In this section we extend Theorem 1.1 to periodic frameworks by showing that L-periodic 2-rigidity, together with a rank condition on the \(\Gamma \)-labelled graph in the case when the framework is d-periodic in \({\mathbb {R}}^d\), implies L-periodic global rigidity. We need several lemmas. The first one is [11, Prop. 13].
Lemma 4.1
Let \(f:{\mathbb {R}}^d\rightarrow {\mathbb {R}}^d\) be a polynomial map with rational coefficients and p be a generic point in \({\mathbb {R}}^d\). Suppose that \(df|_p\) is non-singular. Then for every \(q\in f^{-1}(f(p))\) we have \(\overline{{\mathbb {Q}}(p)}=\overline{{\mathbb {Q}}(q)}\), where \(\overline{{\mathbb {Q}}(p)}\) and \(\overline{{\mathbb {Q}}(q)}\) denote the algebraic closures of \({\mathbb {Q}}(p)\) and \({\mathbb {Q}}(q)\), respectively.
Let \(\Gamma \) be a group isomorphic to \({\mathbb {Z}}^k\), \(t=\max {\{d-k, 1\}}\), \((G,\psi )\) be a \(\Gamma \)-labelled graph with \(|V(G)|\ge t\), and \(L:\Gamma \rightarrow {\mathbb {R}}^d\) be non-singular. For simplicity we suppose that the linear span of \(L(\Gamma )\) is \(\{0\}^{d-k}\times {\mathbb {R}}^k\), the linear subspace spanned by the last k coordinates. We pick any t vertices \(v_1,\ldots , v_t\), and define the augmented function of \(f_{G,L}\) by \(\hat{f}_{G,L}:=(f_{G,L}, g)\), where \(g:{\mathbb {R}}^{d|V|}\rightarrow {\mathbb {R}}^{d+{t\atopwithdelims ()2}}\) is a rational polynomial map given by
$$\begin{aligned} g(p)= & {} \bigl (p_1(v_1), \ldots , p_d(v_1), p_1(v_2), \ldots , p_{t-1}(v_2),\\&\quad \,\, p_1(v_3),\ldots , p_{t-2}(v_3),\ldots , p_1(v_t)\bigr ), \end{aligned}$$
where \(p\in {\mathbb {R}}^{d|V|}\) and \(p_i(v_j)\) denotes the i-th coordinate of \(p(v_j)\). Augmenting \(f_{G,L}\) by g corresponds to “pinning down” some coordinates to eliminate trivial continuous motions. The following lemma is [13, Prop. 3.6].
Lemma 4.2
Let \((G,\psi ,p)\) be a \(\Gamma \)-labelled framework in \({\mathbb {R}}^d\) with rank k periodicity and \(L:\Gamma \rightarrow {\mathbb {R}}^d\) be a non-singular homomorphism such that \(L(\Gamma )\subset \{0\}^{d-k}\times {\mathbb {R}}^k\). Suppose that p is generic and \(|V(G)|\ge \max {\{d-k, 1\}}\). Then
$$\begin{aligned} {\text {rank}} d\hat{f}_{G,L}|_p={\text {rank}} df_{G,L}|_p+d+{d-k\atopwithdelims ()2}. \end{aligned}$$
We also need an adapted version of [13, Lem. 4.5], which is a periodic generalisation of an observation made in [10, 24]. To state this lemma, we require the following definition.
Let \((G,\psi )\) be a \(\Gamma \)-labelled graph and let v be a vertex of G. Suppose that every edge incident to v is directed from v. For each pair of nonparallel edges \(e_1=vu\) and \(e_2=vw\) in \((G,\psi )\), let \(e_1\cdot e_2\) be the edge from u to w with label \(\psi (vu)^{-1}\psi (vw)\). We define \((G_v,\psi _v)\) to be the \(\Gamma \)-labelled graph obtained from \((G,\psi )\) by removing v and inserting \(e_1\cdot e_2\) (unless it is already present) for every pair of nonparallel edges \(e_1, e_2\) incident to v.
Lemma 4.3
Let \((G,\psi ,p)\) be a generic \(\Gamma \)-labelled framework in \({\mathbb {R}}^d\) with rank k periodicity \(\Gamma \) and with \(|V(G)|\ge d-k+2\) and let \(L:\Gamma \rightarrow {\mathbb {R}}^d\) be non-singular. Suppose that the covering \(({\tilde{G}},{\tilde{p}})\) has a vertex v with at least \(d+1\) neighbours \(\gamma _0v_0,\gamma _1v_1,\ldots ,\gamma _dv_d\), where \(v,v_i\in V(G)\) and \(\gamma _i\in \Gamma \), so that the points \({\tilde{p}}(\gamma _0v_0), {\tilde{p}}(\gamma _1v_1),\ldots , {\tilde{p}}(\gamma _dv_d)\) affinely span \({\mathbb {R}}^d\). Suppose further that
-
\((G-v,\psi |_{G-v}, p')\) is L-periodically rigid in \({\mathbb {R}}^d\), with notation \(p'=p|_{V(G)-v}\), and
-
\((G_v, \psi _v, p')\) is L-periodically globally rigid in \({\mathbb {R}}^d\).
Then \((G, \psi , p)\) is L-periodically globally rigid in \({\mathbb {R}}^d\).
Proof
We assume (by rotating the whole space if necessary) that \(L(\Gamma )=\{0\}^{d-k}\times {\mathbb {R}}^k\). We pin the framework \((G,\psi ,p)\) and take any \(q\in \hat{f}_{G,L}^{-1}(\hat{f}_{G,L}(p))\). Since \(|V(G)|\ge d-k+2>\max {\{d-k, 1\}}\), we may assume that v is not “pinned” (i.e., v is different from the vertices selected when augmenting \(f_{G,L}\) to \(\hat{f}_{G,L}\)). Our goal is to show that \(p=q\).
Let \(p'\) and \(q'\) be the restrictions of p and q to \(V(G)-v\), respectively. Since \((G-v,\psi |_{G-v},p')\) is L-periodically rigid, we have \({\text {rank}} df_{G-v,L}|_{p'}=d|V(G-v)|-d-{d-k\atopwithdelims ()2}\) by Theorem 2.2. Then, by Lemma 4.2, we further have \({\text {rank}} d\hat{f}_{G-v,L}|_{p'}=d|V(G-v)|\). Thus we can take a spanning subgraph H of \(G-v\) such that \(d\hat{f}_{H,L}|_{p'}\) has linearly independent rows and is hence non-singular. Since \(q'\in \hat{f}_{H,L}^{-1}(\hat{f}_{H,L}(p'))\), it follows from Lemma 4.1 that \(\overline{{\mathbb {Q}}(p')}=\overline{{\mathbb {Q}}(q')}\). This in turn implies that \(q'\) is generic.
Consider the edges \(e_0=vv_0, e_1=vv_1,\ldots , e_{d}=vv_d\) in \((G,\psi )\) (all assumed to be directed from v) with respective labels \(\psi (e_0)=\gamma _0\), \(\psi (e_1)=\gamma _1\), \(\ldots \), \(\psi (e_d)=\gamma _d\). Note that we may have \(v_i=v_j\) for some i, j. By switching, we may further assume that \(\gamma _0=\text {id}\). For each \(1\le i\le d\), let
$$\begin{aligned} x_i=p(v_i)+L(\gamma _i)-p(v_0), \qquad y_i=q(v_i)+L(\gamma _i)-q(v_0), \end{aligned}$$
and let P and Q be the \(d\times d\)-matrices whose i-th column is \(x_i\) and \(y_i\), respectively. Note that since \(p(v_i)+L(\gamma _i)-p(v_0)=\tilde{p}(\gamma _iv_i)-\tilde{p}(v_0)\), and \(q(v_i)+L(\gamma _i)-q(v_0)=\tilde{q}(\gamma _iv_i)-\tilde{q}(v_0)\), and \(p',q'\) are generic, \(x_1,\ldots , x_d\) and \(y_1,\ldots , y_d\) are, respectively, linearly independent, and hence P and Q are both non-singular.
Let \(x_v=p(v)-p(v_0)\) and \(y_v=q(v)-q(v_0)\). We then have \(\Vert x_v\Vert =\Vert y_v\Vert \) since G has the edge \(vv_0\) with \(\psi (vv_0)=\text {id}\). Due to the existence of the edge \(e_i\) we also have
$$\begin{aligned} 0&=\langle p(v_i)+L(\gamma _i)-p(v), p(v_i)+L(\gamma _i)-p(v)\rangle \\&\quad -\langle q(v_i)+L(\gamma _i)-q(v),q(v_i)+L(\gamma _i)-q(v) \rangle \\&=\langle x_i-x_v, x_i-x_v\rangle - \langle y_i-y_v, y_i-y_v\rangle \\&=\Vert x_i\Vert ^2-\Vert y_i\Vert ^2-2\langle x_i, x_v\rangle +2\langle y_i, y_v\rangle , \end{aligned}$$
where we used \(\Vert x_v\Vert =\Vert y_v\Vert \). Denoting by \(\delta \) the d-dimensional vector whose i-th coordinate is equal to \(\Vert x_i\Vert ^2-\Vert y_i\Vert ^2\), the above d equations can be summarized as
$$\begin{aligned} 0=\delta -2P^T x_v+2Q^T y_v, \end{aligned}$$
which is equivalent to
$$\begin{aligned} y_v=(Q^T)^{-1}P^T x_v -\frac{1}{2}(Q^T)^{-1}\delta . \end{aligned}$$
By putting this into \(\Vert x_v\Vert ^2=\Vert y_v\Vert ^2\), we obtain
$$\begin{aligned} \begin{aligned} x_v^T(I_d-PQ^{-1}(PQ^{-1})^T)x_v&-(\delta ^T Q^{-1}(Q^{-1})^T P^T)x_v\\&+\,\frac{\delta ^T Q^{-1}(Q^{-1})^T\delta }{4}=0, \end{aligned} \end{aligned}$$
(3)
where \(I_d\) denotes the \(d\times d\) identity matrix.
Note that each entry of P is contained in \({\mathbb {Q}}(p')\), and each entry of Q is contained in \({\mathbb {Q}}(q')\). Since \(\overline{{\mathbb {Q}}(p')}=\overline{{\mathbb {Q}}(q')}\), this implies that each entry of \(PQ^{-1}\) is contained in \(\overline{{\mathbb {Q}}(p')}\). On the other hand, since p is generic, the set of coordinates of p(v) (and hence those of \(x_v\)) is algebraically independent over \(\overline{{\mathbb {Q}}(p')}\). Therefore, by regarding the left-hand side of (3) as a polynomial in \(x_v\), the polynomial must be identically zero. In particular, we get
$$\begin{aligned} I_d-PQ^{-1}(PQ^{-1})^T=0. \end{aligned}$$
Thus, \(PQ^{-1}\) is orthogonal. In other words, there is some orthogonal matrix S such that \(P=SQ\), and we get
$$\begin{aligned} \Vert p(v_i)+L(\gamma _i)-p(v_0)\Vert =\Vert x_i\Vert =\Vert Sy_i\Vert =\Vert y_i\Vert =\Vert q(v_i)+L(\gamma _i)-q(v_0)\Vert \end{aligned}$$
for every \(1\le i\le d\). Therefore, \(q'\in f_{G_v, L}^{-1}( f_{G_v, L}(p'))\). Since \((G_v, \psi _v,p)\) is L-periodically globally rigid, this in turn implies that \(f_{V-v, L}(p')=f_{V-v, L}(q')\). Thus we have \(p'=q'\).
Since \(\{p(v_i)+L(\gamma _i): 0\le i\le d\}\) affinely spans \({\mathbb {R}}^d\), there is a unique extension of \(p':V(G)-v\rightarrow {\mathbb {R}}^d\) to \(r:V(G)\rightarrow {\mathbb {R}}^d\) such that \(f_{G, L}(r)=f_{G,L}(p)\). Thus we obtain \(p=q\). \(\square \)
Finally, we need the following special case of [13, Lem. 3.1].
Lemma 4.4
Let \((G, \psi , p)\) be a generic \(\Gamma \)-labelled framework in \({\mathbb {R}}^d\) with \(|V(G)|\ge 2\), rank k periodicity \(\Gamma \), and let \(L:\Gamma \rightarrow {\mathbb {R}}^d\) be a non-singular homomorphism. If \((G, \psi , p)\) is L-periodically globally rigid, then the rank of \((G,\psi )\) is equal to k.
Lemma 4.4 is easily seen to be true, because if the rank of \((G,\psi )\) is less than k, then the covering \((\tilde{G}, {\tilde{p}})\) of \((G,\psi ,p)\) has infinitely many connected components, each of which may be ‘flipped’ individually in a periodic fashion to obtain an L-periodic framework \(({\tilde{G}}, {\tilde{q}})\) which is equivalent, but not congruent to \(({\tilde{G}}, {\tilde{p}})\). Thus, \((G, \psi , p)\) is not L-periodically globally rigid.
This is illustrated by the two equivalent but non-congruent 2-periodic frameworks in \({\mathbb {R}}^2\) shown in Fig. 2 whose \(\Gamma \)-labelled graph \((G,\psi )\) has rank 1. Note, however, that \((G,\psi )\) is L-periodically 2-rigid, since it is L-periodically rigid and the removal of any vertex results in a trivial framework with one vertex orbit and no edges (recall also Theorem 2.2).
It follows that in the case when \(k=d\), L-periodic 2-rigidity is not sufficient for L-periodic global rigidity. In this case we need the added assumption that \({\text {rank}}(G,\psi )=d\). In the case when \(k<d\) and \({\text {rank}}(G,\psi )<k\), \((G,\psi , p)\) can also not be L-periodically globally rigid, by Lemma 4.4. However, in this case, \((G,\psi , p)\) is also not L-periodically 2-rigid.
Theorem 4.5
Let \((G, \psi , p)\) be a generic \(\Gamma \)-labelled framework in \({\mathbb {R}}^d\) with rank k periodicity \(\Gamma \), and let \(L:\Gamma \rightarrow {\mathbb {R}}^d\) be non-singular. If \((G, \psi , p)\) is L-periodically 2-rigid, and if \((G,\psi )\) is also of rank d in the case when \(k=d\), then \((G, \psi , p)\) is L-periodically globally rigid in \({\mathbb {R}}^d\).
Proof
We use induction on |V(G)|. If \(|V(G)|\le d-k+1\), then \((G, \psi , p)\) is L-periodically globally rigid by the L-periodic rigidity of \((G, \psi , p)\) and Corollary 3.3.
Now suppose that \(|V(G)|\ge d-k+2\), and let \(({\tilde{G}}, {\tilde{p}})\) be the covering of \((G,\psi , p)\). By our assumption, \((G-v,\psi |_{G-v},p|_{V(G)-v})\) is L-periodically rigid for any vertex \(v\in V(G)\).
Suppose first that \(|V(G)|= d-k+2\). Then \((G-v,\psi |_{G-v},p|_{V(G)-v})\) is also L-periodically globally rigid by Corollary 3.3. We claim that for any occurrence of any \(v\in V(G)\) in the covering \({\tilde{G}}\), the affine span of the set \(\{\tilde{p}(w): vw \in E(\tilde{G})\}\) is all of \({\mathbb {R}}^d\).
If \(d=k\) (and hence \(|V(G)|=d-k+2=2\)) the claim follows from the fact that \({\text {rank}}(G,\psi )=d\), by our assumption.
If \(d>k\) (and hence \(|V(G)|=d-k+2>2\)), then we suppose for a contradiction that the claim is not true. Then the removal of a neighbour of v (and of all vertices belonging to that same vertex orbit) results in an L-periodic framework with at least two distinct orbits of points (since \(|V(G)|>2\)) and, by our genericity assumption, this framework has the property that all the points connected to \(\tilde{p}(v)\) affinely span a space of dimension at most \(d-2\), so that \(\tilde{p}(v)\) can be rotated about this \((d-2)\)-dimensional axis. Since all copies of points in the same orbit can then also be rotated in a periodic fashion and the affine span of the non-moving points is \((d-1)\)-dimensional (as a k-periodic configuration with \(d-k\) vertex orbits), we obtain a contradiction to the L-periodic 2-rigidity of \(({\tilde{G}},\tilde{p})\).
Thus, the affine span of the points \(\{\tilde{p}(w) : vw \in E(\tilde{G})\}\) is indeed all of \({\mathbb {R}}^d\) as claimed, and it follows from Lemma 4.3 that \((G,\psi , p)\) is L-periodically globally rigid.
We may therefore assume that \(|V(G)|>d-k+2\). We show that \((G_v, \psi _v, p')\) is L-periodically 2-rigid for any \(v\in V(G)\). Suppose for a contradiction that this is not true. Then there is a vertex u whose removal results in an L-periodically flexible framework. As the neighbours of one occurrence of v in \(\tilde{G}\) induce a complete graph in \(\tilde{G_v}\) (where any pair of vertices from the same vertex orbit may always be considered adjacent due to the fixed lattice representation), adding v together with its incident edges to \((G_v-u,\psi _v|_{G_v-u},p|_{V(G)-\{u,v\}})\) still yields an L-periodically flexible framework. This is a contradiction, as \((G-u,\psi |_{G-u},p|_{V(G)-u})\) is an L-periodically rigid \(\Gamma \)-labelled spanning subframework of the framework obtained from \((G_v-u, \psi _v|_{G_v-u}, p|_{V(G)-\{u,v\}})\) by adding v and its incident edges.
Thus \((G_v, \psi _v, p')\) is L-periodically 2-rigid as claimed. Moreover, since \((G,\psi )\) is 2-connected by the L-periodic 2-rigidity of \((G,\psi ,p)\), it follows from the definition of \((G_v,\psi _v)\) that \(\Gamma _G=\Gamma _{G_v}\). Thus, if \((G,\psi )\) is of rank d then so is \((G_v,\psi _v)\). It now follows from the induction hypothesis that \((G_v, \psi _v, p')\) is L-periodically globally rigid. Moreover, by the same argument as above for the case when \(|V(G)|=d-k+2>2\), the affine span of the points \(\{\tilde{p}(w):vw \in E(\tilde{G})\}\) is all of \({\mathbb {R}}^d\). Thus, by Lemma 4.3, \((G, \psi , p)\) is L-periodically globally rigid. \(\square \)