównaj WolframAlpha)
[math]\displaystyle{ \Delta = {\small\frac{B_{10}}{10!}} \cdot | \mathop{\text{DLog}}(8, 10^7) | = 5.632 \cdot 10^{- 63} }[/math]
[math]\displaystyle{ W = 1.28191595049146577908068521816208913078488987854947239276268700691879666704021184913771562 }[/math]
Po uwzględnieniu możliwego błędu znajdujemy wartość stałej z dokładnością [math]\displaystyle{ 61 }[/math] miejsc po przecinku.
[math]\displaystyle{ C = 1.2819159504914657790806852181620891307848898785494723927626870 \ldots }[/math]
Przykład E66
Rozważmy jeszcze raz sumę
- [math]\displaystyle{ \sum_{k = 2}^n \mathop{\text{li}}(k) }[/math]
Wypiszmy dla tej sumy wzór Eulera-Maclaurina dla [math]\displaystyle{ r = 1 }[/math].
- [math]\displaystyle{ \sum_{k = 2}^n \mathop{\text{li}}(k) = \int_2^n \mathop{\text{li}}(t) d t + {\small\frac{1}{2}} \mathop{\text{li}}(n) + {\small\frac{1}{2}} \mathop{\text{li}}(2) + \int_2^n {\small\frac{P_1(t)}{\log t}} d t }[/math]
- [math]\displaystyle{ \;\;\;\: = (x \mathop{\text{li}}(x) - \mathop{\text{li}}(x^2)) \biggr\rvert_{2}^{n} + {\small\frac{1}{2}} \mathop{\text{li}}(n) + {\small\frac{1}{2}} \mathop{\text{li}}(2) + \int_2^n {\small\frac{P_1 (t)}{\log t}} d t }[/math]
- [math]\displaystyle{ \;\;\;\: = n \mathop{\text{li}}(n) - \mathop{\text{li}}(n^2) - 2 \mathop{\text{li}}(2) + \mathop{\text{li}}(4) + {\small\frac{1}{2}} \mathop{\text{li}}(n) + {\small\frac{1}{2}} \mathop{\text{li}}(2) + \int_2^{\infty} {\small\frac{P_1 (t)}{\log t}} d t - \int_n^{\infty} {\small\frac{P_1 (t)}{\log t}} d t }[/math]
- [math]\displaystyle{ \;\;\;\: = \left[ - {\small\frac{3}{2}} \mathop{\text{li}}(2) + \mathop{\text{li}}(4) + \int_2^{\infty} {\small\frac{P_1 (t)}{\log t}} d t \right] + n \mathop{\text{li}}(n) - \mathop{\text{li}}(n^2) + {\small\frac{1}{2}} \mathop{\text{li}}(n) - \int_n^{\infty} {\small\frac{P_1 (t)}{\log t}} d t }[/math]
Wyrażenie w nawiasie kwadratowym jest stałą wstępującą we wzorze Eulera-Maclaurina, zatem
- [math]\displaystyle{ C = - {\small\frac{3}{2}} \mathop{\text{li}}(2) + \mathop{\text{li}}(4) + \int_2^{\infty} {\small\frac{P_1(t)}{\log t}} d t }[/math]
- [math]\displaystyle{ \int_2^{\infty} {\small\frac{P_1 (t)}{\log t}} d t = C + {\small\frac{3}{2}} \mathop{\text{li}}(2) - \mathop{\text{li}}(4) }[/math]
W poprzednim przykładzie wyliczyliśmy wartość stałej [math]\displaystyle{ C }[/math]
- [math]\displaystyle{ C = 1.2819159504914657790806852181620891307848898785494723927626870 \ldots }[/math]
Wynika stąd natychmiast, że
- [math]\displaystyle{ \int_2^{\infty} {\small\frac{P_1 (t)}{\log t}} d t = -0.117923474371345921663180326620119770994144590988603907635106 \ldots }[/math]
Właśnie w taki sposób została obliczona wartość całki niewłaściwej, która występuje w zadaniu E49.
Przykład E67
Rozważmy sumę
- [math]\displaystyle{ \sum_{k = 0}^{n} e^k }[/math]
Mamy
- [math]\displaystyle{ f(x) = e^x }[/math]
- [math]\displaystyle{ F(x) = \int e^x d x = e^x }[/math]
- [math]\displaystyle{ f^{(r)} (x) = {\small\frac{d^r}{d x^r}} e^x = e^x }[/math]
Zatem ze wzoru Eulera-Maclaurina otrzymujemy
- [math]\displaystyle{ \sum_{k = 0}^{n} e^k = e^n - 1 + {\small\frac{1}{2}} (e^n + 1) + \sum_{k = 1}^s {\small\frac{B_{2 k}}{(2 k) !}} (e^n - 1) - {\small\frac{1}{(2 s) !}} \int_0^n P_{2 s} (t) e^t d t }[/math]
- [math]\displaystyle{ \sum_{k = 0}^{n} e^k = 1 + (e^n - 1) + {\small\frac{1}{2}} (e^n - 1) + \sum_{k = 1}^s {\small\frac{B_{2 k}}{(2 k) !}} (e^n - 1) - {\small\frac{1}{(2 s) !}} \int_0^n P_{2 s} (t) e^t d t }[/math]
- [math]\displaystyle{ \sum_{k = 0}^{n} e^k = 1 + (e^n - 1) \left( 1 + {\small\frac{1}{2}} + \sum_{k = 1}^s {\small\frac{B_{2 k}}{(2 k) !}} \right) - {\small\frac{1}{(2 s) !}} \int_0^n P_{2 s} (t) e^t d t }[/math]
Ponieważ dla [math]\displaystyle{ | x | \lt 2 \pi }[/math] prawdziwy jest wzór[18]
- [math]\displaystyle{ {\small\frac{x}{e^x - 1}} = \sum_{k = 0}^{\infty} {\small\frac{B_k \cdot x^k}{k!}} }[/math]
to dla [math]\displaystyle{ x = 1 }[/math] dostajemy
- [math]\displaystyle{ {\small\frac{1}{e - 1}} = \sum_{k = 0}^{\infty} {\small\frac{B_k}{k!}} = 1 - {\small\frac{1}{2}} + \sum_{k = 1}^{\infty} {\small\frac{B_{2 k}}{(2 k) !}} = {\small\frac{1}{2}} + \sum_{k = 1}^{\infty} {\small\frac{B_{2 k}}{(2 k) !}} }[/math]
W granicy, gdy [math]\displaystyle{ s }[/math] dąży do nieskończoności, mamy
- [math]\displaystyle{ \lim_{s \to \infty} \left[ 1 + (e^n - 1) \left( {\small\frac{3}{2}} + \sum_{k = 1}^s {\small\frac{B_{2 k}}{(2 k) !}} \right) \right] = 1 + (e^n - 1) \left( {\small\frac{3}{2}} + {\small\frac{1}{e - 1}} - {\small\frac{1}{2}} \right) = 1 + (e^n - 1) \left( 1 + {\small\frac{1}{e - 1}} \right) = {\small\frac{e^{n + 1} - 1}{e - 1}} }[/math]
W obliczeniu granicy całki dla [math]\displaystyle{ s }[/math] dążącego do nieskończoności pomocne będzie oszacowanie (zobacz E23)
- [math]\displaystyle{ {\small\frac{| B_{2 k} |}{(2 k) !}} \lt {\small\frac{2}{(2 \pi)^{2 k}}} \cdot {\small\frac{1}{1 - 2^{1 - 2 k}}} \leqslant {\small\frac{4}{(2 \pi)^{2 k}}} }[/math]
prawdziwe dla [math]\displaystyle{ k \geqslant 1 }[/math].
Teraz już łatwo znajdujemy
- [math]\displaystyle{ 0 \leqslant {\small\frac{1}{(2 s) !}} \left| \int_0^n P_{2 s} (t) e^t d t \right| \leqslant {\small\frac{1}{(2 s) !}} \int_0^n | P_{2 s} (t) | e^t d t \leqslant {\small\frac{| B_{2 s} |}{(2 s) !}} \int_0^n e^t d t = {\small\frac{| B_{2 s} |}{(2 s) !}} (e^n - 1) \lt {\small\frac{4}{(2 \pi)^{2 s}}} (e^n - 1) }[/math]
Dla dowolnego, ale ustalonego [math]\displaystyle{ n }[/math], jest
- [math]\displaystyle{ \lim_{s \to \infty} {\small\frac{4}{(2 \pi)^{2 s}}} (e^n - 1) = 0 }[/math]
Zatem z twierdzenia o trzech ciągach (zobacz twierdzenia C10 i C8) dostajemy natychmiast
- [math]\displaystyle{ \lim_{s \to \infty} {\small\frac{1}{(2 s) !}} \left| \int_0^n P_{2 s} (t) e^t d t \right| = \lim_{s \to \infty} {\small\frac{1}{(2 s) !}} \int_0^n P_{2 s} (t) e^t d t = 0 }[/math]
Ostatecznie otrzymujemy wzór
- [math]\displaystyle{ \sum_{k = 0}^{n} e^k = {\small\frac{e^{n + 1} - 1}{e - 1}} }[/math]
Znalezienie wzoru na sumę częściową szeregu geometrycznego nie jest odkrywcze, ale z pewnością było pouczające.
Uzupełnienie
Twierdzenie E68
Ogólny wzór na [math]\displaystyle{ n }[/math]-tą pochodną funkcji [math]\displaystyle{ {\small\frac{1}{\log x}} }[/math] ma postać
- [math]\displaystyle{ {\small\frac{d^n}{d x^n}} {\small\frac{1}{\log x}} = (- 1)^n \sum_{k = 1}^{n} {\small\frac{A^n_k}{x^n \log^{k + 1} x}} }[/math]
Liczby naturalne [math]\displaystyle{ A^n_k }[/math] spełniają następujące równania rekurencyjne
- [math]\displaystyle{ A^n_1 = (n - 1) A^{n - 1}_1 }[/math]
- [math]\displaystyle{ A_k^n = k A^{n - 1}_{k - 1} + (n - 1) A^{n - 1}_k \qquad }[/math] dla [math]\displaystyle{ \quad k = 2, \ldots, n - 1 }[/math]
- [math]\displaystyle{ A^n_n = n A^{n - 1}_{n - 1} }[/math]
gdzie [math]\displaystyle{ A^1_1 = 1 }[/math].
Twierdzenie E69
Z równań rekurencyjnych
- [math]\displaystyle{ A^n_1 = (n - 1) A^{n - 1}_1 }[/math]
- [math]\displaystyle{ A_k^n = k A^{n - 1}_{k - 1} + (n - 1) A^{n - 1}_k \qquad }[/math] dla [math]\displaystyle{ \quad k = 2, \ldots, n - 1 }[/math]
- [math]\displaystyle{ A^n_n = n A^{n - 1}_{n - 1} }[/math]
gdzie [math]\displaystyle{ A^1_1 = 1 }[/math], wynikają następujące wzory ogólne
- [math]\displaystyle{ A^n_1 = (n - 1) ! }[/math]
- [math]\displaystyle{ A^n_n = n! }[/math]
oraz
- [math]\displaystyle{ A^n_{n - 1} = {\small\frac{1}{2}} (n - 1) \cdot n! }[/math]
- [math]\displaystyle{ A^n_{n - 2} = {\small\frac{1}{24}} (n - 2) (3 n - 1) \cdot n! }[/math]
- [math]\displaystyle{ A^n_{n - 3} = {\small\frac{1}{48}} n (n - 1) (n - 3) \cdot n! }[/math]
- [math]\displaystyle{ A^n_{n - 4} = {\small\frac{1}{5760}} (n - 4) (15 n^3 - 30 n^2 + 5 n + 2) \cdot n! }[/math]
- [math]\displaystyle{ A^n_2 = 2 (n - 1) ! \cdot \sum_{k = 1}^{n - 1} {\small\frac{1}{k}} }[/math]
Uwaga E70
Z twierdzeń E68 i E69 wynika, że ogólną postać [math]\displaystyle{ n }[/math]-tej pochodnej funkcji [math]\displaystyle{ {\small\frac{1}{\log x}} }[/math] możemy łatwo wypisać
- [math]\displaystyle{ {\small\frac{d^n}{d x^n}} {\small\frac{1}{\log x}} = (- 1)^n \sum_{k = 1}^{n} {\small\frac{A^n_k}{x^n \log^{k + 1} x}} }[/math]
ale nie istnieje wzór ogólny, który pozwoliłby łatwo wyliczać wartości współczynników [math]\displaystyle{ A_k^n }[/math]. W tej sytuacji jedynym wyjściem jest wykorzystanie równania rekurencyjnego
- [math]\displaystyle{ A_k^n = k A^{n - 1}_{k - 1} + (n - 1) A^{n - 1}_k \qquad }[/math] dla [math]\displaystyle{ \quad k = 2, \ldots, n - 1 }[/math]
oraz wzorów
- [math]\displaystyle{ A^n_1 = (n - 1) ! }[/math]
- [math]\displaystyle{ A^n_n = n! }[/math]
Programy odwołujące się do wzorów rekurencyjnych są zazwyczaj niezwykle proste, ale należy ich unikać, bo działają wolno i zużywają duże ilości pamięci. Niżej podajemy przykłady prostych funkcji rekurencyjnych wyliczających silnię i liczby Fibonacciego napisanych w PARI/GP
silnia(n) = if( n == 0, 1, n*silnia(n-1) )
Fibonacci(n) = if( n <= 1, n, Fibonacci(n-1) + Fibonacci(n-2) )
W naszym przypadku rekurencji ominąć nie można, ale rozwiązaniem problemu jest równie prosta funkcja
A(n, k) = if( k == 1 || k == n, k*(n-1)!, k*A(n-1, k-1) + (n-1)*A(n-1, k) )
Dysponując funkcją wyliczającą współczynniki A(n, k)
, możemy łatwo zapisać wzór na [math]\displaystyle{ n }[/math]-tą pochodną funkcji [math]\displaystyle{ {\small\frac{1}{\log x}} }[/math]
DLog(n, x) = (-1)^n * sum(k = 1, n, A(n, k)/( x^n * log(x)^(k+1) ))
Powyższy wzór jest bardzo przydatny przy wyliczaniu wartości [math]\displaystyle{ {\small\frac{d^n}{d x^n}} {\small\frac{1}{\log x}} }[/math] dla większych liczb [math]\displaystyle{ n }[/math]. Jednak [math]\displaystyle{ n }[/math] nie może być zbyt duże – ceną, jaką musimy zapłacić za użycie funkcji rekurencyjnych, jest wydłużenie czasu obliczeń. Przykładowo obliczenie
DLog(26, 10^8) = 7.1305293508389973644228947613613744962 10^(-186)
trwało ponad pół minuty. Zobacz też WolframAlpha
Przypisy
- ↑ Wikipedia, Bernoulli polynomials, (Wiki-en)
- ↑ WolframAlpha, Bernoulli Polynomial, (WolframAlpha)
- ↑ Wolfram MathWorld, Bernoulli Polynomial, (Wolfram)
- ↑ NIST Digital Library of Mathematical Functions, Bernoulli and Euler Polynomials, (LINK)
- ↑ Skocz do: 5,0 5,1 Wikipedia, Twierdzenie Rolle’a, (Wiki-pl), (Wiki-en)
- ↑ D. H. Lehmer, On the Maxima and Minima of Bernoulli Polynomials, The American Mathematical Monthly, Vol. 47, No. 8 (Oct., 1940), pp. 533-538
- ↑ M. Abramowitz and I. A. Stegun (Eds), Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, National Bureau of Standards, Applied Mathematics Series 55, 10th printing, Washington, 1972, (LINK)
- ↑ Wikipedia, Abramowitz and Stegun, (Wiki-en)
- ↑ C. D'Aniello, On some inequalities for the Bernoulli numbers, Rendiconti del Circolo Matematico di Palermo Series II, Volume 43 (1994), pp. 329-332
- ↑ Feng Qi, A double inequality for the ratio of two non-zero neighbouring Bernoulli numbers, Journal of Computational and Applied Mathematics, Volume 351 (2019), pp. 1-5, (LINK)
- ↑ Twierdzenie Weierstrassa: Jeżeli funkcja [math]\displaystyle{ f(x) }[/math] określona w przedziale domkniętym jest ciągła, to jest w nim ograniczona. (Wiki-pl), (Wiki-en)
- ↑ Wikipedia, Euler–Maclaurin formula, (Wiki-en)
- ↑ Wikipedia, Wzór Stirlinga, (Wiki-pl), (Wiki-en)
- ↑ Wikipedia, Logarytm całkowy, (Wiki-pl), (Wiki-en)
- ↑ Wolfram MathWorld, Logarithmic Integral, (Wolfram)
- ↑ Wikipedia, Funkcja całkowo-wykładnicza, (Wiki-pl), (Wiki-en)
- ↑ Wolfram MathWorld, Exponential Integral, (Wolfram)
- ↑ Wikipedia, Liczby Bernoulliego, (Wiki-pl)