Tight upper bound on $d_{\mathrm{GH}}(S^{1},S^{2k+1})$: GPT’s short proof

Abstract.

We describe GPT-5.6 Sol Pro’s short proof that $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})\leq\tfrac{2\pi k}{2k+1}$. This is the odd-dimensional case of a tight upper bound by Harrison and Jeffs on the Gromov–Hausdorff distance between the circle and spheres.

Let $k\geq 1$ be an integer. We give a short proof of the upper bound $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})\leq\tfrac{2\pi k}{2k+1}$. This is the odd-dimensional case of a tight upper bound by Harrison and Jeffs for the Gromov–Hausdorff distance between the circle and spheres [12]. The proof was produced by GPT-5.6 Sol Pro in ChatGPT; see for the prompt and response and see for a PDF of the model’s proof. We edited the writing, not the mathematics.

The equalities $2\cdot d_{\mathrm{GH}}(S^{1},S^{2})=2\cdot d_{\mathrm{GH}}(S^{1},S^{3})=\tfrac{2\pi}{3}$ were proved in [14], in part building on [8]. The lower bounds $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k})\geq\tfrac{2\pi k}{2k+1}$ and $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})\geq\tfrac{2\pi k}{2k+1}$ were proved in [2], in part building on the homotopy types of Vietoris–Rips complexes of the circle [1, 16]. Question 8.1 of [2] asked whether equality holds in these lower bounds. Harrison and Jeffs [12] answered this question by proving matching upper bounds, implying $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})=\tfrac{2\pi k}{2k+1}=2\cdot d_{\mathrm{GH}}(S^{1},S^{2k})$. A second proof appears in [18].

We prove $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})\leq\tfrac{2\pi k}{2k+1}$, and hence $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})=\tfrac{2\pi k}{2k+1}$. We use the symmetric moment curve $g\colon S^{1}\to S^{2k+1}$ defined by

(1) $$g(t)=\tfrac{1}{\sqrt{k+1}}\bigl(\cos t,\sin t,\cos 3t,\sin 3t,\ldots,\cos((2k+1)t),\sin((2k+1)t)\bigr).$$

This curve was defined by Barvinok and Novik in [6]; see also [5, 19]. It was related to Vietoris–Rips complexes of the circle by Adams, Bush, and Frick in [3, 7]. Lim, Mémoli, and Smith used variants of this curve to understand $d_{\mathrm{GH}}(S^{1},S^{2})$ and $d_{\mathrm{GH}}(S^{1},S^{3})$ [14]. Conversations with Frick and Mémoli about this curve helped initiate the polymath collaboration [2]. As described in [15], I proposed combining the symmetric moment curve with Mémoli and Smith’s embedding-projection correspondences to study $d_{\mathrm{GH}}(S^{1},S^{2k+1})$ for all $k$; this proposal is realized here. A variant of the symmetric moment curve to study $d_{\mathrm{GH}}(S^{1},S^{2k})$ was proposed in [15]. The symmetric moment curve also parametrizes the orbit of a point under the diagonal $S^{1}$-action on $S^{1}_{1}*S^{1}_{3}*\ldots*S^{1}_{2k+1}\cong S^{2k+1}$, where $S^{1}_{j}$ denotes a copy of $S^{1}$ with the action $z\cdot w=z^{j}w$; see [4].

The Gromov–Hausdorff distance provides a notion of dissimilarity between metric spaces [9, 10, 11, 21]. It follows from [13] that the Gromov–Hausdorff distance between compact metric spaces can be defined as

(2) $$2\cdot d_{\mathrm{GH}}(X,Y)=\inf_{f,h}\max\{\mathrm{dis}(f),\mathrm{dis}(h),\mathrm{codis}(f,h)\},$$

where $f\colon X\to Y$ and $h\colon Y\to X$ are (possibly discontinuous) functions. The distortion of $f$ is defined as $\mathrm{dis}(f)=\sup_{x,x^{\prime}\in X}|d(x,x^{\prime})-d(f(x),f(x^{\prime}))|$, and the codistortion of $f$ and $h$ is defined as $\mathrm{codis}(f,h)=\sup_{x\in X,y\in Y}|d(x,h(y))-d(f(x),y)|$.

Equip each sphere with its geodesic metric. Let $S^{1}=\mathbb{R}/2\pi\mathbb{Z}$. For $t,t^{\prime}\in S^{1}$, let $d(t,t^{\prime})=\min_{j\in\mathbb{Z}}|t-t^{\prime}+2\pi j|$. For $x,x^{\prime}\in S^{2k+1}$, we have $d(x,x^{\prime})=\arccos\langle x,x^{\prime}\rangle$. Define $g\colon S^{1}\to S^{2k+1}$ by (1). Choose $h\colon S^{2k+1}\to S^{1}$ to be any function satisfying $d(x,g(h(x)))=d(x,g(S^{1}))$ for every $x\in S^{2k+1}$; such an $h$ exists because the continuous function $t\mapsto d(x,g(t))$ attains its minimum on the compact space $S^{1}$.

Theorem 1.

Let $k\geq 1$ be an integer. If $x,x^{\prime}\in S^{2k+1}$ and $t,t^{\prime}\in S^{1}$ satisfy $d(x,g(t))=d(x,g(S^{1}))$ and $d(x^{\prime},g(t^{\prime}))=d(x^{\prime},g(S^{1}))$, then $\bigl|d(x,x^{\prime})-d(t,t^{\prime})\bigr|\leq\tfrac{2\pi k}{2k+1}$.

For $t,t^{\prime}\in S^{1}$ we deduce $\bigl|d(g(t),g(t^{\prime}))-d(t,t^{\prime})\bigr|\leq\tfrac{2\pi k}{2k+1}$; for $x,x^{\prime}\in S^{2k+1}$ we deduce $\bigl|d(x,x^{\prime})-d(h(x),h(x^{\prime}))\bigr|\leq\tfrac{2\pi k}{2k+1}$; and for $x\in S^{2k+1}$ and $t\in S^{1}$ we deduce $\bigl|d(x,g(t))-d(h(x),t)\bigr|\leq\tfrac{2\pi k}{2k+1}$. By (2), we have $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})\leq\max\{\mathrm{dis}(g),\mathrm{dis}(h),\mathrm{codis}(g,h)\}\leq\tfrac{2\pi k}{2k+1}$. Combining with the lower bound [2, Remark 5.5] gives $2\cdot d_{\mathrm{GH}}(S^{1},S^{2k+1})=\tfrac{2\pi k}{2k+1}$.

Proof of Theorem 1.

For a unit vector $y\in S^{2k+1}$, define $F_{y}\colon S^{1}\to\mathbb{R}$ by $F_{y}(s)=\langle y,g(s)\rangle$. Note $d(y,g(s))=d(y,g(S^{1}))$ if and only if $F_{y}(s)=\max_{u}F_{y}(u)$. Since $g(u+\pi)=-g(u)$, we have $F_{y}(u+\pi)=-F_{y}(u)$. Thus, if $M_{y}=\max_{u}F_{y}(u)$, then $\min_{u}F_{y}(u)=-M_{y}$, so $|F_{y}(u)|\leq M_{y}$ for every $u$. Moreover, writing $y=(a_{0},b_{0},\ldots,a_{k},b_{k})$ and using orthogonality of the sine and cosine functions,

$$\tfrac{1}{2\pi}\int_{0}^{2\pi}F_{y}(u)^{2}\,du=\tfrac{1}{2\pi(k+1)}\sum_{j=0}^{k}\int_{0}^{2\pi}(a_{j}^{2}\cos^{2}((2j+1)u)+b_{j}^{2}\sin^{2}((2j+1)u))du=\tfrac{1}{2k+2}\sum_{j=0}^{k}(a_{j}^{2}+b_{j}^{2})=\tfrac{1}{2k+2}.$$

Hence $M_{y}\geq\tfrac{1}{\sqrt{2k+2}}$.

Set $\delta_{2k+1}=d(x,x^{\prime})$ and $\delta_{1}=d(t,t^{\prime})$. We first prove $\delta_{2k+1}\geq\delta_{1}-\tfrac{2\pi k}{2k+1}$. This is automatic when $\delta_{1}\leq\tfrac{2\pi k}{2k+1}$, so suppose $\delta_{1}>\tfrac{2\pi k}{2k+1}$ and write $\delta_{1}=\tfrac{2\pi k}{2k+1}+\beta$ with $0<\beta\leq\tfrac{\pi}{2k+1}$. The block-diagonal orthogonal action that rotates the block of frequency $2j+1$ through angle $(2j+1)\theta$ satisfies $g(u+\theta)=R_{\theta}g(u)$. Applying $R_{-t}$ simultaneously to $x$ and $x^{\prime}$, and then applying the reflection taking $g(u)$ to $g(-u)$ if necessary, preserves all relevant distances and maximizing conditions. We may therefore assume

$$t=0,\qquad t^{\prime}=\delta_{1}=\pi-\eta,\qquad\eta=\tfrac{\pi}{2k+1}-\beta.$$

Let $u_{0}=\pi-\tfrac{\eta}{2}$ be the midpoint, in the chosen coordinate interval, between $t^{\prime}$ and $\pi$. Since $0$ maximizes $F_{x}$, applying Lemma 1 to the trigonometric polynomial $u\mapsto\tfrac{F_{x}(u)}{M_{x}}$ gives $F_{x}(u)\geq M_{x}\cos((2k+1)|u|)$ for $|u|\leq\tfrac{\pi}{2k+1}$. Hence $F_{x}(u_{0})=-F_{x}(\tfrac{-\eta}{2})\leq-M_{x}\cos\tfrac{(2k+1)\eta}{2}$. Since $t^{\prime}$ maximizes $F_{x^{\prime}}$ and $u_{0}-t^{\prime}=\tfrac{\eta}{2}$, applying Lemma 1 to $u\mapsto\tfrac{F_{x^{\prime}}(u+t^{\prime})}{M_{x^{\prime}}}$ gives $F_{x^{\prime}}(u_{0})\geq M_{x^{\prime}}\cos\tfrac{(2k+1)\eta}{2}$. Because $(2k+1)\eta=\pi-(2k+1)\beta$, we have $\cos\tfrac{(2k+1)\eta}{2}=\sin\tfrac{(2k+1)\beta}{2}$. It follows that

$$\|x^{\prime}-x\|\geq\langle x^{\prime}-x,g(u_{0})\rangle=F_{x^{\prime}}(u_{0})-F_{x}(u_{0})\geq(M_{x}+M_{x^{\prime}})\sin\tfrac{(2k+1)\beta}{2}\geq\tfrac{2}{\sqrt{2k+2}}\sin\tfrac{(2k+1)\beta}{2}\geq 2\sin\tfrac{\beta}{2},$$

where we used $M_{x},M_{x^{\prime}}\geq\tfrac{1}{\sqrt{2k+2}}$ and Lemma 2 with $z=\tfrac{\beta}{2}$. Since $\|x^{\prime}-x\|=2\sin\tfrac{\delta_{2k+1}}{2}$ and since sine is increasing on $[0,\frac{\pi}{2}]$, we obtain $\delta_{2k+1}\geq\beta=\delta_{1}-\tfrac{2\pi k}{2k+1}$.

We now prove $\delta_{2k+1}\leq\delta_{1}+\tfrac{2\pi k}{2k+1}$. If $\delta_{1}\geq\tfrac{\pi}{2k+1}$, then $\delta_{1}+\tfrac{2\pi k}{2k+1}\geq\pi$, so this bound is automatic. Suppose $\delta_{1}<\tfrac{\pi}{2k+1}$. Since $d(x^{\prime},g(t^{\prime}))=d(x^{\prime},g(S^{1}))$, we have $d(-x^{\prime},g(t^{\prime}+\pi))=d(-x^{\prime},g(S^{1}))$. Moreover, $d(t,t^{\prime}+\pi)=\pi-\delta_{1}>\tfrac{2\pi k}{2k+1}$. Applying $\delta_{2k+1}\geq\delta_{1}-\tfrac{2\pi k}{2k+1}$ to $x$ and $-x^{\prime}$, with $d(x,g(t))=d(x,g(S^{1}))$ and $d(-x^{\prime},g(t^{\prime}+\pi))=d(-x^{\prime},g(S^{1}))$, gives $\pi-\delta_{2k+1}=d(x,-x^{\prime})\geq d(t,t^{\prime}+\pi)-\tfrac{2\pi k}{2k+1}=(\pi-\delta_{1})-\tfrac{2\pi k}{2k+1}=\tfrac{\pi}{2k+1}-\delta_{1}$. Hence $\delta_{2k+1}\leq\tfrac{2\pi k}{2k+1}+\delta_{1}$. ∎

The above proof used the following two lemmas.

Lemma 1 (p. 69 of [20], Satz 8 of [22], Theorem 3.7 of [17]).

Let $m\geq 1$, and let $T$ be a real trigonometric polynomial of degree at most $m$ such that $|T(u)|\leq 1$ for every $u$. Then $|T^{\prime}(u)|\leq m\sqrt{1-T(u)^{2}}$ for $u\in\mathbb{R}$. If furthermore $T(0)=1$, then $T(u)\geq\cos(m|u|)$ for $|u|\leq\tfrac{\pi}{m}$.

Proof.

Theorem 3.7 of [17] proves $(T^{\prime}(u))^{2}+m^{2}(T(u))^{2}\leq m^{2}$, yielding $|T^{\prime}(u)|\leq m\sqrt{1-T(u)^{2}}$.

Now assume $T(0)=1$. For $0<\varepsilon<1$, put $T_{\varepsilon}=(1-\varepsilon)T$ and $\theta_{\varepsilon}(u)=\arccos T_{\varepsilon}(u)$. The first part of the lemma applied to $T_{\varepsilon}$ gives $|T_{\varepsilon}^{\prime}(u)|\leq m\sqrt{1-T_{\varepsilon}(u)^{2}}$. Since $|T_{\varepsilon}|<1$, the function $\theta_{\varepsilon}$ is differentiable. Hence $|\theta_{\varepsilon}^{\prime}(u)|=\tfrac{|T_{\varepsilon}^{\prime}(u)|}{\sqrt{1-T_{\varepsilon}(u)^{2}}}\leq m$. Therefore $\theta_{\varepsilon}(u)\leq\theta_{\varepsilon}(0)+m|u|$. Letting $\varepsilon\to 0^{+}$ yields $\arccos T(u)\leq m|u|$. If $|u|\leq\tfrac{\pi}{m}$, both sides lie in $[0,\pi]$, on which cosine is decreasing. It follows that $T(u)\geq\cos(m|u|)$ for $|u|\leq\tfrac{\pi}{m}$. ∎

Lemma 2.

For every odd integer $m\geq 3$ and every $0\leq z\leq\tfrac{\pi}{2m}$, we have $\tfrac{\sin(mz)}{\sqrt{m+1}}\geq\sin z$.

Proof.

The claim is clear for $z=0$; let $z>0$. For $m=3$, we have $\tfrac{\sin(3z)}{\sin z}=3-4\sin^{2}z\geq 2=\sqrt{4}$ because $0<z\leq\frac{\pi}{6}$. Now let $m\geq 5$. Since $0\leq mz\leq\frac{\pi}{2}$, concavity of sine on $[0,\frac{\pi}{2}]$ gives $\sin(mz)\geq\tfrac{2mz}{\pi}$, while $\sin z\leq z$. Also $\tfrac{2m}{\pi}\geq\sqrt{m+1}$ for $m\geq 5$; for example, $\pi^{2}<10$ and $4m^{2}-10(m+1)>0$ for every $m\geq 5$. Hence $\sin(mz)\geq\tfrac{2mz}{\pi}\geq\sqrt{m+1}\,z\geq\sqrt{m+1}\sin z$. ∎