Research article Special Issues

Regional forecasting of COVID-19 caseload by non-parametric regression: a VAR epidemiological model

  • Received: 14 November 2020 Accepted: 28 January 2021 Published: 01 February 2021
  • Objectives 

    The COVID-19 pandemic (caused by SARS-CoV-2) has introduced significant challenges for accurate prediction of population morbidity and mortality by traditional variable-based methods of estimation. Challenges to modelling include inadequate viral physiology comprehension and fluctuating definitions of positivity between national-to-international data. This paper proposes that accurate forecasting of COVID-19 caseload may be best preformed non-parametrically, by vector autoregression (VAR) of verifiable data regionally.

    Methods 

    A non-linear VAR model across 7 major demographically representative New York City (NYC) metropolitan region counties was constructed using verifiable daily COVID-19 caseload data March 12–July 23, 2020. Through association of observed case trends with a series of (county-specific) data-driven dynamic interdependencies (lagged values), a systematically non-assumptive approximation of VAR representation for COVID-19 patterns to-date and prospective upcoming trends was produced.

    Results 

    Modified VAR regression of NYC area COVID-19 caseload trends proves highly significant modelling capacity of observed patterns in longitudinal disease incidence (county R2 range: 0.9221–0.9751, all p < 0.001). Predictively, VAR regression of daily caseload results at a county-wide level demonstrates considerable short-term forecasting fidelity (p < 0.001 at one-step ahead) with concurrent capacity for longer-term (tested 11-week period) inferences of consistent, reasonable upcoming patterns from latest (model data update) disease epidemiology.

    Conclusions 

    In contrast to macroscopic variable-assumption projections, regionally-founded VAR modelling may substantially improve projection of short-term community disease burden, reduce potential for biostatistical error, as well as better model epidemiological effects resultant from intervention. Predictive VAR extrapolation of existing public health data at an interdependent regional scale may improve accuracy of current pandemic burden prognoses.

    Citation: Aaron C Shang, Kristen E Galow, Gary G Galow. Regional forecasting of COVID-19 caseload by non-parametric regression: a VAR epidemiological model[J]. AIMS Public Health, 2021, 8(1): 124-136. doi: 10.3934/publichealth.2021010

    Related Papers:

    [1] Yulu Feng, Min Qiu . Some results on p-adic valuations of Stirling numbers of the second kind. AIMS Mathematics, 2020, 5(5): 4168-4196. doi: 10.3934/math.2020267
    [2] Phakhinkon Napp Phunphayap, Prapanpong Pongsriiam . Divisibility of Fibonomial coefficients in terms of their digital representations and applications. AIMS Mathematics, 2022, 7(4): 5314-5327. doi: 10.3934/math.2022296
    [3] Babar Sultan, Mehvish Sultan, Aziz Khan, Thabet Abdeljawad . Boundedness of an intrinsic square function on grand p-adic Herz-Morrey spaces. AIMS Mathematics, 2023, 8(11): 26484-26497. doi: 10.3934/math.20231352
    [4] Tingting Xu, Zaiyong Feng, Tianyang He, Xiaona Fan . Sharp estimates for the p-adic m-linear n-dimensional Hardy and Hilbert operators on p-adic weighted Morrey space. AIMS Mathematics, 2025, 10(6): 14012-14031. doi: 10.3934/math.2025630
    [5] Naqash Sarfraz, Muhammad Bilal Riaz, Qasim Ali Malik . Some new characterizations of boundedness of commutators of p-adic maximal-type functions on p-adic Morrey spaces in terms of Lipschitz spaces. AIMS Mathematics, 2024, 9(7): 19756-19770. doi: 10.3934/math.2024964
    [6] Naqash Sarfraz, Muhammad Aslam . Some weighted estimates for the commutators of p-adic Hardy operator on two weighted p-adic Herz-type spaces. AIMS Mathematics, 2021, 6(9): 9633-9646. doi: 10.3934/math.2021561
    [7] Pham Thi Kim Thuy, Kieu Huu Dung . Hardy–Littlewood maximal operators and Hausdorff operators on p-adic block spaces with variable exponents. AIMS Mathematics, 2024, 9(8): 23060-23087. doi: 10.3934/math.20241121
    [8] Yanlong Shi, Xiangxing Tao . Rough fractional integral and its multilinear commutators on p-adic generalized Morrey spaces. AIMS Mathematics, 2023, 8(7): 17012-17026. doi: 10.3934/math.2023868
    [9] Qianjun He, Xiang Li . Necessary and sufficient conditions for boundedness of commutators of maximal function on the p-adic vector spaces. AIMS Mathematics, 2023, 8(6): 14064-14085. doi: 10.3934/math.2023719
    [10] Phakhinkon Phunphayap, Prapanpong Pongsriiam . Extremal orders and races between palindromes in different bases. AIMS Mathematics, 2022, 7(2): 2237-2254. doi: 10.3934/math.2022127
  • Objectives 

    The COVID-19 pandemic (caused by SARS-CoV-2) has introduced significant challenges for accurate prediction of population morbidity and mortality by traditional variable-based methods of estimation. Challenges to modelling include inadequate viral physiology comprehension and fluctuating definitions of positivity between national-to-international data. This paper proposes that accurate forecasting of COVID-19 caseload may be best preformed non-parametrically, by vector autoregression (VAR) of verifiable data regionally.

    Methods 

    A non-linear VAR model across 7 major demographically representative New York City (NYC) metropolitan region counties was constructed using verifiable daily COVID-19 caseload data March 12–July 23, 2020. Through association of observed case trends with a series of (county-specific) data-driven dynamic interdependencies (lagged values), a systematically non-assumptive approximation of VAR representation for COVID-19 patterns to-date and prospective upcoming trends was produced.

    Results 

    Modified VAR regression of NYC area COVID-19 caseload trends proves highly significant modelling capacity of observed patterns in longitudinal disease incidence (county R2 range: 0.9221–0.9751, all p < 0.001). Predictively, VAR regression of daily caseload results at a county-wide level demonstrates considerable short-term forecasting fidelity (p < 0.001 at one-step ahead) with concurrent capacity for longer-term (tested 11-week period) inferences of consistent, reasonable upcoming patterns from latest (model data update) disease epidemiology.

    Conclusions 

    In contrast to macroscopic variable-assumption projections, regionally-founded VAR modelling may substantially improve projection of short-term community disease burden, reduce potential for biostatistical error, as well as better model epidemiological effects resultant from intervention. Predictive VAR extrapolation of existing public health data at an interdependent regional scale may improve accuracy of current pandemic burden prognoses.


    Abbreviations

    COVID-19/COVID:

    coronavirus disease 2019; 

    SARS-CoV-2:

    Severe acute respiratory syndrome coronavirus 2; 

    VAR:

    vector autoregression; 

    NYC:

    New York City; 

    S-I-R/SIR:

    susceptible, infectious, removed (immune) framework of compartmental disease modelling; 

    SUQC:

    susceptible, unquarantined infected, quarantined infected, confirmed infected framework of compartmental disease modelling; 

    AIC:

    Akaike Information Criteria; 

    COVID+:

    confirmed positive COVID-19 case; 

    MAE:

    mean absolute error; 

    CDC:

    Centers for Disease Control and Prevention

    The Fibonacci sequence (Fn)n1 is given by the recurrence relation Fn=Fn1+Fn2 for n3 with the initial values F1=F2=1. For each m1 and 1km, the Fibonomial coefficients (mk)F is defined by

    (mk)F=F1F2F3Fm(F1F2F3Fk)(F1F2F3Fmk)=Fmk+1Fmk+2FmF1F2F3Fk.

    Similar to the binomial coefficients, we define (mk)F=1 if k=0 and (mk)F=0 if k>m, and it is well-known that (mk)F is always an integer for every m1 and k0.

    Recently, there has been an increasing interest in the study of Fibonomial coefficients. Marques and Trojovský [25,26] start the investigation on the divisibility of Fibonomial coefficients by determining the integers n1 such that (pnn)F is divisible by p for p=2,3. Marques, Sellers, and Trojovský [24] show that p divides (pa+1pa)F for p±2(mod5) and a1. Marques and Trojovsk' [27] and Trojovský [42] extend their results further and obtained the p-adic valuation of (pa+1pa)F in [42]. Then Ballot [2,Theorem 2] generalizes the Kummer-like theorem of Knuth and Wilf [22] and uses it to give a generalization of Marques and Trojovský's results. In particular, Ballot [2,Theorems 3.6,5.2,and 5.3] finds all integers n such that p(pnn)U for any nondegenerate fundamental Lucas sequence U and p=2,3 and for p=5,7 in the case U=F. Phunphayap and Pongsriiam [31] provide the most general formula for the p-adic valuation of Fibonomial coefficients in the most general form (mn)F. For other recent results on the divisibility properties of the Fibonacci numbers, the Fibonomial coefficients, and other combinatorial numbers, see for example [3,4,5,11,12,13,16,17,28,30,32,33,34,37,38,41,43]. For some identities involving Fibonomial coefficients and generalizations, we refer the reader to the work of Kilic and his coauthors [7,8,18,19,20,21]. For the p-adic valuations of Eulerian, Bernoulli, and Stirling numbers, see [6,9,14,23,40]. Hence the relation p(pann)F has been studied only in the case p=2,3,5,7 and a=1.

    In this article, we extend the investigation on (pann)F to the case of any prime p and any positive integer a. Replacing n by pa and pa by p, this becomes Marques and Trojovský's results [27,42]. Substituting a=1, p{2,3,5,7}, and letting n be arbitrary, this reduces to Ballot's theorems [2]. So our results are indeed an extension of those previously mentioned. To obtain such the general result for all p and a, the calculation is inevitably long but we try to make it as simple as possible. As a reward, we can easily show in Corollaries 3.3 and 3.4 that (4nn)F is odd if and only if n is a nonnegative power of 2, and (8nn)F is odd if and only if n=(1+32k)/7 for some k1(mod3).

    We organize this article as follows. In Section 2, we give some preliminaries and results which are needed in the proof of the main theorems. In Section 3, we calculate the p-adic valuation of (pann)F for all a, p, and n, and use it to give a characterization of the positive integers n such that (pann)F is divisible by p where p is any prime which is congruent to ±2(mod5). Remark that there also is an interesting pattern in the p-adic representation of the integers n such that (pnn)F is divisible by p. The proof is being prepared but it is a bit too long to include in this paper. We are trying to make it simpler and shorter and will publish it in the future. For more information and some recent articles related to the Fibonacci numbers, we refer the readers to [15,35,36,39] and references therein.

    Throughout this article, unless stated otherwise, x is a real number, p is a prime, a,b,k,m,n,q are integers, m,n1, and q2. The p-adic valuation (or p-adic order) of n, denoted by νp(n), is the exponent of p in the prime factorization of n. In addition, the order (or the rank) of appearance of n in the Fibonacci sequence, denoted by z(n), is the smallest positive integer m such that nFm, x is the largest integer less than or equal to x, {x} is the fractional part of x given by {x}=xx, x is the smallest integer larger than or equal to x, and amodm is the least nonnegative residue of a modulo m. Furthermore, for a mathematical statement P, the Iverson notation [P] is defined by

    [P]={1,if P holds;0,otherwise.

    We define sq(n) to be the sum of digits of n when n is written in base q, that is, if n=(akak1a0)q=akqk+ak1qk1++a0 where 0ai<q for every i, then sq(n)=ak+ak1++a0. Next, we recall some well-known and useful results for the reader's convenience.

    Lemma 1. Let p5 be a prime. Then the following statements hold.

    (i) nFm if and only if z(n)m

    (ii) z(p)p+1 if and only if p±2(mod5) and z(p)p1, otherwise.

    (iii) gcd(z(p),p)=1.

    Proof. These are well-known. See, for example, in [31,Lemma 1] for more details.

    Lemma 2. (Legendre's formula) Let n be a positive integer and let p be a prime. Then

    νp(n!)=k=1npk=nsp(n)p1.

    We will deal with a lot of calculations involving the floor function. So we recall the following results, which will be used throughout this article, sometimes without reference.

    Lemma 3. For kZ and xR, the following holds

    (i) k+x=k+x,

    (ii) {k+x}={x},

    (iii) x+x={1,if xZ0,if xZ

    (iv) 0{x}<1 and {x}=0 if and only if xZ.

    (v) x+y={x+y,if {x}+{y}<1x+y+1,if {x}+{y}1

    (vi) xk=xk for k1.

    Proof. These are well-known and can be proved easily. For more details, see in [10,Chapter 3]. We also refer the reader to [1,29] for a nice application of these properties.

    The next three theorems given by Phunphayap and Pongsriiam [31] are important tools for obtaining the main results of this article.

    Theorem 4. [31,Theorem 7] Let p be a prime, a0, 0, and m1. Assume that p±1(modm) and δ=[0(modm)] is the Iverson notation. Then

    νp(pam!)={(pa1)m(p1)a{m}+νp(m!),if p1(modm);(pa1)m(p1)a2δ+νp(m!),if p1(modm) and a is even;(pa1)m(p1)a12δ{m}+νp(m!),if p1(modm) and a is odd.

    Theorem 5. [31,Theorem 11 and Corollary 12] Let 0km be integers. Then the following statements hold.

    (i) Let A2=ν2(m6!)ν2(k6!)ν2(mk6!). If r=mmod6 and s=kmod6, then

    ν2((mk)F)={A2,if rs and (r,s)(3,1),(3,2),(4,2);A2+1,if (r,s)=(3,1),(3,2),(4,2);A2+3,if r<s and (r,s)(0,3),(1,3),(2,3)(1,4),(2,4),(2,5);A2+2,if (r,s)=(0,3),(1,3),(2,3),(1,4),(2,4)(2,5).

    (ii) ν5((mk)F)=ν5((mk)).

    (iii) Suppose that p is a prime, p2, and p5. If m=mz(p), k=kz(p), r=mmodz(p), and s=kmodz(p), then

    νp((mk)F)=νp((mk))+[r<s](νp(mk+z(p)z(p))+νp(Fz(p))).

    Theorem 6. [31,Theorem 13] Let a, b, 1, and 2 be positive integers and ba. For each p5, assume that 1pb>2pa and let mp=1pbaz(p) and kp=2z(p). Then the following statements hold.

    (i) If ab(mod2), then ν2((12b22a)F) is equal to

    {ν2((m2k2)),if 12(mod3) or 20(mod3);a+2+ν2(m2k2)+ν2((m2k2)),if 10(mod3) and 20(mod3);a2+1+ν2(m2k2)+ν2((m2k2)),if 11(mod3) and 22(mod3);a+12+ν2((m2k2)),if 12(mod3) and 21(mod3)

    and if ab(mod2), then ν2((12b22a)F) is equal to

    {ν2((m2k2)),if 12(mod3) or 20(mod3);a+2+ν2(m2k2)+ν2((m2k2)),if 10(mod3) and 20(mod3);a+12+ν2((m2k2)),if 11(mod3) and 21(mod3);a2+1+ν2(m2k2)+ν2((m2k2)),if 12(mod3) and 22(mod3).

    (ii) Let p5 be an odd prime and let r=1pbmodz(p) and s=2pamodz(p). If p±1(mod5), then

    νp((1pb2pa)F)=[r<s](a+νp(mpkp)+νp(Fz(p)))+νp((mpkp)),

    and if p±2(mod5), then νp((1pb2pa)F) is equal to

    {νp((mpkp)),if r=s or 20(modz(p));a+νp(Fz(p))+νp(mpkp)+νp((mpkp)),if 10(modz(p)) and20(modz(p));a2+νp((mpkp)),if r>s1,20(modz(p))and a is even;a2+νp(Fz(p))+νp(mpkp)+νp((mpkp)),if r<s1,20(modz(p))and a is even;a+12+νp(mpkp)+νp((mpkp)),if r>s1,20(modz(p))and a is odd;a12+νp(Fz(p))+νp((mpkp)),if r<s1,20(modz(p))and a is odd.

    In fact, Phunphayap and Pongsriiam [31] obtain other results analogous to Theorems 2.5 and 2.6 too but we do not need them in this article.

    We begin with the calculation of the 2-adic valuation of (2ann)F and then use it to determine the integers n such that (2nn)F,(4nn)F,(8nn)F are even. Then we calculate the p-adic valuation of (pann)F for all odd primes p. For binomial coefficients, we know that ν2((2nn))=s2(n). For Fibonomial coefficients, we have the following result.

    Theorem 7. Let a and n be positive integers, ε=[n0(mod3)], and A=(2a1)n32ν2(n). Then the following statements hold.

    (i) If a is even, then

    ν2((2ann)F)=δ+Aa2εν2(A!)=δ+s2(A)a2ε, (3.1)

    where δ=[nmod6=3,5]. In other words, δ=1 if n3,5(mod6) and δ=0 otherwise.

    (ii) If a is odd, then

    ν2((2ann)F)=δ+Aa12εν2(A!)=δ+s2(A)a12ε, (3.2)

    where δ=(nmod6)12[2n]+ν2(n)+3nmod32[nmod6=2,4]. In other words, δ=(nmod6)12 if n is odd, δ=0 if n0(mod6), δ=ν2(n)2+1 if n4(mod6), and δ=ν2(n)+12 if n2(mod6).

    Proof. The second equalities in (3.1) and (3.2) follow from Legendre's formula. So it remains to prove the first equalities in (3.1) and (3.2). To prove (ⅰ), we suppose that a is even and divide the consideration into two cases.

    Case 1. 2n. Let r=2anmod6 and s=nmod6. Then s{1,3,5}, r2an4n4s(mod6), and therefore (r,s)=(4,1),(0,3),(2,5). In addition, A=(2a1)n3=(2a1)n3 and δ=[s=3,5]. By Theorem 5(ⅰ), the left–hand side of (3.1) is A2 if s=1 and A2+2 if s=3,5, where A2=ν2(2an6!)ν2(n6!)ν2((2a1)n6!). We obtain by Theorem 4 that

    ν2(2an6!)=ν2(2a1n3!)=(2a11)n3a22ε{n3}+ν2(n3!).

    By Legendre's formula and Lemma 3, we have

    ν2(n6!)=ν2(n3!)n6,
    ν2((2a1)n6!)=ν2((2a1)n3!)(2a1)n6=ν2(A!)(2a1)n6,
    n6+(2a1)n6=ns6+2anr6ns6+rs6=2anr6[s{3,5}].

    From the above observation, we obtain

    A2=(2a11)n3a22ε{n3}+2anr6[s{3,5}]ν2(A!)=Aa22ε{n3}r6[s{3,5}]ν2(A!)={Aa2ν2(A!),if s=1;Aν2(A!)1,if s=3;Aa2ν2(A!)1,if s=5.

    It is now easy to check that A2 (if s=1), A2+2 (if s=3,5) are the same as δ+Aa2εν2(A!) in (3.1). So (3.1) is verified.

    Case 2. 2n. We write n=2b where 2 and let m=2a3, k=3, r=2amod3, and s=mod3. Since a is even, r=s. Then we apply Theorem 6(ⅰ) to obtain

    ν2((2ann)F)=ν2((2a+b2b)F)=ν2((mk))=ν2(m!)ν2(k!)ν2((mk)!). (3.3)

    We see that 0(mod3) if and only if n0(mod3). In addition, A=(2a1)3 and δ=0. By Theorem 4, we have

    ν2(m!)=Aa2ε+ν2(k!).

    In addition,

    mk=2a33=2ar3s3=2a3=A.

    So ν2((mk)!)=ν2(A!). Substituting these in (3.3), we obtain (3.1). This completes the proof of (ⅰ).

    To prove (ⅱ), we suppose that a is odd and divide the proof into two cases.

    Case 1. 2n. This case is similar to Case 1 of the previous part. So we let r=2anmod6 and s=nmod6. Then s{1,3,5}, r2an2n2s(mod6), (r,s)=(2,1),(0,3),(4,5), δ=s12, and the left–hand side of (3.2) is A2 if s=1, A2+2 if s=3, and A2+3 if s=5, where A2=ν2(2an6!)ν2(n6!)ν2((2a1)n6!). In addition, we have

    ν2(2an6!)=(2a11)n3a12ε+ν2(n3!),
    ν2(n6!)=ν2(n3!)n6,
    ν2((2a1)n6!)=ν2(A!)(2a1)n6,
    n6+(2a1)n6=2anr6[s{3,5}].

    Therefore

    A2=(2a11)n3a12ε+2anr6[s{3,5}]ν2(A!).

    Furthermore,

    A=(2a1)n3=2anr3ns3+rs3={(2a1)n313,if s=1;(2a1)n3,if s=3;(2a1)n323,if s=5

    which implies that A=(2a1)n3r6. Then

    A2=Aa12ε[s{3,5}]ν2(A!).

    It is now easy to check that A2 (if s=1), A2+2 (if s=3), and A2+3 (if s=5), are the same as δ+Aa12εν2(A!) in (3.2). So (3.2) is verified.

    Case 2. 2n. This case is similar to Case 2 of the previous part. So we write n=2b where 2 and let m=2a3, k=3, r=2amod3, and s=mod3. We obtain by Theorem 6 that ν2((2ann)F) is equal to

    ν2((2a+b2b)F)={ν2((mk)),if 0(mod3);b+12+ν2((mk)),if 1(mod3);b2+1+ν2(mk)+ν2((mk)),if 2(mod3). (3.4)

    By Theorem 4, we have

    ν2(m!)=(2a1)3a12ε{3}+ν2(k!).

    Since (2a1)(mod3), {(2a1)3}={3}. This implies that ν2(m!)=Aa12ε+ν2(k!). In addition, (r,s)=(0,0),(2,1),(1,2), and

    mk=2a33=2ar3s3=(2a1)(rs)3=A+[s=2].

    From the above observation, we obtain

    ν2((mk))=ν2(m!)ν2(k!)ν2((mk)!)={Aa12εν2(A!),if s=0,1;Aa12εν2((A+1)!),if s=2.

    Substituting this in (3.4), we see that

    ν2((2ann)F)={Aν2(A!),if 0(mod3);b+12+Aa12ν2(A!),if 1(mod3);b2+1+Aa12ν2(A!),if 2(mod3). (3.5)

    Recall that n=2b(1)b(mod3). So (3.5) implies that

    ν2((2ann)F)={Aν2(A!),if n0(mod3);b2+1+Aa12ν2(A!),if n1(mod3) and b is even;b+12+1+Aa12ν2(A!),if n1(mod3) and b is odd;b2+1+Aa12ν2(A!),if n2(mod3) and b is even;b+12+Aa12ν2(A!),if n2(mod3) and b is odd, 

    which is the same as (3.2). This completes the proof.

    We can obtain the main result of Maques and Trojovský [25] as a corollary.

    Corollary 8. (Marques and Trojovský [25]) (2nn)F is even for all n2.

    Proof. Let n2 and apply Theorem 7 with a=1 to obtain ν2((2nn)F)=δ+s2(A). If n0,1(mod6), then δ>0. If n0(mod6), then n32ν2(n), and so A1 and s2(A)>0. If n1(mod6), then A=n3>1 and so s2(A)>0. In any case, ν2((2nn)F)>0. So (2nn)F is even.

    Corollary 9. Let n2. Then (4nn)F is even if and only if n is not a power of 2. In other words, for each nN, (4nn)F is odd if and only if n=2k for some k0.

    Proof. Let δ, ε, and A be as in Theorem 7. If n=2k for some k1, then we apply Theorem 7 with a=2, δ=0, ε=1, A=1 leading to ν2((4nn)F)=0, which implies that (4nn)F is odd.

    Suppose n is not a power of 2. By Theorem 7, ν2((4nn)F)=δ+s2(A)εs2(A)1. Since n is not a power of 2, the sum s2(n)2. It is easy to see that s2(m)=s2(2cm) for any c,mN. Therefore s2(A)=s2(n2ν2(n))=s2(2ν2(n)n2ν2(n))=s2(n)2, which implies ν2((4nn)F)1, as required.

    Observe that 2,22,23 are congruent to 2,4,1(mod7), respectively. This implies that if k1 and k1(mod3), then (1+32k)/7 is an integer. We can determine the integers n such that (8nn)F is odd as follows.

    Corollary 10. (8nn)F is odd if and only if n=1+32k7 for some k1(mod3).

    Proof. Let a,δ,A,ε be as in Theorem 7. We first suppose n=(1+32k)/7 where k1 and k1(mod3). Then n7n1+32k1(mod6). Then a=3, ε=1, δ=0, A=2k, and so ν2((8nn)F)=0. Therefore (8nn)F is odd. Next, assume that (8nn)F is odd. Observe that A2 and s2(A)>0. If n0(mod3), then ε=0 and ν2((8nn)F)=δ+s2(A)>0, which is not the case. Therefore n1,2(mod3), and so ε=1. If n0(mod2), then δ=ν2(n)+3nmod321, and so ((8nn))Fs2(A)>0, which is a contradiction. So n1(mod2). This implies n1,5(mod6). But if n5(mod6), then δ2 and ν2((8nn)F)>0, a contradiction. Hence n1(mod6). Then δ=0. Since s2(A)1=ν2((8nn)F)=0, we see that A=2k for some k1. Then 7n13=7n3=A=2k, which implies n=1+32k7, as required.

    Theorem 11. For each a,nN, ν5((5ann)F)=ν5((5ann))=s5((5a1)n)4. In particular, (5ann)F is divisible by 5 for every a,nN.

    Proof. The first equality follows immediately from Theorem 5(ⅱ). By Legendre's formula, ν5((nk))=s5(k)+s5(nk)s5(n)4 for all nk1. So ν5((5ann)F) is

    s5(n)+s5(5ann)s5(5an)4=s5((5a1)n)4.

    Theorem 12. Let p2,5, a,nN, r=panmodz(p), s=nmodz(p), and A=n(pa1)pνp(n)z(p). Then the following statements hold.

    (i) If p±1(mod5), then νp((pann)F) is equal to

    Ap1a{npνp(n)z(p)}νp(A!)=sp(A)p1a{npνp(n)z(p)}. (3.6)

    (ii) If p±2(mod5) and a is even, then νp((pann)F) is equal to

    Ap1a2[s0]νp(A!)=sp(A)p1a2[s0]. (3.7)

    (iii) If p±2(mod5) and a is odd, then νp((pann)F) is equal to

    Ap1a12[s0]νp(A!)+δ, (3.8)

    where δ=(νp(n)2+[2νp(n)][r>s]+[r<s]νp(Fz(p)))[rs], or equivalently, δ=0 if r=s, δ=νp(n)2+νp(Fz(p)) if r<s, and δ=νp(n)2 if r>s.

    Proof. We first prove (ⅰ) and (ⅱ). So we suppose that the hypothesis of (ⅰ) or (ⅱ) is true. By writing νp(A!)=Asp(A)p1, we obtain the equalities in (3.6) and (3.7). By Lemma 1(ⅱ), pa1(modz(p)). Then r=s.

    Case 1. pn. Let m=panz(p) and k=nz(p). Then we obtain by Theorem 5(ⅲ) that

    νp((pann)F)=νp((mk))=νp(m!)νp(k!)νp((mk)!). (3.9)

    By Lemma 1(ⅱ) and Theorem 4, we see that if p±1(mod5), then p1(modz(p)) and

    νp(m!)=νp(npaz(p)!)=n(pa1)z(p)(p1)a{nz(p)}+νp(k!), (3.10)

    and if p±2(mod5) and a is even, then p1(modz(p)) and

    νp(m!)=n(pa1)z(p)(p1)a2[s0]+νp(k!). (3.11)

    Since z(p)pa1 and pn, A=n(pa1)z(p). Therefore

    mk=panz(p)nz(p)=panrz(p)nsz(p)=n(pa1)z(p)=A. (3.12)

    Substituting (3.10), (3.11), and (3.12) in (3.9), we obtain (3.6) and (3.7).

    Case 2. pn. Let n=pb where p, m=paz(p), and k=z(p). Since r=s, we obtain by Theorem 6 that νp((pann)F) is equal to

    νp((pa+bpb)F)=νp((mk))=νp(m!)νp(k!)νp((mk)!). (3.13)

    Since gcd(p,z(p))=1, we see that 0(modz(p))n0(modz(p))s=0. Similar to Case 1, we have νp(m!)=(pa1)z(p)(p1)a{z(p)}+νp(k!) if p±1(mod5), νp(m!)=(pa1)z(p)(p1)a2[s0]+νp(k!) if p±2(mod5) and a is even, pa(modz(p)), A=(pa1)z(p), and mk=A. So (3.13) leads to (3.6) and (3.7). This proves (ⅰ) and (ⅱ).

    To prove (ⅲ), suppose that p±2(mod5) and a is odd. By Lemma 1(ⅱ), p1(modz(p)). In addition, pa1p1=pa1+pa2++11(modz(p)). We divide the consideration into two cases.

    Case 1. pn. This case is similar to Case 1 of the previous part. So we apply Theorems 4 and 5(ⅲ). Let m=panz(p) and k=nz(p). Then

    νp(m!)=n(pa1)z(p)(p1)a12[s0]{nz(p)}+νp(k!),
    mk=panrz(p)nsz(p)=n(pa1)(rs)z(p),
    A=nparz(p)nsz(p)+rsz(p)=mk+rsz(p).

    Since pa1p11(modz(p)), n(pa1)p1n(modz(p)). This implies that {n(pa1)z(p)(p1)}={nz(p)}. Therefore

    νp(m!)=n(pa1)z(p)(p1)a12[s0]+νp(k!)=Ap1a12[s0]+νp(k!).

    From the above observation, if rs, then A=mk and

    νp((pann)F)=νp((mk))=Ap1a12[s0]νp(A!),

    which leads to (3.8). If r<s, then A=mk1, pann+z(p)z(p)=A+1, and νp((pann)F) is equal to

    Ap1a12[s0]νp((A+1)!)+νp(A+1)+νp(Fz(p))=Ap1a12[s0]νp(A!)+νp(Fz(p)),

    which is the same as (3.8).

    Case 2. pn. Let n=pb where p, m=paz(p), and k=z(p). Similar to Case 1, s=00(modz(p)). In addition, (pa1)p1(modz(p)), and so we obtain by Theorem 4 that νp(m!)=Ap1a12[s0]+νp(k!). The calculation of νp((pann)F)=νp((pa+bpb)F) is done by the applications of Theorem 6 and is divided into several cases. Suppose r=s. Then pa+bpanrsnpb(modz(p)). Since (p,z(p))=1, this implies pa(modz(p)). Therefore A=paz(p)=paz(p)=mk and

    νp((pann)F)=νp((mk))=νp(m!)νp(k!)νp((mk)!),

    which is (3.8). Obviously, if 0(modz(p)), then r=s, which is already done. So from this point on, we assume that rs and 0(modz(p)). Recall that p1(modz(p)) and a is odd. So if b is odd, then

    rnpanpb(modz(p)),  snpbpa(modz(p)),  and
    A=pasz(p)rz(p)+srz(p)=pasz(p)rz(p)+srz(p)=mk+srz(p).

    Similarly, if b is even, then r=pamodz(p), s=modz(p), and A=mk+rsz(p). Let R=Ap1a12[s0]νp(A!)+δ be the quantity in (3.8). From the above observation and the application of Theorem 6, we obtain νp((pann)F) as follows. If r>s and b is even, then A=mk and

    νp((pann)F)=b2+νp((mk))=b2+Ap1a12[s0]νp(A!)=R.

    If r>s and b is odd, then A=mk1 and

    νp((pann)F)=b+12+νp(A+1)+νp((mk))=b+12+Ap1a12[s0]νp(A!)=R.

    If r<s and b is even, then A=mk1 and

    νp((pann)F)=b2+νp(Fz(p))+νp(A+1)+νp((mk))=b2+νp(Fz(p))+Ap1a12[s0]νp(A!)=R.

    If and is odd, then and

    This completes the proof.

    In the next two corollaries, we give some characterizations of the integers such that is divisible by .

    Corollary 13. Let be a prime and let and be positive integers. If , then .

    Proof. We first consider the case . Assume that and , , , and are as in Theorem 12. Then , , and . Every case in Theorem 12 leads to , which implies . If , then the result follows immediately from Theorem 11. If , then every case of Theorem 7 leads to , which implies the desired result.

    Corollary 14. Let be a prime and let , , , , and be as in Theorem 12. Assume that and . Then the following statements hold.

    Assume that is even. Then if and only if .

    Assume that is odd and . If , then . If , then if and only if .

    Assume that is odd and . If , then . If , then if and only if .

    Proof. We use Lemmas 2 and 3 repeatedly without reference. For (ⅰ), we obtain by (3.7) that

    This proves (ⅰ). To prove (ⅱ) and (ⅲ), we let be as in Theorem 12 and divide the consideration into two cases.

    Case 1. . If , then we obtain by Theorem 5(ⅲ) that . Suppose . Then and (3.8) is

    If , then (3.8) implies that

    Similarly, if , then . This proves (ⅱ).

    Case 2. . We write where . Then . Recall that . If , then Theorem 6 implies that if is even and it is if is odd. In any case, . So . If , then and we obtain as in Case 1 that if and only if . This proves (ⅲ).

    Corollary 15. Let be a prime and let . Assume that . Then if and only if .

    Proof. We remark that by Lemma 1(ⅱ), is an integer. Let . We apply Theorem 12(ⅰ) with . If , then (3.6) implies that . If , then . This completes the proof.

    We give exact formulas for the -adic valuations of Fibonomial coefficients of the form for all primes and . Then we use it to characterize the integers such that is divisible by .

    Phakhinkon Phunphayap receives a scholarship from Science Achievement Scholarship of Thailand(SAST). This research was jointly supported by the Thailand Research Fund and the Faculty of Science Silpakorn University, grant number RSA5980040.

    The authors declare that there is no conflict of interests regarding the publication of this article.



    Conflict of interest



    All authors declare no conflicts of interest in this paper.

    [1] World Health Organization Coronavirus disease (‎‎COVID-19)‎‎: Weekly Epidemiological Report, 27 January 2021 (2021) .Available from: https://www.who.int/publications/m/item/weekly-epidemiological-update---27-january-2021.
    [2] Bai Y, Yao L, Wei T, et al. (2020) Presumed asymptomatic carrier transmission of COVID-19. JAMA 323: 1406-1407. doi: 10.1001/jama.2020.2565
    [3] Bastos ML, Tavaziva G, Abidi SK, et al. (2020) Diagnostic accuracy of serological tests for covid-19: systematic review and meta-analysis. BMJ 1: 370.
    [4] Roda WC, Varughese MB, Han D, et al. (2020) Why is it difficult to accurately predict the COVID-19 epidemic? Infect Dis Modell 5: 271-281. doi: 10.1016/j.idm.2020.03.001
    [5] Naudé W (2020) Artificial intelligence vs COVID-19: limitations, constraints and pitfalls. AI Soc 35.
    [6] Volpert V, Banerjee M, Petrovskii S (2020) On a quarantine model of coronavirus infection and data analysis. Math Modell Nat Phenom 15: 24. doi: 10.1051/mmnp/2020006
    [7] Zhao S, Chen H (2020) Modeling the epidemic dynamics and control of COVID-19 outbreak in China. Quant Biol 11: 1-9.
    [8] Shen CY (2020) A logistic growth model for COVID-19 proliferation: experiences from China and international implications in infectious diseases. Int J Infect Dis .
    [9] Elliott G, Stock JH (2001) Confidence intervals for autoregressive coefficients near one. J Econometrics 103: 155-181. doi: 10.1016/S0304-4076(01)00042-2
    [10] Hsiao WC, Huang HY, Ing CK (2018) Interval Estimation for a First-Order Positive Autoregressive Process. J Time Ser Anal 39: 447-467. doi: 10.1111/jtsa.12297
    [11] Branas CC, Rundle A, Pei S, et al. (2020) Flattening the curve before it flattens us: hospital critical care capacity limits and mortality from novel coronavirus (SARS-CoV2) cases in US counties. medRxiv .
    [12] Biswas K, Khaleque A, Sen P (2003) Covid-19 spread: Reproduction of data and prediction using a SIR model on Euclidean network. arXiv preprint arXiv:2003.07063 2020 Mar 16.
    [13] Postnikov EB (2020) Estimation of COVID-19 dynamics “on a back-of-envelope”: Does the simplest SIR model provide quantitative parameters and predictions? Chaos, Solitons Fractals 135: 109841. doi: 10.1016/j.chaos.2020.109841
    [14] Metcalf CJ, Lessler J (2017) Opportunities and challenges in modeling emerging infectious diseases. Science 357: 149-152. doi: 10.1126/science.aam8335
    [15] Funk S, Camacho A, Kucharski AJ, et al. (2018) Real-time forecasting of infectious disease dynamics with a stochastic semi-mechanistic model. Epidemics 22: 56-61. doi: 10.1016/j.epidem.2016.11.003
    [16] He ZL, Li JG, Nie L, et al. (2017) Nonlinear state-dependent feedback control strategy in the SIR epidemic model with resource limitation. Adv Differ Equ 2017: 1-8. doi: 10.1186/s13662-016-1057-2
    [17] Dubey B, Dubey P, Dubey US (2015) Dynamics of an SIR Model with Nonlinear Incidence and Treatment Rate. Appl Appl Math 10: 718-737.
    [18] Harjule P, Tiwari V, Kumar A (2021) Mathematical models to predict COVID-19 outbreak: An interim review. J Interdiscip Math 13: 1-26. doi: 10.1080/09720502.2020.1848316
    [19] Eker S (2020) Validity and usefulness of COVID-19 models. Humanit Soc Sci Commun 7: 1-5. doi: 10.1057/s41599-020-00553-4
    [20] Iwasaki A, Yang Y (2020) The potential danger of suboptimal antibody responses in COVID-19. Nat Rev Immunol 21: 1-3.
    [21] To KK, Tsang OT, Leung WS, et al. (2020) Temporal profiles of viral load in posterior oropharyngeal saliva samples and serum antibody responses during infection by SARS-CoV-2: an observational cohort study. Lancet Infect Dis .
    [22] Bertozzi AL, Franco E, Mohler G, et al. (2020) The challenges of modeling and forecasting the spread of COVID-19. arXiv preprint arXiv:2004.04741 .
    [23] Nakamura G, Grammaticos B, Deroulers C, et al. (2020) Effective epidemic model for COVID-19 using accumulated deaths. arXiv preprint arXiv:2007.02855 .
    [24] Bogg T, Milad E Slowing the Spread of COVID-19: Demographic, personality, and social cognition predictors of guideline adherence in a representative US sample (2020) .Available from: https://www.researchgate.net/publication/340427042_Slowing_the_Spread_of_COVID-19_Demographic_Personality_and_Social_Cognition_Predictors_of_Guideline_Adherence_in_a_Representative_US_Sample.
    [25] Dowd JB, Andriano L, Brazel DM, et al. (2020) Demographic science aids in understanding the spread and fatality rates of COVID-19. P Natl Acad Sci USA 117: 9696-9698. doi: 10.1073/pnas.2004911117
  • publichealth-08-01-010-s001.pdf
  • This article has been cited by:

    1. Kritkhajohn Onphaeng, Prapanpong Pongsriiam, Exact divisibility by powers of the integers in the Lucas sequence of the first kind, 2020, 5, 2473-6988, 6739, 10.3934/math.2020433
    2. Phakhinkon Napp Phunphayap, Prapanpong Pongsriiam, Jeffrey Shallit, Sumsets associated with Beatty sequences, 2022, 345, 0012365X, 112810, 10.1016/j.disc.2022.112810
    3. Phakhinkon Napp Phunphayap, Prapanpong Pongsriiam, Divisibility of Fibonomial coefficients in terms of their digital representations and applications, 2022, 7, 2473-6988, 5314, 10.3934/math.2022296
    4. Kritkhajohn Onphaeng, Prapanpong Pongsriiam, Exact divisibility by powers of the integers in the Lucas sequences of the first and second kinds, 2021, 6, 2473-6988, 11733, 10.3934/math.2021682
    5. Christian J.-C. Ballot, Hugh C. Williams, 2023, Chapter 5, 978-3-031-37237-7, 105, 10.1007/978-3-031-37238-4_5
    6. Prapanpong Pongsriiam, Sums of divisors on arithmetic progressions, 2024, 88, 0031-5303, 443, 10.1007/s10998-023-00566-x
  • Reader Comments
  • © 2021 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(3871) PDF downloads(137) Cited by(7)

Figures and Tables

Figures(2)  /  Tables(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog