Dôkazy ekvivalentných podmienok pre Riemannovu hypotézu
Autor: Ing. Robert Polák (Robopol)
Dátum: 17.03.2022
Abstrakt
Hľadanie dôkazov ekvivalentných podmienok pre Riemannovu hypotézu. Ekvivalentné podmienky boli vytvorené v minulosti autormi: Srinivasa Ramanujan, Lagarias, Gronwall, Robin.
Úvod
Táto publikácia je venovaná hľadaniu dôkazov v prospech Riemannovej hypotézy. Nadväzuje priamo na články referencia (7),(8),(9). Články referencia (7),(8),(9) obsahujú postupnosť mojich úvah ako nájsť dôkaz ekvivalentných podmienok pre RH, ako urobil v minulosti Srinivasa Ramanujan (Ramanudžan), Lagarias, Gronwall ,Robin (viď. referencie (1),(2)). Túto publikáciu treba vnímať v kontexte článkov - referencie (7),(8),(9). V publikácii budem postupne predkladať kroky, postupy, ktoré vedú na dôkaz ekvivalentných podmienok pre RH.
POZNÁMKA: log x=ln x, vo vzťahoch nižšie je log x - prirodzený logaritmus.
Referencie (1), (2), (5)
Citácia:
Funkcia súčtu deliteľov \(\sigma\) je definovaná ako
\[ \sigma(n) := \sum_{d | n} d \]
Napríklad, \(\sigma(4) = 7\) a \(\sigma(p^n) = (p + 1)\sigma(n)\), ak \(p\) je prvočíslo nedeliteľné \(n\). V roku 1913 švédsky matematik Thomas Gronwall našiel maximálny poriadok \(\sigma\).
Teorém 1. (Gronwall)
\[ G(n) := \frac{\sigma(n)}{n \log (\log n)} \]
spĺňa \(\lim\limits_{n \to \infty} \sup G(n) = e^{\gamma} = 1.78107...\), kde \(\gamma\) je Euler-Mascheroniho konštanta.
Teorém 2. (Ramanujan)
Ak platí Riemannova hypotéza, potom
\[ G(n) < e^{\gamma} \quad (n \gg 1) \]
Kde \(n \gg 1\) znamená pre všetky dostatočne veľké \(n\). V roku 1984 francúzsky matematik Guy Robin dokázal, že silnejšie tvrdenie o funkcii \(G\) je ekvivalentné s RH.
Teorém 3. (Robin)
Riemannova hypotéza platí práve vtedy, ak
\[ G(n) < e^{\gamma} , \quad (n > 5040) \]
Teorém 4. (Lagarias)
Riemannova hypotéza platí práve vtedy, ak
\[ \sigma(n) < H_n + \exp(H_n) \log(H_n) \quad (n > 1) \]
kde \(H_n\) označuje \(n\)-té harmonické číslo:
\[ H_n = \sum_{n=1}^{\infty} \frac{1}{n} = 1 + \frac{1}{2} + \frac{1}{3} + \frac{1}{4} + ... \]
\[ \sum_{n=1}^{k} \frac{1}{n} = \ln k + \gamma + \epsilon_k \leq \ln k + 1, \quad \epsilon_k \sim \frac{1}{2k} \]
Numerické testy
Numerické testy sú podrobne popísané v článku – referencia (7), a pokročilé testy – referencia (8)
obr.1 Guy Robin test pre menšie a pokročilé testovanie.
Poznámka
Ideálne čísla sú v literatúre nazývané highly composite numbers.
Final verzia Riemann
Referencia (10):
Stiahnutie súboru v Pythone – Github: riemann_hypothesis_final.py
POPIS
Algoritmus obsahuje výpočet Guy Robin rovnice pre rôzne sekvencie, pre preverenie platnosti Riemannovej hypotézy. Optimálne trvá výpočet pre sekvenciu 3: 40000 ideálnych čísel zhruba 1–2 minúty.
Pokročilé testovanie
Verzia Riemann test (zdrojový kód programu)
Stiahnutie súboru v Pythone – Github: Riemann_test.py
Výsledky
Hodnota \(e^\gamma\) nebola prekročená. Test bol vykonaný až pre vysoko-zložené číslo pozostávajúce z prvých 500 tisíc prvočísiel násobených medzi sebou. Číslo "N" malo hodnotu až 3 201 675 číslic.
Výpočet sigma
Vo vzťahoch pre \(G(n)\) v zmysle Gronwall, Robin vystupuje:
\[ G(n): \]
Príklad: \( n = 12, \; \sigma(n) = 1 + 2 + 3 + 4 + 6 + 12 = 28 \)
Každé číslo je možné rozložiť na prvočíselné činitele:
\[ n = \prod_{i,j} p_i^{j_i} \quad , \; p_i \in \text{prvočísla}, \; j_i \in \mathbb{N} \quad \text{(1.1)} \]
Príklad: \( n = 2 \cdot 3 \cdot 5 \cdot 7 \)
Definujme jednoduchú postupnosť:
\[ n = \prod_{i} p_i \quad , \; p_i \in \text{prvočísla} \quad \text{(1.2)} \]
Potom platí:
\[ \sigma(n) = \prod_{p_i \in \text{prime}} (p_i + 1) \quad \text{(1.3)} \]
\[ \frac{\sigma(n)}{n} = \prod_{p_i \in \text{prime}} \left( 1 + \frac{1}{p_i} \right) \quad \text{(1.4)} \]
Príklad: \( n = 2 \cdot 3 \cdot 5 \cdot 7 \cdot 11 \cdot 13 = 30030 \)
Pre (1.1) platí:
\[ \sigma(n) = \prod_{p_i \in \text{prime}} \left(1 + p_i + p_i^2 + p_i^3 + \dots + p_i^{j_i} \right) \quad \text{(1.5)}\]
\[ \frac{\sigma(n)}{n} = \prod_{p_i \in \text{prime}} \frac{1 + p_i + p_i^2 + p_i^3 + \dots + p_i^{j_i}}{p_i^{j_i}} \quad \text{(1.6)}\]
Príklad: \( n = 2^4 \cdot 3^2 \cdot 5 \cdot 7 = 5040 \)
\( \sigma(n) = (1 + 2 + 2^2 + 2^3 + 2^4) \cdot (1 + 3 + 3^2) \cdot (5 + 1) \cdot (7 + 1) = 19344 \)
Vysoko zložené čísla
Definujme postupnosť (3) – vysoko zložené čísla. Vysoko zložené čísla sú také, kde maximalizujeme vzťah \(\sigma(n)/n\).
\[ \sup \frac{\sigma(n)}{n} = \sup \prod_{p_i \in \text{prime}} \frac{(1+p_i + p_i^2 + p_i^3 + \dots+ p_i^{j_i})}{p_i^{j_i}} \quad \text{(1.7)} \]
Príklad: \( n = 2^4 \cdot 3^2 \cdot 5 \cdot 7 \cdot 11 \cdot 13 = 720720 \)
\( \sigma(n) = 3249792,\quad \sigma(n)/n = 4.509 \)
Úprava rovnice (1.5):
Podľa referencie (2), strana 9:
\[ \sigma(n) = \prod_{p_i \in \text{prime}} \left(1 + p_i + p_i^2 + p_i^3 + \dots+ p_i^{j_i} \right) = \prod_{p_i \in \text{prime}} \frac{p_i^{j_i+1} - 1}{p_i - 1} \quad \text{(1.8)} \]
Dosadenie do rovnice (1.7):
\[ \sup \frac{\sigma(n)}{n} = \sup \prod_{p_i \in \text{prime}} \frac{p_i^{j_i+1} - 1}{(p_i - 1) p_i^{j_i}} \quad \text{(1.9)} \]
\[ \frac{\sigma(n)}{n} = (2 - 2^{-j_2}) (3/2 - 3^{-j_3}/2) (5/4 - 5^{-j_5}/4) (7/6 - 7^{-j_7}/6) \dots \quad \text{(1.10)} \]
\[ \frac{\sigma(n)}{n} = \prod_{p_i \in \text{prime}} \left(\frac{p_i}{p_i - 1} - \frac{p_i^{-j_i}}{p_i - 1} \right) \quad \text{(2.0)} \]
Definícia \(\beta(n)\):
\[ \beta(n) := \prod_{p_i \in \text{prime}} \frac{p_i}{p_i - 1} , \quad \beta(n) > \sup \frac{\sigma(n)}{n} \quad \text{(2.1)} \]
Reformulácia podmienok RH
Z empirického testovania s využitím programovacieho jazyka Python je zobrazený priebeh pre sekvenciu (1) a sekvenciu (3). (Viac v článku – referencia (8)).
tab. č.1 sekvencia (1)
tab. č.2 sekvencia (3)
Testovacia sekvencia (1)
Počiatočný test:
\[ n = \prod_{p_1 \in \text{prime}}^{p_{10000}} p_i, \quad \text{posledné prvočíslo} = p_{1000} = 104729 \]
Konečný test:
\[ n = \prod_{p_1 \in \text{prime}}^{p_{500000}} p_i, \quad \text{posledné prvočíslo} = p_{500000} = 7368787 \]
Testovacia sekvencia (3) – vysoko zložené čísla
V zmysle rovnice (1.7) a tabuľky č.2, pre vysoko zložené čísla:
Počiatočný test:
\[ n = \sup \prod_{p_1 \in \text{prime}}^{p_{10000}} p_i^{j_i}, \quad \text{posledné prvočíslo} = p_{1000} = 104729 \]
Konečný test:
\[ n = \sup \prod_{p_1 \in \text{prime}}^{p_{300000}} p_i^{j_i}, \quad \text{posledné prvočíslo} = p_{300000} = 4256233 \]
Výsledky numerických testov
Numerické testovanie ukazuje, že nasledujúce tvrdenia zrejme platia:
\[ \text{posledné prvočíslo} = p_n, \quad \log(n) \approx p_n \quad \text{(2.2)} \]
Pre sekvenciu (1):
\[ \log(n) < \text{posledné prvočíslo}, \quad \text{alebo} \quad \log(n) < p_n \quad \text{(2.3)} \]
Pre sekvenciu (3) – vysoko zložené čísla:
\[ \log(n) > \text{posledné prvočíslo}, \quad \text{alebo} \quad \log(n) > p_n \quad \text{(2.4)} \]
Dôkazy pre (2.3) a (2.4) sú v dodatku, viď. https://www.poling.sk/Riemann/Appendix_RH.html
Lagariasov teorém
\[ \sigma(n) < \log(n) + \gamma + \varepsilon + e^{\ln(n)+\gamma+\varepsilon} \log(\log(n) + \gamma + \varepsilon) \quad \text{(2.5)} \]
Toto vytvára limitu pre Gronwall teorém, pričom pre \(N \to \infty\), \( \varepsilon = 0 \):
\[ \lim_{n \to \infty} \frac{\sigma(n)}{n \log(\log(n))} = \lim_{n \to \infty} \frac{\log(n) + \gamma + n e^{\gamma} e^{\varepsilon} \log(\log(n)+\gamma)}{n \log(\log(n))} = e^{\gamma} \quad \text{(2.7)} \]
Guy Robin teorém
\[ \frac{\sigma(n)}{n} < e^{\gamma} \log(\log n) \quad \text{pre} \quad n > 5040 \quad \text{(2.8)} \]
V tejto stati vznikne silnejší teorém a pokiaľ sa ho podarí preukázať, platia aj zvyšné tvrdenia (Robin, Lagarias). Tento teorém je naviazaný na vysoko–zložené čísla, kde každé číslo \(n\) je práve vysoko–zložené číslo.
Robopol teorém
Pre vysoko zložené čísla v zmysle rovníc (1.7) a (2.1) dostávame:
\[ \beta(n) < e^\gamma \log(\log n) \quad \text{(3.0)} \]
\[ \prod_{p_i \in \text{prime}}^{p_n} \frac{p_i}{p_i - 1} < e^\gamma \log(\log(n)) \quad \text{(3.1)} \]
Pre vysoko zložené čísla – \(n\), ak \( p_n \geq p_{10} \).
Pre vysoko zložené čísla v zmysle rovnice (2.4), tabuľka č.2 dostávame:
\[ \prod_{p_i \in \text{prime}}^{p_n} \frac{p_i}{p_i - 1} < e^\gamma \log(p_n) \quad \text{(3.2)} \]
Pre vysoko zložené čísla – \(n\), ak \( p_n \geq p_{100} \).
Silnejšie tvrdenie ako rovnica (2.8)
Rovnice (3.1) a (3.2) sú silnejším tvrdením ako rovnica (2.8), pretože:
\[ \beta(n) > \sup \frac{\sigma(n)}{n} \quad \text{a} \quad \log(n) > p_n \quad \text{pre vysoko zložené číslo } n \quad \text{(3.3)} \]
Test Robopol teorému
Program (v Pythone) testuje Robopol teorém:
Referencia (10)
Stiahnutie súboru na GitHub: sigma-max_test.py
Tabuľka č.3 ukazuje výsledky testov pre veľmi veľké čísla:
tab. No. 3 Test robopol teorému pre veľmi veľké čísla.
Z tabuľky a grafu je vidieť, že je teorém veľmi tesne splnený a smerom k nekonečnu sa mierne priebehy rozchádzajú. To znamená, že teorém by mohol platiť (ako silnejšie tvrdenie oproti Robin teorému) do nekonečna. Na to však potrebujeme dôkaz. Samozrejme, toto platí v zmysle rovnice (3.2) pre vysoko–zložené čísla, ktoré sú už dostatočne veľké.
Dôkaz Robopol teorému iným spôsobom obsahuje dodatok č. 2.
Aproximácia \( \pi(x) \)
Obrázok č.2: Hodnoty \(\pi(n)\) pre prvých 60 kladných celých čísel, zdroj: Wikipedia.
Funkcia počítania prvočísel
\[ \pi(x) \approx \frac{x}{\log x}, \quad \text{rok: 1792} \quad \text{(3.3)} \]
\[ \lim_{x \to \infty} \frac{\pi(x)}{x / \log(x)} = 1 \quad \text{(3.4)} \]
Presnejšia aproximácia \(\pi(x)\)
\[ \pi(x) \approx \text{Li}(x) := \int_2^x \frac{dt}{\log t} \quad \text{(3.5)} \]
\[ \lim_{x \to \infty} \frac{\text{Li}(x)}{\pi(x)} = 1 \quad \text{(3.6)} \]
Obrázok č.3: Graf ukazujúci pomer funkcie počítania prvočísel \(\pi(x)\) k dvom jej aproximáciám: \(x / \log x\) a \(\text{Li}(x)\). Zdroj: Wikipedia.
Podľa zdroja (3) platí:
V práci Pierrea Dusarta existujú silnejšie verzie tohto typu nerovnosti, ktoré sú platné pre väčšie \(x\). Neskôr v roku 2010 Dusart dokázal:
\[ \frac{x}{\log(x)-1} < \pi(x) < \frac{x}{\log(x)-1.1} \quad \text{(3.7)} \]
De la Vallée Poussinov dôkaz naznačuje nasledujúce tvrdenie: Pre každé \(\varepsilon > 0\) existuje \(S\), také že pre všetky \(x > S\) platí:
\[ \frac{x}{\log(x)-(1-\varepsilon)} < \pi(x) < \frac{x}{\log(x)-(1+\varepsilon)} \quad \text{(3.8)} \]
Vzťah (3.8) je veľmi dôležitý pre následné dokazovanie.
Hovorí, že aproximácia \( \frac{x}{\log(x)-1} \) je tá, ktorá by mala byť približne rovná \( \pi(x) \) smerom k nekonečnu. Teda ide o kritickú hranicu (stred, linku). Zapíšme to:
\[ \pi(x) \approx \frac{x}{\log(x)-1} \quad \text{(3.9)} \]
\[ \lim_{x \to \infty} \pi(x) = \lim_{x \to \infty} \frac{x}{\log(x)-1} \quad \text{(3.10)} \]
Príklad
Pre \(x = 10^{25}\):
\[ \pi(10^{25}) = 176846309399143769411680 \]
\[ \frac{10^{25}}{\log(10^{25})-(1+0.019)} \approx \pi(10^{20}) \sim 2.02482 \cdot 10^{18} \]
\[ \frac{10^{25}}{\log(10^{25})-(1+0.019)} > \pi(10^{20}) \]
Robopol teorém súvislosti
V zmysle rovnice (3.2) odvodíme ďalšie súvislosti, ktoré by mali platiť.
Obrázok č.4: Priebeh grafu – Robopolov teorém (pre veľmi veľké čísla).
Tvrdenie (1.0)
Nech platí na celom definičnom obore rovnica (3.2):
\[ \beta(p_{k+1}) - \beta(p_k) \approx e^{\gamma} \log(p_{k+1}) - e^{\gamma} \log(p_k) \quad \text{(4.0)} \]
Obrázok č.5: Geometrická interpretácia rovnice 4.0.
Modifikácia rovnice (4.0)
\[ \prod_{p_i \in \text{prime}}^{p_{k+1}} \frac{p_i}{p_i - 1} - \prod_{p_i \in \text{prime}}^{p_k} \frac{p_i}{p_i - 1} \approx e^{\gamma} \log(p_{k+1}) - e^{\gamma} \log(p_k) \quad \text{(4.1)} \]
\[ \varepsilon_k = \prod_{p_i \in \text{prime}}^{p_k} \frac{p_i}{p_i - 1} , \quad \omega_k = \log(p_k) e^{\gamma} \quad \text{(4.2)} \]
Identifikácia počiatočných bodov
V zmysle obrázka č.6 stotožníme počiatočné body v \( p_k \):
\[ p_k^* = \varepsilon_k = \omega_k \]
Obrázok č.6: Grafické znázornenie počiatočných bodov v \( p_k \).
\[ \Delta \varepsilon_k = \varepsilon_k \frac{p_{k+1}}{p_{k+1} - 1} - \varepsilon_k = \varepsilon_k \left(\frac{p_{k+1}}{p_{k+1} - 1} - 1\right) \quad \text{(4.3)} \]
\[ \Delta \omega_k = e^{\gamma} (\log(p_{k+1}) - \log(p_k)) \quad \text{(4.4)} \]
V zmysle rovnice (4.1):
\[ \Delta \omega_k \approx \Delta \varepsilon_k, \quad e^{\gamma} (\ln(p_{k+1}) - \ln(p_k)) \approx \varepsilon_k \frac{p_{k+1}}{p_{k+1}-1} - \varepsilon_k \quad \text{(4.5)} \]
Po dosadení do rovnice (4.5) \( \varepsilon_k = \omega_k \) dostávame:
\[ e^{\gamma} \log(p_{k+1}) - e^{\gamma} \log(p_k) \approx \log(p_{k+1}) e^{\gamma} \frac{p_{k+1}}{p_{k+1} - 1} - \log(p_k) e^{\gamma} \quad \text{(4.6)} \]
\[ \log(p_{k+1}) \approx \log(p_k) \frac{p_{k+1}}{p_{k+1}-1} \quad \text{(4.7)} \]
Príklad
Pre veľmi veľké prvočísla:
\[ p_{999,999,999} = 22801763477 \]
\[ p_{1,000,000,000} = 22801763489 \]
\[ \log(22801763489) \approx \log(22801763477) \cdot \frac{22801763489}{22801763489-1} \]
Tvrdenie (2.0)
Nech existuje hladká, spojitá funkcia \( g(x) = f(x) \), ktorá aproximuje \( \pi(x) \) tak, že platí smerom k nekonečnu:
\[ \lim_{x \to \infty} g(x) = \lim_{x \to \infty} \pi(x) \quad \text{(4.8)} \]
Tvrdeniu (2.0) vyhovuje v zmysle rovníc (3.8) a (3.9):
\[ g(x) = \frac{x}{\log(x)-1} \quad \text{(4.9)} \]
Tvrdenie (3.0)
Nech platí na definičnom obore funkcie \( g(x) \) pre všetky \( x \geq 100 \) v zmysle rovnice (4.7) táto rovnica:
\[ \log(x + \Delta x) \geq \log(x) \frac{x+\Delta x}{x+\Delta x-1} \quad \text{(4.10)} \]
Kde \( \Delta x \) predstavuje horizontálnu vzdialenosť podľa obrázka č.7.
Obrázok č.7: Horizontálna vzdialenosť \( \Delta x \) v zmysle rovnice (4.10).
Zadefinujme najskôr rovnicu v zmysle obrázka č.7:
\[ \frac{x}{\log(x)-1} + 1 = \frac{x+\Delta x}{\log(x+\Delta x)-1} \quad \text{(5.0)} \]
Alebo alternatívne:
\[ \frac{x}{\log(x)-1} = 1 + \frac{x-\Delta x}{\log(x-\Delta x)-1} \quad \text{(5.1)} \]
Pointa tvrdení (1.0), (2.0) a (3.0)
Pointa tvrdení (1.0), (2.0) a (3.0) je v tom, že ak je splnená rovnica (4.10) na celom definičnom obore \( x > 100 \) - pre všetky \( \Delta x \), ktoré vypočítame z rovnice (5.1) – potom nutne platí aj teorém, tj. rovnice (3.0) a (3.2). Zadefinujeme to do tvrdenia (4.0).
Tvrdenie (4.0)
Ak platí rovnica (4.10) pre všetky \( x > 100 \), pričom \( \Delta x \) vypočítame z rovnice (5.1), potom nutne platí aj teorém - rovnica (3.0) a (3.2).
Vysvetlenie
Rovnica (4.10) je pomerne jednoduchá a intuitívna. Je už spojitá oproti pôvodnej (4.7), ktorá obsahovala iba približné hodnoty, pretože prvočísla v \(x\) vykazujú drobné fluktuácie od strednej hodnoty vyhladenej funkcie, ktorá by ich dokázala dokonale aproximovať (ako hladká krivka bez ozubenia).
\[ \log(p_{k+1}) > \log(p_k) \frac{p_{k+1}}{p_{k+1}-1} \quad \text{(5.2)} \]
Teda, aby sme potvrdili platnosť rovníc (3.0) a (3.2), preskúmame (5.2) v ľubovoľnom bode \(p_k\) a jeho suseda \(p_{k+1}\) - a to na vyhladenej funkcii \(g(x)\) pre všetky \(x > 100\). Ak preukážeme platnosť pre všetky \(x\), potom je zrejmé, že aj rovnice (3.0) a (3.2) musia platiť.
Niekto by mohol namietať, že \(g(x)\) neaproximuje \(\pi(x)\) dobre pre malé hodnoty, teda pre malé \(x\) platí, že \(g(x) < \pi(x)\). No zároveň vieme, že limitne \(g(x)\) doháňa \(\pi(x)\) (v zmysle rovnice (4.8)). Teda ak limitne \(g(x)\) doháňa \(\pi(x)\), je zjavné, že smerom k nekonečnu – sklon \(g(x)\) rastie viac než sklon \(\pi(x)\).
Z numerických testov (tabuľka č. 3) je preukázané, že rovnice (3.0) a (3.2) platia pre veľmi veľké čísla - konkrétne pre vysoko zložené čísla s \(p_n\) uvedenými v tabuľke č. 3. Súčasne vidíme, že krivky sa mierne rozchádzajú, pričom má stále viac navrch \( e^{\log(p_n)} \).
V nasledujúcich statiach sa teda pokúsim dokázať, že kľúčové tvrdenie (4.0) naozaj platí.
Upravenie rovnice (4.10)
Upravme rovnicu (4.10) nasledovne:
\[ \log(x + \Delta x) \geq \log(x) \frac{x+\Delta x}{x+\Delta x-1} \]
Použijeme substitúciu: \( x + \Delta x = t \).
\[ \log(t) \geq \log(t - \Delta x) \frac{t}{t-1} \]
\[ \log(t - \Delta x) \geq \frac{\log(t) \cdot (t-1)}{t} \]
Eliminujeme logaritmus:
\[ t - \Delta x \geq t^{(t-1)/t} \quad \text{(5.3)} \]
\[ \Delta x \geq t - t^{(t-1)/t} \]
Definícia \( \Delta x_{min} \)
\[ \Delta x_{min} := t - t^{(t-1)/t} \quad \text{(5.4)} \]
\[ \Delta x \geq \Delta x_{min} \quad \text{(5.5)} \]
Derivácia funkcie
\[ \frac{d}{dt} \Bigl(t - t^{(t-1)/t}\Bigr) = t^{-(t+1)/t} \left( t + \log(t) - 1 \right) \quad \text{(5.6)} \]
Limita funkcie
\[ \lim_{t \to \infty} t^{-(t+1)/t} \left( t + \log(t) - 1 \right) = 1 \quad \text{(5.7)} \]
Príklad
Pre hodnotu \( t = 1000 \):
\[ \Delta x = 1000 - 1000^{(1000-1)/1000} = 6.883951579\ldots \]
Výpočet hodnoty \( x \):
\[ x = t - \Delta x_{min} = 1000 - 6.883951579 = 993.116048420\ldots \]
Overenie rovnice:
\[ \log(x + \Delta x_{min}) = \log(x) \frac{x+\Delta x_{min}}{x+\Delta x_{min}-1} \]
Po dosadení hodnôt:
\[ \log(1000) = \log(993.116048420) \cdot \frac{993.116048420+6.883951579}{993.116048420+6.883951579-1} \]
Výsledok:
\[ 6.907755279\ldots = 6.907755279\ldots \]
Analýza aproximácie \( \pi(x)=\frac{x}{\log(x)} \)
Najskôr preveríme jednoduchší prípad aproximácie \( \pi(x)=\frac{x}{\log(x)} \).
V zmysle obrázka č.7 (pre aproximáciu \( \pi(x) = \frac{x}{\log(x)} \)) dostávame rovnicu:
\[ \frac{x}{\log x} = 1 + \frac{x - \Delta x}{\log (x - \Delta x)} \quad \text{(6.0)} \]
Analytické riešenie rovnice
Skutočné analytické riešenie tejto rovnice je:
\[ \Delta x = x - e^{-W_{-1}\Bigl(\frac{\log(x)}{\log(x)-x}\Bigr)} , \quad \Delta x > 0 \quad \text{(6.1)} \]
Alternatívne:
\[ \Delta x = x - \frac{\log(x)-x}{\log(x)-1}\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right), \quad \Delta x > 0 \quad \text{(6.2)} \]
Kde \(W_{-1}(z)\) je Lambertova funkcia (referencia (4)).
Derivácia funkcie
\[ \frac{d}{dx} \left( \frac{\log(x)-x}{\log(x)}\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right) \right)
= -\frac{(\log(x)-1)\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)^2}{\log^2(x) \Bigl(W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)+1\Bigr)} \quad \text{(6.3)} \]
Limita funkcie
\[ \lim_{x \to \infty} -\frac{(\log(x)-1)\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)^2}{\log^2(x) \Bigl(W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)+1\Bigr)} \]
Z hľadiska štandardných matematických funkcií sa nenašiel žiadny uzavretý výsledok.
V zmysle rovnice (5.5) pre \( x > 100 \) musí platiť:
\[ x - \frac{\log(x)-x}{\log(x)}\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-x}\right) \geq x - x^{(x-1)/x} \quad \text{(6.4)} \]
Príklady výpočtov
Príklad 1: \( x = 10^6 \)
\[ \Delta x \approx 14.89353360214\ldots \]
\[ \Delta x_{min} \approx 13.815415124237\ldots \]
\[ \Delta x - \Delta x_{min} \approx 1.078118477907\ldots \]
Príklad 2: \( x = 10^{200} \)
\[ \Delta x \approx 461.519194796772488\ldots \]
\[ \Delta x_{min} \approx 460.517018598809\ldots \]
\[ \Delta x - \Delta x_{min} \approx 1.00217\ldots \]
Z numerických výpočtov by sme očakávali, že limita:
\[ \lim_{x \to \infty} (\Delta x - \Delta x_{min}) = 1 \]
Úprava rovnice (6.0)
Upravme rovnicu (6.0) na tento požadovaný tvar:
\[ \frac{x}{\log x} - 1 - \frac{x-\Delta x}{\log(x-\Delta x)} = 0 \quad \text{(6.5)} \]
Ďalší krok:
\[ \frac{x\,\log(x-\Delta x) - \log(x)\,\log(x-\Delta x) - \log(x)(x-\Delta x)}{\log(x)\,\log(x-\Delta x)} = 0 \]
Zjednodušenie:
\[ (x-\log(x))\,\log(x-\Delta x) - \log(x)(x-\Delta x) = 0 \quad \text{(6.6)} \]
Taylorov rozvoj
Urobme Taylorov rozvoj podľa \( x \) a dostaneme:
\[
\begin{aligned}
& (\Delta x + \Delta x \log(x) - \log^2(x)) - \frac{\Delta x (\Delta x - 2 \log(x))}{2x} \\
& + \frac{\Delta x^2 (3 \log(x) - 2 \Delta x)}{6x^2} + O\left(\frac{1}{x^3}\right)
\end{aligned}
\quad \text{(6.7)}
\]
Pre \( x \to \infty \) sa rovnica redukuje na:
\[ \Delta x + \Delta x \log(x) - \log^2(x) = 0 \quad \text{(6.8)} \]
Vyjadrenie \( \Delta x \)
Vyjadríme \( \Delta x \) z rovnice:
\[ \Delta x = \frac{\log^2(x)}{\log(x)-1} \quad \text{(6.9)} \]
Výpočet limity
Podľa (6.4) vypočítame limitu:
\[ \lim_{x \to \infty} \Bigl(\Delta x - \Delta x_{min}\Bigr)
= \lim_{x \to \infty} \left[\frac{\log^2(x)}{\log(x)-1} - \Bigl(x - x^{(x-1)/x}\Bigr)\right] = 1 \quad \text{(6.10)} \]
Výsledok
Rovnica (6.10) ukázala, že rovnica \( \Delta x \geq \Delta x_{min} \) platí až do nekonečna.
Analýza aproximácie \( \pi(x)=\frac{x}{\log(x)-1} \)
Teraz preveríme aproximáciu \( \pi(x) = \frac{x}{\log(x)-1} \) v zmysle rovníc (3.9) a (3.10).
Riešením rovnice (5.1) je:
\[
\Delta x = \frac{(x - \log(x) + 1) \cdot W_{-1} \left(-\frac{\sqrt[1+x-\log(x)]{\frac{e^{x+1}}{x} (\log(x)-1)}}{\,x - \log(x) + 1}\right)}{\log(x)-1} + x \quad \text{(7.0)}
\]
V zmysle rovnice (5.5) pre \( x > 100 \) musí platiť:
\[ \Delta x \geq \Delta x_{min} \]
\[
\frac{(x - \log(x) + 1) \cdot W_{-1} \left(-\frac{\sqrt[\,x+1-\log(x)]{\frac{e^{x+1}}{x} - (\log(x)-1)}}{\,x - \log(x) + 1}\right)}{\log(x)-1} + x \geq x - x^{(x-1)/x} \quad \text{(7.1)}
\]
Príklad výpočtu
Pre hodnotu \( x = 10^{180} \) máme:
\[
\Delta x = \dots = 414.467741185201\ldots
\]
Minimálna hodnota \( \Delta x \):
\[
\Delta x_{min} = 10^{180} - \Bigl(10^{180}\Bigr)^{(10^{180}-1)/10^{180}} = 414.465316\ldots
\]
Rozdiel medzi \( \Delta x \) a \( \Delta x_{min} \):
\[
\Delta x > \Delta x_{min}, \quad \Delta x - \Delta x_{min} \sim 0
\]
Z numerických výpočtov by sme očakávali, že limita:
\[
\lim_{x \to \infty} \Delta x - \Delta x_{min} = 0
\]
Teraz preveríme aproximáciu \( \pi(x) = \frac{x}{\log(x)-1} \) v zmysle rovníc (3.9) a (3.10).
Úprava rovnice (5.1)
Upravme rovnicu (5.1) na tento požadovaný tvar:
\[
\frac{x}{\log x-1} - 1 - \frac{x-\Delta x}{\log(x-\Delta x)-1} = 0 \quad \text{(7.2)}
\]
\[
\frac{x}{\log(x)-1} - \frac{x}{\log(x-\Delta x)-1} + \frac{\Delta x}{\log(x-\Delta x)-1} - 1 = 0 \quad \text{(7.3)}
\]
Rozvoj podľa \( x \)
Puiseuxov rad:
\[
-\frac{2 \Delta x - (\Delta x - 2) \log(x) + \log^2(x) + 1}{(\log(x)-1)^2}
+ \frac{\Delta x^2 (\log(x)-3)}{2 x (\log(x)-1)^3} + O\left(\frac{1}{x^2}\right)
\quad \text{(7.4)}
\]
Pre \( x \to \infty \) sa rovnica redukuje na:
\[
-\frac{2 \Delta x - (\Delta x - 2)\log(x) + \log^2(x) + 1}{(\log(x)-1)^2} = 0 \quad \text{(7.5)}
\]
Vyjadrenie z rovnice pre \( \Delta x \)
\[
\Delta x = \frac{(\log(x)-1)^2}{\log(x)-2} \quad \text{(7.6)}
\]
Výpočet limity
Vypočítame limitu:
\[
\lim_{x \to \infty} \Delta x - \Delta x_{min} = \lim_{x \to \infty} \left[\frac{(\log(x)-1)^2}{\log(x)-2} - \Bigl(x - x^{(x-1)/x}\Bigr)\right] = 0 \quad \text{(7.9)}
\]
Rozvoj série pre \( x \to \infty \):
\[
\frac{1}{\log(x)-2} + \frac{\log^2(x)}{2x} + O\left(\frac{1}{x^2}\right)
\quad \text{(7.10)}
\]
Výsledok
Rovnica (7.9) ukázala, že rovnica \( \Delta x \geq \Delta x_{min} \) platí až do nekonečna.
Analýza aproximácie \( \pi(x)=\frac{x}{\log(x)-\epsilon} \)
Pre aproximáciu platí:
\[
\pi(x) = \frac{x}{\log(x)-\epsilon}, \quad \epsilon = 1+\varepsilon, \quad \varepsilon>0 \quad \text{(8.0)}
\]
Riešením rovnice (8.0) je:
\[
\Delta x = \frac{(x+\log(x)+\epsilon) \cdot W_{-1}\!\left(\frac{\left(e^{\frac{\epsilon^2}{x+\epsilon}} \cdot x^{-\frac{\epsilon}{x}}\right)^{\frac{x}{x+\epsilon-\log(x)}} \cdot (\epsilon-\log(x))}{x-\log(x)+\epsilon}\right)}{\log(x)-\epsilon} + x \quad \text{(8.1)}
\]
V zmysle rovnice (5.5) pre \( x > 100 \) musí platiť:
\[ \Delta x \geq \Delta x_{min} \quad \text{(8.2)} \]
\[
\frac{(x+\log(x)+\epsilon) \cdot W_{-1}\!\left(\frac{\left(e^{\frac{\epsilon^2}{x+\epsilon}} \cdot x^{-\frac{\epsilon}{x}}\right)^{\frac{x}{x+\epsilon-\log(x)}} \cdot (\epsilon-\log(x))}{x-\log(x)+\epsilon}\right)}{\log(x)-\epsilon} + x \geq x - x^{(x-1)/x} \quad \text{(8.2)}
\]
Príklady
Príklad 1:
\[ x = 10^{15}, \quad \varepsilon = 1.0000000000001 \]
\[ \Delta x = \frac{277}{8} = 34.625 \]
\[ \Delta x_{min} = 10^{15} - (10^{15})^{(10^{15}-1)/10^{15}} = 34.538 \]
\( \Delta x > \Delta x_{min} \)
Príklad 2:
\[ x = 10^{16}, \quad \varepsilon = 1.0000000000001 \]
\[ \Delta x = 36 \]
\[ \Delta x_{min} = 10^{16} - (10^{16})^{(10^{16}-1)/10^{16}} = 36.841 \]
\( \Delta x < \Delta x_{min} \) – teda rovnica (8.2) nie je splnená.
Analýza výsledku
Rovnica (8.2) pre aproximáciu (8.0) nebude všeobecne splnená. Aproximácia (8.0) je však v zmysle rovnice (3.8) väčšia ako \( \pi(x) \).
Referencie / References:
Proofs of Equivalent Conditions for the Riemann Hypothesis
Author: Ing. Robert Polák (Robopol)
Date: 17.03.2022
Abstract
Searching for proofs of equivalent conditions for the Riemann Hypothesis. These equivalent conditions were previously established by authors such as Srinivasa Ramanujan, Lagarias, Gronwall, and Robin.
Introduction
This publication is devoted to finding evidence in favor of the Riemann hypothesis. It follows up directly on the articles with the reference (7), (8), (9). Articles with the reference (7), (8), (9) contain a sequence of my thoughts on how to find evidence of equivalent conditions for RH, as Srinivasa Ramanujan (Ramanujan), Lagarias, Gronwall, Robin did in the past (see references (1), (2). )). This publication should be read in the context of the articles - references (7), (8), (9). In the publication, I will gradually present the steps and procedures that lead to the proof of equivalent conditions for RH.
NOTE: log x = ln x, log x - the natural logarithm.
References (1), (2), (5)
Citation:
The sum-of-divisors function \(\sigma\) is defined as
\[ \sigma(n) := \sum_{d | n} d \]
For example, \(\sigma(4) = 7\) and \(\sigma(p^n) = (p+1)\sigma(n)\) if \(p\) is a prime that does not divide \(n\). In 1913, the Swedish mathematician Thomas Gronwall determined the maximal order of \(\sigma\).
Theorem 1 (Gronwall)
\[ G(n) := \frac{\sigma(n)}{n \log (\log n)} \]
It satisfies \(\lim_{n \to \infty} \sup G(n) = e^{\gamma} = 1.78107\ldots\), where \(\gamma\) is the Euler–Mascheroni constant.
Theorem 2 (Ramanujan)
If the Riemann Hypothesis holds, then
\[ G(n) < e^{\gamma} \quad (n \gg 1) \]
where \(n \gg 1\) means for all sufficiently large \(n\). In 1984, the French mathematician Guy Robin proved that a stronger statement about \(G(n)\) is equivalent to RH.
Theorem 3 (Robin)
The Riemann Hypothesis holds if and only if
\[ G(n) < e^{\gamma}, \quad (n > 5040) \]
Theorem 4 (Lagarias)
The Riemann Hypothesis holds if and only if
\[ \sigma(n) < H_n + \exp(H_n) \log(H_n) \quad (n > 1) \]
where \(H_n\) denotes the \(n\)th harmonic number:
\[ H_n = \sum_{n=1}^{\infty} \frac{1}{n} = 1 + \frac{1}{2} + \frac{1}{3} + \frac{1}{4} + \dots \]
\[ \sum_{n=1}^{k} \frac{1}{n} = \ln k + \gamma + \epsilon_k \leq \ln k + 1, \quad \epsilon_k \sim \frac{1}{2k} \]
Numerical Tests
The numerical tests are described in detail in reference (7), and the advanced tests in reference (8).
Fig.1: Guy Robin test for basic and advanced testing.
Note
"Ideal numbers" are referred to in the literature as highly composite numbers.
Final Version Riemann
Reference (10):
Download the Python file – GitHub: riemann_hypothesis_final.py
Description
The algorithm computes the Guy Robin formula for various sequences to verify the validity of the Riemann Hypothesis. Optimally, the computation for sequence 3 (about 40,000 ideal numbers) takes around 1–2 minutes.
Advanced Testing
Version Riemann Test (source code)
Download the Python file – GitHub: Riemann_test.py
Results
The value \(e^\gamma\) was not exceeded. The test was performed on a highly composite number obtained by multiplying together the first 500,000 primes. The number "N" had as many as 3,201,675 digits.
Computation of \(\sigma\)
In the formulas for \(G(n)\) in the sense of Gronwall and Robin, we have:
\[ G(n): \]
Example: \( n = 12, \; \sigma(n) = 1 + 2 + 3 + 4 + 6 + 12 = 28 \)
Every number can be factorized into its prime factors:
\[ n = \prod_{i,j} p_i^{j_i} \quad , \; p_i \in \text{primes}, \; j_i \in \mathbb{N} \quad \text{(1.1)} \]
Example: \( n = 2 \cdot 3 \cdot 5 \cdot 7 \)
Define the simple sequence:
\[ n = \prod_{i} p_i \quad , \; p_i \in \text{primes} \quad \text{(1.2)} \]
Then it holds that:
\[ \sigma(n) = \prod_{p_i \in \text{prime}} (p_i + 1) \quad \text{(1.3)} \]
\[ \frac{\sigma(n)}{n} = \prod_{p_i \in \text{prime}} \left( 1 + \frac{1}{p_i} \right) \quad \text{(1.4)} \]
Example: \( n = 2 \cdot 3 \cdot 5 \cdot 7 \cdot 11 \cdot 13 = 30030 \)
For (1.1) it holds that:
\[ \sigma(n) = \prod_{p_i \in \text{prime}} \left(1 + p_i + p_i^2 + p_i^3 + \dots + p_i^{j_i}\right) \quad \text{(1.5)} \]
\[ \frac{\sigma(n)}{n} = \prod_{p_i \in \text{prime}} \frac{1 + p_i + p_i^2 + p_i^3 + \dots + p_i^{j_i}}{p_i^{j_i}} \quad \text{(1.6)} \]
Example: \( n = 2^4 \cdot 3^2 \cdot 5 \cdot 7 = 5040 \)
\( \sigma(n) = (1+2+2^2+2^3+2^4) \cdot (1+3+3^2) \cdot (5+1) \cdot (7+1) = 19344 \)
Highly Composite Numbers
Define sequence (3) – highly composite numbers. These are numbers that maximize the ratio \(\sigma(n)/n\).
\[ \sup \frac{\sigma(n)}{n} = \sup \prod_{p_i \in \text{prime}} \frac{(1+p_i + p_i^2 + p_i^3 + \dots+ p_i^{j_i})}{p_i^{j_i}} \quad \text{(1.7)} \]
Example: \( n = 2^4 \cdot 3^2 \cdot 5 \cdot 7 \cdot 11 \cdot 13 = 720720 \)
\( \sigma(n) = 3249792,\quad \sigma(n)/n = 4.509 \)
Modification of Equation (1.5):
According to reference (2), page 9:
\[ \sigma(n) = \prod_{p_i \in \text{prime}} \left(1+p_i+p_i^2+p_i^3+\dots+p_i^{j_i}\right)
= \prod_{p_i \in \text{prime}} \frac{p_i^{j_i+1}-1}{p_i-1} \quad \text{(1.8)} \]
Substitution into Equation (1.7):
\[ \sup \frac{\sigma(n)}{n} = \sup \prod_{p_i \in \text{prime}} \frac{p_i^{j_i+1}-1}{(p_i-1)p_i^{j_i}} \quad \text{(1.9)} \]
\[ \frac{\sigma(n)}{n} = (2-2^{-j_2})\,(3/2-3^{-j_3}/2)\,(5/4-5^{-j_5}/4)\,(7/6-7^{-j_7}/6)\dots \quad \text{(1.10)} \]
\[ \frac{\sigma(n)}{n} = \prod_{p_i \in \text{prime}} \left(\frac{p_i}{p_i-1} - \frac{p_i^{-j_i}}{p_i-1} \right) \quad \text{(2.0)} \]
Definition of \(\beta(n)\):
\[ \beta(n) := \prod_{p_i \in \text{prime}} \frac{p_i}{p_i-1}, \quad \beta(n) > \sup \frac{\sigma(n)}{n} \quad \text{(2.1)} \]
Reformulation of the RH Conditions
Empirical testing using Python shows the behavior for sequence (1) and sequence (3) (see reference (8) for details).
Table no.1: sequence (1)
Table no.2: sequence (3)
Test Sequence (1)
Initial test:
\[ n = \prod_{p_1 \in \text{prime}}^{p_{10000}} p_i, \quad \text{last prime} = p_{1000} = 104729 \]
Final test:
\[ n = \prod_{p_1 \in \text{prime}}^{p_{500000}} p_i, \quad \text{last prime} = p_{500000} = 7368787 \]
Test Sequence (3) – Highly Composite Numbers
In the sense of (1.7) and Table no.2, for highly composite numbers:
Initial test:
\[ n = \sup \prod_{p_1 \in \text{prime}}^{p_{10000}} p_i^{j_i}, \quad \text{last prime} = p_{1000} = 104729 \]
Final test:
\[ n = \sup \prod_{p_1 \in \text{prime}}^{p_{300000}} p_i^{j_i}, \quad \text{last prime} = p_{300000} = 4256233 \]
Numerical Test Results
Numerical experiments suggest that the following statements likely hold:
\[ \text{last prime} = p_n, \quad \log(n) \approx p_n \quad \text{(2.2)} \]
For sequence (1):
\[ \log(n) < \text{last prime}, \quad \text{or} \quad \log(n) < p_n \quad \text{(2.3)} \]
For sequence (3) – highly composite numbers:
\[ \log(n) > \text{last prime}, \quad \text{or} \quad \log(n) > p_n \quad \text{(2.4)} \]
Proofs for (2.3) and (2.4) can be found in the appendix, see https://www.poling.sk/Riemann/Appendix_RH.html
Lagarias' Theorem
\[ \sigma(n) < \log(n) + \gamma + \varepsilon + e^{\ln(n)+\gamma+\varepsilon} \log(\log(n)+\gamma+\varepsilon) \quad \text{(2.5)} \]
This sets a limit for Gronwall's theorem, and as \( N \to \infty \) with \( \varepsilon = 0 \):
\[ \lim_{n \to \infty} \frac{\sigma(n)}{n \log(\log(n))} = \lim_{n \to \infty} \frac{\log(n) + \gamma + n e^{\gamma} e^{\varepsilon}\log(\log(n)+\gamma)}{n \log(\log(n))} = e^{\gamma} \quad \text{(2.7)} \]
Guy Robin Theorem
\[ \frac{\sigma(n)}{n} < e^{\gamma} \log(\log n) \quad \text{for } n > 5040 \quad \text{(2.8)} \]
In this section, an even stronger theorem emerges; if it can be proven then the other statements (Robin, Lagarias) also hold. This theorem is tied to highly composite numbers, where each number \( n \) is a highly composite number.
Robopol Theorem
For highly composite numbers in the sense of (1.7) and (2.1), we obtain:
\[ \beta(n) < e^\gamma \log(\log n) \quad \text{(3.0)} \]
\[ \prod_{p_i \in \text{prime}}^{p_n} \frac{p_i}{p_i-1} < e^\gamma \log(\log(n)) \quad \text{(3.1)} \]
For a highly composite number \( n \), if \( p_n \geq p_{10} \).
For highly composite numbers in the sense of (2.4), Table no.2 gives:
\[ \prod_{p_i \in \text{prime}}^{p_n} \frac{p_i}{p_i-1} < e^\gamma \log(p_n) \quad \text{(3.2)} \]
For a highly composite number \( n \), if \( p_n \geq p_{100} \).
A Stronger Statement than (2.8)
Equations (3.1) and (3.2) are stronger than (2.8) because:
\[ \beta(n) > \sup \frac{\sigma(n)}{n} \quad \text{and} \quad \log(n) > p_n \quad \text{for a highly composite number } n \quad \text{(3.3)} \]
Robopol Theorem Test
A Python program tests the Robopol theorem:
Reference (10)
Download from GitHub: sigma-max_test.py
Table no.3 shows test results for very large numbers:
Table no.3: Test of the Robopol theorem for very large numbers.
From the table and graph, it is evident that the theorem is barely satisfied and that as we approach infinity the curves diverge slightly. This suggests that the theorem (as a stronger statement than Robin's theorem) may hold to infinity. However, a formal proof is needed. Note that this applies in the sense of (3.2) for sufficiently large highly composite numbers.
Alternative Proof of Robopol Theorem via Mertens is included in Appendix No. 2.
Approximation of \( \pi(x) \)
Fig.2: Values of \(\pi(n)\) for the first 60 positive integers (source: Wikipedia).
Prime-Counting Function
\[ \pi(x) \approx \frac{x}{\log x}, \quad \text{year: 1792} \quad \text{(3.3)} \]
\[ \lim_{x \to \infty} \frac{\pi(x)}{x / \log(x)} = 1 \quad \text{(3.4)} \]
More Accurate Approximation of \(\pi(x)\)
\[ \pi(x) \approx \mathrm{Li}(x) := \int_2^x \frac{dt}{\log t} \quad \text{(3.5)} \]
\[ \lim_{x \to \infty} \frac{\mathrm{Li}(x)}{\pi(x)} = 1 \quad \text{(3.6)} \]
Fig.3: Graph showing the ratio of the prime-counting function \(\pi(x)\) to its two approximations: \(x/\log x\) and \(\mathrm{Li}(x)\). Source: Wikipedia.
According to source (3):
Pierre Dusart's work provides stronger versions of these inequalities valid for larger \(x\). Later in 2010, Dusart proved:
\[ \frac{x}{\log(x)-1} < \pi(x) < \frac{x}{\log(x)-1.1} \quad \text{(3.7)} \]
De la Vallée Poussin's proof suggests that for every \(\varepsilon > 0\) there exists an \(S\) such that for all \(x > S\) it holds:
\[ \frac{x}{\log(x)-(1-\varepsilon)} < \pi(x) < \frac{x}{\log(x)-(1+\varepsilon)} \quad \text{(3.8)} \]
The relation (3.8) is crucial for the subsequent arguments.
This indicates that the approximation \( \frac{x}{\log(x)-1} \) is the critical boundary for \(\pi(x)\) as \(x \to \infty\). In other words:
\[ \pi(x) \approx \frac{x}{\log(x)-1} \quad \text{(3.9)} \]
\[ \lim_{x \to \infty} \pi(x) = \lim_{x \to \infty} \frac{x}{\log(x)-1} \quad \text{(3.10)} \]
Example
For \( x = 10^{25} \):
\[ \pi(10^{25}) = 176846309399143769411680 \]
\[ \frac{10^{25}}{\log(10^{25})-(1+0.019)} \approx \pi(10^{20}) \sim 2.02482 \cdot 10^{18} \]
\[ \frac{10^{25}}{\log(10^{25})-(1+0.019)} > \pi(10^{20}) \]
Robopol Theorem Connections
From equation (3.2) we derive additional relationships that should hold.
Fig.4: Graphical behavior of the Robopol theorem (for very large numbers).
Statement (1.0)
Assume that equation (3.2) holds over its entire domain:
\[ \beta(p_{k+1}) - \beta(p_k) \approx e^{\gamma}\,\log(p_{k+1}) - e^{\gamma}\,\log(p_k) \quad \text{(4.0)} \]
Fig.5: Geometric interpretation of equation (4.0).
Modification of Equation (4.0)
\[ \prod_{p_i \in \text{prime}}^{p_{k+1}} \frac{p_i}{p_i-1} - \prod_{p_i \in \text{prime}}^{p_k} \frac{p_i}{p_i-1} \approx e^{\gamma}\,\log(p_{k+1}) - e^{\gamma}\,\log(p_k) \quad \text{(4.1)} \]
\[ \varepsilon_k = \prod_{p_i \in \text{prime}}^{p_k} \frac{p_i}{p_i-1}, \quad \omega_k = \log(p_k) e^{\gamma} \quad \text{(4.2)} \]
Identifying Initial Points
In the sense of Fig.6, we identify the initial points at \(p_k\):
\[ p_k^* = \varepsilon_k = \omega_k \]
Fig.6: Graphical depiction of the initial points at \(p_k\).
\[ \Delta \varepsilon_k = \varepsilon_k \frac{p_{k+1}}{p_{k+1}-1} - \varepsilon_k = \varepsilon_k \left(\frac{p_{k+1}}{p_{k+1}-1} - 1\right) \quad \text{(4.3)} \]
\[ \Delta \omega_k = e^{\gamma} (\log(p_{k+1}) - \log(p_k)) \quad \text{(4.4)} \]
Thus, from (4.1):
\[ \Delta \omega_k \approx \Delta \varepsilon_k, \quad e^{\gamma} (\ln(p_{k+1}) - \ln(p_k)) \approx \varepsilon_k \frac{p_{k+1}}{p_{k+1}-1} - \varepsilon_k \quad \text{(4.5)} \]
Substituting \(\varepsilon_k = \omega_k\) into (4.5) yields:
\[ e^{\gamma}\,\log(p_{k+1}) - e^{\gamma}\,\log(p_k) \approx \log(p_{k+1}) e^{\gamma} \frac{p_{k+1}}{p_{k+1} - 1} - \log(p_k) e^{\gamma} \quad \text{(4.6)} \]
\[ \log(p_{k+1}) \approx \log(p_k) \frac{p_{k+1}}{p_{k+1}-1} \quad \text{(4.7)} \]
Example
For very large primes:
\[ p_{999,999,999} = 22801763477 \]
\[ p_{1,000,000,000} = 22801763489 \]
\[ \log(22801763489) \approx \log(22801763477) \cdot \frac{22801763489}{22801763489-1} \]
Statement (2.0)
Suppose there exists a smooth, continuous function \( g(x) = f(x) \) approximating \(\pi(x)\) such that in the limit it holds:
\[ \lim_{x \to \infty} g(x) = \lim_{x \to \infty} \pi(x) \quad \text{(4.8)} \]
Tvrdeniu (2.0) vyhovuje v zmysle rovníc (3.8) a (3.9):
\[ g(x) = \frac{x}{\log(x)-1} \quad \text{(4.9)} \]
Statement (3.0)
Nech platí na definičnom obore funkcie \( g(x) \) pre všetky \( x \ge 100 \) v zmysle rovnice (4.7) táto rovnica:
\[ \log(x + \Delta x) \geq \log(x) \frac{x+\Delta x}{x+\Delta x-1} \quad \text{(4.10)} \]
Kde \( \Delta x \) predstavuje horizontálnu vzdialenosť podľa obrázka č.7.
Obrázok č.7: Horizontálna vzdialenosť \( \Delta x \) v zmysle rovnice (4.10).
Zadefinujme najskôr rovnicu v zmysle obrázka č.7:
\[ \frac{x}{\log(x)-1} + 1 = \frac{x+\Delta x}{\log(x+\Delta x)-1} \quad \text{(5.0)} \]
Alebo alternatívne:
\[ \frac{x}{\log(x)-1} = 1 + \frac{x-\Delta x}{\log(x-\Delta x)-1} \quad \text{(5.1)} \]
The Point of Statements (1.0), (2.0) and (3.0)
The point is that if equation (4.10) is satisfied for all \( x > 100 \) – for all \( \Delta x \) computed from (5.1) – then the theorem (i.e. equations (3.0) and (3.2)) necessarily holds. This is our Statement (4.0).
Statement (4.0)
If equation (4.10) holds for all \( x > 100 \), with \( x \) computed from equation (5.1), then the theorem – i.e., equations (3.0) and (3.2) – must necessarily hold.
Explanation
Equation (4.10) is quite simple and intuitive. It is already continuous compared to the original (4.7), which contained only approximate values because the primes in \( x \) exhibit small fluctuations from the average of a smoothed function that would perfectly approximate them (like a smooth curve without serrations).
\[ \log(p_{k+1}) > \log(p_k) \frac{p_{k+1}}{p_{k+1}-1} \quad \text{(5.2)} \]
Teda, aby sme potvrdili platnosť rovníc (3.0) a (3.2), preskúmame (5.2) v ľubovoľnom bode \(p_k\) a jeho suseda \(p_{k+1}\) - a to na vyhladenej funkcii \(g(x)\) pre všetky \(x > 100\). Ak preukážeme platnosť pre všetky \(x\), potom je zrejmé, že aj rovnice (3.0) a (3.2) musia platiť.
Niekto by mohol namietať, že \(g(x)\) neaproximuje \(\pi(x)\) dobre pre malé hodnoty, teda pre malé \(x\) platí, že \(g(x) < \pi(x)\). No zároveň vieme, že limitne \(g(x)\) doháňa \(\pi(x)\) (v zmysle rovnice (4.8)). Teda ak limitne \(g(x)\) doháňa \(\pi(x)\), je zjavné, že smerom k nekonečnu – sklon \(g(x)\) rastie viac než sklon \(\pi(x)\).
Z numerických testov (tabuľka č. 3) je preukázané, že rovnice (3.0) a (3.2) platia pre veľmi veľké čísla - konkrétne pre vysoko zložené čísla s \(p_n\) uvedenými v tabuľke č. 3. Súčasne vidíme, že krivky sa mierne rozchádzajú, pričom má stále viac navrch \( e^{\log(p_n)} \).
V nasledujúcich statiach sa teda pokúsim dokázať, že kľúčové tvrdenie (4.0) naozaj platí.
Upravenie rovnice (4.10)
Upravme rovnicu (4.10) nasledovne:
\[ \log(x + \Delta x) \geq \log(x) \frac{x+\Delta x}{x+\Delta x-1} \]
Použijeme substitúciu: \( x + \Delta x = t \).
\[ \log(t) \geq \log(t - \Delta x) \frac{t}{t-1} \]
\[ \log(t - \Delta x) \geq \frac{\log(t) \cdot (t-1)}{t} \]
Eliminujeme logaritmus:
\[ t - \Delta x \geq t^{(t-1)/t} \quad \text{(5.3)} \]
\[ \Delta x \geq t - t^{(t-1)/t} \]
Definícia \( \Delta x_{min} \)
\[ \Delta x_{min} := t - t^{(t-1)/t} \quad \text{(5.4)} \]
\[ \Delta x \geq \Delta x_{min} \quad \text{(5.5)} \]
Derivácia funkcie
\[ \frac{d}{dt} \Bigl(t - t^{(t-1)/t}\Bigr) = t^{-(t+1)/t} \left( t + \log(t) - 1 \right) \quad \text{(5.6)} \]
Limita funkcie
\[ \lim_{t \to \infty} t^{-(t+1)/t} \left( t + \log(t) - 1 \right) = 1 \quad \text{(5.7)} \]
Príklad
Pre hodnotu \( t = 1000 \):
\[ \Delta x = 1000 - 1000^{(1000-1)/1000} = 6.883951579\ldots \]
Then, compute \( x \):
\[ x = t - \Delta x_{min} = 1000 - 6.883951579 = 993.116048420\ldots \]
Verification of the equation:
\[ \log(x + \Delta x_{min}) = \log(x) \frac{x+\Delta x_{min}}{x+\Delta x_{min}-1} \]
After substitution:
\[ \log(1000) = \log(993.116048420) \cdot \frac{993.116048420+6.883951579}{993.116048420+6.883951579-1} \]
Result:
\[ 6.907755279\ldots = 6.907755279\ldots \]
Analysis of the Approximation \( \pi(x)=\frac{x}{\log(x)} \)
First, let us check the simpler approximation \( \pi(x)=\frac{x}{\log(x)} \).
According to Fig.7 (for \( \pi(x)=\frac{x}{\log(x)} \)) we have:
\[ \frac{x}{\log x} = 1 + \frac{x - \Delta x}{\log (x - \Delta x)} \quad \text{(6.0)} \]
Analytical Solution of the Equation
The exact analytical solution of this equation is:
\[ \Delta x = x - e^{-W_{-1}\left(\frac{\log(x)}{\log(x)-x}\right)}, \quad \Delta x > 0 \quad \text{(6.1)} \]
Alternatively:
\[ \Delta x = x - \frac{\log(x)-x}{\log(x)-1}\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right), \quad \Delta x > 0 \quad \text{(6.2)} \]
where \( W_{-1}(z) \) is the Lambert W function (reference (4)).
Derivácia funkcie
\[ \frac{d}{dx} \!\left(\frac{\log(x)-x}{\log(x)}\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)\right)
= -\frac{(\log(x)-1)\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)^2}{\log^2(x) \Bigl(W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)+1\Bigr)} \quad \text{(6.3)} \]
Limita funkcie
\[ \lim_{x \to \infty} -\frac{(\log(x)-1)\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)^2}{\log^2(x) \Bigl(W_{-1}\!\left(\frac{\log(x)}{\log(x)-1}\right)+1\Bigr)} \]
No simple closed-form result was found in terms of standard functions.
In view of (5.5) for \( x > 100 \) it must hold that:
\[ x - \frac{\log(x)-x}{\log(x)}\,W_{-1}\!\left(\frac{\log(x)}{\log(x)-x}\right) \geq x - x^{(x-1)/x} \quad \text{(6.4)} \]
Example Calculations
Example 1: \( x = 10^6 \)
\[ \Delta x \approx 14.89353360214\ldots \]
\[ \Delta x_{min} \approx 13.815415124237\ldots \]
\[ \Delta x - \Delta x_{min} \approx 1.078118477907\ldots \]
Example 2: \( x = 10^{200} \)
\[ \Delta x \approx 461.519194796772488\ldots \]
\[ \Delta x_{min} \approx 460.517018598809\ldots \]
\[ \Delta x - \Delta x_{min} \approx 1.00217\ldots \]
Thus, numerically we expect that:
\[ \lim_{x \to \infty} \Bigl(\Delta x - \Delta x_{min}\Bigr) = 1 \]
Modification of Equation (6.0)
Rewrite equation (6.0) into the desired form:
\[ \frac{x}{\log x} - 1 - \frac{x-\Delta x}{\log(x-\Delta x)} = 0 \quad \text{(6.5)} \]
Next step:
\[ \frac{x\,\log(x-\Delta x) - \log(x)\,\log(x-\Delta x) - \log(x)(x-\Delta x)}{\log(x)\,\log(x-\Delta x)} = 0 \]
Simplification:
\[ (x-\log(x))\,\log(x-\Delta x) - \log(x)(x-\Delta x) = 0 \quad \text{(6.6)} \]
Taylor Expansion
Expanding in \( x \) we obtain:
\[
\begin{aligned}
& \Bigl(\Delta x + \Delta x\,\log(x) - \log^2(x)\Bigr)
- \frac{\Delta x\Bigl(\Delta x-2\log(x)\Bigr)}{2x} \\
& + \frac{\Delta x^2\Bigl(3\log(x)-2\Delta x\Bigr)}{6x^2} + O\Bigl(\frac{1}{x^3}\Bigr)
\end{aligned}
\quad \text{(6.7)}
\]
For \( x \to \infty \) this reduces to:
\[ \Delta x + \Delta x\,\log(x) - \log^2(x) = 0 \quad \text{(6.8)} \]
Solving for \( \Delta x \)
Solve for \( \Delta x \):
\[ \Delta x = \frac{\log^2(x)}{\log(x)-1} \quad \text{(6.9)} \]
Limit Evaluation
Thus, by (6.4):
\[ \lim_{x \to \infty} \Bigl(\Delta x - \Delta x_{min}\Bigr)
= \lim_{x \to \infty} \left[\frac{\log^2(x)}{\log(x)-1} - \Bigl(x - x^{(x-1)/x}\Bigr)\right] = 1 \quad \text{(6.10)} \]
Conclusion
Equation (6.10) shows that \( \Delta x \geq \Delta x_{min} \) holds up to infinity.
Analysis of the Approximation \( \pi(x)=\frac{x}{\log(x)-1} \)
Now, let us check the approximation \( \pi(x) = \frac{x}{\log(x)-1} \) in the sense of equations (3.9) and (3.10).
Solving equation (5.1) yields:
\[
\Delta x = \frac{(x - \log(x) + 1) \cdot W_{-1}\!\left(-\frac{\sqrt[1+x-\log(x)]{\frac{e^{x+1}}{x} (\log(x)-1)}}{\,x - \log(x) + 1}\right)}{\log(x)-1} + x \quad \text{(7.0)}
\]
According to (5.5), for \( x > 100 \) it must hold that:
\[ \Delta x \geq \Delta x_{min} \]
\[
\frac{(x - \log(x) + 1) \cdot W_{-1}\!\left(-\frac{\sqrt[\,x+1-\log(x)]{\frac{e^{x+1}}{x} - (\log(x)-1)}}{\,x - \log(x) + 1}\right)}{\log(x)-1} + x \geq x - x^{(x-1)/x} \quad \text{(7.1)}
\]
Example Calculation
For \( x = 10^{180} \) we have:
\[
\Delta x = \dots = 414.467741185201\ldots
\]
The minimum value is:
\[
\Delta x_{min} = 10^{180} - \Bigl(10^{180}\Bigr)^{(10^{180}-1)/10^{180}} = 414.465316\ldots
\]
The difference between \( \Delta x \) and \( \Delta x_{min} \):
\[
\Delta x > \Delta x_{min}, \quad \Delta x - \Delta x_{min} \sim 0
\]
Thus, numerically we expect that:
\[
\lim_{x \to \infty} \Bigl(\Delta x - \Delta x_{min}\Bigr) = 0
\]
Now, we reexamine the approximation \( \pi(x) = \frac{x}{\log(x)-1} \) in the sense of (3.9) and (3.10).
Modification of Equation (5.1)
Rewrite equation (5.1) into the desired form:
\[
\frac{x}{\log x-1} - 1 - \frac{x-\Delta x}{\log(x-\Delta x)-1} = 0 \quad \text{(7.2)}
\]
\[
\frac{x}{\log(x)-1} - \frac{x}{\log(x-\Delta x)-1} + \frac{\Delta x}{\log(x-\Delta x)-1} - 1 = 0 \quad \text{(7.3)}
\]
Expansion in \( x \)
Using a Puiseux series expansion:
\[
-\frac{2\Delta x - (\Delta x-2)\log(x) + \log^2(x) + 1}{(\log(x)-1)^2}
+ \frac{\Delta x^2(\log(x)-3)}{2x(\log(x)-1)^3} + O\left(\frac{1}{x^2}\right)
\quad \text{(7.4)}
\]
For \( x \to \infty \) this reduces to:
\[
-\frac{2\Delta x - (\Delta x-2)\log(x) + \log^2(x) + 1}{(\log(x)-1)^2} = 0 \quad \text{(7.5)}
\]
Solving for \( \Delta x \)
\[
\Delta x = \frac{(\log(x)-1)^2}{\log(x)-2} \quad \text{(7.6)}
\]
Limit Evaluation
Thus, we compute the limit:
\[
\lim_{x \to \infty} \Bigl(\Delta x - \Delta x_{min}\Bigr)
= \lim_{x \to \infty} \left[\frac{(\log(x)-1)^2}{\log(x)-2} - \Bigl(x - x^{(x-1)/x}\Bigr)\right] = 0 \quad \text{(7.9)}
\]
Series expansion for \( x \to \infty \):
\[
\frac{1}{\log(x)-2} + \frac{\log^2(x)}{2x} + O\left(\frac{1}{x^2}\right)
\quad \text{(7.10)}
\]
Conclusion
Equation (7.9) shows that \( \Delta x \geq \Delta x_{min} \) holds up to infinity.
Analysis of the Approximation \( \pi(x)=\frac{x}{\log(x)-\epsilon} \)
For the approximation, it holds that:
\[
\pi(x) = \frac{x}{\log(x)-\epsilon}, \quad \epsilon = 1+\varepsilon, \quad \varepsilon > 0 \quad \text{(8.0)}
\]
The solution of equation (8.0) yields:
\[
\Delta x = \frac{(x+\log(x)+\epsilon) \cdot W_{-1}\!\left(\frac{\left(e^{\frac{\epsilon^2}{x+\epsilon}} \cdot x^{-\frac{\epsilon}{x}}\right)^{\frac{x}{x+\epsilon-\log(x)}} \cdot (\epsilon-\log(x))}{x-\log(x)+\epsilon}\right)}{\log(x)-\epsilon} + x \quad \text{(8.1)}
\]
In view of (5.5) for \( x > 100 \) it must hold that:
\[ \Delta x \geq \Delta x_{min} \quad \text{(8.2)} \]
\[
\frac{(x+\log(x)+\epsilon) \cdot W_{-1}\!\left(\frac{\left(e^{\frac{\epsilon^2}{x+\epsilon}} \cdot x^{-\frac{\epsilon}{x}}\right)^{\frac{x}{x+\epsilon-\log(x)}} \cdot (\epsilon-\log(x))}{x-\log(x)+\epsilon}\right)}{\log(x)-\epsilon} + x \geq x - x^{(x-1)/x} \quad \text{(8.2)}
\]
Príklady
Príklad 1:
\[ x = 10^{15}, \quad \varepsilon = 1.0000000000001 \]
\[ \Delta x = \frac{277}{8} = 34.625 \]
\[ \Delta x_{min} = 10^{15} - (10^{15})^{(10^{15}-1)/10^{15}} = 34.538 \]
\( \Delta x > \Delta x_{min} \)
Príklad 2:
\[ x = 10^{16}, \quad \varepsilon = 1.0000000000001 \]
\[ \Delta x = 36 \]
\[ \Delta x_{min} = 10^{16} - (10^{16})^{(10^{16}-1)/10^{16}} = 36.841 \]
\( \Delta x < \Delta x_{min} \) – teda rovnica (8.2) nie je splnená.
Analýza výsledku
Rovnica (8.2) pre aproximáciu (8.0) nebude všeobecne splnená. Aproximácia (8.0) je však v zmysle rovnice (3.8) väčšia ako \( \pi(x) \).
Referencie / References: