On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations

The theorem on a normal limit (n → ∞) distribution of the number of false solutions of a system of nonlinear Boolean equations with independent random coefficients is proved. In particular, we assume that each equation has coefficients that take value 1 with probability that varies in some neighborh...

Full description

Saved in:
Bibliographic Details
Date:2007
Main Authors: Masol, V., Slobodyan, S.
Format: Article
Language:English
Published: Інститут математики НАН України 2007
Online Access:https://nasplib.isofts.kiev.ua/handle/123456789/4485
Tags: Add Tag
No Tags, Be the first to tag this record!
Journal Title:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Cite this:On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations / V. Masol, S. Slobodyan // Theory of Stochastic Processes. — 2007. — Т. 13 (29), № 1-2. — С. 144-151. — Бібліогр.: 5 назв.— англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
_version_ 1859726531468722176
author Masol, V.
Slobodyan, S.
author_facet Masol, V.
Slobodyan, S.
citation_txt On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations / V. Masol, S. Slobodyan // Theory of Stochastic Processes. — 2007. — Т. 13 (29), № 1-2. — С. 144-151. — Бібліогр.: 5 назв.— англ.
collection DSpace DC
description The theorem on a normal limit (n → ∞) distribution of the number of false solutions of a system of nonlinear Boolean equations with independent random coefficients is proved. In particular, we assume that each equation has coefficients that take value 1 with probability that varies in some neighborhood of the point 1/2; the system has a solution with the number of ones equals ρ(n), ρ(n) → ∞ as n → ∞. The proof is constructed on the check of auxiliary statement conditions which in turn generalizes one well-known result.
first_indexed 2025-12-01T11:16:55Z
format Article
fulltext Theory of Stochastic Processes Vol.13 (29), no.1-2, 2007, pp.144-151 VOLODYMYR MASOL AND SVITLANA SLOBODYAN ON THE ASYMPTOTIC NORMALITY OF THE NUMBER OF FALSE SOLUTIONS OF A SYSTEM OF NONLINEAR RANDOM BOOLEAN EQUATIONS The theorem on a normal limit (n → ∞) distribution of the number of false solutions of a system of nonlinear Boolean equations with independent random coefficients is proved. In particular, we assume that each equation has coefficients that take value 1 with probability that varies in some neighborhood of the point 1 2 ; the system has a solution with the number of ones equals ρ(n), ρ(n) → ∞ as n → ∞. The proof is constructed on the check of auxiliary statement conditions which in turn generalizes one well-known result. 1. Introduction Let us consider a system of equations over the field GF(2) consisting of two elements gi(n)∑ k=1 ∑ 1≤j1<...<jk≤n a (i) j1...jk xj1 . . . xjk = bi, i = 1, . . . , N, (1) that satisfies condition (A). Condition (A): 1) Coefficients a (i) j1...jk , 1 ≤ j1 < ... < jk ≤ n, k = 1, ..., gi(n), i = 1, ..., N , are independent random variables that take value 1 with probability P { a (i) j1...jk = 1 } = pik and value 0 with probability P { a (i) j1...jk = 0 } = 1−pik. 2) Elements bi, i = 1, ..., N , are the result of the substitution of a fixed n-dimensional vector x̄0, which has ρ (n) /n − ρ (n)/, components equal to one /zero/ into the left-hand side of the system (1). 3) Function gi(n), i = 1, ..., N, is nonrandom, gi(n) ∈ {2, ..., n} , i = 1, ..., N. 2000 Mathematics Subject Classifications. Primary 60C05, 15A52, 15A03. Key words and phrases. The nonlinear random Boolean equations, normal limit distribution, number of false solutions 144 THE NORMAL LIMIT DISTRIBUTION 145 Denote by νn the number of false solutions of the system (1), i.e. the number of solutions of the system (1) different from the vector x̄0. We are interested in the conditions under which the random variable νn has a normal limit (n → ∞) distribution. 2. Formulation of the theorem Theorem. Let condition (A) hold, and moreover [λ] = 2m, (2) where m = n − N, [·] is the sign of the integral part, λ = 1 2(1 + α + ω) log2 ρ(n) ϕ(n) lnn , ϕ(n) > 0, (3) λ → ∞, (4) λ (α ln α − α − 1) → ∞, (5) ω √ λ → ∞ (6) as n → ∞; let for any arbitrary i, i = 1, ..., N, there exist a nonempty set Ti such that for all sufficiently large values of n Ti ⊆ {2, 3, ..., gi(n)} , Ti �= ∅, δit(n) ≤ pit ≤ 1 − δit(n), t ∈ Ti; (7) lim n→∞ (α + ω)λ B (ρ(n) − 1, 1) < ∞, (8) B (X, Y ) = N∑ i=1 exp { −2 ∑ t∈Ti δit(n)Ct−Y X } ; (2 + (1 + α + ω) ln 2)λ − ln λ 2 + ln B (εϕ(n), 0) → −∞ (n → ∞), (9) where ε = const, 0 < ε < 1; lim n→∞ (− ln N + ln B (εϕ(n), 1)) < 0 (n → ∞). (10) Then distribution function of the random variable νn−λ√ λ tends to the stan- dard normal distribution function . 146 VOLODYMYR MASOL AND SVITLANA SLOBODYAN 3. Auxiliary statements Lemma 1. Let ξ and η be random variables that take non-negative integer values. If max 1≤r≤T ∣∣∣M(ξ)r(M(η)r) −1 − 1 ∣∣∣ = εT < 1; (11) M(η)r ≤ Cλ r 1 , 1 ≤ r ≤ T, (12) where M(ζ)r denotes r–factorial moment of a random variable ζ, r ≥ 1, then for an arbitrary t, 0 ≤ t ≤ T − αλ1 and α > 1, |P {ξ ≥ t} − P {η ≥ t}| ≤ ≤ C√ 2π max(1,λ1−1) ( εT e2λ1 + 1+εT√ 2π max(1,αλ1) exp {(t + λ1 − T )u(α)} ) , (13) where u(α) = (α−1)−1(α ln α−α−1) for 2 > ln α > 0, and u(α) = ln α−1 for ln α ≥ 2. Proof. By virtue of Bonferrony inequalities, for any arbitrary random vari- able ζ that takes non-negative integer values, and for any arbitrary integers t > 0 and d ≥ 0 t+2d+β∑ r=t (−1)r+tCt−1 r−1 1 r! M(ζ)r ≤ P{ζ ≥ t} ≤ t+2d∑ r=t (−1)r+tCt−1 r−1 1 r! M(ζ)r, (14) where min (M(ζ)t+2d+β , M(ζ)t+2d) < ∞, β ∈ {−1, 1}. (Proof of relation (14) for β = −1, M(ζ)t+2d−1 < ∞ look, for example, in ([1], p.136, 223)) Let numbers t and T have identical parity. Then, using (14) for β = −1, we receive P{ξ ≥ t} − P{η ≥ t} ≤ Γ(T ) + M(ξ)T Ct−1 T−1 1 T ! , (15) P{ξ ≥ t} − P{η ≥ t} ≥ Γ(T ) − M(η)T Ct−1 T−1 1 T ! , (16) where Γ(T ) = T−1∑ r=t (−1)r+tCt−1 r−1 1 r! M(η)r ( M(ξ)r M(η)r − 1 ) . If the difference P{ξ ≥ t}−P{η ≥ t} is non-negative /non-positive/that we obtain |P{ξ ≥ t} − P{η ≥ t}| ≤ |Γ(T )| + 1 T ! Ct−1 T−1 max (M(ξ)T , M(η)T ) (17) by virtue of conditions (15), (16). It is easy to check up that max (M(ξ)T , M(η)T ) ≤ M(η)T ( 1 + ∣∣∣∣∣M(ξ)r M(η)r − 1 ∣∣∣∣∣ ) . (18) THE NORMAL LIMIT DISTRIBUTION 147 Using (17), (18) and conditions (11), (12), we obtain |P{ξ ≥ t} − P{η ≥ t}| ≤ ≤ C ( εT T−1∑ r=t Ct−1 r−1 1 r! λr 1 + (1 + εT )λT 1 1 T ! Ct−1 T−1 ) . (19) Hence |P{ξ ≥ t} − P{η ≥ t}| ≤ C ( λt 1 t! eλ1εT + (1 + εT ) λT−t 1 (T − t)! λt 1 t! ) . (20) Below the following relations will be established for the integer u, u ≥ 1, λu 1 u! ≤ (2π max(1, λ1 − 1))−1/2 eλ1 ; (21) for the integer N, N ≥ max(1, αλ1), λN 1 N ! ≤ (2π max(1, αλ1)) −1/2 exp{(λ1 − N)u(α) − λ1}, (22) where u(α) = (α − 1)−1(α ln α − α − 1) for 2 > lnα > 0, u(α) = ln α − 1 for ln α ≥ 2. Relations (20)–(22) prove (13), when t and T have identical parity . Let now parameters t and T have different parity. Let us show, that we can obtain (17) in this case. Using (14) for some d ≥ 0, β = 1 and t + 2d = T − 1, we receive P{ξ ≥ t} − P{η ≥ t} ≤ Γ(T ) + M(η)T 1 T ! Ct−1 T−1, (23) P{ξ ≥ t} − P{η ≥ t} ≥ Γ(T ) − M(ξ)T 1 T ! Ct−1 T−1. (24) By virtue of (23), (24), we obtain (17) similarly when we used inequalities (15) and (16). To complete the proof of Lemma 1 it is, therefore, enough to establish (21) and (22). Let us check (21). Indeed, it follows from Stirling formula that u! ≥ (u/e)u √ 2πu. Hence λu 1 u! ≤ ( λ1e u )u 1√ 2πu . (25) Let ϕ(u) = ( λ1e u )u and let us show that max u≥λ1−1 ϕ(u) = ϕ(λ1). (26) Indeed, the first derivative ϕ′(u) = ϕ(u)(lnλ1 − ln u) and ϕ′(u) = 0 at u = λ1. Since the second derivative ϕ′′(u) = ϕ(u)((lnλ1 − ln u)2 − u−1) 148 VOLODYMYR MASOL AND SVITLANA SLOBODYAN is negative at u = λ1, ϕ′′(λ1) < 0, relation (26) holds. Using (26), we establish (21) for u ≥ max(1, λ1 − 1). Let further 1 ≤ u ≤ λ1 − 1; and let ψ(u) = ( λ1e u )u 1√ u , then max 1≤u≤λ1−1 ψ(u) = ψ(λ1 − 1). (27) Indeed, ψ′(u) = ψ(u)(lnλ1 − f(u)), (28) where f(u) = lnu + (2u)−1. Let us show that function f(u) takes its maximal value on an interval 1 ≤ u ≤ λ1 − 1 at u = λ1 − 1, max 1≤u≤λ1−1 f(u) = f(λ1 − 1). (29) Indeed, f ′(u) = u−1 − 1 2 u−2, and f ′(u) = 0 at u = 1 2 . At the same time f ′′(u)|u= 1 2 = (−u−2 + u−3)|u= 1 2 = 4. Therefore, function f(u) increases for u > 1 2 and (29) holds on the interval 1 ≤ u ≤ λ1 − 1. As a result we get ψ′(u) > 0 for 1 ≤ u ≤ λ1 − 1. (30) Indeed, taking into account (29), ln λ1 − f(u) ≥ ≥ ln ( 1 + 1 λ1−1 ) − 1 2(λ1−1) ≥ 1 2(λ1−1) ( 1 − 1 λ1−1 ) > 0 (31) for λ1 > 2. (Here the inequality ln(1 + x) > x − 1 2 x2 has been used for x > 0.) Relations (28) and (31) prove (30). Estimate (30) allows, apparently, to conclude that equality (27) holds. With the help of (25) and (27) we find λu 1 u! ≤ ( λ1e λ1 − 1 )λ1−1 1√ 2π(λ1 − 1) ≤ eλ1√ 2π(λ1 − 1) (32) for 1 ≤ u ≤ λ1 − 1. Estimate (32) proves (21) for 1 ≤ u ≤ λ1 − 1. Relation (21) is proved. Let us check (22). With the help of Stirling formula and inequality λ1 N ≤ 1 α , we can obtain λN 1 N ! ≤ 1√ 2πN eλ1(1−ln α)× × exp {( 1 − ln α + 2−ln α α−1 ) (N − λ1) } exp {−λ1(2 − ln α)} . (33) By virtue of conditions N ≥ αλ1, 2 − ln α > 0, and α > 1, the right-hand side of the inequality (33) can be estimate as follows 1√ 2πN eλ1(1−ln α) exp {( 1 − ln α + 2−ln α α−1 ) (N − λ1) } × × exp {−λ1(2 − ln α)} ≤ 1√ 2πN e−λ1 exp { (λ1 − N)α lnα−α−1 α−1 } . (34) THE NORMAL LIMIT DISTRIBUTION 149 From relations (33) and (34) the estimate (22) follows for α < e2. Let now α ≥ e2. Then λN 1 N ! ≤ 1√ 2πN e−λ1 exp {−(N − λ1)(lnα − 1)} . Relation (22) is proved for all α > 1. Lemma 1 is proved. Lemma 2. Let X and Y be random variables that take non-negative integer values, and MX = λ∗. If for all r ≤ (α + γ)λ∗ M(Y )r ≤ C(λ∗)r (35) with some constant C, and λ∗ (α ln α − α − 1) → ∞, (36) γ ≥ 0, (37) max 1≤r≤(α+γ)λ ∗ ∣∣∣M(X)r(M(Y )r) −1 − 1 ∣∣∣ e2λ ∗ √ λ∗ → 0 (38) as λ∗ → ∞, then max 0≤t≤γλ ∗ |P {X ≥ t} − P {Y ≥ t}| → 0 (λ∗ → ∞). (39) Proof. Assumptions (35) and (38) imply the conditions of Lemma 1, by virtue of which (13) holds for 0 ≤ t ≤ γλ ∗, α > 1. Using (36) and (37) it is easy to show that exp{(t + λ ∗ − (α + γ)λ ∗) u(α)} → 0 as λ∗ → ∞ uniformly for 0 ≤ t ≤ γλ ∗. Taking into account (38), it follows from the last statement that the right-hand side of the inequality (13) tends to zero as λ∗ → ∞ uniformly, for 0 ≤ t ≤ γλ ∗. The left-hand side of the inequality (13) tends, therefore, to zero too for λ ∗ and t mentioned above, which proves, obviously, (39). Lemma 2 is proved. Remark. The lemma 2 (for α = 5 and γ = 2) follows from the lemma 3 in [2]. 4. Proof of the theorem Let us show that under the conditions of the theorem we can use Lemma 2. Let the random variable Y in the mentioned lemma have a Poisson dis- tribution with parameter 2m, while the distribution of the random vari- able X coincides with the distribution of the random variable νn. Then 150 VOLODYMYR MASOL AND SVITLANA SLOBODYAN M(Y )r = 2mr, r ≥ 1, while expectation Mνn can, by virtue of its explicit form obtained in [3], be presented in the following way Mνn = 2m ( 1 − 1 2n ) M̃, (40) where exp {−B(ρ(n) − 1, 1)} ≤ M̃ ≤ exp {B(ρ(n) − 1, 1)}. Now condition (35) becomes 2mr ≤ C(Mνn)r. (41) It follows from (40) that inequality (41) holds true for r ≤ (1 + α + ω)Mνn and C ≥ ( 1 − 1 2n )−(1+α+ω)Mνn exp{(1 + α + ω)B(ρ(n) − 1, 1)Mνn}. (42) By virtue of conditions (3), (8), and equality (40), the right-hand side of (42) is limited as n → ∞. Therefore, it is possible to choose a limited constant C < ∞ such that condition (35) holds true. Further we note that under conditions (3)–(10) relation (38) is estab- lished in [4] for α = 5 and ω = 1. For arbitrary α and ω satisfying conditions of the theorem, verification of the relation (38) can be executed similarly. By virtue of Lemma 2, we obtain max 0≤t≤(1+ω)λ ∗ |P {νn ≥ t} − P {Y ≥ t}| → 0 as n → ∞, (43) where λ∗ = Mνn according to the notations introduced above. By virtue of (40) and conditions (3) and (8), the last equality allows to present λ∗ as λ∗ = [λ] (1 + r(n)) , (44) where r(n) = O (B(ρ(n) − 1, 1)) + O ( 1 2n ) (45) and r(n) → 0 as n → ∞. We can write relation (43) in the following way max −√ λ∗≤l≤ω √ λ ∗ ∣∣∣∣∣P { νn − λ∗ √ λ∗ ≥ l } − P { Y − λ∗ √ λ∗ ≥ l }∣∣∣∣∣ → 0, n → ∞, (46) where l = t−λ∗√ λ∗ . Let us show that distributions of the random variables νn−λ∗√ λ∗ and νn−λ√ λ coincide as n → ∞. Indeed, νn − λ∗ √ λ∗ = νn − λ√ λ + ηn, (47) THE NORMAL LIMIT DISTRIBUTION 151 where ηn = νn−λ√ λ ( O ( ε(n) λ ) + O(r(n)) ) − λr(n)−ε(n)(1+r(n))√ λ∗ , [λ] = λ − ε(n), 0 ≤ ε(n) < 1. The random variable ηn tends in probability to zero as n → ∞. Indeed, for an arbitrary ε > 0 P {|ηn| > ε} ≤ 1 ε M |ηn| ≤ 1 ε (∣∣∣∣∣O ( 1√ λ )∣∣∣∣∣ + ∣∣∣O (√ λr(n) )∣∣∣ ) (48) and, by virtue of (3), (8), and (45), the right-hand side of (48) tends to zero as n increases, i.e. P {|ηn| > ε} → 0, n → ∞. (49) Relations (47), (49), and theorem ([5], p.157) prove that distributions of the random variables νn−λ∗√ λ∗ and νn−λ√ λ coincide as n → ∞. Similarly we can verify that distributions of Y −λ∗√ λ∗ and Y −[λ]√ [λ] are the same as n → ∞. Thus, relation (46) can be writen as max −√ λ∗≤l≤ω √ λ ∗ ∣∣∣∣∣∣P { νn − λ√ λ ≥ l } − P ⎧⎨ ⎩Y − [λ]√ [λ] ≥ l ⎫⎬ ⎭ ∣∣∣∣∣∣ → 0, n → ∞. (50) Finally we notice that the random variable Y −[λ]√ [λ] has the standard normal distribution as λ → ∞. The theorem is proved. Bibliography 1. Zubkov A.M., Sevastianov B.A., Chistyakov V.P., The collection of prob- lems in probability theory, Moscow, Nauka,(1989). (in Russian) 2. Zubkov A.M., Inequalities for transitions with prohibitions , The mathe- matical collection, (1979), v.109 (151), no.4 (11), 491–532. (in Russian) 3. Masol V.I., Moments of the number of solutions of system of random Boolean equations, Random Oper. and Stoch. Equations, (1993), vol. 1, no. 2, 171–179. 4. Masol V.I., Slobodyan S.Y., On the convergence to the normal limit dis- tribution of the number of false solutions of a system of nonlinear random Boolean equations, PT&MS, (2007), (Submitted). (in Ukrainian) 5. Chistyakov V.P., Course of probability theory, Moscow, Nauka, (2003). (in Russian) Department of Probability Theory and Mathematical Statistics, Kyiv National Taras Shevchenko University, Kyiv, Ukraine. E-mail address: vimasol@ukr.net Department of Probability Theory and Mathematical Statistics, Kyiv National Taras Shevchenko University, Kyiv, Ukraine. E-mail address: sv yaros@rambler.ru
id nasplib_isofts_kiev_ua-123456789-4485
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
issn 0321-3900
language English
last_indexed 2025-12-01T11:16:55Z
publishDate 2007
publisher Інститут математики НАН України
record_format dspace
spelling Masol, V.
Slobodyan, S.
2009-11-19T10:18:19Z
2009-11-19T10:18:19Z
2007
On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations / V. Masol, S. Slobodyan // Theory of Stochastic Processes. — 2007. — Т. 13 (29), № 1-2. — С. 144-151. — Бібліогр.: 5 назв.— англ.
0321-3900
https://nasplib.isofts.kiev.ua/handle/123456789/4485
The theorem on a normal limit (n → ∞) distribution of the number of false solutions of a system of nonlinear Boolean equations with independent random coefficients is proved. In particular, we assume that each equation has coefficients that take value 1 with probability that varies in some neighborhood of the point 1/2; the system has a solution with the number of ones equals ρ(n), ρ(n) → ∞ as n → ∞. The proof is constructed on the check of auxiliary statement conditions which in turn generalizes one well-known result.
en
Інститут математики НАН України
On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations
Article
published earlier
spellingShingle On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations
Masol, V.
Slobodyan, S.
title On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations
title_full On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations
title_fullStr On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations
title_full_unstemmed On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations
title_short On the asymptotic normality of the number of false solutions of a system of nonlinear random Boolean equations
title_sort on the asymptotic normality of the number of false solutions of a system of nonlinear random boolean equations
url https://nasplib.isofts.kiev.ua/handle/123456789/4485
work_keys_str_mv AT masolv ontheasymptoticnormalityofthenumberoffalsesolutionsofasystemofnonlinearrandombooleanequations
AT slobodyans ontheasymptoticnormalityofthenumberoffalsesolutionsofasystemofnonlinearrandombooleanequations