Vibrational Control of Cluster Synchronization: Connections with Deep Brain Stimulation

Yuzhen Qin, Danielle S. Bassett and Fabio Pasqualetti

arXiv: 2204.00678v2 · PDF · 5 April 2022
Categories: eess.SY, cs.SY, math.OC, physics.bio-ph
Licence: CC BY 4.0 The site shows the text.

Abstract

Cluster synchronization underlies various functions in the brain. Abnormal patterns of cluster synchronization are often associated with neurological disorders. Deep brain stimulation (DBS) is a neurosurgical technique used to treat several brain diseases, which has been observed to regulate neuronal synchrony patterns. Despite its widespread use, the mechanisms of DBS remain largely unknown. In this paper, we hypothesize that DBS plays a role similar to vibrational control since they both highly rely on high-frequency excitation to function. Under the framework of Kuramoto-oscillator networks, we study how vibrations introduced to network connections can stabilize cluster synchronization. We derive some sufficient conditions and also provide an effective approach to design vibrational control. Also, a numerical example is presented to demonstrate our theoretical findings.

Items 14

Notation

GM0U

We denote the unit circle as \mathbb{S}^1, a point of which is a phase. The n-dimensional torus is denoted by \mathbb{T}^n = \mathbb{S}\times\dots\times\mathbb{S}. Given a set \mathcal{S}, |\mathcal{S}| stands for its cardinality. For two sets \mathcal{S}_1 and \mathcal{S}_2, \mathcal{S}_1\backslash\mathcal{S}_2 is their set-theoretic difference. Given A\in\mathbb{R}^{n\times m}, B:=[A]^+ is the matrix where b_{ij}=a_{ij} if a_{ij}>0 and b_{ij}=0 otherwise. Given matrices A_1,\dots,A_r, {\rm blkdiag}(A_1,\dots,A_r) is the block-diagonal matrix of them.

Definition

Definition 1 (Synchronization manifold)

5N1A

For the the network \mathcal{G}=(\mathcal{V},\mathcal{E}), consider the partition \mathcal{C}:=\{\mathcal{C}_1,\mathcal{C}_2,\dots,\mathcal{C}_r\}, where \mathcal{C}_k\subset \mathcal{V}, \mathcal{C}_k \cap \mathcal{C}_\ell =\emptyset for any k\neq \ell, and \cup_{k=1}^r \mathcal{C}_k=\mathcal{V}. The cluster synchronization manifold associated with the partition \mathcal{C} is defined as

\mathcal{M}:= \{\theta \in \mathbb{T}^n: \theta_i=\theta_j, \forall i,j \in \mathcal{C}_k, k= 1,\dots,r\}.\hspace{12pt}\null \hfill\ensuremath{\triangle}

Used by Theorem 1

Assumption

Assumption 1 (Invariance)

9QTO

For k=1,2,\dots, r: i) the natural frequencies satisfy \omega_i=\omega_j for any i,j \in \mathcal{C}_k; and ii) the coupling strengths satisfy that, for any \ell\in\{1,2,\dots,r\}\backslash\{k\}, \sum_{p\in \mathcal{C}_\ell}(a_{ip}-a_{jp})=0 for any i,j \in \mathcal{C}_k.

Lemma

Lemma 1 (Connecting the stability of Systems (?) and (?))

TBMY

If the equilibrium x=0 is exponentially stable uniformly in y for the system (?), it is also exponentially stable uniformly in y for the system (?).

Theorem

Theorem 1 (Sufficient condition for vibrational stabilization)

984W

Assume that \bar J={\rm blkdiag}(\bar J^{(1)},\dots,\bar J^{(r)}) in Eq. (?) is Hurwitz. Let \bar X_k be the solution to the Lyapunov equation

\begin{aligned} {(\bar J^{(k)})^\top \bar X_k+\bar X_k^\top \bar J^{(k)}=-I}. \end{aligned}

Define the matrix S=[s_{k \ell}]_{r\times r} with

\begin{aligned} s_{k\ell}=\Big\{\begin{matrix*}[l] \lambda_{\max}^{-1}(\bar X_k)-\bar\gamma_{k k} , &\text{ if }k=\ell,\\ s_{k\ell}= -\bar \gamma_{k \ell}, &\text{ if }k \neq\ell. \end{matrix*} \end{aligned}

If S is an M-matrix, then there exists \varepsilon^*>0 such that, for any \varepsilon<\varepsilon^*:

(i) the equilibrium x=0 of the system (?) is exponentially stable uniformly in y;

(ii) the cluster synchronization manifold \mathcal{M} of the system (?) is exponentially stable.

Uses Definition 1 · Used by Remark 1

Remark

Remark 1 (Synchronization robustness)

OABQ

For a stable linear system \dot x=Ax, previous studies propose that \lambda_{\max}(X), where X is the solution to the Lyapunov equation A^\top X+X A=-2I, can measure the robustness of this system [PRV-TM:80, YR:85]. In our case, from (?), the uncontrolled intra-dynamics around the manifold \mathcal{M} are described by

\begin{aligned} &\dot x_k=J^{(k)} x_k+f^{(k)}_{\rm inter}(x,y),&k=1,\dots,r, \end{aligned}

where J^{(k)} is stable and f^{(k)}_{\rm inter}(x,y) is taken as the perturbation. Here, x_k=0 means synchronization of the oscillators in the kth cluster. Let X_k be the solution to

\begin{aligned} \scalebox{0.85}{\mbox{\ensuremath{\displaystyle ( J^{(k)})^\top X_k+X_k^\top J^{(k)}=-I}}}. \end{aligned}

Likewise, one can use \lambda^{-1}_{\max}(X_k) to measure the robustness of synchronization in the kth cluster. If the intra-cluster synchronization is sufficiently robust (i.e., large \lambda^{-1}_{\max}(X_k)’s) to dominate the perturbations resulted from inter-cluster connections, the cluster synchronization is stable. A sufficient condition is constructed in [TM-GB-DSB-FP:18, Th. 3.2]. By contrast, if \lambda_{\max}(X_k)’s are not large enough, the cluster synchronization can lose its stability. Yet, the robustness of the intra-cluster synchronization can be reshaped by introducing vibrations to the local network connections. The new robustness is instead measured by \lambda^{-1}_{\max}( \bar X_k), where \bar X_k is the solution to Eq. (1).

Uses Theorem 1

Example

Example 1 (Improving robustness by vibrational control)

KV3S

We restrict our attention to only one cluster in this example. Specifically, consider that J^{(k)} in (1) is

\begin{aligned} \scalebox{0.85}{\mbox{\ensuremath{\displaystyle J^{(k)}=\begin{bmatrix} -1& 4\\0 &-2 \end{bmatrix}}}}. \end{aligned}

Suppose that there is a vibrational control U^{(k)}(t) to the kth cluster that results in the state transition matrix

\begin{aligned} {\Phi^{(k)}(t,t_0)=\begin{bmatrix} 1&0\\ \cos(t_0)-\cos(t)&1 \end{bmatrix}}. \end{aligned}

It can be calculated that

\begin{aligned} \scalebox{0.9}{\mbox{\ensuremath{\displaystyle \bar J^{(k)} =\frac{1}{2\pi}\int_{0}^{2 \pi} \big(\Phi^{(k)}(t,t_0)\big)^{-1} J^{(k)} \Phi^{(k)}(t,t_0)dt=\begin{bmatrix} -1& 4\\-2 &-2 \end{bmatrix}}}}. \end{aligned}

Solving the equations (1) and (1), one can obtain \lambda^{-1}_{\max}( X_k)\approx 0.52 and \lambda^{-1}_{\max}(\bar X_k)=2, which means that the synchronization in this cluster becomes more robust due to the vibrational control.

Lemma

Lemma 2 (Necessary condition for robustness improvement)

ZKAF

Let \lambda^{-1}_{\max}(\bar X_k) and \lambda^{-1}_{\max}( X_k) be the solutions to the equations (1) and (1), respectively. There exists a vibrational control input such that \lambda^{-1}_{\max}(\bar X_k)> \lambda^{-1}_{\max}(X_k) only if \mathcal{G}_k is not a uniformly weighted complete graph(A uniformly weighted complete graph is a complete graph in which every edge has the same weight.).

Lemma

Lemma 3 (Lower triangular transition matrix)

Z6PC

Assume that the vibrational control is such that P(t) in the system (?) satisfies (?). Then, the state transition matrix \Phi(t,t_0)={\rm blkdiag}(\Phi_1,\dots,\Phi_r) satisfies

\begin{aligned} \scalebox{0.9}{\mbox{\ensuremath{\displaystyle \Phi_k=\begin{bmatrix} 1 &0 &0 &\cdots &0 &0\\ \Phi_{21}^{(k)} &1 &0& \cdots &0 &0\\ \Phi_{31}^{(k)} &\Phi_{32}^{(k)} &1 &\cdots &0 &0\\ \vdots&\vdots &\vdots&\ddots&\vdots&\vdots\\ \Phi_{\bar n_k,1}^{(k)} &\Phi_{\bar n_k,2}^{(k)} &\Phi_{\bar n_k,3}^{(k)} &\cdots &\Phi_{\bar n_k,\bar n_k-1}^{(k)} &1 \end{bmatrix}}}}, \end{aligned}

where \scalebox{1}{\mbox{\ensuremath{\displaystyle \Phi_{ij}^{(k)}(t)=-u_{ij}^{(k)}(\cos (t)-\cos(t_0))}}}.

Lemma

Lemma 4 (Existence of vibrational control)

PMRJ

For each intra-cluster network \mathcal{G}_k,k=1,\dots,r, there always exist vibrational control inputs such that P^{(k)}(t) is strictly lower triangular and satisfies P^{(k)}(t)\neq 0.

Lemma

Lemma 5

F92N

For the incidence matrices B and \hat B, there exists R={ \@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\scriptsize}{8}{9pt}}{8}{9pt}}{8}{9pt}}{8}{9pt}}{8}{9pt}}{8}{9pt}}{8}{9pt}\arraycolsep=0.4\arraycolsep\ensuremath{\begin{bmatrix} R_1&0\\ R_2&R_3\end{bmatrix}}} such that B^\top =R\hat B^\top, where

\begin{aligned} \scalebox{0.85}{\mbox{\ensuremath{\displaystyle R_1= {\rm blkdiag}\big((B^{(1)}_{{\rm intra}})^\top ((\hat B^{(1)}_{{\rm intra}})^\top)^\dagger,\dots,(B^{(r)}_{{\rm intra}})^\top ((\hat B^{(r)}_{{\rm intra}})^\top)^\dagger\big),}}} \end{aligned}

and { \@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\small}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}\arraycolsep=0.9\arraycolsep\ensuremath{\begin{matrix}R_2=B_{\rm inter}^\top (\hat B_{\rm intra} ^\top P_{\rm inter})^\dagger\end{matrix}}}, { \@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\small}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}\arraycolsep=0.9\arraycolsep\ensuremath{\begin{matrix}R_3= B_{\rm inter}^\top (\hat B_{\rm inter} P_{\rm intra})^\dagger\end{matrix}}} with { \@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\small}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}\arraycolsep=0.9\arraycolsep\ensuremath{\begin{matrix}P_{\rm intra}=I_n-\hat B_{\rm intra} \hat B_{\rm intra}^\dagger\end{matrix}}} and { \@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\@setfontsize{\small}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}}{10}{12pt}\arraycolsep=0.9\arraycolsep\ensuremath{\begin{matrix}P_{\rm inter}=I_n-\hat B_{\rm inter} \hat B_{\rm inter}^\dagger\end{matrix}}}.

Proof of Lemma 5

The expressions of R_2 and R_3 follow from [KR-IH:2021]. Recall that {B_{\rm intra}={\rm blkdiag}( B_{{\rm intra}}^{(1)},\dots, B_{{\rm intra}}^{(r))}}and {\hat B_{\rm intra}={\rm blkdiag}(\hat B^{(1)}_{{\rm intra}},\dots,\hat B^{(r)}_{{\rm intra}})}. Since {(B^{(k)}_{{\rm intra}})^\top } has full row rank, we have

\begin{aligned} (B^{(k)}_{{\rm intra}})^\top = (B^{(k)}_{{\rm intra}})^\top ((\hat B^{(k)}_{{\rm intra}})^\top)^\dagger(\hat B^{(k)}_{{\rm intra}})^\top, \end{aligned}

which completes the proof.

Lemma

Lemma 6 (Growth bound of perturbations)

O9F7

There exist some constants \bar \gamma_{k\ell}>0, k,\ell=1,\dots,r, such that, for any k, it holds that

\begin{aligned} \|\Phi^{-1} f^{(k)}_{\rm inter}( \Phi z, y)\|\le \sum\nolimits_{\ell=1}^{r}\bar \gamma_{k\ell} \|z_\ell\|.\hspace{20pt}\null \hfill\ensuremath{\triangle} \end{aligned}

Proof

AE9H

We construct the proof by considering a specific configuration of vibrational control.

Without loss of generality, we assume that each B_{{\rm intra}}^{(k)} and \hat B_{{\rm intra}}^{(k)} are ordered in a way such that their first \bar n_k=n_k-1 columns are the same, and the first two column are e_2-e_1 and e_3-e_2, respectively, where e_i is the ith column of I_{ n_k}. Then, there exists a matrix \tilde{R}^{(k)}_1 such that R_1^{(k)}={\begin{bmatrix} I_{\bar n_k}\\ \tilde{R}^{(k)}_1 \end{bmatrix}}.

Let U^{(k)}_{1}(t)=u e_1e_1^\top \sin(t) and U^{(k)}_{2}(t)=-u e_1e_1^\top\sin(t). It can be calculated that

\begin{aligned} &P^{(k)}(t)\\ &= -(\hat B^{(k)}_{{\rm intra}})^\top [B^{(k)}_{{\rm intra}}]^+ U^{(k)}_{1}(t)-[-B^{(k)}_{{\rm intra}}]^{+} U^{(k)}_{2}(t) R_1^{(k)} \\ &\scalebox{1}{\mbox{\ensuremath{\displaystyle =-(\hat B^{(k)}_{{\rm intra}})^\top u(e_1+e_2)e_1^\top \begin{bmatrix} I_{\bar n_k}\\ \tilde{R}^{(k)}_1\end{bmatrix}\sin(t)}}}\\ &\scalebox{1}{\mbox{\ensuremath{\displaystyle =-u(\hat B^{(k)}_{{\rm intra}})^\top (e_1+e_2) (e_1')^\top\sin(t) }}}, \end{aligned}

where the second equality follows from e_1^\top \scalebox{0.7}{\mbox{\ensuremath{\displaystyle \begin{bmatrix} I_{\bar n_k}\\ \tilde{R}^{(k)}_1 \end{bmatrix}}}}=(e_1')^\top with e_1' being the first column of I_{\bar n_k}. Then, it can be seen that the first two rows of P^{(k)}(t) are -u(e_2-e_1)^\top (e_1+e_2) (e_1')^\top \sin(t)=0, and -u(e_3-e_2)^\top (e_1+e_2) (e_1')^\top=(e'_1)^\top \sin(t), respectively. The rest rows of P^{(k)}(t) can only have non-zero values at the first column. Therefore, P^{(k)}(t) have the strictly lower triangular form and satisfies P^{(k)}(t)\neq 0, which completes the proof.

Source and licence

The text on this page is from arXiv:2204.00678v2 by Yuzhen Qin, Danielle S. Bassett and Fabio Pasqualetti, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2204/2204.00678.jsonl.