betaFeedback

WorksMathematics

Optimal bounds for design curves on the torus

Abstract

A closed rectifiable curve on the torus Td=Rd/Zd\mathbb T^d=\mathbb R^d/\mathbb Z^d is a tt-design curve if normalized arclength integration is exact for trigonometric polynomials e2πi⟨m,x⟩e^{2πi \langle m,x\rangle} with frequencies m∈Zdm\in\mathbb Z^d satisfying ∥m∥∞≤t\|m\|_\infty\le t. For the arclength LL of such curves, we establish the sharp lower bound \begin{equation}\label{eq:abs} L\geq \Big(\frac{(t+1)^{2d}-1}{t(t+2)}\Big)^{1/2}. \end{equation} Equality is attained by the family of Korobov curves described in \cite{Kunis,Lindblad}. For even tt, we also derive an explicit lower bound on the length of total-degree tt-design curves, defined by the frequency condition ∥m∥1≤t\|m\|_1\leq t. This bound is sharp for t=2t=2 in every dimension and for every even tt in dimension two. We determine the minimum length of hyperbolic cross t2t^2-design curves in dimension two. For d≥3d\geq3, we establish lower bounds of optimal order for the length of hyperbolic cross tdt^d-design curves. Finally, we obtain lower bounds of optimal order for the length of tt-design curves associated with sparse frequency sets.

Introduction

Exact integration along closed curves was used in [4] to approximate measures on the torus and the sphere, with exactness imposed on trigonometric polynomials and spherical polynomials, respectively. The notion of a tt-design curve was subsequently introduced on the sphere in [5], requiring normalized arclength integration to be exact for all spherical polynomials of degree at most tt. Such curves are analogues of spherical tt-design point sets [2], which have been studied extensively for several decades [1]. More recently, asymptotically optimal tt-design curves were constructed by Lindblad [11] on S3⊆R4S^3 \subseteq\mathbb{R}^4 and on tori, and by Kunis [10] on tori. King and Mixon [9] obtained explicit spectral lower bounds on the length and exact optimality results for 2-design curves on odd-dimensional spheres S2d−1⊆R2dS^{2d-1} \subseteq\mathbb{R}^{2d}.

We concentrate on the torus Td=Rd/Zd\mathbb{T}^d = \mathbb{R}^d/\mathbb{Z}^d.

Definition 1.1. Let tt be a positive integer. A rectifiable curve γ:[0,L]→Rd\gamma: [0,L] \to\mathbb{R}^d with γ(L)−γ(0)∈Zd\gamma(L)-\gamma(0) \in\mathbb{Z}^d, parametrized by arclength, is called a tt-design curve on Td\mathbb{T}^d if

1L∫0Le2πi⟨m,γ(s)⟩ ds=∫Tde2πi⟨m,x⟩ dx,(1)\frac{1}{L}\int_0^L e^{2\pi i\langle m,\gamma(s)\rangle}\,ds = \int_{\mathbb{T}^d} e^{2\pi i\langle m,x\rangle}\,dx, \tag*{(1)}

for every m∈Zdm \in\mathbb{Z}^d satisfying ∥m∥∞≤t\lVert m\rVert_\infty\le t.

2020 Mathematics Subject Classification. 41A55, 41A63, 26B15, 65D32.

Key words and phrases. tt-designs, curves, torus, trigonometric polynomials.

Kunis [10] obtained explicit lower bounds on the length needed for exact trigonometric quadrature as in (1). The bounds are of optimal order in tt and allow additional weighting in the arclength integral, hence, hold for a larger class than tt-design curves. For even tt, a simplified version in [10 Theorem 3.10] gives L≥0.942d+1(t+1)d−1L \ge\frac{0.94}{\sqrt{2d+1}}(t+1)^{d-1}. Here, in the stricter unweighted setting of tt-design curves, we derive the estimate (1) that implies L≥(t+1)d−1L \ge(t+1)^{d-1}. This bound turns out to be sharp.

We adapt King and Mixon’s spectral approach on the sphere [9] to the torus:

(a) The design property makes a suitable family of exponential functions along the curve orthonormal.

(b) Fourier analysis and Bessel’s inequality yield a lower bound for the averaged L2L^2-energy of derivatives of this family.

(c) On the torus, the energy and the lower bound can be evaluated explicitly.

We derive results for four frequency shapes:

(1) tt-design curve. The resulting estimate (1) is sharp. The Korobov curves described in [10, 11] attain this bound and therefore have minimum length among all tt-design curves on Td\mathbb{T}^d. These main results are derived in § 2.

(2) Total degree tt-design curve. We also consider the related notion of total-degree tt-design curves in § 3, where the frequencies are {m∈Zd:∥m∥1≤t}\{m \in\mathbb{Z}^d : \lVert m\rVert_1 \le t\}. Curves of minimal length are constructed for total-degree 1 and 2 in every dimension and for every even total degree in dimension two.

(3) Hyperbolic cross tt-design curves. We discuss the hyperbolic cross [3, 8],

Hd(t):={m∈Zd:∏j=1dmax⁡(1,∣mj∣)≤t},H_d(t) := \left\{m \in\mathbb{Z}^d : \prod_{j=1}^{d} \max(1,\lvert m_j\rvert) \le t\right\},

as an alternative to tensor-product frequency sets in § 4. We derive sharp bounds for exact arclength integration of frequencies in the hyperbolic cross in dimension d=2d=2. For d≥3d \ge3, we obtain sharp lower bounds on the order of the length. This higher-dimensional hyperbolic-cross bound is attained by a construction that uses a uniform lattice-point counting argument and a hyperbolic-cross estimate.

(4) Sparse tt-design curves. Finally, we consider the sparse frequency sets

Ωd,r(t)={m∈Zd:∥m∥∞≤t,#supp⁡(m)≤r},1≤r≤d.\Omega_{d,r}(t) = \left\{m \in\mathbb{Z}^d : \lVert m\rVert_\infty\le t,\quad\#\operatorname{supp}(m) \le r\right\},\qquad1 \le r \le d.

The corresponding trigonometric polynomials are sums of terms depending on at most rr coordinates. For fixed dd and rr, we establish lower bounds of order tr−1t^{r-1} for the length of design curves associated with these frequencies and verify the existence of simple closed geodesics with matching length order. Thus the optimal length exponent is determined by the sparsity parameter rr, rather than by the ambient dimension dd.

Sharp length bounds for tt-design curves

Let dd be a positive integer and denote the dd-dimensional torus by Td=Rd/Zd\mathbb{T}^d = \mathbb{R}^d/\mathbb{Z}^d, endowed with the normalized Haar measure dx\mathrm{d}x. We write ds\mathrm{d}s for the Lebesgue measure on the real line. A closed rectifiable curve on Td\mathbb{T}^d of length L>0L>0 is represented by a path γ ⁣:[0,L]→Rd\gamma\colon[0,L]\to\mathbb{R}^d, parametrized by arclength, satisfying

γ(L)−γ(0)∈Zd.\gamma(L)-\gamma(0)\in\mathbb{Z}^d.

The corresponding closed curve on Td\mathbb{T}^d is obtained by reduction modulo Zd\mathbb{Z}^d. The uniform condition ∥m∥∞≤t\lVert m\rVert_\infty\leq t follows the exact integration conditions used in [10]. Since the Fourier basis is orthonormal and the integral on the right-hand side of (1) is the inner product with the constant 11, we have ∫Tde2πi⟨m,x⟩ dx=δm,0\int_{\mathbb{T}^d} e^{2\pi i\langle m,x\rangle}\,\mathrm{d}x=\delta_{m,0}.

Theorem 2.1. Suppose tt is a positive integer. If γ\gamma is a tt-design curve on Td\mathbb{T}^d of length LL, then

L≥((t+1)2d−1t(t+2))1/2(2)L\geq\left(\frac{(t+1)^{2d}-1}{t(t+2)}\right)^{1/2} \tag*{(2)}

There are tt-design curves on Td\mathbb{T}^d that attain this bound.

The tt-design curves attaining this bound appeared in [10, 11], and we refer to them as Korobov curves. Let dd and tt be positive integers, and define the Korobov vector

vd,t:=(1,t+1,…,(t+1)d−1)T∈Zd.(3)v_{d,t}:=\left(1,t+1,\ldots,(t+1)^{d-1}\right)^{\mathsf{T}}\in\mathbb{Z}^d. \tag*{(3)}

The associated curve γ~d,t ⁣:[0,1]→Rd\widetilde{\gamma}_{d,t}\colon[0,1]\to\mathbb{R}^d, γ~d,t(s)=s vd,t\widetilde{\gamma}_{d,t}(s)=s\,v_{d,t} parametrizes the line segment from the origin to vd,tv_{d,t}. Its length is

L=∥vd,t∥2=((t+1)2d−1t(t+2))1/2.(4)L=\lVert v_{d,t}\rVert_2=\left(\frac{(t+1)^{2d}-1}{t(t+2)}\right)^{1/2}. \tag*{(4)}

The arclength parametrization is γd,t ⁣:[0,L]→Rd\gamma_{d,t}\colon[0,L]\to\mathbb{R}^d, s↦s∥vd,t∥2vd,ts\mapsto\frac{s}{\lVert v_{d,t}\rVert_2}v_{d,t}, so that γd,t(L)−γd,t(0)=vd,t∈Zd\gamma_{d,t}(L)-\gamma_{d,t}(0)=v_{d,t}\in\mathbb{Z}^d. It was mentioned in [11] and verified in [10] that they form tt-design curves. Thus, (3) is sharp.

The proof of the lower bound (3) in Theorem 2.1 needs some preparation. Let γ ⁣:[0,L]→Rd\gamma\colon[0,L]\to\mathbb{R}^d be a rectifiable curve, parametrized by arclength. For every a∈Rda\in\mathbb{R}^d, define a function fa ⁣:[0,L]→Cf_a\colon[0,L]\to\mathbb{C},

fa(s):=e2πi⟨a,γ(s)⟩.f_a(s):=e^{2\pi i\langle a,\gamma(s)\rangle}.

We restrict aa to a suitable cube of side length tt centered at 00. For every positive integer tt, the set

[t]:={0,1,…,t}−t/2[t]:=\{0,1,\ldots,t\}-t/2

has cardinality t+1t+1, and the Cartesian product [t]d[t]^d has (t+1)d(t+1)^d elements. If tt is even, then [t]d⊆Zd[t]^d\subseteq\mathbb{Z}^d, and tt odd yields [t]d⊆(Z+12)d[t]^d\subseteq(\mathbb{Z}+\frac{1}{2})^d. The difference satisfies [t]d−[t]d⊆[−t,t]d∩Zd[t]^d-[t]^d\subseteq[-t,t]^d\cap\mathbb{Z}^d, no matter if tt is even or odd.

Invariance of [t]d[t]^d under coordinate sign changes and permutations implies that the covariance matrix of the product lattice [t]d[t]^d is a scalar multiple of the identity IdI_d. Further direct calculation for the discrete uniform distribution as in [7 Section 2.3.3] shows

1(t+1)d∑a∈[t]daaT=t(t+2)12Id.(5)\frac{1}{(t+1)^d}\sum_{a\in[t]^d} aa^{\mathsf T}=\frac{t(t+2)}{12}I_d. \tag*{(5)}

This isotropy property is key to the following lemma.

Lemma 2.2. Suppose tt is a positive integer and let γ:[0,L]→Rd\gamma: [0,L] \to\mathbb{R}^d be a rectifiable curve, parametrized by arclength. Then fa′∈L2([0,L],ds/L)f'_a \in L^2([0,L],ds/L), for all a∈[t]da\in[t]^d, and

1(t+1)d∑a∈[t]d1L∫0L∣fa′(s)∣2ds=4π2t(t+2)12.(6)\frac{1}{(t+1)^d}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L |f'_a(s)|^2ds=4\pi^2\frac{t(t+2)}{12}. \tag*{(6)}

The identity (7) in Lemma 2.2 holds independently of the design property but relies on the isotropy condition (6).

Proof of Lemma 2.2. Since γ\gamma is rectifiable and parametrized by arclength, it is 1-Lipschitz. Consequently, faf_a is Lipschitz as well. Therefore, both are differentiable almost everywhere with

∥γ′∥2=1,fa′=2πi⟨a,γ′⟩fa.(7)\|\gamma'\|_2=1,\qquad f'_a=2\pi i\langle a,\gamma'\rangle f_a. \tag*{(7)}

This also implies almost everywhere

∣fa′∣2=4π2∣⟨a,γ′⟩∣2=4π2γ′TaaTγ′.(8)|f'_a|^2=4\pi^2|\langle a,\gamma'\rangle|^2=4\pi^2\gamma'^{\mathsf T}aa^{\mathsf T}\gamma'. \tag*{(8)}

We have ∥fa∥L∞([0,L])=1\|f_a\|_{L^\infty([0,L])}=1 and ∥fa′∥L∞([0,L])≤2π∥a∥2\|f'_a\|_{L^\infty([0,L])}\leq2\pi\|a\|_2, so that, in particular, fa′∈L2([0,L],ds/L)f'_a\in L^2([0,L],ds/L).

The relations (9) and (6) lead to

1(t+1)d∑a∈[t]d1L∫0L∣fa′(s)∣2ds=4π2(t+1)dL∑a∈[t]d∫0Lγ′(s)TaaTγ′(s) ds=4π2L∫0Lγ′(s)Tt(t+2)12Idγ′(s) ds.\begin{aligned} \frac{1}{(t+1)^d}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L |f'_a(s)|^2ds =\frac{4\pi^2}{(t+1)^dL}\sum_{a\in[t]^d}\int_0^L \gamma'(s)^{\mathsf T}aa^{\mathsf T}\gamma'(s)\,ds \\ &=\frac{4\pi^2}{L}\int_0^L \gamma'(s)^{\mathsf T}\frac{t(t+2)}{12}I_d\gamma'(s)\,ds. \end{aligned}

Since γ′(s)Tγ′(s)=∥γ′(s)∥22=1\gamma'(s)^{\mathsf T}\gamma'(s)=\|\gamma'(s)\|_2^2=1 almost everywhere by (8), we arrive at (7). □

We establish orthonormality of the functions faf_a for a∈[t]da\in[t]^d as a consequence of the tt-design property.

Lemma 2.3. Let tt be a positive integer. If γ\gamma is a tt-design curve on Td\mathbb{T}^d of length LL, then the functions {fa:a∈[t]d}\{f_a:a\in[t]^d\} are orthonormal in L2([0,L],ds/L)L^2([0,L],ds/L).

Proof of Lemma 2.3. For a,b∈[t]da,b\in[t]^d, we have a−b∈Zda-b\in\mathbb{Z}^d and ∥a−b∥∞≤t\|a-b\|_\infty\leq t, so that the tt-design property (2) yields

1L∫0Lfa(s)‾fb(s) ds=1L∫0Le2πi⟨a−b,γ(s)⟩ ds=∫Tde2πi⟨a−b,x⟩ dx=δa,b.\frac{1}{L}\int_0^L \overline{f_a(s)}f_b(s)\,ds =\frac{1}{L}\int_0^L e^{2\pi i\langle a-b,\gamma(s)\rangle}\,ds =\int_{\mathbb{T}^d}e^{2\pi i\langle a-b,x\rangle}\,dx =\delta_{a,b}.

Thus, the functions {fa:a∈[t]d}\{f_a:a\in[t]^d\} are orthonormal in L2([0,L],ds/L)L^2([0,L],ds/L). □

The orthonormality is key to deriving a lower bound on the left-hand side of (7) for tt-design curves.

Proposition 2.4. Let tt be a positive integer. If γ\gamma is a tt-design curve on Td\mathbb{T}^d of length LL, then

1(t+1)d∑a∈[t]d1L∫0L∣fa′(s)∣2 ds≥(2πL)2(t+1)2d−112.(9)\frac{1}{(t+1)^d}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L |f'_a(s)|^2\,ds \ge\left(\frac{2\pi}{L}\right)^2\frac{(t+1)^{2d}-1}{12}. \tag*{(9)}

Theorem 2.1 follows from combining Lemma 2.2 with Proposition 2.4.

Proof of Theorem 2.1. We estimate the left-hand side of (6) by the lower bound in (9) and obtain

4π2t(t+2)12≥(2πL)2(t+1)2d−112.4\pi^2\frac{t(t+2)}{12}\ge\left(\frac{2\pi}{L}\right)^2\frac{(t+1)^{2d}-1}{12}.

Canceling 4π2/124\pi^2/12 and rearranging yields

L2≥(t+1)2d−1t(t+2).L^2\ge\frac{(t+1)^{2d}-1}{t(t+2)}.

Taking square roots leads to (2).

To complete the proof of Theorem 2.1, we verify Proposition 2.4.

Proof of Proposition 2.4. Let N=#[t]dN=\#[t]^d and v=γ(L)−γ(0)∈Zdv=\gamma(L)-\gamma(0)\in\mathbb{Z}^d. The parity of tt and v1+⋯+vdv_1+\cdots+v_d lead to three cases.

(i) We first treat the case when tt is even, so that [t]d⊆Zd[t]^d\subseteq\mathbb{Z}^d. For each a∈[t]da\in[t]^d, the function faf_a satisfies fa(L)=fa(0)f_a(L)=f_a(0), and we denote its Fourier coefficients by

ck(fa)=1L∫0Lfa(s)e−2πiks/L ds,k∈Z.c_k(f_a)=\frac{1}{L}\int_0^L f_a(s)e^{-2\pi iks/L}\,ds,\qquad k\in\mathbb{Z}.

Since faf_a is Lipschitz continuous, it is absolutely continuous and the Fourier coefficients of fa′f'_a are determined by the standard relation

ck(fa′)=2πikLck(fa),k∈Z.(10)c_k(f'_a)=\frac{2\pi ik}{L}c_k(f_a),\qquad k\in\mathbb{Z}. \tag*{(10)}

Parseval’s identity turns the L2L^2-norm of fa′f'_a into a weighted sequence norm of Fourier coefficients of faf_a,

1(t+1)d∑a∈[t]d1L∫0L∣fa′(s)∣2 ds=1(t+1)d∑a∈[t]d(2πL)2∑k∈Zk2∣ck(fa)∣2.\frac{1}{(t+1)^d}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L |f'_a(s)|^2\,ds =\frac{1}{(t+1)^d}\sum_{a\in[t]^d}\left(\frac{2\pi}{L}\right)^2\sum_{k\in\mathbb{Z}} k^2|c_k(f_a)|^2.

We define the total contribution of the kk-th Fourier coefficient by

pk:=∑a∈[t]d∣ck(fa)∣2.p_k:=\sum_{a\in[t]^d}|c_k(f_a)|^2.

To bound ∑k∈Zk2pk\sum_{k\in\mathbb{Z}} k^2p_k from below, we use that the functions {fa:a∈[t]d}\{f_a:a\in[t]^d\} are orthonormal by Lemma 2.3, so that Bessel’s inequality yields

pk=∑a∈[t]d∣ck(fa)∣2≤1.p_k=\sum_{a\in[t]^d}|c_k(f_a)|^2\le1.

Parseval’s identity for the Fourier basis implies

∑k∈Zpk=∑k∈Z∑a∈[t]d∣ck(fa)∣2=N.\sum_{k\in\mathbb{Z}}p_k=\sum_{k\in\mathbb{Z}}\sum_{a\in[t]^d}|c_k(f_a)|^2=N.

Since tt is even, NN is odd, and M:=(N−1)/2M := (N-1)/2 is an integer. We make use of

∑∣k∣≤M(1−pk)=∑∣k∣>Mpk,\sum_{|k|\le M}(1-p_k)=\sum_{|k|>M}p_k,

to estimate ∑k∈Zk2pk\sum_{k\in\mathbb{Z}} k^2p_k from below,

∑k∈Zk2pk−∑∣k∣≤Mk2=∑∣k∣>Mk2pk−∑∣k∣≤Mk2(1−pk)≥∑∣k∣>MM2pk−∑∣k∣≤MM2(1−pk)=0.\sum_{k\in\mathbb{Z}} k^2p_k-\sum_{|k|\le M}k^2 =\sum_{|k|>M}k^2p_k-\sum_{|k|\le M}k^2(1-p_k) \ge\sum_{|k|>M}M^2p_k-\sum_{|k|\le M}M^2(1-p_k)=0.

Consequently, ∑k∈Zk2pk≥∑∣k∣≤Mk2\sum_{k\in\mathbb{Z}} k^2p_k\ge\sum_{|k|\le M}k^2, which then implies the lower bound

1N∑a∈[t]d1L∫0L∣fa′(s)∣2 ds≥(2πL)21N∑∣k∣≤Mk2.(11)\frac{1}{N}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L |f_a'(s)|^2\,ds \ge\left(\frac{2\pi}{L}\right)^2\frac{1}{N}\sum_{|k|\le M}k^2. \tag*{(11)}

The sum evaluates to

1N∑∣k∣≤Mk2=2N∑k=1(N−1)/2k2=N2−112,\frac{1}{N}\sum_{|k|\le M}k^2 =\frac{2}{N}\sum_{k=1}^{(N-1)/2}k^2 =\frac{N^2-1}{12},

so that (12) with N=(t+1)dN=(t+1)^d leads to the claimed estimate (10) in Proposition 2.4.

(ii) Let tt be odd and suppose that v=γ(L)−γ(0)v=\gamma(L)-\gamma(0) is such that v1+⋯+vdv_1+\cdots+v_d is even. For a∈[t]d⊂(Z+12)da\in[t]^d\subset(\mathbb{Z}+\frac{1}{2})^d, this still implies ⟨a,v⟩∈Z\langle a,v\rangle\in\mathbb{Z}, as when tt is even. Hence,

fa(L)=e2πi⟨a,γ(L)⟩=e2πi⟨a,v+γ(0)⟩=fa(0).f_a(L)=e^{2\pi i\langle a,\gamma(L)\rangle} =e^{2\pi i\langle a,v+\gamma(0)\rangle} =f_a(0).

Thus, the functions faf_a still satisfy (11), and we can follow the case (i) using the standard Fourier coefficients. Since tt is odd, NN is even, and M:=N/2M:=N/2 is an integer. To keep sums of length NN, we switch from ∑∣k∣≤M\sum_{|k|\le M} to ∑k=−MM−1\sum_{k=-M}^{M-1}. Computations analogous to the case (i) lead to

∑k∈Zk2pk≥∑k=−MM−1k2.\sum_{k\in\mathbb{Z}}k^2p_k\ge\sum_{k=-M}^{M-1}k^2.

This yields the lower bound

1N∑a∈[t]d1L∫0L∣fa′(s)∣2 ds≥(2πL)21N∑k=−MM−1k2.\frac{1}{N}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L |f_a'(s)|^2\,ds \ge\left(\frac{2\pi}{L}\right)^2\frac{1}{N}\sum_{k=-M}^{M-1}k^2.

The sum can be computed exactly by

1N∑k=−MM−1k2=N2+212≥N2−112.\frac{1}{N}\sum_{k=-M}^{M-1}k^2 =\frac{N^2+2}{12}\ge\frac{N^2-1}{12}.

This gives a slightly stronger bound in this case.

(iii) For tt odd and v1+⋯+vdv_1+\cdots+v_d odd, we have ⟨a,v⟩∈Z+12\langle a,v\rangle\in\mathbb{Z}+\frac{1}{2} for all a∈[t]da\in[t]^d, hence, the functions faf_a satisfy

fa(L)=−fa(0).f_a(L)=-f_a(0).

Thus, the derivative relation (11) for Fourier coefficients no longer holds with the standard Fourier basis, and we must change to an anti-periodic basis. We use

ϕk(s)=e2πi(k+12)s/L,k∈Z.\phi_k(s)=e^{2\pi i(k+\frac{1}{2})s/L},\qquad k\in\mathbb{Z}.

This is an orthonormal basis for L2([0,L],ds/L)L^2([0,L],\mathrm{d}s/L), and the half-integer Fourier coefficients of faf_a are

ck+12(fa):=1L∫0Lfa(s)e−2πi(k+12)s/L ds,k∈Z.c_{k+\frac{1}{2}}(f_a) := \frac{1}{L}\int_0^L f_a(s)e^{-2\pi i(k+\frac{1}{2})s/L}\,\mathrm{d}s,\qquad k\in\mathbb{Z}.

Since fa(L)=−fa(0)f_a(L)=-f_a(0) and e−2πi(k+12)=−1e^{-2\pi i(k+\frac{1}{2})}=-1, the boundary term in integration by parts vanishes, and hence

ck+12(fa′)=2πi(k+12)Lck+12(fa).c_{k+\frac{1}{2}}(f_a')=\frac{2\pi i(k+\frac{1}{2})}{L}c_{k+\frac{1}{2}}(f_a).

Thus, we obtain

1(t+1)d∑a∈[t]d1L∫0L∣fa′(s)∣2 ds=1(t+1)d∑a∈[t]d(2πL)2∑k∈Z(k+12)2∣ck+12(fa)∣2.\frac{1}{(t+1)^d}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L|f_a'(s)|^2\,\mathrm{d}s=\frac{1}{(t+1)^d}\sum_{a\in[t]^d}\left(\frac{2\pi}{L}\right)^2\sum_{k\in\mathbb{Z}}\left(k+\frac{1}{2}\right)^2|c_{k+\frac{1}{2}}(f_a)|^2.

The total contribution of the k+12k+\frac{1}{2}-th Fourier coefficient is

pk+12:=∑a∈[t]d∣ck+12(fa)∣2,k∈Z.p_{k+\frac{1}{2}}:=\sum_{a\in[t]^d}|c_{k+\frac{1}{2}}(f_a)|^2,\qquad k\in\mathbb{Z}.

Bessel’s inequality and Parseval’s identity again give

pk+12=∑a∈[t]d∣ck+12(fa)∣2≤1,∑k∈Zpk+12=N.p_{k+\frac{1}{2}}=\sum_{a\in[t]^d}|c_{k+\frac{1}{2}}(f_a)|^2\leq1,\qquad\sum_{k\in\mathbb{Z}}p_{k+\frac{1}{2}}=N.

The same minimization argument as in case (i), now for half-integer frequencies and M=N/2M=N/2, yields

1N∑a∈[t]d1L∫0L∣fa′(s)∣2 ds≥(2πL)21N∑k=−MM−1(k+12)2=(2πL)2N2−112.\frac{1}{N}\sum_{a\in[t]^d}\frac{1}{L}\int_0^L|f_a'(s)|^2\,\mathrm{d}s\geq\left(\frac{2\pi}{L}\right)^2\frac{1}{N}\sum_{k=-M}^{M-1}\left(k+\frac{1}{2}\right)^2=\left(\frac{2\pi}{L}\right)^2\frac{N^2-1}{12}.

This concludes the proof of Proposition 2.4. □

Remark 2.5. Case (ii) in the proof of Proposition 2.4 yields a stronger bound under an additional parity assumption. If γ\gamma is a tt-design curve on Td\mathbb{T}^d of length LL, tt is odd, and v1+⋯+vdv_1+\cdots+v_d is even for v=γ(L)−γ(0)v=\gamma(L)-\gamma(0), then L≥((t+1)2d+2t(t+2))1/2L\geq\left(\frac{(t+1)^2d+2}{t(t+2)}\right)^{1/2}.

Total-degree tt-design curves

Definition 1.1 addressed the uniform frequency range ∥m∥∞≤t\lVert m\rVert_\infty\leq t. Design curves are considered in [11] for trigonometric polynomials of total degree at most tt, corresponding to the frequency condition ∥m∥1≤t\lVert m\rVert_1\leq t.

Definition 3.1. Let tt be a positive integer. A rectifiable curve γ:[0,L]→Rd\gamma:[0,L]\to\mathbb{R}^d with γ(L)−γ(0)∈Zd\gamma(L)-\gamma(0)\in\mathbb{Z}^d, parametrized by arclength, is called a total-degree tt-design curve on Td\mathbb{T}^d if

1L∫0Le2πi⟨m,γ(s)⟩ ds=∫Tde2πi⟨m,x⟩ dx,(12)\frac{1}{L}\int_0^L e^{2\pi i\langle m,\gamma(s)\rangle}\,\mathrm{d}s=\int_{\mathbb{T}^d}e^{2\pi i\langle m,x\rangle}\,\mathrm{d}x, \tag*{(12)}

for every m∈Zdm\in\mathbb{Z}^d satisfying ∥m∥1≤t\lVert m\rVert_1\leq t.

Every tt-design curve in the sense of Definition 1.1 is also a total-degree tt-design curve, because ∥m∥1≤t\lVert m\rVert_1 \le t implies ∥m∥∞≤t\lVert m\rVert_\infty\le t.

The curve γd,t\gamma_{d,t} induced by the Korobov vector (3) is a total-degree (t+1)(t+1)-design curve. The latter is seen as follows. For 0<∥m∥1≤t+10 < \lVert m\rVert_1 \le t+1, either ∥m∥∞≤t\lVert m\rVert_\infty\le t, in which case exactness follows from the Korobov curve being a tt-design, or there is mi=±(t+1)m_i=\pm(t+1), in which case ⟨m,vd,t⟩≠0\langle m,v_{d,t}\rangle\ne0.

Total-degree one

We proceed by first determining the minimum length for total-degree 11-design curves. For even degrees, we then derive a general lower bound and show that it is attained for t=2t=2 in every dimension and for every even tt in dimension two.

Lemma 3.2. The curve γ:[0,d]→Rd\gamma:[0,\sqrt{d}]\to\mathbb{R}^d, s↦sd(1,…,1)⊤s\mapsto\frac{s}{\sqrt{d}}(1,\ldots,1)^\top is a total-degree 11-design curve. Its length d\sqrt{d} is minimal among all total-degree 11-design curves on Td\mathbb{T}^d.

Proof. The design property follows from the observation that every nonzero m∈Zdm\in\mathbb{Z}^d with ∥m∥1≤1\lVert m\rVert_1\le1 satisfies ⟨m,(1,…,1)⊤⟩≠0\langle m,(1,\ldots,1)^\top\rangle\ne0.

To prove the lower bound on the length, we again follow the approach of [9], which takes a particularly simple form in this case. Let γ:[0,L]→Rd\gamma:[0,L]\to\mathbb{R}^d be a total-degree 11-design curve. For j=1,…,dj=1,\ldots,d, define fj:=fejf_j:=f_{e_j}, so that

fj(s)=e2πi⟨ej,γ(s)⟩=e2πiγj(s),0≤s≤L.f_j(s)=e^{2\pi i\langle e_j,\gamma(s)\rangle}=e^{2\pi i\gamma_j(s)},\qquad0\le s\le L.

According to ∣fj′∣2=4π2∣γj′∣2\lvert f_j'\rvert^2=4\pi^2\lvert\gamma_j'\rvert^2 and 1=∥γ′∥22=∑j=1d∣γj′∣21=\lVert\gamma'\rVert_2^2=\sum_{j=1}^d\lvert\gamma_j'\rvert^2 almost everywhere we obtain

1d∑j=1d1L∫0L∣fj′(s)∣2 ds=4π2d.\frac{1}{d}\sum_{j=1}^d\frac{1}{L}\int_0^L\lvert f_j'(s)\rvert^2\,\mathrm{d}s=\frac{4\pi^2}{d}.

Since ∣fj(s)∣=1\lvert f_j(s)\rvert=1, the functions fjf_j have unit norm, 1L∫0L∣fj(s)∣2 ds=1\frac{1}{L}\int_0^L\lvert f_j(s)\rvert^2\,\mathrm{d}s=1. The design property does not imply orthonormality of f1,…,fdf_1,\ldots,f_d, but they are orthogonal to the constant function, 1L∫0Lfj(s) ds=0\frac{1}{L}\int_0^L f_j(s)\,\mathrm{d}s=0. A Fourier coefficient argument using (10) leads to

1L∫0L∣fj′(s)∣2 ds≥(2πL)2.\frac{1}{L}\int_0^L\lvert f_j'(s)\rvert^2\,\mathrm{d}s\ge\left(\frac{2\pi}{L}\right)^2.

Thus, we obtain 4π2d≥(2πL)2\frac{4\pi^2}{d}\ge\left(\frac{2\pi}{L}\right)^2, which yields L≥dL\ge\sqrt{d}. □

Even total-degree

Assume t=2nt=2n is even. Set

Ad,n:={a∈Zd:∥a∥1≤n},Nd,n:=#Ad,n.A_{d,n}:=\{a\in\mathbb{Z}^d:\lVert a\rVert_1\le n\},\qquad N_{d,n}:=\#A_{d,n}.

This yields Ad,n−Ad,n⊆{m∈Zd:∥m∥1≤t}A_{d,n}-A_{d,n}\subseteq\{m\in\mathbb{Z}^d:\lVert m\rVert_1\le t\}, and the quantity

σd,n:=1Nd,n∑a∈Ad,naj2,j=1,…,d,\sigma_{d,n}:=\frac{1}{N_{d,n}}\sum_{a\in A_{d,n}}a_j^2,\qquad j=1,\ldots,d,

is independent of jj. Invariance of Ad,nA_{d,n} under coordinate sign changes and permutations implies

1Nd,n∑a∈Ad,naa⊤=σd,nId,\frac{1}{N_{d,n}}\sum_{a\in A_{d,n}}aa^\top=\sigma_{d,n}I_d,

that leads to a lower bound on the length of total-degree tt-design curves.

Theorem 3.3. Suppose t=2nt = 2n is a positive even integer. If γ\gamma is a total-degree tt-design curve on Td\mathbb{T}^d of length LL, then

L≥(Nd,n2−112σd,n)1/2.(13)L \ge\left(\frac{N_{d,n}^2 - 1}{12\sigma_{d,n}}\right)^{1/2}. \tag*{(13)}

Proof. We repeat the argument of Section 2 with the functions fa(s)=e2πi⟨a,γ(s)⟩f_a(s) = e^{2\pi i\langle a,\gamma(s)\rangle} but the centered cube [t]d[t]^d replaced by Ad,nA_{d,n}. The isotropy condition allows the computation of

1Nd,n∑a∈Ad,n1L∫0L∣fa′(s)∣2 ds=4π2σd,n,\frac{1}{N_{d,n}}\sum_{a\in A_{d,n}}\frac{1}{L}\int_0^L |f'_a(s)|^2\,ds = 4\pi^2\sigma_{d,n},

and the total-degree tt-design property (13) ensures that the functions {fa:a∈Ad,n}\{f_a : a\in A_{d,n}\} are orthonormal in L2([0,L],ds/L)L^2([0,L],ds/L).

The set Ad,nA_{d,n} consists of the origin and pairs {a,−a}\{a,-a\}, so that Nd,nN_{d,n} is odd, and we may follow case (i) in the proof of Proposition 2.4. It leads to

1Nd,n∑a∈Ad,n1L∫0L∣fa′(s)∣2 ds≥(2πL)2Nd,n2−112.\frac{1}{N_{d,n}}\sum_{a\in A_{d,n}}\frac{1}{L}\int_0^L |f'_a(s)|^2\,ds \ge\left(\frac{2\pi}{L}\right)^2\frac{N_{d,n}^2-1}{12}.

Thus, we arrive at

4π2σd,n≥(2πL)2Nd,n2−112,4\pi^2\sigma_{d,n} \ge\left(\frac{2\pi}{L}\right)^2\frac{N_{d,n}^2-1}{12},

which leads to the estimate (14).

The quantities Nd,n=#Ad,nN_{d,n} = \#A_{d,n} are known as Delannoy numbers [12],

Nd,n=∑j=0min⁡(d,n)2j(dj)(nj).N_{d,n} = \sum_{j=0}^{\min(d,n)} 2^j \binom{d}{j}\binom{n}{j}.

This is a polynomial of degree dd in nn with leading coefficient 2d/d!2^d/d!. Hence, for large nn, the asymptotics Nd,n∼2dnd/d!N_{d,n}\sim2^d n^d/d! hold.

Lemma 3.4. For every d,n≥1d,n\ge1, we have σd,n=12(Nd+2,nNd,n−1)\sigma_{d,n} = \frac{1}{2}\left(\frac{N_{d+2,n}}{N_{d,n}}-1\right), and n→∞n\to\infty yields the leading term σd,n∼2(d+1)(d+2)n2\sigma_{d,n}\sim\frac{2}{(d+1)(d+2)}n^2, so that

(Nd,n2−112σd,n)1/2∼2dd!((d+1)(d+2)24)1/2nd−1.(14)\left(\frac{N_{d,n}^2-1}{12\sigma_{d,n}}\right)^{1/2}\sim\frac{2^d}{d!}\left(\frac{(d+1)(d+2)}{24}\right)^{1/2}n^{d-1}. \tag*{(14)}

Proof. Fixing the first coordinate at a1=ra_1 = r in σd,n=1Nd,n∑a∈Ad,na12\sigma_{d,n} = \frac{1}{N_{d,n}}\sum_{a\in A_{d,n}}a_1^2 leaves exactly Nd−1,n−∣r∣N_{d-1,n-|r|} choices for the remaining coordinates, so that

σd,n=1Nd,n∑r=−nnr2Nd−1,n−∣r∣=2Nd,n∑r=1nr2Nd−1,n−r,\sigma_{d,n} = \frac{1}{N_{d,n}}\sum_{r=-n}^{n}r^2N_{d-1,n-|r|} = \frac{2}{N_{d,n}}\sum_{r=1}^{n}r^2N_{d-1,n-r},

where we use the convention N0,p=Nq,0=1N_{0,p}=N_{q,0}=1 for p,q≥0p,q\ge0.

For r≥1r\ge1, the number of triplets (a1,a2,a3)∈Z3(a_1,a_2,a_3)\in\mathbb{Z}^3 satisfying ∣a1∣+∣a2∣+∣a3∣=r|a_1|+|a_2|+|a_3|=r can be counted by adding the numbers for exactly one, two, or three nonzero coordinates

and is 6+12(r−1)+8(r−12)=4r2+26 + 12(r - 1) + 8\binom{r-1}{2} = 4r^2 + 2. According to the sum rr of the absolute values of their last three coordinates, we obtain

Nd+2,n=Nd−1,n+∑r=1n(4r2+2)Nd−1,n−r.N_{d+2,n} = N_{d-1,n} + \sum_{r=1}^{n}(4r^2 + 2)N_{d-1,n-r}.

Similarly, we determine

Nd,n=Nd−1,n+2∑r=1nNd−1,n−r.N_{d,n} = N_{d-1,n} + 2\sum_{r=1}^{n}N_{d-1,n-r}.

Subtracting one from the other yields

Nd+2,n−Nd,n=4∑r=1nr2Nd−1,n−r=2Nd,nσd,n,N_{d+2,n} - N_{d,n} = 4\sum_{r=1}^{n}r^2N_{d-1,n-r} = 2N_{d,n}\sigma_{d,n},

so that σd,n=12(Nd+2,nNd,n−1)\sigma_{d,n} = \frac{1}{2}\left(\frac{N_{d+2,n}}{N_{d,n}} - 1\right). The observation Nd,n∼2dnd/d!N_{d,n} \sim2^d n^d/d! implies σd,n∼2(d+1)(d+2)n2\sigma_{d,n} \sim\frac{2}{(d+1)(d+2)}n^2 and combining both gives (15).

Minimal length for d=2d = 2 or t=2t = 2

We specify Theorem 3.3 for two distinct cases.

Corollary 3.5. Suppose n≥1n \ge1 is an integer.

(d=2)(d = 2) If γ\gamma is a total-degree 2n2n-design curve on T2\mathbb{T}^2 of length LL, then

L≥2n2+2n+1.(15)L \ge\sqrt{2n^2 + 2n + 1}. \tag*{(15)}

(t=2)(t = 2) If γ\gamma is a total-degree 22-design curve on Td\mathbb{T}^d of length LL, then

L≥16d(d+1)(2d+1).(16)L \ge\sqrt{\frac{1}{6}d(d+1)(2d+1)}. \tag*{(16)}

We will see that there are curves that attain these bounds.

Proof. (a) For d=2d = 2, we have N2,n=2n2+2n+1N_{2,n} = 2n^2 + 2n + 1 and N4,n−N2,n=23n(n+1)(n2+n+1)N_{4,n} - N_{2,n} = \frac{2}{3}n(n+1)(n^2 + n + 1). We can calculate σ2,n\sigma_{2,n} by

σ2,n=N4,n−N2,n2N2,n=n(n+1)(n2+n+1)3(2n2+2n+1).\sigma_{2,n} = \frac{N_{4,n} - N_{2,n}}{2N_{2,n}} = \frac{n(n+1)(n^2+n+1)}{3(2n^2+2n+1)}.

Since N2,n2−1=4n(n+1)(n2+n+1)N_{2,n}^2 - 1 = 4n(n+1)(n^2+n+1), substitution into (14) gives L2≥N2,n2−112σ2,n=2n2+2n+1L^2 \ge\frac{N_{2,n}^2-1}{12\sigma_{2,n}} = 2n^2 + 2n + 1.

(b) For t=2t = 2, the Delannoy numbers simplify to Nd,1=2d+1N_{d,1} = 2d + 1, and we obtain σd,1=2/(2d+1)\sigma_{d,1} = 2/(2d+1), so that Nd,12−112σd,1=16d(d+1)(2d+1)\frac{N_{d,1}^2-1}{12\sigma_{d,1}} = \frac{1}{6}d(d+1)(2d+1).

Example 3.6. Let t=2t = 2. The curve associated to the vector

wd=(1,2,…,d)Tw_d = (1,2,\ldots,d)^{\mathsf{T}}

is a total-degree 22-design curve on Td\mathbb{T}^d. Indeed, if mm has one nonzero entry, then ⟨m,wd⟩≠0\langle m,w_d\rangle\ne0. If it has two nonzero entries, both have modulus one, and the corresponding entries of wdw_d are positive and distinct, giving ⟨m,wd⟩≠0\langle m,w_d\rangle\ne0. Its length is Ld=16d(d+1)(2d+1)L_d = \sqrt{\frac{1}{6}d(d+1)(2d+1)}, which is minimal among all total-degree 22-design curves on Td\mathbb{T}^d by Corollary 3.5.

The following construction provides total-degree 2n2n-design curves in every dimension and attains the bound (16) of Corollary 3.5 when d=2d=2. Let tt be a positive even integer, so that t=2nt=2n. The curve associated to the vector

wd,n=(nd−1,nd−2(n+1),…,(n+1)d−1)Tw_{d,n}=\left(n^{d-1},n^{d-2}(n+1),\ldots,(n+1)^{d-1}\right)^{\mathsf{T}}

is βd,n(s)=s∥wd,n∥2wd,n\beta_{d,n}(s)=\frac{s}{\lVert w_{d,n}\rVert_2}w_{d,n}, for 0≤s≤∥wd,n∥2=Ld,n0\leq s\leq\lVert w_{d,n}\rVert_2=L_{d,n}.

Lemma 3.7. The curve βd,n\beta_{d,n} is a total-degree 2n2n-design curve on Td\mathbb{T}^d of length

Ld,n=((n+1)2d−n2d2n+1)1/2.L_{d,n}=\left(\frac{(n+1)^{2d}-n^{2d}}{2n+1}\right)^{1/2}.

For large nn, the asymptotic is Ld,n∼d nd−1L_{d,n}\sim\sqrt{d}\,n^{d-1}.

Proof. Suppose ⟨m,wd,n⟩=0\langle m,w_{d,n}\rangle=0 for some nonzero m∈Zdm\in\mathbb{Z}^d. Since every entry of wd,nw_{d,n} is positive, mm has at least two nonzero entries. Let r<sr<s be their first and last indices. Dividing the relation by nd−s(n+1)r−1n^{d-s}(n+1)^{r-1} gives

∑j=rsmjns−j(n+1)j−r=0.\sum_{j=r}^{s}m_j n^{s-j}(n+1)^{j-r}=0.

Every term with j>rj>r contains a factor n+1n+1. Since the sum is zero, n+1n+1 must also divide the first term mrns−rm_r n^{s-r}. As nn and n+1n+1 are coprime, this gives (n+1)∣mr(n+1)\mid m_r. Similarly, every term with j<sj<s contains a factor nn, so nn divides the last term ms(n+1)s−rm_s(n+1)^{s-r}, and hence n∣msn\mid m_s. Both entries are nonzero, so

∥m∥1≥∣mr∣+∣ms∣≥2n+1.\lVert m\rVert_1\geq|m_r|+|m_s|\geq2n+1.

Thus, whenever 0<∥m∥1≤2n0<\lVert m\rVert_1\leq2n, the integer ⟨m,wd,n⟩\langle m,w_{d,n}\rangle is nonzero. It follows that βd,n\beta_{d,n} is a total-degree 2n2n-design curve on Td\mathbb{T}^d.

To compute the length Ld,n=∥wd,n∥2L_{d,n}=\lVert w_{d,n}\rVert_2, we recall the standard identity

(a−b)∑j=0d−1ad−1−jbj=ad−bd,a≠b.(a-b)\sum_{j=0}^{d-1}a^{d-1-j}b^j=a^d-b^d,\qquad a\ne b.

For a=(n+1)2a=(n+1)^2 and b=n2b=n^2, this yields

∥wd,n∥22=∑j=0d−1(n+1)2(d−1−j)n2j=(n+1)2d−n2d2n+1.\lVert w_{d,n}\rVert_2^2=\sum_{j=0}^{d-1}(n+1)^{2(d-1-j)}n^{2j}=\frac{(n+1)^{2d}-n^{2d}}{2n+1}.

Example 3.8. For d=2d=2 and nn a positive integer, β2,n\beta_{2,n} is a total-degree 2n2n-design curve on T2\mathbb{T}^2 with w2,n=(n,n+1)Tw_{2,n}=(n,n+1)^{\mathsf{T}} and length L2,n=2n2+2n+1L_{2,n}=\sqrt{2n^2+2n+1}. The length is minimal among all total-degree 2n2n-design curves on T2\mathbb{T}^2 by Corollary 3.5.

Examples 3.6 and 3.8 show that the bounds in Corollary 3.5 are sharp.

Hyperbolic cross design curves

Hyperbolic crosses provide a standard alternative to tensor-product frequency sets [3, 8]. We denote the dd-dimensional hyperbolic cross by

Hd(t):={m∈Zd:∏j=1dmax⁡(1,∣mj∣)≤t}.H_d(t):=\left\{m\in\mathbb{Z}^d:\prod_{j=1}^{d}\max(1,|m_j|)\leq t\right\}.

and define design curves for frequencies in Hd(t)H_d(t).

Definition 4.1. Let tt be a positive integer. A rectifiable curve γ:[0,L]→Rd\gamma: [0,L] \to\mathbb{R}^{d} with γ(L)−γ(0)∈Zd\gamma(L)-\gamma(0) \in\mathbb{Z}^{d}, parametrized by arclength, is called a hyperbolic cross tt-design curve on Td\mathbb{T}^{d} if

1L∫0Le2πi⟨m,γ(s)⟩ ds=∫Tde2πi⟨m,x⟩ dx,\frac{1}{L}\int_{0}^{L}e^{2\pi i\langle m,\gamma(s)\rangle}\,\mathrm{d}s=\int_{\mathbb{T}^{d}}e^{2\pi i\langle m,x\rangle}\,\mathrm{d}x,

for every m∈Hd(t)m \in H_d(t).

Sharp bounds for d=2d=2

For the hyperbolic cross in dimension two, we obtain a sharp lower bound on the length of such design curves.

Theorem 4.2. Let n≥1n \ge1 be an integer and d=2d=2.

(a) If γ\gamma is a hyperbolic cross n2n^2-design curve on T2\mathbb{T}^{2}, then its length LL satisfies

L≥2n2+2n+1.L \ge\sqrt{2n^2+2n+1}.

(b) The curve β2,n\beta_{2,n} in Example 3.8 is a hyperbolic cross n2n^2-design and attains this bound.

It is a feature of the hyperbolic cross that the design strength can be n2n^2 although the length is only of order nn.

Proof. (a) Suppose first that n≥2n \ge2. Consider m∈Z2m \in\mathbb{Z}^{2} such that ∥m∥1≤2n\lVert m\rVert_{1} \le2n. If both coordinates are nonzero, then ∣m1m2∣≤(∥m∥1/2)2≤n2|m_1m_2| \le(\lVert m\rVert_1/2)^2 \le n^2. Otherwise we have

max⁡{1,∣m1∣}max⁡{1,∣m2∣}≤2n≤n2.\max\{1,|m_1|\}\max\{1,|m_2|\}\le2n\le n^2.

Consequently,

{m∈Z2:∥m∥1≤2n}⊆H2(n2),\{m\in\mathbb{Z}^{2}:\lVert m\rVert_1\le2n\}\subseteq H_2(n^2),

and every hyperbolic cross n2n^2-design curve on T2\mathbb{T}^{2} is also a total-degree 2n2n-design curve. Corollary 3.5 therefore yields L≥2n2+2n+1L \ge\sqrt{2n^2+2n+1}.

For n=1n=1, the hyperbolic cross is H2(1)={−1,0,1}2H_2(1)=\{-1,0,1\}^{2}, so γ\gamma is a 11-design curve in the sense of Definition 1.1. Theorem 2.1 gives L≥(24−13)1/2=5L\ge\left(\frac{2^4-1}{3}\right)^{1/2}=\sqrt{5}, which is the claimed bound for n=1n=1.

(b) The curve in Example 3.8 is β2,n(s)=sL2,nw2,n\beta_{2,n}(s)=\frac{s}{L_{2,n}}w_{2,n} with w2,n=(n,n+1)Tw_{2,n}=(n,n+1)^{\mathsf{T}} and its length L2,n=2n2+2n+1L_{2,n}=\sqrt{2n^2+2n+1} matches the bound in Part (a). To verify the design property, it is sufficient to check that

⟨m,w2,n⟩≠0for every m∈H2(n2)∖{0}.\langle m,w_{2,n}\rangle\ne0\qquad\text{for every }m\in H_2(n^2)\setminus\{0\}.

Indeed, if m≠0m\ne0 and nm1+(n+1)m2=0nm_1+(n+1)m_2=0, then (n+1)m2=−nm1(n+1)m_2=-nm_1. Since nn and n+1n+1 are coprime, n+1n+1 must divide m1m_1, so that there is k∈Z∖{0}k\in\mathbb{Z}\setminus\{0\} such that k(n+1)=m1k(n+1)=m_1. Thus, we obtain

0=nk(n+1)+(n+1)m2=(n+1)(nk+m2),0=nk(n+1)+(n+1)m_2=(n+1)(nk+m_2),

so that m2=−nkm_2=-nk. This also implies m=k(n+1,−n)Tm=k(n+1,-n)^{\mathsf{T}}. Consequently,

max⁡{1,∣m1∣}max⁡{1,∣m2∣}=k2n(n+1)≥n(n+1)>n2,\max\{1,|m_1|\}\max\{1,|m_2|\}=k^2n(n+1)\ge n(n+1)>n^2,

which leads to m∉H2(n2)m\notin H_2(n^2).

Thus, for every nonzero m∈H2(n2)m\in H_2(n^2), the integer ⟨m,w2,n⟩\langle m,w_{2,n}\rangle is nonzero, and this implies exact integration of e2πi⟨m,x⟩e^{2\pi i\langle m,x\rangle}. For m=0m=0, the normalized integral equals 1. Hence β2,n\beta_{2,n} is a hyperbolic cross n2n^2-design curve. □

Sharp asymptotic bounds for d≥3d \ge3

We determine a sharp asymptotic bound on the length when the dimension is bigger than two.

Theorem 4.3. Let d≥3d \ge3 be an integer.

(a) If γ\gamma is a hyperbolic cross ndn^d-design curve on Td\mathbb{T}^d, for some positive integer nn, then its length LL satisfies

L≥nd−1.L \ge n^{d-1}.

(b) There is a constant Cd>0C_d > 0 such that, for all positive integers nn, there exists a simple, hyperbolic cross ndn^d-design curve on Td\mathbb{T}^d of length L≤Cdnd−1L \le C_d n^{d-1}.

Before we can prove this theorem, we need to verify some estimates regarding the hyperbolic cross.

Lemma 4.4. Let n≥1n \ge1 be an integer.

(i) For all integers d≥2d \ge2,

∑m∈Hd(nd)∖{0}1∥m∥∞≤(3d)dd−1nd−1.(17)\sum_{m \in H_d(n^d)\setminus\{0\}} \frac{1}{\lVert m\rVert_\infty} \le\frac{(3d)^d}{d-1}n^{d-1}. \tag*{(17)}

(ii) For all integers d≥3d \ge3, there is a constant cd>0c_d > 0 such that

#Hd(nd)≤cdn2(d−1).(18)\#H_d(n^d) \le c_d n^{2(d-1)}. \tag*{(18)}

Proof of Lemma 4.4. (i) Let N=ndN = n^d. For m∈Hd(N)∖{0}m \in H_d(N) \setminus\{0\}, the substitution aj=max⁡(1,∣mj∣)a_j = \max(1, |m_j|) has at most 3d3^d preimages and satisfies ∥a∥∞=∥m∥∞\lVert a\rVert_\infty= \lVert m\rVert_\infty, so that we derive

∑m∈Hd(N)∖{0}1∥m∥∞≤3d∑a1,…,ad≥1a1⋯ad≤N1∥a∥∞.\sum_{\substack{m \in H_d(N)\setminus\{0\}}} \frac{1}{\lVert m\rVert_\infty} \le3^d \sum_{\substack{a_1,\ldots,a_d\ge1\\a_1\cdots a_d\le N}} \frac{1}{\lVert a\rVert_\infty}.

For every summation index aa, we define the boxes Qa:=∏j=1d(aj−1,aj]Q_a := \prod_{j=1}^d (a_j-1,a_j], so that x∈Qax \in Q_a satisfies

x1⋯xd≤N,1∥x∥∞≥1∥a∥∞.x_1\cdots x_d \le N,\qquad\frac{1}{\lVert x\rVert_\infty} \ge\frac{1}{\lVert a\rVert_\infty}.

Since these boxes are disjoint and have volume one,

∑a1,…,ad≥1a1⋯ad≤N1∥a∥∞≤∫x∈(0,∞)dx1⋯xd≤Ndx∥x∥∞.\sum_{\substack{a_1,\ldots,a_d\ge1\\a_1\cdots a_d\le N}} \frac{1}{\lVert a\rVert_\infty} \le\int_{\substack{x\in(0,\infty)^d\\x_1\cdots x_d\le N}} \frac{dx}{\lVert x\rVert_\infty}.

Except on a set of measure zero, every xx has a unique largest coordinate. Thus we can split the integration domain into dd regions, according to which coordinate is largest. By symmetry, we may take xd=rx_d = r as the largest coordinate and write xj=ryjx_j = r y_j, with 0<yj<10 < y_j < 1, for j<dj < d. According to dx=rd−1dydrdx = r^{d-1}dydr and xd=rx_d = r, we obtain

∫x∈(0,∞)dx1⋯xd≤Ndx∥x∥∞=d∫x∈(0,∞)dx1⋯xd≤Nx1,…,xd−1≤xddxxd=d∫(0,1)d−1∫0(N/(y1⋯yd−1))1/drd−2 dr dy=dd−1N(d−1)/d∫(0,1)d−1(y1⋯yd−1)−(d−1)/d dy=dd−1N(d−1)/d(∫01s−(d−1)/d ds)d−1\begin{aligned} \int_{\substack{x\in(0,\infty)^d\\x_1\cdots x_d\le N}}\frac{dx}{\|x\|_\infty} =d\int_{\substack{x\in(0,\infty)^d\\x_1\cdots x_d\le N\\x_1,\ldots,x_{d-1}\le x_d}}\frac{dx}{x_d} \\ &=d\int_{(0,1)^{d-1}}\int_0^{(N/(y_1\cdots y_{d-1}))^{1/d}}r^{d-2}\,dr\,dy \\ &=\frac{d}{d-1}N^{(d-1)/d}\int_{(0,1)^{d-1}}(y_1\cdots y_{d-1})^{-(d-1)/d}\,dy \\ &=\frac{d}{d-1}N^{(d-1)/d}\left(\int_0^1s^{-(d-1)/d}\,ds\right)^{d-1} \end{aligned}

Since ∫01s−(d−1)/d ds=[d s1/d]01=d\int_0^1s^{-(d-1)/d}\,ds=\left[d\,s^{1/d}\right]_0^1=d, we derive

∫x∈(0,∞)dx1⋯xd≤Ndx∥x∥∞=ddd−1N(d−1)/d.\int_{\substack{x\in(0,\infty)^d\\x_1\cdots x_d\le N}}\frac{dx}{\|x\|_\infty} =\frac{d^d}{d-1}N^{(d-1)/d}.

Multiplication by 3d3^d yields (17).

(ii) All implicit constants1 below depend only on dd. According to [3], we note that

#Hd(N)≲dN(1+log⁡N)d−1\#H_d(N)\lesssim_d N(1+\log N)^{d-1}

provided that N≥1N\ge1 and d≥3d\ge3. Since (1+log⁡N)d−1≲dN(d−2)/d(1+\log N)^{d-1}\lesssim_d N^{(d-2)/d}, we obtain

#Hd(N)≲dN1+(d−2)/d=N2(d−1)/d.\#H_d(N)\lesssim_d N^{1+(d-2)/d}=N^{2(d-1)/d}.

Proof of Theorem 4.3. (a) We observe that [−n,n]d∩Zd⊆Hd(nd)[-n,n]^d\cap\mathbb{Z}^d\subseteq H_d(n^d), so that Theorem 2.1 yields

L2≥(n+1)2d−1n(n+2)=∑j=0d−1(n+1)2j≥(n+1)2(d−1).L^2\ge\frac{(n+1)^{2d}-1}{n(n+2)}=\sum_{j=0}^{d-1}(n+1)^{2j}\ge(n+1)^{2(d-1)}.

This implies L≥(n+1)d−1≥nd−1L\ge(n+1)^{d-1}\ge n^{d-1}.

(b) We aim to find v∈Zdv\in\mathbb{Z}^d such that ∥v∥2≤Cdnd−1\|v\|_2\le C_dn^{d-1} and ⟨v,m⟩≠0\langle v,m\rangle\ne0, for all m∈Hd(nd)∖{0}m\in H_d(n^d)\setminus\{0\}. Then the straight line s↦s∥v∥2vs\mapsto\frac{s}{\|v\|_2}v constitutes the hyperbolic cross ndn^d-design curve.

We say that m∈Zd∖{0}m\in\mathbb{Z}^d\setminus\{0\} is primitive if its coordinates have no common divisor greater than 1, meaning gcd⁡(∣m1∣,…,∣md∣)=1\gcd(|m_1|,\ldots,|m_d|)=1. Let Pd(nd)P_d(n^d) be the primitive vectors in Hd(nd)∖{0}H_d(n^d)\setminus\{0\}. Every forbidden hyperplane is represented by a vector in Pd(nd)P_d(n^d), so we may define, for every integer k≥1k\ge1,

Bk,m:={v∈[−k,k]d∩Zd:⟨v,m⟩=0},m∈Pd(nd).B_{k,m}:=\{v\in[-k,k]^d\cap\mathbb{Z}^d:\langle v,m\rangle=0\},\qquad m\in P_d(n^d).

The complete forbidden set is Bk:=⋃m∈Pd(nd)Bk,mB_k:=\bigcup_{m\in P_d(n^d)}B_{k,m}.

The hyperplane-counting bound [6 Equation (14)] is

#Bk,m≤(2k∥m∥∞+1)(2k+1)d−2.(19)\#B_{k,m}\le\left(\frac{2k}{\|m\|_\infty}+1\right)(2k+1)^{d-2}. \tag*{(19)}

The union bound #Bk≤∑m∈Pd(nd)#Bk,m\#\mathcal{B}_k \le\sum_{m\in\mathcal{P}_d(n^d)} \#\mathcal{B}_{k,m} leads to

#Bk≤∑m∈Pd(nd)(2k∥m∥∞+1)(2k+1)d−2=(2k+1)d−2(#Pd(nd)+∑m∈Pd(nd)2k∥m∥∞).\begin{aligned} \#\mathcal{B}_k \le\sum_{m\in\mathcal{P}_d(n^d)} \left(\frac{2k}{\lVert m\rVert_\infty}+1\right)(2k+1)^{d-2} \\ &= (2k+1)^{d-2}\left(\#\mathcal{P}_d(n^d)+\sum_{m\in\mathcal{P}_d(n^d)}\frac{2k}{\lVert m\rVert_\infty}\right). \end{aligned}

The cardinality of the integer cube [−k,k]d∩Zd[-k,k]^d\cap\mathbb{Z}^d is (2k+1)d(2k+1)^d, so that the proportion of forbidden vectors in the cube is

#Bk(2k+1)d≤#Pd(nd)+2k∑m∈Pd(nd)1∥m∥∞(2k+1)2.\frac{\#\mathcal{B}_k}{(2k+1)^d}\le\frac{\#\mathcal{P}_d(n^d)+2k\sum_{m\in\mathcal{P}_d(n^d)}\frac{1}{\lVert m\rVert_\infty}}{(2k+1)^2}.

Lemma 4.4 with Pd(nd)⊆Hd(nd)\mathcal{P}_d(n^d)\subseteq\mathcal{H}_d(n^d) implies the estimates

#Bk(2k+1)d≲dn2(d−1)+2knd−1(2k+1)2≤(nd−1k)2+nd−1k.\begin{aligned} \frac{\#\mathcal{B}_k}{(2k+1)^d}\lesssim_d\frac{n^{2(d-1)}+2kn^{d-1}}{(2k+1)^2} \\ &\le\left(\frac{n^{d-1}}{k}\right)^2+\frac{n^{d-1}}{k}. \end{aligned}

Thus, there is a constant Ad>0A_d>0 such that this proportion is smaller than one for k=⌈Adnd−1⌉k=\lceil A_dn^{d-1}\rceil. Hence, there is a vector w∈[−k,k]d∩Zdw\in[-k,k]^d\cap\mathbb{Z}^d with w∉Bkw\notin\mathcal{B}_k. This implies that ⟨w,m⟩≠0\langle w,m\rangle\ne0 for all m∈Pd(nd)m\in\mathcal{P}_d(n^d) and therefore also for all m∈Hd(nd)∖{0}m\in\mathcal{H}_d(n^d)\setminus\{0\}. Its norm satisfies

∥w∥2≤d∥w∥∞≤dk≤d(Ad+1)nd−1.\lVert w\rVert_2\le\sqrt{d}\lVert w\rVert_\infty\le\sqrt{d}k\le\sqrt{d}(A_d+1)n^{d-1}.

We may put Cd=d(Ad+1)C_d=\sqrt{d}(A_d+1). To obtain a simple curve, we use

v:=w/gcd⁡(∣w1∣,…,∣wd∣).v:=w/\gcd(|w_1|,\ldots,|w_d|).

The length bound L:=∥v∥2≤Cdnd−1L:=\lVert v\rVert_2\le C_dn^{d-1} still holds and so do the nonorthogonality conditions. Thus, the curve γ:[0,L]→Td\gamma:[0,L]\to\mathbb{T}^d, s↦s∥v∥2vs\mapsto\frac{s}{\lVert v\rVert_2}v is simple and a hyperbolic ndn^d-design curve on Td\mathbb{T}^d.

Sparse frequency sets

For integers d≥2d\ge2, 1≤r≤d1\le r\le d, and t≥1t\ge1, define the sparse frequency set

Ωd,r(t):={m∈Zd:∥m∥∞≤t, #supp⁡(m)≤r},\Omega_{d,r}(t):=\{m\in\mathbb{Z}^d:\lVert m\rVert_\infty\le t,\ \#\operatorname{supp}(m)\le r\},

where supp⁡(m):={j:mj≠0}\operatorname{supp}(m):=\{j:m_j\ne0\}.

Definition 5.1. For integers d≥2d\ge2, 1≤r≤d1\le r\le d, and t≥1t\ge1, a rectifiable curve γ:[0,L]→Rd\gamma:[0,L]\to\mathbb{R}^d with γ(L)−γ(0)∈Zd\gamma(L)-\gamma(0)\in\mathbb{Z}^d, parametrized by arclength, is called an rr-sparse tt-design curve on Td\mathbb{T}^d if

1L∫0Le2πi⟨m,γ(s)⟩ ds=∫Tde2πi⟨m,x⟩ dx,∀m∈Ωd,r(t).\frac{1}{L}\int_0^L e^{2\pi i\langle m,\gamma(s)\rangle}\,ds=\int_{\mathbb{T}^d}e^{2\pi i\langle m,x\rangle}\,dx,\qquad\forall m\in\Omega_{d,r}(t).

For fixed dd and rr, we obtain lower bounds of optimal order in tt for the length of these curves.

Theorem 5.2. Let d≥2d\ge2 and 1≤r≤d1\le r\le d be integers.

(a) If γ\gamma is an rr-sparse tt-design curve on Td\mathbb{T}^{d}, for some positive integer tt, then its length LL satisfies

L≥(dr(t+1)2r−1t(t+2))1/2≥tr−1.L \ge\left(\frac{d}{r}\frac{(t+1)^{2r}-1}{t(t+2)}\right)^{1/2} \ge t^{r-1}.

(b) There is a constant Cd,r>0C_{d,r}>0 such that, for every positive integer tt, there exists a simple rr-sparse tt-design curve on Td\mathbb{T}^{d} of length L≤Cd,rtr−1L\le C_{d,r}t^{r-1}.

Proof. (a) Since ∥γ′(s)∥2=1\|\gamma'(s)\|_{2}=1 almost everywhere, the terms

Ej:=L∫0L∣γj′(s)∣2 ds,j=1,…,d,E_j:=L\int_{0}^{L}|\gamma'_j(s)|^{2}\,ds,\qquad j=1,\ldots,d,

add to ∑j=1dEj=L2\sum_{j=1}^{d}E_j=L^{2}.

Fix I⊂{1,…,d}I\subset\{1,\ldots,d\} with #I=r\#I=r, and define

AI:={a∈Rd:aj∈[t] for j∈I,aj=0 for j∉I}.A_I:=\{a\in\mathbb{R}^{d}:a_j\in[t]\text{ for }j\in I,\quad a_j=0\text{ for }j\notin I\}.

Then N:=#AI=(t+1)rN:=\#A_I=(t+1)^r and AI−AI⊂Ωd,r(t)A_I-A_I\subset\Omega_{d,r}(t). Consequently, the functions

fa(s):=e2πi⟨a,γ(s)⟩,a∈AI,f_a(s):=e^{2\pi i\langle a,\gamma(s)\rangle},\qquad a\in A_I,

are orthonormal in L2([0,L],ds/L)L^{2}([0,L],ds/L).

The covariance identity (6), applied to the coordinates in II, gives

1N∑a∈AIaaT=t(t+2)12∑j∈IejejT.\frac{1}{N}\sum_{a\in A_I}aa^{\mathsf T}=\frac{t(t+2)}{12}\sum_{j\in I}e_je_j^{\mathsf T}.

Using fa′=2πi⟨a,γ′⟩faf'_a=2\pi i\langle a,\gamma'\rangle f_a, we obtain

1N∑a∈AI1L∫0L∣fa′(s)∣2 ds=(2πL)2t(t+2)12∑j∈IEj.\frac{1}{N}\sum_{a\in A_I}\frac{1}{L}\int_{0}^{L}|f'_a(s)|^{2}\,ds =\left(\frac{2\pi}{L}\right)^{2}\frac{t(t+2)}{12}\sum_{j\in I}E_j.

Write w=γ(L)−γ(0)∈Zdw=\gamma(L)-\gamma(0)\in\mathbb{Z}^{d}. All the functions faf_a, a∈AIa\in A_I, satisfy the same boundary condition,

fa(L)=(−1)t∑j∈Iwjfa(0).f_a(L)=(-1)^{t\sum_{j\in I}w_j}f_a(0).

Thus the periodic or antiperiodic Fourier argument in the proof of Proposition 2.4 applies to this orthonormal family. Bessel’s inequality bounds the total squared Fourier coefficient at each frequency by one, Parseval’s identity gives total mass NN, and the sum of the NN smallest squared frequencies is at least N(N2−1)/12N(N^{2}-1)/12, so that we obtain

1N∑a∈AI1L∫0L∣fa′(s)∣2 ds≥(2πL)2N2−112.\frac{1}{N}\sum_{a\in A_I}\frac{1}{L}\int_{0}^{L}|f'_a(s)|^{2}\,ds \ge\left(\frac{2\pi}{L}\right)^{2}\frac{N^{2}-1}{12}.

This implies

∑j∈IEj≥(t+1)2r−1t(t+2).\sum_{j\in I}E_j\ge\frac{(t+1)^{2r}-1}{t(t+2)}.

Summing over all (dr)\binom{d}{r} choices of II, and observing that each coordinate belongs to (d−1r−1)\binom{d-1}{r-1} of them, we obtain

(d−1r−1)L2≥(dr)(t+1)2r−1t(t+2).\binom{d-1}{r-1}L^{2}\ge\binom{d}{r}\frac{(t+1)^{2r}-1}{t(t+2)}.

The observation (dr)/(d−1r−1)=d/r\binom{d}{r}/\binom{d-1}{r-1}=d/r yields the first bound in Part (a). The second follows from

(t+1)2r−1t(t+2)=∑ℓ=0r−1(t+1)2ℓ≥t2r−2\frac{(t+1)^{2r}-1}{t(t+2)}=\sum_{\ell=0}^{r-1}(t+1)^{2\ell}\ge t^{2r-2}

and d/r≥1d/r\ge1.

(b) For r=1r=1, the primitive vector v=(1,…,1)Tv=(1,\ldots,1)^{\mathsf{T}} gives a simple 1-sparse tt-design geodesic of length d\sqrt{d}.

Suppose r≥2r\ge2. We use the same counting argument as in the proof of Part (b) in Theorem 4.3. For every integer 1≤s≤t1\le s\le t, counting frequencies by their support gives

#{m∈Ωd,r(t):∥m∥∞=s}=∑h=1r(dh)2h(sh−(s−1)h)≲d,rsr−1.\#\{m\in\Omega_{d,r}(t):\lVert m\rVert_{\infty}=s\}=\sum_{h=1}^{r}\binom{d}{h}2^{h}(s^{h}-(s-1)^{h})\lesssim_{d,r}s^{r-1}.

Consequently,

#Ωd,r(t)≲d,rtr,∑m∈Ωd,r(t)∖{0}1∥m∥∞≲d,r∑s=1tsr−2≲d,rtr−1.\#\Omega_{d,r}(t)\lesssim_{d,r}t^{r},\qquad\sum_{m\in\Omega_{d,r}(t)\setminus\{0\}}\frac{1}{\lVert m\rVert_{\infty}}\lesssim_{d,r}\sum_{s=1}^{t}s^{r-2}\lesssim_{d,r}t^{r-1}.

Let P\mathcal{P} be the primitive vectors in Ωd,r(t)∖{0}\Omega_{d,r}(t)\setminus\{0\}, and define

Bk:=⋃m∈P{w∈[−k,k]d∩Zd:⟨w,m⟩=0}.\mathcal{B}_{k}:=\bigcup_{m\in\mathcal{P}}\{w\in[-k,k]^{d}\cap\mathbb{Z}^{d}:\langle w,m\rangle=0\}.

Dividing a frequency by the greatest common divisor of its coordinates preserves membership in Ωd,r(t)\Omega_{d,r}(t). Thus every forbidden hyperplane is represented by an element of P\mathcal{P}.

By the hyperplane-counting bound (19) and the union bound,

#Bk(2k+1)d≤2k∑m∈P∥m∥∞−1+#P(2k+1)2≲d,rtr−1k+trk2.\frac{\#\mathcal{B}_{k}}{(2k+1)^{d}}\le\frac{2k\sum_{m\in\mathcal{P}}\lVert m\rVert_{\infty}^{-1}+\#\mathcal{P}}{(2k+1)^{2}}\lesssim_{d,r}\frac{t^{r-1}}{k}+\frac{t^{r}}{k^{2}}.

Choose k=⌈cd,rtr−1⌉k=\lceil c_{d,r}t^{r-1}\rceil. Since r≥2r\ge2 and t≥1t\ge1,

tr−1k+trk2≤1cd,r+t2−rcd,r2≤1cd,r+1cd,r2.\frac{t^{r-1}}{k}+\frac{t^{r}}{k^{2}}\le\frac{1}{c_{d,r}}+\frac{t^{2-r}}{c_{d,r}^{2}}\le\frac{1}{c_{d,r}}+\frac{1}{c_{d,r}^{2}}.

For sufficiently large cd,rc_{d,r}, the forbidden proportion is therefore less than one. Hence there exists w∈[−k,k]d∩Zdw\in[-k,k]^{d}\cap\mathbb{Z}^{d} such that

⟨w,m⟩≠0,∀m∈Ωd,r(t)∖{0}.\langle w,m\rangle\ne0,\qquad\forall m\in\Omega_{d,r}(t)\setminus\{0\}.

The primitive vector v:=w/gcd⁡(∣w1∣,…,∣wd∣)v:=w/\gcd(|w_{1}|,\ldots,|w_{d}|) satisfies the same nonorthogonality conditions and leads to a simple curve of length

L:=∥v∥2≤d k≤d(cd,r+1)tr−1.L:=\lVert v\rVert_{2}\le\sqrt{d}\,k\le\sqrt{d}(c_{d,r}+1)t^{r-1}.

Thus, Cd,r=d(cd,r+1)C_{d,r}=\sqrt{d}(c_{d,r}+1) is the required constant. □

For r=dr=d, the frequency set Ωd,r(t)\Omega_{d,r}(t) is the full tensor-product box, and the lower bound in Theorem 5.2 reduces to Theorem 2.1. For r=1r=1, the diagonal geodesic has minimum length d\sqrt{d}, independently of tt. For intermediate values of rr, Theorem 5.2 shows that the optimal order of length is tr−1t^{r-1}. In particular, the power of tt is determined by the sparsity parameter rr, rather than by the ambient dimension dd.

Acknowledgement of AI assistance

ChatGPT (OpenAI; model: GPT-6 Astra) was used for discussion of selected proof steps and language editing. The author takes full responsibility for the content of the paper.

Faculty of Mathematics, University of Vienna, Oskar-Morgenstern-Platz 1, A-1090 Vienna, Austria

Email address: [email protected]

References

  1. [1]A. Bondarenko, D. Radchenko, M. Viazovska. Optimal asymptotic bounds for spherical designs. Ann. of Math., 178(2):443–452, 2013.arxiv.org/abs/1009.4407
  2. [2]P. Delsarte, J. M. Goethals, J. J. Seidel. Spherical codes and designs. Geom. Dedicata, 6(3):363–388, 1977.
  3. [3]D. Dung, V. Temlyakov, T. Ullrich. Hyperbolic Cross Approximation. Advanced Courses in Mathematics – CRM Barcelona, Birkhäuser, 2018.
  4. [4]M. Ehler, M. Gräf, S. Neumayer, G. Steidl. Curve based approximation of measures on manifolds by discrepancy minimization. Found. Comput. Math., 21:1595–1642, 2021.
  5. [5]M. Ehler, K. Gröchenig. t-design curves and mobile sampling on the sphere. Forum Math. Sigma, 11:e105, 2023.arxiv.org/abs/2306.13152
  6. [6]M. Forst, L. Fukhansky. Counting basis extensions in a lattice. Proc. Amer. Math. Soc., 150(8):3199–3213, 2022.arxiv.org/abs/2011.05307
  7. [7]R. J. Guenther, W. J. Wilson, D. L. Mohr. Statistical Methods. Academic Press, 3rd ed., 2010.
  8. [8]L. Kämmerer, S. Kunis, D. Potts. Interpolation lattices for hyperbolic cross trigonometric polynomials. J. Complexity, 28(1):76–92, 2012.DOI
  9. [9]E. J. King, D. G. Mixon. Short spherical t-design curves. arXiv:2607.15386, 2026.arxiv.org/abs/2607.15386
  10. [10]S. Kunis. Positive quadrature and mobile sampling of multivariate trigonometric polynomials. arXiv:2608.11915, 2026.arxiv.org/abs/2608.11915
  11. [11]A. Lindblad. Asymptotically optimal t-design curves on S³. arXiv:2408.04044, 2025.arxiv.org/abs/2408.04044
  12. [12]OEIS Foundation Inc. A008288. The On-Line Encyclopedia of Integer Sequences.

Paper details

vlnc:W_39gbx2sx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT

View statement
Looking for related papers…
Contents