We study combinatoric convolution sums of certain divisor functions involving even indices. We express them as a linear combination of divisor functions and Euler polynomials and obtain identities D2k(n)=(1/4)σ2k+1,0(n;2)-2·42kσ2k+1(n/4)-(1/2)[∑d|n,d≡1(4){E2k(d)+E2k(d-1)}+22k∑d|n,d≡1(2)E2k((d+(-1)(d-1)/2)/2)], U2k(p,q)=22k-2[-((p+q)/2)E2k((p+q)/2+1)+((q-p)/2)E2k((q-p)/2)-E2k((p+1)/2)-E2k((q+1)/2)+E2k+1((p+q)/2 +1)-E2k+1((q-p)/2)], and F2k(n)=(1/2){σ2k+1†(n)-σ2k†(n)}. As applications of these identities, we give several concrete interpretations in terms of the procedural modelling method.
1. Statement of Main Theoretical Results
Bernoulli polynomials Bn(x) and Euler polynomials En(x) are usually defined by means of the following generating functions: ueux/(eu-1)=∑k=0∞Bk(x)(uk/k!) and 2ueux/(eu+1)=∑k=0∞Ek(x)(uk/k!).
Bernoulli numbers and Euler numbers are Bk=Bk(0) and Ek=2kEk(1/2). We recall the fundamental properties as follows [1, 2]:
(1)Bk(1-x)=(-1)kBk(x),Ek(1-x)=(-1)kEk(x),(2)Ek(x)=2k+1{Bk+1(x)-2k+1Bk+1(x2)},(3)Bn+1(x+1)-Bn+1(x)=(n+1)xn,En(x+1)+En(x)=2xn,(4)Bn(x+y)=∑k=0n(nk)Bk(x)yn-k,En(x+y)=∑k=0n(nk)Ek(x)yn-k.
For n,N∈N and i,k,l∈Z, we define the following:
(5)σk(n)=∑d∣ndk,σk,l(n;2)=∑d∣n,d≡l(2)dk,σk*(n)=∑d∣n,n/dodddk,σk†(n)=∑d∣n,n/d≡1(4)dk-(-1)k∑d∣n,n/d≡-1(4)dk,σk††(n)=∑d∣n,d≡1(4)dk-(-1)k∑d∣n,d≡-1(4)dk.
We are motivated by the Besge [3] and Ramanujan [4] identity
(6)∑k=1n-1σ1(k)σ1(n-k)=512σ3(n)+(112-12n)σ1(n)
and also by the recent development on the convolution sums. In this paper, we investigate the three combinatoric convolution sums with “even indices.” Consider
(7)D2k(n):=∑r=0k(2k2r)∑m=1n-1σ2k-2r††(m)σ2r††(n-m),F2k(n):=∑r=0k(2k2r)∑m=1n-1σ2k-2r†(m)σ2r†(n-m),U2k(p,q):=∑a,b,c,deven,a+b≥21≤l≤p-1,1≤l′≤q-1(2ka,b,c,d)11111111111×σa††(l)σb††(p-l)σc††(l′)11111111111×σd††(q-l′),
where (2ka,b,c,d)=(2k)!/a!b!c!d! with a+b+c+d=2k. Now, we state our main results.
Theorem 1.
For any positive integers k≥1 and n≥2, one has
(8)D2k(n)=14σ2k+1,0(n;2)-2·42kσ2k+1(n4)-12[∑d∣n,d≡1(2)E2k(d+(-1)(d-1)/22)∑d∣n,d≡1(4){E2k(d)+E2k(d-1)}+22k∑d∣n,d≡1(2)E2k(d+(-1)(d-1)/22)].
Corollary 2.
Let k≥4 be integer and p and q primes with p≡q≡-1(mod4). Then one has
(9)U2k(p,q)=22k-2[-(p+q2)E2k(p+q2+1)+(q-p2)E2k(q-p2)-E2k(p+12)-E2k(q+12)+E2k+1(p+q2+1)-E2k+1(q-p2)(p+q2)].
Theorem 3.
For any positive integers k≥1 and n≥2, one has
(10)F2k(n)=12{σ2k+1†(n)-σ2k†(n)}.
Theorem 3 can be obtained by [5, 6]. Therefore, we omit the proof.
2. Application and Perspective 2.1. Branch and Leaves Pattern
In real-time virtual system, realistic and efficient generation and expression of plants composing a broad terrain is a continuing problem [7, 8]. The procedural modelling method using convolution sums of divisor functions (MCD) was suggested for a variety of natural trees in a virtual ecosystem [9]. The basic structure of MCD is that it defines the growth grammar including the branch propagation, a growth pattern of branches and leaves, and a process of growth deformation for various generations of tree. In Figure 1, we illustrate the MCD for the divisor trees.
MCD tree.
For more examples see [10]. Theorems 1 and 3 give us basic background for efficient and diverse generations and expressions of trees composing virtual ecosystem or real-time animation processing. In order to apply MCD to the growth structure of a tree model, (5) is modified and expressed in
(11)∑r=0l(2l2r)∑k-1N-1Di(Bi(x,y)k)Di(Li(x,y)(N-k)),
where D represents various divisor functions, i is the current growth step (gs), and N-1 is the final iteration number of the ith growth step. Here, Di(Bi(x,y)k) is a divisor function that determines the pattern of the number of branches and Di(Li(x,y)(N-k) is a divisor function that determines the number of leaves with l+1 different types of trees and grasses in the virtual system. Further, Bi(x,y),Li(x,y) are the ith exponential function of branches and leaves. Through this model, we verified the possibility of natural and efficient tree growth [9, 11]. We approximately fix a certain tree model for MCD satisfying the patterns of branch and leaves in virtual system or real tree wood; then using formulae (8), (9), and (10), we obtain the approximate total number for MCD trees leaves by taking two kinds of trees with approximate MCD (Theorem 3), that is, branch pattern Bi(x,y)=σ0† (resp., σ2†) and leaves pattern Li(x,y)=σ2† (resp., σ0†). We give the first eleven values of the approximate total number T(n) for MCD tree leaves with n-1 steps (see Table 1).
n
2
3
4
5
6
7
8
9
10
11
12
T(n)
2
10
24
50
96
148
224
342
452
606
832
2.2. Color of Leaves
Leaves are important elements that constitute plants [12]. In [13] we see the leaves model of convolution sum of divisor functions. First, we fix the random color variation step N and then we compute convolution sums by use of (6) as shown in Figure 2. We denote by (fs,fd) the start and destination colors; output colors in phase are determined by interpolating convolution sums. Next, based on the minimum and maximum values of the elements of convolution sums of divisor functions, normalized weighted values between 0 and 1 are calculated and then each color table is organized as follows:
(12)t=d-dmindmax-dmin,f=(1-t)fs+fdt,dstep=(cnmax-cnmin)dsize+1,idx=(cnij-cnmin)dstep,(13)cleafij=cdidx.
The noise map that utilizes Perlin noise is used in the creation of cloud textures for soft color changes. Here, dsize is the number of values that are not duplicated among the elements of convolution sums calculated in the selected table; cnmax, cnmin are the maximum and minimum noise values; and dstep indicates the section of noise values influenced by a single element of convolution sums. Once these parameters are fixed, the leaf color is determined by finding the index “idx” and the corresponding color “cdidx” in the color table comprised of divisor functions that correspond to the noise color in ith row and jth column [13]. Figure 2 represents the image of a color table set created by calculating the color table of each section. Theorem 1 gives the number of values of leaf color and Theorem 3 gives pattern of branch and leaves in the same model. In real-time ecosystem, efficient and fast time is requested. When n is large, the MCD that uses (7) is complicate to compute directly in reasonable time. Our results (8), (9), and (10) give simple formulas for efficient time for them. In the forthcoming papers, we will investigate more applications of real-time virtual ecosystem.
Color of leaves.
2.3. Proof of Main Results
To prove Theorem 1, we establish Lemmas 4, 6, 7, and 9 and Corollary 5 and we recall Proposition 8.
Lemma 4.
Let k≥1. Set B~k:=∑j=0k(kj)2jBj=2kBk(1/2). Then one has
(14)B2k+1(d+12)=122k+1∑i=0k(2k+12i+1)B~2k-2id2i+1.
Proof.
By formula (4) we obtain
(15)B2k+1(1+d2)-B2k+1(1-d2)=122k∑i=0k(2k+12i)B2i(12)22id2k+1-2i.
By (15) and (1) we get the lemma.
By [14, Theorem 3] and Lemma 4 we get the corollary.
Corollary 5.
Let k≥1 and n≥2. Then
(16)∑s=0k-1(2k2s+1)∑m=1n-1σ2k-2s-1,1(m;2)σ2s+1,1(n-m;2)=22k-1σ2k+1(n2)+22k2k+1∑d∣ndoddB2k+1(d+12).
Lemma 6.
For any integers α and k≥1, one has
(17)∑u,v,w≥0(ku,v,w)(4α+1)u(-1)w4vBv=4kBk(α),(18)22k+1∑u,v,w≥0(2k+1u,2v,w)(4α+1)u(-1)w42v-1B2v=2·42k2k+1B2k+1(α)+(4α)2k.
Proof.
Write the series as follows:
(19)∑k=0∞tkk!{∑u+v+w=k(ku,v,w)(4α+1)u(-1)w4vBv}=(∑u=0∞(4α+1)utuu!)(∑v=0∞Bv(4t)vv!)×(∑w=0∞(-1)wtww!)=(e(4α+1)t)(4te4t-1)(e-t)=∑k=0∞(4t)kk!Bk(α)=∑k=0∞tkk!(4kBk(α)).
This completes the proof of (17). On the other hand, (18) may be rewritten as
(20)22k+1∑u+2v+w=2k+1(2k+1u,2v,w)(4α+1)u(-1)w42v-1B2v=22k+1×∑u+v+w=2k+1(2k+1u,v,w)(4α+1)u(-1)w4v-1Bv-22k+1∑u+w=2k(2k+1u,1,w)(4α+1)u(-1)w41-1B1.
From (17) we obtain
(21)22k+1∑u+2v+w=2k+1(2k+1u,2v,w)(4α+1)u(-1)w42v-1B2v=12(2k+1)∑u+v+w=2k+1(2k+1u,v,w)(4α+1)u(-1)w4vBv+∑j=02k(2kj)(-1)j(4α+1)2k-j=2·42k2k+1B2k+1(α)+(4α)2k.
This completes the proof of the lemma.
Lemma 7.
For any integers k≥1 and n≥2, one has
(22)-22k+1∑u,v,w≥0(2k+1u,2v,w)(-1)w42v-1B2vσu††(n)=-∑d∣nd≡1(4)(2·42k2k+1B2k+1(d-14)+(d-1)2k)+∑d∣nd≡-1(4)(2·42k2k+1B2k+1(-d+14)+(d+1)2k).
Proof.
We observe that
(23)-22k+1∑u,v,w≥0(2k+1u,2v,w)(-1)w42v-1B2vσu††(n)=-22k+1∑4α+1∣n(2k+1u,2v,w)(-1)w42v-1×B2v(4α+1)u+22k+1∑4α-1∣n(2k+1u,2v,w)(-1)w42v-1B2v(-4α+1)u.
From (18), we get
(24)-∑4α+1∣n(2·42k2k+1B2k+1(α)+(4α)2k)+∑4α-1∣n(2·42k2k+1B2k+1(-α)+(4α)2k)=-∑d∣n,d≡1(4)(2·42k2k+1B2k+1(d-14)+(d-1)2k)+∑d∣n,d≡-1(4)(2·42k2k+1B2k+1(-d+14)+(d+1)2k).
We then get lemma.
We recall Proposition 8 to be modified in Lemma 9.
Proposition 8 (see [15]).
Let N≥3, 1≤i≤N-1, and n,k∈N. One has
(25)∑s=02k(2ks)(-1)s∑m=1n-1σ2k-s♭(m;i,N)σs♭(n-m;i,N)=-2N2kσ2k+1(nN)-∑u=02k(2ku)(-i)2k-uσu♭(n;i,N)-22k+1×∑u,v,w∈N∪{0}(2k+1u,2v,w)N2v-1B2v(-i)wσu♭(n;i,N),
where σr♭(n;i,N):=∑d∣nd≡i(N)dr-(-1)r∑d∣nd≡-i(N)dr.
Using Lemma 7, we modify Proposition 8 as follows.
Lemma 9.
For any n,k∈N one has(26)∑s=02k(2ks)(-1)s∑m=1n-1σ2k-s††(m)σs††(n-m)=-2·42kσ2k+1(n4)+2·42k2k+1×(-∑d∣n,d≡1(4)B2k+1(d-14)+∑d∣n,d≡-1(4)B2k+1(-d+14))+2(-∑d∣n,d≡1(4)(d-1)2k+∑d∣n,d≡-1(4)(d+1)2k).
Proof.
First, we check that
(27)∑u=02k(2ku)(-1)2k-uσu††(n)=∑u=02k(2ku)(-1)2k-u×(∑d∣n,d≡1(4)du-(-1)u∑d∣n,d≡-1(4)du)=∑d∣n,d≡1(4)(d-1)2k-∑d∣n,d≡-1(4)(d+1)2k.
From Lemma 7 and Proposition 8, our claim follows.
To end the proof of Theorem 1, we observe that
(28)D2k(n)=∑s=02k(2ks)(-1)s∑m=1n-1σ2k-s††(m)σs††(n-m)+∑s=0k-1(2k2s+1)×∑m=1n-1σ2k-2s-1,1(m;2)σ2s+1,1(n-m;2).
Comparing Corollary 5 and Lemma 9, we see that
(29)D2k(n)=14σ2k+1,0(n;2)-2·42kσ2k+1(n4)-2∑d∣n,d≡1(2)(-1)(d-1)/2(d-(-1)(d-1)/2)2k+22k2k+1×[∑d∣n,d≡1(4)B2k+1(d+12)+∑d∣n,d≡-1(4)B2k+1(d+12)-22k+1∑d∣n,d≡1(4)B2k+1(d-14)-22k+1∑d∣n,d≡-1(4)B2k+1(d+54)].
From (3) we have
(30)D2k(n)=14σ2k+1,0(n;2)-2·42kσ2k+1(n4)-2∑d∣n,d≡1(2)(-1)(d-1)/2(d-(-1)(d-1)/2)2k+22k2k+1×[∑d∣n,d≡1(4){B2k+1(d-12)-22k+1B2k+1(d-14)}+∑d∣nd≡-1(4){B2k+1(d+12)-22k+1B2k+1(d+14)}+∑d∣n,d≡1(4)(2k+1)(d-12)2k-∑d∣n,d≡-1(4)2(2k+1)(d+12)2k].
By (2) and (3) we get the theorem.
Proof of Corollary 2.
From ([16, page 150]) we can derive
(31)∑r=1k-1(2k2r)E2r(x)E2k-2r(y)=(1-x-y)E2k(x+y)+E2k+1(x+y)+(y-x)E2k(y-x)-E2k+1(y-x)-E2k(x)-E2k(y).
We set x=(p+1)/2 and y=(q+1)/2 in (31) and get the corollary.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
ErdelyiA.19531New York, NY, USAMcGraw HillSrivastavaH. M.PinterA.Remarks on some relationships between the Bernoulli and Euler polynomials200417437538010.1016/S0893-9659(04)90077-8MR20457402-s2.0-1842790483BesgeM.Extrait d'une lettre de M. Besge à M. Liouville18627256RamanujanS.On certain arithmetical functions191622159184ChoB.KimD.ParkH.Evaluation of a certain combinatorial convolution sum in higher level cases2013406120321010.1016/j.jmaa.2013.04.052MR30624122-s2.0-84878795894KimD.BayadA.Convolution identities for twisted Eisenstein series and twisted divisor functions20132013, article 8110.1186/1687-1812-2013-81MR3055852PalubickiW.HorelK.LongayS.RunionsA.LaneB.MěchR.PrusinkiewiczP.Self-organizing tree models for image synthesis2009283, article 5810.1145/1531326.15313642-s2.0-70349684624TanP.FangT.XiaoJ.ZhaoP.QuanL.Single image tree modeling2008275, article 10810.1145/1409060.14090612-s2.0-57749204351KimJ.KimD.ChoH.Procedural modeling of trees based on convolution sums of divisor functions for real-time virtual ecosystems2013243-423724610.1002/cav.15062-s2.0-84877855437http://open.nims.re.kr/JrDr/KimJ.KimD.ChoH.Tree growth model design for realistic game landscape production (Korean)20131324958XiaoH.ChenX.Modeling and simulation of curled dry leaves2011722107941080210.1039/c1sm05998j2-s2.0-84255190217KimJ.KimD.The procedural modeling and visualization of multiple leavesSubmittedKimD.ParkY. K.Bernoulli identities and combinatoric convolution sums with odd divisor functions20142014889097310.1155/2014/890973MR3166664ChoB.KimD.ParkH.Certain combinatorial convolu tion sums for divisor functions and Bernoulli numbersaccepted to Bulletin of the Korean Mathematical SocietyChuW.ZhouR. R.Convolution of Bernoulli and Euler polynomials2010618147163MR2757610