1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
use crate::allocators::{BiDimAllocator, DimAllocator};
use crate::connectivity::Connectivity;
use crate::nalgebra::MatrixViewMut;
use crate::{Real, SmallDim};
use fenris_geometry::AxisAlignedBoundingBox;
use fenris_optimize::newton::NewtonSettings;
use nalgebra::allocator::Allocator;
use nalgebra::OPoint;
use nalgebra::{DVectorView, DVectorViewMut, DimName, Dyn};
use nalgebra::{DefaultAllocator, DimMin, OMatrix, OVector, Scalar, U1};
use num::Zero;
use numeric_literals::replace_float_literals;
use std::error::Error;
use std::fmt::Debug;
mod hexahedron;
mod quadrilateral;
mod segment;
mod tetrahedron;
mod triangle;
pub use hexahedron::*;
pub use quadrilateral::*;
pub use segment::*;
pub use tetrahedron::*;
pub use triangle::*;
pub trait ReferenceFiniteElement<T>
where
T: Scalar,
DefaultAllocator: DimAllocator<T, Self::ReferenceDim>,
{
type ReferenceDim: SmallDim;
/// Returns the number of nodes in the element.
fn num_nodes(&self) -> usize;
/// Evaluates each basis function at the given reference coordinates. The result is given
/// in a row vector where each entry is the value of the corresponding basis function.
///
/// TODO: Document that it should panic if the result does not have exactly the correct
/// number of columns (==nodes)
fn populate_basis(&self, basis_values: &mut [T], reference_coords: &OPoint<T, Self::ReferenceDim>);
/// Given nodal weights, construct a matrix whose columns are the
/// gradients of each shape function in the element.
fn populate_basis_gradients(
&self,
basis_gradients: MatrixViewMut<T, Self::ReferenceDim, Dyn>,
reference_coords: &OPoint<T, Self::ReferenceDim>,
);
}
/// Reference finite elements with a number of nodes fixed at compile-time.
///
/// This is essentially equivalent to the old ReferenceFiniteElement trait, and exists only
/// to make existing code written with the old API in mind still work without having to
/// immediately re-write everything. It will be removed once everything has been ported over
/// to the new API (`ReferenceFiniteElement`).
///
/// TODO: Remove this trait once all elements and tests have been ported over to
/// `ReferenceFiniteElement`
pub trait FixedNodesReferenceFiniteElement<T>
where
T: Scalar,
DefaultAllocator: DimAllocator<T, Self::ReferenceDim>
+ Allocator<T, U1, Self::NodalDim>
+ Allocator<T, Self::ReferenceDim, Self::NodalDim>,
{
type ReferenceDim: SmallDim;
type NodalDim: SmallDim;
/// Evaluates each basis function at the given reference coordinates. The result is given
/// in a row vector where each entry is the value of the corresponding basis function.
fn evaluate_basis(&self, reference_coords: &OPoint<T, Self::ReferenceDim>) -> OMatrix<T, U1, Self::NodalDim>;
/// Given nodal weights, construct a matrix whose columns are the
/// gradients of each shape function in the element.
fn gradients(
&self,
reference_coords: &OPoint<T, Self::ReferenceDim>,
) -> OMatrix<T, Self::ReferenceDim, Self::NodalDim>;
}
/// Implements `ReferenceFiniteElement` for any element that implements
/// `FixedNodesReferenceFiniteElement`.
///
/// This could be done with a blanket impl, but this prevents
/// us from implementing `ReferenceFiniteElement` in some generic contexts, so we instead
/// invoke this macro for each element that this applies to. This is a temporary solution
/// that we use because it would take some work reworking the tests in order to remove the
/// `FixedNodesReferenceFiniteElement` trait altogether.
macro_rules! impl_reference_finite_element_for_fixed {
($element:ty) => {
impl<T> ReferenceFiniteElement<T> for $element
where
T: Scalar,
$element: FixedNodesReferenceFiniteElement<T>,
DefaultAllocator: BiDimAllocator<
T,
<$element as FixedNodesReferenceFiniteElement<T>>::NodalDim,
<$element as FixedNodesReferenceFiniteElement<T>>::ReferenceDim,
>,
{
type ReferenceDim = <Self as FixedNodesReferenceFiniteElement<T>>::ReferenceDim;
fn num_nodes(&self) -> usize {
use nalgebra::DimName;
<Self as FixedNodesReferenceFiniteElement<T>>::NodalDim::dim()
}
fn populate_basis(&self, result: &mut [T], reference_coords: &OPoint<T, Self::ReferenceDim>) {
let basis_values = <Self as crate::element::FixedNodesReferenceFiniteElement<T>>::evaluate_basis(
self,
reference_coords,
);
result.clone_from_slice(&basis_values.as_slice());
}
fn populate_basis_gradients(
&self,
mut result: nalgebra::MatrixViewMut<T, Self::ReferenceDim, nalgebra::Dyn>,
reference_coords: &OPoint<T, Self::ReferenceDim>,
) {
let gradients =
<Self as crate::element::FixedNodesReferenceFiniteElement<T>>::gradients(self, reference_coords);
result.copy_from(&gradients);
}
}
};
}
impl_reference_finite_element_for_fixed!(Tri3d2Element<T>);
impl_reference_finite_element_for_fixed!(Tri6d2Element<T>);
impl_reference_finite_element_for_fixed!(Quad4d2Element<T>);
impl_reference_finite_element_for_fixed!(Quad9d2Element<T>);
impl_reference_finite_element_for_fixed!(Segment2d1Element<T>);
impl_reference_finite_element_for_fixed!(Segment2d2Element<T>);
impl_reference_finite_element_for_fixed!(Tet4Element<T>);
impl_reference_finite_element_for_fixed!(Hex8Element<T>);
impl_reference_finite_element_for_fixed!(Hex27Element<T>);
impl_reference_finite_element_for_fixed!(Hex20Element<T>);
impl_reference_finite_element_for_fixed!(Tri3d3Element<T>);
impl_reference_finite_element_for_fixed!(Tet10Element<T>);
impl_reference_finite_element_for_fixed!(Tet20Element<T>);
pub trait FiniteElement<T>: ReferenceFiniteElement<T>
where
T: Scalar,
DefaultAllocator: BiDimAllocator<T, Self::GeometryDim, Self::ReferenceDim>,
{
type GeometryDim: SmallDim;
/// Compute the Jacobian of the transformation from the reference element to the given
/// element at the given reference coordinates.
fn reference_jacobian(
&self,
reference_coords: &OPoint<T, Self::ReferenceDim>,
) -> OMatrix<T, Self::GeometryDim, Self::ReferenceDim>;
/// Maps reference coordinates to physical coordinates in the element.
fn map_reference_coords(&self, reference_coords: &OPoint<T, Self::ReferenceDim>) -> OPoint<T, Self::GeometryDim>;
/// The diameter of the finite element.
///
/// The diameter of a finite element is defined as the largest distance between any two
/// points in the element, i.e.
/// h = min |x - y| for x, y in K
/// where K is the element and h is the diameter.
fn diameter(&self) -> T;
}
/// TODO: Do we *really* need the Debug bound?
pub trait ElementConnectivity<T>: Debug + Connectivity
where
T: Scalar,
DefaultAllocator: BiDimAllocator<T, Self::GeometryDim, Self::ReferenceDim>,
{
type Element: FiniteElement<T, GeometryDim = Self::GeometryDim, ReferenceDim = Self::ReferenceDim>;
type GeometryDim: SmallDim;
type ReferenceDim: SmallDim;
/// Returns the finite element associated with this connectivity.
///
/// The vertices passed in should be the collection of *all* vertices in the mesh.
fn element(&self, all_vertices: &[OPoint<T, Self::GeometryDim>]) -> Option<Self::Element>;
/// TODO: Move this out of the trait itself?
fn populate_element_variables<'a, SolutionDim>(
&self,
mut u_local: MatrixViewMut<T, SolutionDim, Dyn>,
u_global: impl Into<DVectorView<'a, T>>,
) where
T: Zero,
SolutionDim: DimName,
{
let u_global = u_global.into();
let indices = self.vertex_indices();
let sol_dim = SolutionDim::dim();
for (i_local, i_global) in indices.iter().enumerate() {
u_local
.index_mut((.., i_local))
.copy_from(&u_global.index((sol_dim * i_global..sol_dim * i_global + sol_dim, ..)));
}
}
}
/// A finite element whose geometry dimension and reference dimension coincide.
pub trait VolumetricFiniteElement<T>: FiniteElement<T, ReferenceDim = <Self as FiniteElement<T>>::GeometryDim>
where
T: Scalar,
DefaultAllocator: BiDimAllocator<T, Self::GeometryDim, Self::ReferenceDim>,
{
}
impl<T, E> VolumetricFiniteElement<T> for E
where
T: Scalar,
E: FiniteElement<T, ReferenceDim = <Self as FiniteElement<T>>::GeometryDim>,
DefaultAllocator: BiDimAllocator<T, Self::GeometryDim, Self::ReferenceDim>,
{
}
pub trait SurfaceFiniteElement<T>: FiniteElement<T>
where
T: Scalar,
DefaultAllocator: BiDimAllocator<T, Self::GeometryDim, Self::ReferenceDim>,
{
/// Compute the normal at the point associated with the provided reference coordinate.
fn normal(&self, xi: &OPoint<T, Self::ReferenceDim>) -> OVector<T, Self::GeometryDim>;
}
// TODO: Move these?
pub type ElementForConnectivity<T, Connectivity> = <Connectivity as ElementConnectivity<T>>::Element;
pub type ConnectivityGeometryDim<T, Conn> = <Conn as ElementConnectivity<T>>::GeometryDim;
pub type ConnectivityReferenceDim<T, Conn> = <Conn as ElementConnectivity<T>>::ReferenceDim;
pub type ElementGeometryDim<T, Element> = <Element as FiniteElement<T>>::GeometryDim;
/// Linear basis function on the interval [-1, 1].
///
///`alpha == -1` denotes the basis function associated with the node at `x == -1`,
/// and `alpha == 1` for `x == 1`.
#[replace_float_literals(T::from_f64(literal).unwrap())]
#[inline(always)]
fn phi_linear_1d<T>(alpha: T, xi: T) -> T
where
T: Real,
{
(1.0 + alpha * xi) / 2.0
}
/// Gradient for the linear basis function on the interval [-1, 1].
///
/// See `phi_linear_1d` for the meaning of `alpha`.
#[replace_float_literals(T::from_f64(literal).unwrap())]
#[inline(always)]
fn phi_linear_1d_grad<T>(alpha: T) -> T
where
T: Real,
{
alpha / 2.0
}
/// Quadratic basis function on the interval [-1, 1].
///
/// `alpha == -1` denotes the basis function associated with the node at `x == -1`,
/// `alpha == 0` denotes the basis function associated with the node at `x == 0`,
/// and `alpha == 1` for `x == 1`.
#[replace_float_literals(T::from_f64(literal).unwrap())]
#[inline(always)]
fn phi_quadratic_1d<T>(alpha: T, xi: T) -> T
where
T: Real,
{
// The compiler should hopefully be able to use constant propagation to
// precompute all expressions involving constants and alpha
let alpha2 = alpha * alpha;
let xi2 = xi * xi;
(3.0 / 2.0 * alpha2 - 1.0) * xi2 + 0.5 * alpha * xi + 1.0 - alpha2
}
/// Derivative of quadratic basis function on the interval [-1, 1].
///
/// `alpha == -1` denotes the basis function associated with the node at `x == -1`,
/// `alpha == 0` denotes the basis function associated with the node at `x == 0`,
/// and `alpha == 1` for `x == 1`.
#[replace_float_literals(T::from_f64(literal).unwrap())]
#[inline(always)]
fn phi_quadratic_1d_grad<T>(alpha: T, xi: T) -> T
where
T: Real,
{
// The compiler should hopefully be able to use constant propagation to
// precompute all expressions involving constants and alpha
let alpha2 = alpha * alpha;
2.0 * (3.0 / 2.0 * alpha2 - 1.0) * xi + 0.5 * alpha
}
/// Maps physical coordinates `x` to reference coordinates `xi` by solving the equation
/// x - T(xi) = 0 using Newton's method.
///
pub fn map_physical_coordinates<T, Element, GeometryDim>(
element: &Element,
x: &OPoint<T, GeometryDim>,
) -> Result<OPoint<T, GeometryDim>, Box<dyn Error>>
where
T: Real,
Element: FiniteElement<T, GeometryDim = GeometryDim, ReferenceDim = GeometryDim>,
GeometryDim: DimName + DimMin<GeometryDim, Output = GeometryDim>,
DefaultAllocator: DimAllocator<T, GeometryDim>,
{
use fenris_optimize::calculus::VectorFunctionBuilder;
use fenris_optimize::newton::newton;
let f = VectorFunctionBuilder::with_dimension(GeometryDim::dim())
.with_function(|f, xi| {
// Need to create stack-allocated xi
let xi = OPoint::from(
xi.generic_view((0, 0), (GeometryDim::name(), U1::name()))
.clone_owned(),
);
f.copy_from(&(element.map_reference_coords(&xi).coords - &x.coords));
})
.with_jacobian_solver(
|sol: &mut DVectorViewMut<T>, xi: &DVectorView<T>, rhs: &DVectorView<T>| {
let xi = OPoint::from(
xi.generic_view((0, 0), (GeometryDim::name(), U1::name()))
.clone_owned(),
);
let j = element.reference_jacobian(&xi);
let lu = j.full_piv_lu();
sol.copy_from(rhs);
if lu.solve_mut(sol) {
Ok(())
} else {
Err(Box::<dyn Error>::from(
"LU decomposition failed. Jacobian not invertible?",
))
}
},
);
// We solve the equation T(xi) = x, i.e. we seek reference coords xi such that when
// transformed to physical coordinates yield x. We note here that what Newton's method solves
// is the system T(xi) - x = 0, which can be re-interpreted as finding xi such that
// T_trans(xi) = 0, with T_trans(xi) = T(xi) - x.
// This means that we seek xi such that the translated transformation transforms xi to
// the zero vector. Since x should be a point in the element, it follows that we can expect
// the diameter of the element to give us a representative scale of the "size" of x,
// so we can construct our convergence criterion as follows:
// ||T(x_i) - x|| <= eps * diameter
// with eps some small constant.
let settings = NewtonSettings {
// Note: Max iterations is entirely random at this point. Should of course
// be made configurable. TODO
max_iterations: Some(20),
// TODO: eps is here hard-coded without respect to the type T, so it will not be appropriate
// across e.g. different floating point types. Fix this!
tolerance: T::from_f64(1e-12).unwrap() * element.diameter(),
};
let mut xi = OVector::<T, GeometryDim>::zeros();
let mut f_val = OVector::<T, GeometryDim>::zeros();
let mut dx = OVector::<T, GeometryDim>::zeros();
// Because we cannot prove to the compiler that the strides of `OVector<T, GeometryDim>`
// are compatible (in a `DimEq` sense) without nasty additional trait bounds,
// we first take slices of the vectors so that the stride is dynamic. At this point,
// it is known that `DimEq<Dyn, U1>` works, so we can use it with `newton`,
// `which expects `Into<DMatrixViewMut<T>>`.
macro_rules! slice {
($e:expr) => {
$e.generic_view_with_steps_mut((0, 0), (GeometryDim::name(), U1::name()), (0, 0))
};
}
newton(f, &mut slice!(xi), &mut slice!(f_val), &mut slice!(dx), settings)?;
Ok(OPoint::from(xi))
}
/// Projects physical coordinates `x` to reference coordinates `xi` by solving the equation
/// x - T(xi) = 0 using a generalized form of Newton's method.
///
/// Unlike `map_physical_coordinates`, this method is also applicable to e.g. surface finite
/// elements, in which the reference dimension and geometry dimension differ.
///
/// The method panics if `ReferenceDim` is greater than `GeometryDim`.
///
/// TODO: This method is totally misleading as is, because it does not take into account
/// the geometry of the reference element in reference coordinates, so it will happily
/// return points outside of the reference geometry.
#[allow(non_snake_case)]
pub fn project_physical_coordinates<T, Element>(
element: &Element,
x: &OPoint<T, Element::GeometryDim>,
) -> Result<OPoint<T, Element::ReferenceDim>, Box<dyn Error>>
where
T: Real,
Element: FiniteElement<T>,
Element::ReferenceDim: DimName + DimMin<Element::ReferenceDim, Output = Element::ReferenceDim>,
DefaultAllocator: BiDimAllocator<T, Element::GeometryDim, Element::ReferenceDim>,
{
assert!(
Element::ReferenceDim::dim() <= Element::GeometryDim::dim(),
"ReferenceDim must be smaller or equal to GeometryDim."
);
// See comments in `map_physical_coordinates` for why this is a reasonable tolerance.
let tolerance = T::from_f64(1e-12).unwrap() * element.diameter();
// We wish to solve the system
// f(xi) - x = 0,
// but xi and x have different dimensions. To overcome this difficulty, we use a modified
// version of Newton's method in which we solve the normal equations for the Jacobian equation
// instead of solving the Jacobian system directly (which is underdetermined in this case).
//
// Our stopping condition is based on the optimality condition for the
// least-squares problem min || x - f(xi) ||, whose geometrical interpretation at the
// minimum is exactly that of a projection onto the surface.
let x = &x.coords;
let mut xi = OPoint::<T, Element::ReferenceDim>::origin();
let mut f = element.map_reference_coords(&xi).coords;
let mut j = element.reference_jacobian(&xi);
let mut jT = j.transpose();
let mut iter = 0;
// TODO: Do we need to alter the tolerance due to the jT term?
while (&jT * (&f - x)).norm() > tolerance {
let jTj = &jT * j;
let lu = jTj.full_piv_lu();
let rhs = -jT * (&f - x);
if let Some(sol) = lu.solve(&rhs) {
xi += sol;
} else {
return Err(Box::from(
"LU decomposition failed. Normal equation for Jacobian not invertible?",
));
}
f = element.map_reference_coords(&xi).coords;
j = element.reference_jacobian(&xi);
jT = j.transpose();
iter += 1;
// TODO: Should better handle degenerate/problematic cases. For now we just want to
// avoid infinite loops
if iter > 1000 {
eprintln!("Exceeded 1000 iterations for project_physical_coordinates");
}
}
Ok(OPoint::from(xi))
}
/// The result of a [`ClosestPointInElement`] query.
#[derive(Debug, Clone, PartialEq)]
pub enum ClosestPoint<T, D>
where
T: Scalar,
D: DimName,
DefaultAllocator: Allocator<T, D>,
{
/// The point is inside the element.
InElement(OPoint<T, D>),
/// The closest point in the element to the query point.
ClosestPoint(OPoint<T, D>),
}
impl<T, D> ClosestPoint<T, D>
where
T: Scalar,
D: DimName,
DefaultAllocator: Allocator<T, D>,
{
pub fn point(&self) -> &OPoint<T, D> {
match self {
ClosestPoint::InElement(point) | ClosestPoint::ClosestPoint(point) => point,
}
}
}
/// A finite element you can query for the closest point to an arbitrary point.
pub trait ClosestPointInElement<T: Scalar>: FiniteElement<T>
where
DefaultAllocator: BiDimAllocator<T, Self::GeometryDim, Self::ReferenceDim>,
{
fn closest_point(&self, p: &OPoint<T, Self::GeometryDim>) -> ClosestPoint<T, Self::ReferenceDim>;
}
/// A finite element that can be queried for its bounding box.
pub trait BoundsForElement<T: Scalar>: FiniteElement<T>
where
DefaultAllocator: BiDimAllocator<T, Self::GeometryDim, Self::ReferenceDim>,
{
fn element_bounds(&self) -> AxisAlignedBoundingBox<T, Self::GeometryDim>;
}