From 227927a0c7f8e99395ec7f597a9ccf92f9fd88ee Mon Sep 17 00:00:00 2001 From: Leslie Greengard Date: Wed, 21 Feb 2024 12:41:45 -0500 Subject: [PATCH] edited on Leslie's machine --- docs/math.rst | 126 ++++++++++++++++++++++++-------------------------- 1 file changed, 60 insertions(+), 66 deletions(-) diff --git a/docs/math.rst b/docs/math.rst index 944c8fb..729f724 100644 --- a/docs/math.rst +++ b/docs/math.rst @@ -1,6 +1,6 @@ Definitions =========== -Let $x_{j} = (x_{j,1}, x_{j,2}) \in \mathbb{R}^{2}$, $i=1,2,\ldots N$, denote a collection +Let $x_{j} = (x_{j,1}, x_{j,2}) \in \mathbb{R}^{2}$, $j=1,2,\ldots N$, denote a collection of source locations and let $x = (x_{1}, x_{2}) \in \mathbb{R}^{2}$ denote a target location. Unless stated otherwise, the gradients are with respect to the target variable $x$. @@ -9,26 +9,23 @@ Laplace FMM ************* The Laplace FMM comes in three varieties -* rfmm2d: Charges, dipole strengths, potentials, their +* rfmm2d: charges, dipole strengths, potentials, their gradients, and Hessians are all double precision -* lfmm2d: All of the above quantities are double complex -* cfmm2d: Same as lfmm2d except that there are no dipole orientation vectors, - and the dipole kernel is given by $1/(z - \xi_{j})$ where $z, \xi_{j}$ - are the target, and source locations viewed as points in the complex - plane given by - $z = x_{1} + i\cdot x_{2}$, and $\xi_{j} = x_{j,1} + i \cdot x_{j,2}$. - Moreover, the gradients, and Hessians - in this case are derivatives with respect to $z$, and in a slight abuse of - notation, we set $\frac{\mathrm{d}}{\mathrm{d} z} \log(\|z\|) = 1/z$ +* lfmm2d: all of the above quantities are double complex +* cfmm2d: all quantities are double complex, except there are no dipole orientation vectors. Instead, the dipole kernel is assumed to be of the form $1/(z - \xi_{j})$ where $z, \xi_{j}$ are the target and source locations, viewed as points in the complex + plane given by $z = x_{1} + i\cdot x_{2}$, and $\xi_{j} = x_{j,1} + i \cdot x_{j,2}$. + In this case, instead of gradients and Hessians, the first and second derivatives of + the potential with respect to $z$ are returned and, with a slight abuse of + notation, we set $\frac{\mathrm{d}}{\mathrm{d} z} \log(\|z\|) = 1/z$. Laplace FMM (rfmm2d) --------------------- Let $c_{j} \in \mathbb{R}$, $j=1,2,\ldots N$, -denote a collection of charge strengths, $v_{j} \in \mathbb{R}$, +denote a collection of charge strengths, let $v_{j} \in \mathbb{R}$, $j=1,2,\ldots N$, -denote a collection of dipole strengths, and $d_{j} \in \mathbb{R}^{2}$, +denote a collection of dipole strengths, and let $d_{j} \in \mathbb{R}^{2}$, $j=1,2,\ldots N$, denote the corresponding dipole orientation vectors. The Laplace FMM (rfmm2d) computes @@ -42,16 +39,16 @@ given by and its gradient $\nabla u(x) \in \mathbb{R}^{2}$ at the source and target locations. When $x=x_{j}$, the term -corresponding to $x_{j}$ is dropped from the sum. +corresponding to $x_{j}$ (the self-interaction) is omitted from the sum. Laplace FMM (lfmm2d) --------------------- Let $c_{j} \in \mathbb{C}$, $j=1,2,\ldots N$, -denote a collection of charge strengths, $v_{j} \in \mathbb{C}$, +denote a collection of charge strengths, let $v_{j} \in \mathbb{C}$, $j=1,2,\ldots N$, -denote a collection of dipole strengths, and $d_{j} \in \mathbb{R}^{2}$, +denote a collection of dipole strengths, and let $d_{j} \in \mathbb{R}^{2}$, $j=1,2,\ldots N$, denote the corresponding dipole orientation vectors. The Laplace FMM (rfmm2d) computes @@ -64,19 +61,22 @@ given by u(x) = \sum_{j=1}^{N} c_{j} \log{(\|x-x_{j}\|)} - v_{j} d_{j}\cdot \nabla \log{(\|x-x_{j}\|)} \, , and its gradient $\nabla u(x) \in \mathbb{C}^{2}$ -at the source and target locations. When $x=x_{j}$, the term -corresponding to $x_{j}$ is dropped from the sum. +at the source and target locations. (The outputs are complex-valued in this case because +the $\{ c_j \}$ and $\{ v_j \}$ are complex.) +When $x=x_{j}$, the term +corresponding to $x_{j}$ is omitted from the sum. Laplace FMM (cfmm2d) --------------------- Let $c_{j} \in \mathbb{C}$, $j=1,2,\ldots N$, -denote a collection of charge strengths, and $v_{j} \in \mathbb{C}$, +denote a collection of charge strengths, and let $v_{j} \in \mathbb{C}$, $j=1,2,\ldots N$, denote a collection of dipole strengths. -The Laplace FMM (cfmm2d) computes +For $z, \xi_{j} \in \mathbb{C}$, +the Laplace FMM (cfmm2d) computes the potential $u(z) \in \mathbb{C}$ given by @@ -85,24 +85,24 @@ given by u(z) = \sum_{j=1}^{N} c_{j} \log{(\|z-\xi_{j}\|)} - \frac{v_{j}}{z-\xi_{j}} \, , -and its gradient $\frac{\mathrm{d}}{\mathrm{d}z} u(z) \in +and its +derivatives $u'(z), u''(z) \in \mathbb{C}$ at the source and target locations. When $z=\xi_{j}$, the term -corresponding to $\xi_{j}$ is dropped from the sum. Here -$z = x(1) + i x(2)$, and $\xi_{j} = x_{j,1} + ix_{j,2}$, are the complex -numbers corresponding to $x$ and $x_{j}$ respectively. +corresponding to $\xi_{j}$ is omitted from the sum. +As noted above, we define $\frac{\mathrm{d}}{\mathrm{d} z} \log(\|z\|) = 1/z$. Helmholtz FMM ************* Let $c_{j} \in \mathbb{C}$, $j=1,2,\ldots N$, -denote a collection of charge strengths, $v_{j} \in \mathbb{C}$, +denote a collection of charge strengths, let $v_{j} \in \mathbb{C}$, $j=1,2,\ldots N$, -denote a collection of dipole strengths, and $d_{j} \in \mathbb{R}^{2}$, +denote a collection of dipole strengths, and let $d_{j} \in \mathbb{R}^{2}$, $j=1,2,\ldots N$, denote the corresponding dipole orientation vectors. -Let $k\in\mathbb{C}$ denote the wave number or the Helmholtz -parameter. +Let $k\in\mathbb{C}$ denote the wave number (the Helmholtz +parameter). The Helmholtz FMM computes the potential $u(x) \in \mathbb{C}$ @@ -116,27 +116,27 @@ given by and its gradient $\nabla u(x) \in \mathbb{C}^{2}$ at the source and target locations, where $H_{0}^{(1)}$ is the Hankel function of the first kind of order $0$. When $x=x_{j}$, the term -corresponding to $x_{j}$ is dropped from the sum. +corresponding to $x_{j}$ is omitted from the sum. Biharmonic FMM *************** Let $c_{j} = (c_{j,1}, c_{j,2})\in \mathbb{C}^2$, $j=1,2,\ldots N$, -denote a collection of charge strengths, and +denote a collection of charge strengths, and let $v_{j} = (v_{j,1}, v_{j,2}, v_{j,3}) \in \mathbb{C}^{3}$, $j=1,2,\ldots N$, denote a collection of dipole strengths. -The Biharmonic FMM computes -the potential $u(x)$ and its `gradient` = +For $z, \xi_j \in \mathbb{C}$, the biharmonic FMM computes +the potential $u(z)$ and its `gradient` = $(P_{z} \frac{\mathrm{d}}{\mathrm{d}z}, P_{\overline{z}} \frac{\mathrm{d}}{\mathrm{d}z}, \frac{\mathrm{d}}{\mathrm{d}\overline{z}})$ given by .. math:: :label: biharm_nbody - u(z) &= \sum_{j=1}^{N} c_{j,1} \log{\|z - \xi_{j}\|} + + u(z) &= \sum_{j=1}^{N} 2 \, c_{j,1} \log{\|z - \xi_{j}\|} + c_{j,2} \frac{z-\xi_{j}}{\overline{z-\xi_{j}}} + \frac{v_{j,1}}{z - \xi_{j}} + \frac{v_{j,3}}{\overline{z-\xi_{j}}} + v_{j,2} \frac{z - \xi_{j}}{(\overline{z-\xi_{j}})^2} \, , \\ @@ -154,16 +154,12 @@ given by at the source and target locations. When $z=\xi_{j}$, the term -corresponding to $\xi_{j}$ is dropped from the sum. Here -$z = x(1) + i x(2)$, and $\xi_{j} = x_{j}(1) + ix_{j}(2)$, are the complex -numbers corresponding to $x$ and $x_{j}$ respectively. -When $x=x_{j}$, the term -corresponding to $x_{j}$ is dropped from the sum. -In the above, $(P_{z} \frac{\mathrm{d}}{\mathrm{d}z}$ -is the part of the gradient $\frac{\mathrm{d}}{\mathrm{d} z}$ -which is purely a function of $z$, and similarly -$(P_{\overline{z}} \frac{\mathrm{d}}{\mathrm{d}z}$ -is the part of the gradient $\frac{\mathrm{d}}{\mathrm{d} z}$ +corresponding to $\xi_{j}$ is omitted from the sum. +The expression $P_{z} \frac{\mathrm{d}}{\mathrm{d}z}$ +denotes the component of the derivative $\frac{\mathrm{d}}{\mathrm{d} z}$ +which is purely a function of $z$, and the expression +$P_{\overline{z}} \frac{\mathrm{d}}{\mathrm{d}z}$ +denotes the component of the derivative $\frac{\mathrm{d}}{\mathrm{d} z}$ which is purely a function of $\overline{z}$. @@ -177,26 +173,24 @@ Green's function given by G^{\textrm{mbh}}(x,y) = \frac{1}{2\pi \beta^2}\left(K_{0}(\beta \|x-y \|) - \log{\|x-y\|}\right) where $K_{0}$ is the modified Bessel function of order $0$, and $\beta$ is the -Modified Biharmonic wavenumber. +modified biharmonic wavenumber. Let $c_{j} \in \mathbb{R}$, -denote a collection of charge strengths, +denote a collection of charge strengths, let $v_{j} \in \mathbb{R}$, -denote a collection of dipole strengths, +denote a collection of dipole strengths, let $d_{j} = (d_{j,1}, d_{j,2})$ denote a collection -of dipole vectors, +of dipole vectors, let $q_{j} \in \mathbb{R}$ denote a collection of -quadrupole strengths, +quadrupole strengths, let $w_{j} = (w_{j,1}, w_{j,2}, w_{j,3}) \in \mathbb{R}^{3}$, -denote a collection of quadrupole orientation vectors, -$o_{j}$ denote a collection of octopole strengths, and +denote a collection of quadrupole three-vectors, let +$o_{j}$ denote a collection of octopole strengths, and let $p_{j} = (p_{j,1}, p_{j,2}, p_{j,3}, p_{j,4}) \in \mathbb{R}^{4}$, -denote a collection of octopole strengths. For all the vectors, -$j=1,2,\ldots N$. +denote a collection of octopole four-vectors. -The Modified Biharmonic FMM computes -the potential $u(x)\in \mathbb{R}$ +The modified biharmonic FMM computes the potential $u(x)\in \mathbb{R}$ given by .. math:: @@ -214,7 +208,7 @@ given by and its gradients $\nabla u(x) \in \mathbb{R}^{2}$ at the source and target locations. When $x=x_{j}$, the term -corresponding to $x_{j}$ is dropped from the sum. +corresponding to $x_{j}$ is omitted from the sum. Stokes FMM ************ @@ -230,15 +224,15 @@ Let $G^{\textrm{stok}}(x,y)$ denote the Stokeslet given by -\log{\|x-y \|} + \frac{(x_{2}-y_{2})^2}{\|x-y \|^2} \end{bmatrix} \, , -$P^{\textrm{stok}}(x,y)$ denote the associated pressure tensor +and let $P^{\textrm{stok}}(x,y)$ denote the associated pressure tensor .. math:: - P(x,y) = \frac{1}{\|x-y \|^2}\begin{bmatrix} + P^{\textrm{stok}}(x,y) = \frac{1}{\|x-y \|^2}\begin{bmatrix} (x_{1}-y_{1}) & (x_{2}-y_{2}) - \end{bmatrix} \, , + \end{bmatrix} \, . -$T^{\textrm{stok}}(x,y)$ denote the Stresslet whose action of a vector $v$ +Let $T^{\textrm{stok}}(x,y)$ denote the Stresslet whose action on a vector $v$ is given by .. math:: @@ -248,20 +242,20 @@ is given by (x_{1}-y_{1})(x_{2}-y_{2}) & (x_{2} - y_{2})^2 \end{bmatrix} \, , -and finally let $\Pi^{\textrm{stok}} (x,y)$ denote its associated pressure tensor given by +and let $\Pi^{\textrm{stok}} (x,y)$ denote its associated pressure tensor given by .. math:: v\cdot \Pi(x,y)^{\textrm{stok}} = -\frac{v}{\|x-y \|^2} + \frac{2 v \cdot(x-y)}{\|x-y \|^4} \begin{bmatrix} (x_{1}-y_{1}) \\ (x_{2}-y_{2}) - \end{bmatrix} + \end{bmatrix} \, . Let $c_{j} \in \mathbb{R}^2$, $j=1,2,\ldots N$, -denote a collection of Stokeslet strengths, $v_{j} \in \mathbb{R}^2$, +denote a collection of Stokeslet strengths, let $v_{j} \in \mathbb{R}^2$, $j=1,2,\ldots N$, -denote a collection of Stresslet strengths, and $d_{j} \in \mathbb{R}^{2}$, +denote a collection of Stresslet strengths, and let $d_{j} \in \mathbb{R}^{2}$, $j=1,2,\ldots N$, denote the corresponding Stresslet orientation vectors. The Stokes FMM computes @@ -277,7 +271,7 @@ the potential $u(x) \in \mathbb{R}^2$, its gradient $\nabla u(x) \in \Pi^{\textrm{stok}}(x,x_{j}) \cdot v_{j}^{T} at the source and target locations. When $x=x_{j}$, the term -corresponding to $x_{j}$ is dropped from the sum. +corresponding to $x_{j}$ is omitted from the sum. Vectorized versions ******************* @@ -310,8 +304,8 @@ at the source and target locations. In double precision arithmetic, two numbers which are within machine precision of each other cannot be distinguished. In order to account for this, suppose that the sources - and targets are contained in a cube with side length $L$, then + and targets are contained in a square with side length $L$, then for all $x$ such that $\| x-x_{j} \| \leq L \varepsilon_{\textrm{mach}}$, - the term corresponding to $x_{j}$ is dropped from the sum. + the term corresponding to $x_{j}$ is omitted from the sum. Here $\varepsilon_{\textrm{mach}} = 2^{-52}$ is machine precision.