matterviz 0.3.4 → 0.3.6

package/dist/src/lib/math.ts

@@ -0,0 +1,1081 @@
+import type { LatticeParams, Pbc } from '$lib/structure/index'
+
+export type Vec2 = [number, number]
+export type Vec3 = [number, number, number]
+export type Vec4 = [number, number, number, number]
+export type Vec9 = [number, number, number, number, number, number, number, number, number]
+export type Point2D = { x: number; y: number }
+export type Point3D = Point2D & { z: number }
+export type Matrix3x3 = [Vec3, Vec3, Vec3]
+export type Matrix4x4 = [Vec4, Vec4, Vec4, Vec4]
+export type NdVector = number[]
+
+export const is_finite_vec3_like = (
+  values: ArrayLike<unknown> | undefined,
+): values is ArrayLike<number> => {
+  if (values?.length !== 3) return false
+  return [0, 1, 2].every(
+    (idx) => typeof values[idx] === `number` && Number.isFinite(values[idx]),
+  )
+}
+
+export const finite_vec3_from_values = (
+  values: ArrayLike<unknown> | undefined,
+): Vec3 | undefined => {
+  if (!is_finite_vec3_like(values)) return undefined
+  return [values[0], values[1], values[2]]
+}
+
+// Column-major 4x4 matrix as flat 16-element tuple (for Three.js/WebGL)
+// oxfmt-ignore
+export type Matrix4Tuple = [
+  number, number, number, number,
+  number, number, number, number,
+  number, number, number, number,
+  number, number, number, number,
+]
+
+// Generate all k-element combinations from an array.
+export function combinations<T>(arr: T[], k: number): T[][] {
+  if (k === 0) return [[]]
+  if (arr.length < k) return []
+  const [first, ...rest] = arr
+  return [
+    ...combinations(rest, k - 1).map((combo) => [first, ...combo]),
+    ...combinations(rest, k),
+  ]
+}
+
+export const LOG_EPS = 1e-9
+export const EPS = 1e-10
+export const RAD_TO_DEG = 180 / Math.PI
+export const DEG_TO_RAD = Math.PI / 180
+const MAX_MIN_IMAGE_CANDIDATES = 100_000
+
+export const to_degrees = (radians: number): number => radians * RAD_TO_DEG
+export const to_radians = (degrees: number): number => degrees * DEG_TO_RAD
+
+// Calculate all lattice parameters in a single efficient pass
+export function calc_lattice_params(matrix: Matrix3x3): LatticeParams & { volume: number } {
+  const [a_vec, b_vec, c_vec] = matrix
+
+  // Calculate vector lengths (lattice parameters a, b, c)
+  const a = Math.hypot(a_vec[0], a_vec[1], a_vec[2])
+  const b = Math.hypot(b_vec[0], b_vec[1], b_vec[2])
+  const c = Math.hypot(c_vec[0], c_vec[1], c_vec[2])
+
+  // Calculate volume using scalar triple product
+  const volume = Math.abs(
+    a_vec[0] * (b_vec[1] * c_vec[2] - b_vec[2] * c_vec[1]) +
+      a_vec[1] * (b_vec[2] * c_vec[0] - b_vec[0] * c_vec[2]) +
+      a_vec[2] * (b_vec[0] * c_vec[1] - b_vec[1] * c_vec[0]),
+  )
+
+  // Calculate dot products for angles (only once each)
+  const dot_ab = a_vec[0] * b_vec[0] + a_vec[1] * b_vec[1] + a_vec[2] * b_vec[2]
+  const dot_ac = a_vec[0] * c_vec[0] + a_vec[1] * c_vec[1] + a_vec[2] * c_vec[2]
+  const dot_bc = b_vec[0] * c_vec[0] + b_vec[1] * c_vec[1] + b_vec[2] * c_vec[2]
+
+  // Convert to angles in degrees
+  const alpha = Math.acos(dot_bc / (b * c)) * RAD_TO_DEG
+  const beta = Math.acos(dot_ac / (a * c)) * RAD_TO_DEG
+  const gamma = Math.acos(dot_ab / (a * b)) * RAD_TO_DEG
+
+  return { a, b, c, alpha, beta, gamma, volume }
+}
+
+export const scale = <T extends NdVector>(vec: T, factor: number): T =>
+  vec.map((component) => component * factor) as T
+
+export const euclidean_dist = (vec1: NdVector, vec2: NdVector): number => {
+  if (vec1.length !== vec2.length) {
+    throw new Error(`Vectors must be of same length`)
+  }
+  return Math.hypot(...vec1.map((x, idx) => x - vec2[idx]))
+}
+
+const vec3_norm_sq = (vec: Vec3): number => vec[0] ** 2 + vec[1] ** 2 + vec[2] ** 2
+
+// Exact minimum-image displacement for row-vector lattices.
+// Rounded fractional wrapping is only approximate for highly skewed cells, so
+// we use it as a starting guess and then search the small set of shifts that
+// can still beat that Cartesian radius.
+export function min_image_displacement(
+  from: Vec3,
+  to: Vec3,
+  lattice_matrix: Matrix3x3,
+  converters?: LatticeConverters,
+  pbc: Pbc = [true, true, true],
+): Vec3 {
+  const { cart_to_frac, frac_to_cart, reciprocal_axis_norms } =
+    converters ?? create_lattice_converters(lattice_matrix)
+  const frac_from = cart_to_frac(from)
+  const frac_to = cart_to_frac(to)
+  const frac_diff: Vec3 = [
+    frac_to[0] - frac_from[0],
+    frac_to[1] - frac_from[1],
+    frac_to[2] - frac_from[2],
+  ]
+  const wrapped_frac_diff: Vec3 = [
+    pbc[0] ? frac_diff[0] - Math.round(frac_diff[0]) : frac_diff[0],
+    pbc[1] ? frac_diff[1] - Math.round(frac_diff[1]) : frac_diff[1],
+    pbc[2] ? frac_diff[2] - Math.round(frac_diff[2]) : frac_diff[2],
+  ]
+
+  let best_displacement = frac_to_cart(wrapped_frac_diff)
+  let best_dist_sq = vec3_norm_sq(best_displacement)
+  const search_radius = Math.sqrt(best_dist_sq) + EPS
+  const candidate_shift_ranges = ([0, 1, 2] as const).map((axis_idx) => {
+    if (!pbc[axis_idx]) return [0, 0] as const
+    const axis_bound = reciprocal_axis_norms[axis_idx] * search_radius
+    return [
+      Math.ceil(-frac_diff[axis_idx] - axis_bound),
+      Math.floor(-frac_diff[axis_idx] + axis_bound),
+    ] as const
+  })
+  let candidate_count = 1
+  for (const [shift_min, shift_max] of candidate_shift_ranges) {
+    candidate_count *= shift_max - shift_min + 1
+    if (candidate_count > MAX_MIN_IMAGE_CANDIDATES) {
+      throw new Error(
+        `Minimum-image search would test >${MAX_MIN_IMAGE_CANDIDATES} candidates ` +
+          `for lattice ${JSON.stringify(lattice_matrix)}; reciprocal norms=` +
+          `${JSON.stringify(reciprocal_axis_norms)} ranges=${JSON.stringify(candidate_shift_ranges)}`,
+      )
+    }
+  }
+  const [[i_min, i_max], [j_min, j_max], [k_min, k_max]] = candidate_shift_ranges
+
+  // Only test integer shifts that reciprocal-space bounds say could still win.
+  for (let ii = i_min; ii <= i_max; ii++) {
+    for (let jj = j_min; jj <= j_max; jj++) {
+      for (let kk = k_min; kk <= k_max; kk++) {
+        const candidate_frac_diff: Vec3 = [
+          frac_diff[0] + ii,
+          frac_diff[1] + jj,
+          frac_diff[2] + kk,
+        ]
+        const candidate_displacement = frac_to_cart(candidate_frac_diff)
+        const candidate_dist_sq = vec3_norm_sq(candidate_displacement)
+        if (candidate_dist_sq < best_dist_sq) {
+          best_dist_sq = candidate_dist_sq
+          best_displacement = candidate_displacement
+        }
+      }
+    }
+  }
+
+  return best_displacement
+}
+
+// Calculate the minimum distance between two points considering periodic boundary conditions.
+export const pbc_dist = (
+  pos1: Vec3,
+  pos2: Vec3,
+  lattice_matrix: Matrix3x3,
+  converters?: LatticeConverters,
+  pbc: Pbc = [true, true, true],
+): number => Math.hypot(...min_image_displacement(pos1, pos2, lattice_matrix, converters, pbc))
+
+export function matrix_inverse_3x3(matrix: Matrix3x3): Matrix3x3 {
+  const [[m11, m12, m13], [m21, m22, m23], [m31, m32, m33]] = matrix
+
+  const det = det_3x3(matrix)
+
+  if (!Number.isFinite(det) || Math.abs(det) < EPS) {
+    throw new Error(`Matrix is singular or ill-conditioned; cannot invert`)
+  }
+
+  const inv_det = 1 / det
+
+  return [
+    [
+      (m22 * m33 - m23 * m32) * inv_det,
+      (m13 * m32 - m12 * m33) * inv_det,
+      (m12 * m23 - m13 * m22) * inv_det,
+    ],
+    [
+      (m23 * m31 - m21 * m33) * inv_det,
+      (m11 * m33 - m13 * m31) * inv_det,
+      (m13 * m21 - m11 * m23) * inv_det,
+    ],
+    [
+      (m21 * m32 - m22 * m31) * inv_det,
+      (m12 * m31 - m11 * m32) * inv_det,
+      (m11 * m22 - m12 * m21) * inv_det,
+    ],
+  ]
+}
+
+// Multiply a 3x3 matrix by a 3D vector
+export function mat3x3_vec3_multiply(matrix: Matrix3x3, vector: Vec3): Vec3 {
+  const [a, b, c] = matrix
+  const [x, y, z] = vector
+  const a_new = a[0] * x + a[1] * y + a[2] * z
+  const b_new = b[0] * x + b[1] * y + b[2] * z
+  const c_new = c[0] * x + c[1] * y + c[2] * z
+  return [a_new, b_new, c_new]
+}
+
+export function add<T extends NdVector>(...vecs: T[]): T {
+  // add up any number of same-length vectors
+  if (vecs.length === 0) throw new Error(`Cannot add zero vectors`)
+
+  const first_vec = vecs[0]
+  const length = first_vec.length
+
+  // Validate all vectors have the same length
+  for (const vec of vecs) {
+    if (vec.length !== length) {
+      throw new Error(`All vectors must have the same length`)
+    }
+  }
+
+  const result = Array.from<number>({ length }).fill(0)
+  for (const vec of vecs) {
+    for (let idx = 0; idx < length; idx++) {
+      result[idx] += vec[idx]
+    }
+  }
+  return result as T
+}
+
+export function subtract<T extends NdVector>(vec1: T, vec2: T): T {
+  if (vec1.length !== vec2.length) {
+    throw new Error(`Vectors must be of same length`)
+  }
+  return vec1.map((val, idx) => val - vec2[idx]) as T
+}
+
+// Validate matrix structure and return column count
+function validate_matrix(mat: number[][], name: string): number {
+  // Check for empty matrix (no rows)
+  if (mat.length === 0) {
+    throw new Error(`${name} must have at least one row`)
+  }
+
+  if (!mat.every((row) => Array.isArray(row))) {
+    throw new Error(`${name} must contain only array rows (no undefined/non-array elements)`)
+  }
+
+  const cols = mat[0]?.length
+  if (!Number.isFinite(cols)) throw new Error(`${name} has no columns`)
+
+  // Check for zero columns
+  if (cols === 0) {
+    throw new Error(`${name} must have at least one column`)
+  }
+
+  if (!mat.every((row) => row.length === cols)) {
+    throw new Error(`${name} must be rectangular`)
+  }
+  return cols
+}
+
+export function dot(vec1: NdVector, vec2: NdVector): number
+export function dot(vec1: NdVector[], vec2: NdVector): number[]
+export function dot(vec1: NdVector[], vec2: NdVector[]): number[][]
+export function dot(
+  vec1: NdVector | NdVector[],
+  vec2: NdVector | NdVector[],
+): number | number[] | number[][] {
+  const vec1_is_matrix = vec1.some((entry) => Array.isArray(entry))
+  const vec2_is_matrix = vec2.some((entry) => Array.isArray(entry))
+
+  // Vector dot product
+  if (!vec1_is_matrix && !vec2_is_matrix) {
+    const left_vec = vec1 as number[]
+    const right_vec = vec2 as number[]
+    if (left_vec.length !== right_vec.length) {
+      throw new Error(`Vectors must be of same length`)
+    }
+    return left_vec.reduce((sum, val, idx) => sum + val * right_vec[idx], 0)
+  }
+
+  // Matrix-vector multiplication
+  if (vec1_is_matrix && !vec2_is_matrix) {
+    const mat = vec1 as number[][]
+    const vec = vec2 as number[]
+    const cols = validate_matrix(mat, `Matrix`)
+    if (cols !== vec.length) {
+      throw new Error(`Matrix columns must equal vector length`)
+    }
+    return mat.map((row) => row.reduce((sum, val, idx) => sum + val * vec[idx], 0))
+  }
+
+  // Matrix-matrix multiplication
+  if (vec1_is_matrix && vec2_is_matrix) {
+    const mat1 = vec1 as number[][]
+    const mat2 = vec2 as number[][]
+    const mat1_cols = validate_matrix(mat1, `First matrix`)
+    const mat2_cols = validate_matrix(mat2, `Second matrix`)
+    if (mat1_cols !== mat2.length) {
+      throw new Error(`First matrix columns must equal second matrix rows`)
+    }
+    return mat1.map((_row, ii) =>
+      Array.from({ length: mat2_cols }, (_col, jj) =>
+        mat1[ii].reduce((sum, _val, kk) => sum + mat1[ii][kk] * mat2[kk][jj], 0),
+      ),
+    )
+  }
+
+  throw new Error(`Unsupported input types for dot product`)
+}
+
+// Conversion utilities for vectors and tensors below
+
+// Convert 3x3 symmetric tensor to 6-element Voigt notation vector
+// Voigt notation maps: (1,1)->1, (2,2)->2, (3,3)->3, (2,3)->4, (1,3)->5, (1,2)->6
+export function to_voigt(tensor: number[][]): number[] {
+  if (tensor.length !== 3 || !tensor.every((row) => row.length === 3)) {
+    throw new Error(`Expected 3x3 tensor, got ${tensor.length}x${tensor[0]?.length ?? `n/a`}`)
+  }
+  const [t11, t12, t13, _t21, t22, t23, _t31, _t32, t33] = tensor.flat()
+  return [t11, t22, t33, t23, t13, t12]
+}
+
+// Convert 6-element Voigt notation vector to 3x3 symmetric tensor
+export function from_voigt(voigt: number[]): number[][] {
+  if (voigt.length !== 6) {
+    throw new Error(`Expected 6-element Voigt vector, got ${voigt.length} elements`)
+  }
+  const [v1, v2, v3, v4, v5, v6] = voigt
+
+  return [
+    [v1, v6, v5],
+    [v6, v2, v4],
+    [v5, v4, v3],
+  ]
+}
+
+// Convert flat 9-element array to 3x3 tensor (row-major order)
+export function vec9_to_mat3x3(flat_array: number[]): number[][] {
+  if (flat_array.length !== 9) {
+    throw new Error(`Expected 9-element array, got ${flat_array.length} elements`)
+  }
+  const [a1, a2, a3, a4, a5, a6, a7, a8, a9] = flat_array
+  return [
+    [a1, a2, a3],
+    [a4, a5, a6],
+    [a7, a8, a9],
+  ]
+}
+
+// Convert 3x3 tensor to flat 9-element array (row-major order)
+export function tensor_to_flat_array(tensor: number[][]): number[] {
+  if (tensor.length !== 3 || !tensor.every((row) => row.length === 3)) {
+    throw new Error(`Expected 3x3 tensor, got ${tensor.length}x${tensor[0]?.length ?? `n/a`}`)
+  }
+
+  const [t11, t12, t13, t21, t22, t23, t31, t32, t33] = tensor.flat()
+  return [t11, t12, t13, t21, t22, t23, t31, t32, t33]
+}
+
+// Transpose a 3x3 matrix
+export const transpose_3x3_matrix = (matrix: Matrix3x3): Matrix3x3 => [
+  [matrix[0][0], matrix[1][0], matrix[2][0]],
+  [matrix[0][1], matrix[1][1], matrix[2][1]],
+  [matrix[0][2], matrix[1][2], matrix[2][2]],
+]
+
+// Scale each row of a 3x3 matrix by the corresponding element of a Vec3.
+// Used to scale lattice vectors by supercell factors.
+export function scale_lattice_matrix(
+  orig_matrix: Matrix3x3,
+  scaling_factors: Vec3,
+): Matrix3x3 {
+  const [nx, ny, nz] = scaling_factors
+  const [a, b, c] = orig_matrix
+  return [
+    [a[0] * nx, a[1] * nx, a[2] * nx],
+    [b[0] * ny, b[1] * ny, b[2] * ny],
+    [c[0] * nz, c[1] * nz, c[2] * nz],
+  ]
+}
+
+const create_cart_to_frac_matrix = (lattice: Matrix3x3): Matrix3x3 =>
+  matrix_inverse_3x3(transpose_3x3_matrix(lattice))
+
+// Curried fractional→Cartesian converter (caches transposed matrix)
+export const create_frac_to_cart = (lattice: Matrix3x3) => {
+  const transposed = transpose_3x3_matrix(lattice)
+  return (frac: Vec3): Vec3 => mat3x3_vec3_multiply(transposed, frac)
+}
+
+// Curried Cartesian→fractional converter (caches inverse transpose)
+export const create_cart_to_frac = (lattice: Matrix3x3) => {
+  const cart_to_frac_mat = create_cart_to_frac_matrix(lattice)
+  return (cart: Vec3): Vec3 => mat3x3_vec3_multiply(cart_to_frac_mat, cart)
+}
+
+// Paired converters for a lattice — the safe way to do cart↔frac conversion.
+// Encapsulates the transpose convention so callers never touch raw matrices.
+export type LatticeConverters = {
+  cart_to_frac: (v: Vec3) => Vec3
+  frac_to_cart: (v: Vec3) => Vec3
+  reciprocal_axis_norms: Vec3
+}
+
+export const create_lattice_converters = (lattice: Matrix3x3): LatticeConverters => {
+  const cart_to_frac_mat = create_cart_to_frac_matrix(lattice)
+  return {
+    cart_to_frac: (cart: Vec3): Vec3 => mat3x3_vec3_multiply(cart_to_frac_mat, cart),
+    frac_to_cart: create_frac_to_cart(lattice),
+    reciprocal_axis_norms: cart_to_frac_mat.map((row) => Math.hypot(...row)) as Vec3,
+  }
+}
+
+// Convert unit cell parameters to lattice matrix (crystallographic convention)
+export function cell_to_lattice_matrix(
+  a: number,
+  b: number,
+  c: number,
+  alpha: number,
+  beta: number,
+  gamma: number,
+): Matrix3x3 {
+  // Convert angles to radians
+  const alpha_rad = alpha * DEG_TO_RAD
+  const beta_rad = beta * DEG_TO_RAD
+  const gamma_rad = gamma * DEG_TO_RAD
+
+  const cos_alpha = Math.cos(alpha_rad)
+  const cos_beta = Math.cos(beta_rad)
+  const cos_gamma = Math.cos(gamma_rad)
+  const sin_gamma = Math.sin(gamma_rad)
+
+  // Calculate volume factor for triclinic system
+  const vol_factor = Math.sqrt(
+    1 - cos_alpha ** 2 - cos_beta ** 2 - cos_gamma ** 2 + 2 * cos_alpha * cos_beta * cos_gamma,
+  )
+
+  // Standard crystallographic lattice vectors
+  const c_x = c * cos_beta
+  const c_y = (c * (cos_alpha - cos_beta * cos_gamma)) / sin_gamma
+  const c_z = (c * vol_factor) / sin_gamma
+  return [
+    [a, 0, 0],
+    [b * cos_gamma, b * sin_gamma, 0],
+    [c_x, c_y, c_z],
+  ]
+}
+
+export function det_3x3(matrix: Matrix3x3): number {
+  // |A| = a(ei − fh) − b(di − fg) + c(dh − eg)
+  // where matrix = [[a, b, c], [d, e, f], [g, h, i]]
+  const [[m00, m01, m02], [m10, m11, m12], [m20, m21, m22]] = matrix
+  return (
+    m00 * (m11 * m22 - m12 * m21) -
+    m01 * (m10 * m22 - m12 * m20) +
+    m02 * (m10 * m21 - m11 * m20)
+  )
+}
+
+export function get_coefficient_of_variation(values: number[]): number {
+  if (values.length <= 1) return 0
+  const mean = values.reduce((sum, val) => sum + val, 0) / values.length
+  const variance = values.reduce((sum, val) => sum + (val - mean) ** 2, 0) / values.length
+  return Math.abs(mean) > 1e-10 ? Math.sqrt(variance) / Math.abs(mean) : Math.sqrt(variance)
+}
+
+// Compute 4x4 determinant (used for 4D barycentric coordinates)
+export function det_4x4(matrix: Matrix4x4): number {
+  const [a_row, b_row, c_row, d_row] = matrix
+  const [a0, a1, a2, a3] = a_row
+  const [b0, b1, b2, b3] = b_row
+  const [c0, c1, c2, c3] = c_row
+  const [d0, d1, d2, d3] = d_row
+  return (
+    a0 * (b1 * (c2 * d3 - c3 * d2) - b2 * (c1 * d3 - c3 * d1) + b3 * (c1 * d2 - c2 * d1)) -
+    a1 * (b0 * (c2 * d3 - c3 * d2) - b2 * (c0 * d3 - c3 * d0) + b3 * (c0 * d2 - c2 * d0)) +
+    a2 * (b0 * (c1 * d3 - c3 * d1) - b1 * (c0 * d3 - c3 * d0) + b3 * (c0 * d1 - c1 * d0)) -
+    a3 * (b0 * (c1 * d2 - c2 * d1) - b1 * (c0 * d2 - c2 * d0) + b2 * (c0 * d1 - c1 * d0))
+  )
+}
+
+// Compute NxN determinant using LU decomposition with partial pivoting
+// More numerically stable than cofactor expansion for N > 4
+// Returns 0 for singular/near-singular matrices (pivot < EPS ≈ 1e-10)
+export function det_nxn(matrix: number[][]): number {
+  const mat_size = matrix.length
+  if (mat_size === 0) return 1
+  if (!matrix.every((row) => row.length === mat_size)) {
+    throw new Error(`det_nxn requires a square matrix`)
+  }
+
+  // Fast paths for small matrices
+  if (mat_size === 1) return matrix[0][0]
+  if (mat_size === 2) return matrix[0][0] * matrix[1][1] - matrix[0][1] * matrix[1][0]
+  if (mat_size === 3) return det_3x3(matrix as Matrix3x3)
+  if (mat_size === 4) return det_4x4(matrix as Matrix4x4)
+
+  // LU decomposition with partial pivoting
+  // Create a working copy to avoid mutating input
+  const lu = matrix.map((row) => [...row])
+  let swaps = 0
+
+  for (let col = 0; col < mat_size; col++) {
+    // Find pivot (largest absolute value in column)
+    let [max_row, max_val] = [col, Math.abs(lu[col][col])]
+    for (let row = col + 1; row < mat_size; row++) {
+      const val = Math.abs(lu[row][col])
+      if (val > max_val) {
+        max_val = val
+        max_row = row
+      }
+    }
+
+    // Singular matrix (or nearly so)
+    if (max_val < EPS) return 0
+
+    // Swap rows if needed
+    if (max_row !== col) {
+      ;[lu[col], lu[max_row]] = [lu[max_row], lu[col]]
+      swaps++
+    }
+
+    // Eliminate below pivot
+    const pivot = lu[col][col]
+    for (let row = col + 1; row < mat_size; row++) {
+      const factor = lu[row][col] / pivot
+      lu[row][col] = 0
+      for (let k = col + 1; k < mat_size; k++) {
+        lu[row][k] -= factor * lu[col][k]
+      }
+    }
+  }
+
+  // Determinant is product of diagonal elements × (-1)^swaps
+  let det = swaps % 2 === 0 ? 1 : -1
+  for (let idx = 0; idx < mat_size; idx++) {
+    det *= lu[idx][idx]
+  }
+  return det
+}
+
+// 3D cross product
+export const cross_3d = (vec1: Vec3, vec2: Vec3): Vec3 => [
+  vec1[1] * vec2[2] - vec1[2] * vec2[1],
+  vec1[2] * vec2[0] - vec1[0] * vec2[2],
+  vec1[0] * vec2[1] - vec1[1] * vec2[0],
+]
+
+// Scalar linear interpolation
+export const lerp = (start: number, end: number, t: number): number =>
+  start + t * (end - start)
+
+// Vec3 linear interpolation
+export const lerp_vec3 = (start: Vec3, end: Vec3, t: number): Vec3 => [
+  start[0] + t * (end[0] - start[0]),
+  start[1] + t * (end[1] - start[1]),
+  start[2] + t * (end[2] - start[2]),
+]
+
+// Centered fractional part: offset from nearest integer, returns value in [-0.5, 0.5)
+// Useful for wrapping coordinates to first Brillouin zone or similar periodic domains
+export const centered_frac = (val: number): number => {
+  let wrapped = val - Math.round(val)
+  // Handle floating point edge cases at boundaries (range is [-0.5, 0.5), exclusive at +0.5)
+  if (wrapped < -0.5) wrapped += 1
+  if (wrapped >= 0.5) wrapped -= 1
+  return wrapped || 0 // normalize -0 to 0
+}
+
+// Element-wise equality check for two optional Vec3s.
+// Returns true if both are the same reference, or both are defined with equal components.
+export const vecs_equal = (vec_a?: Vec3, vec_b?: Vec3): boolean =>
+  vec_a === vec_b ||
+  (!!vec_a &&
+    !!vec_b &&
+    vec_a[0] === vec_b[0] &&
+    vec_a[1] === vec_b[1] &&
+    vec_a[2] === vec_b[2])
+
+// Normalize a Vec3 to unit length, returns zero vector if input is zero
+export function normalize_vec3(vec: Vec3, fallback?: Vec3): Vec3 {
+  const len = Math.hypot(vec[0], vec[1], vec[2])
+  if (len < EPS) return fallback ?? [0, 0, 0]
+  return [vec[0] / len, vec[1] / len, vec[2] / len]
+}
+
+// Compute orthonormal basis vectors in a plane perpendicular to `normal`.
+// Uses Gram-Schmidt orthogonalization + cross product.
+export function compute_in_plane_basis(normal: Vec3): [Vec3, Vec3] {
+  let ref_vec: Vec3 = [1, 0, 0]
+  if (Math.abs(normal[0]) > 0.9) ref_vec = [0, 1, 0]
+
+  const dot_nr = dot(normal, ref_vec)
+  const u_raw: Vec3 = [
+    ref_vec[0] - dot_nr * normal[0],
+    ref_vec[1] - dot_nr * normal[1],
+    ref_vec[2] - dot_nr * normal[2],
+  ]
+  const u_vec = normalize_vec3(u_raw, [0, 1, 0])
+  const v_vec = cross_3d(normal, u_vec)
+  return [u_vec, v_vec] // u, v basis vectors
+}
+
+// Check whether N 3D points all lie on the same plane within tolerance.
+// Fewer than 3 points are trivially coplanar.
+// Uses cross product to find a plane normal from non-collinear edges,
+// then checks all remaining points have zero distance to that plane.
+export function are_coplanar(points: number[][], tolerance = 1e-6): boolean {
+  if (points.length < 3) return true
+  const origin = points[0]
+  // Find first pair of edges from origin that are not collinear
+  let normal: Vec3 | null = null
+  for (let idx = 1; idx < points.length - 1; idx++) {
+    const edge_a: Vec3 = [
+      points[idx][0] - origin[0],
+      points[idx][1] - origin[1],
+      points[idx][2] - origin[2],
+    ]
+    for (let jdx = idx + 1; jdx < points.length; jdx++) {
+      const edge_b: Vec3 = [
+        points[jdx][0] - origin[0],
+        points[jdx][1] - origin[1],
+        points[jdx][2] - origin[2],
+      ]
+      const cross = cross_3d(edge_a, edge_b)
+      const len = Math.hypot(cross[0], cross[1], cross[2])
+      if (len > tolerance) {
+        normal = [cross[0] / len, cross[1] / len, cross[2] / len]
+        break
+      }
+    }
+    if (normal) break
+  }
+  // All edges are collinear -> all points lie on a line -> coplanar
+  if (!normal) return true
+  const plane_d = dot(normal, origin)
+  for (let idx = 1; idx < points.length; idx++) {
+    const dist = Math.abs(dot(normal, points[idx]) - plane_d)
+    if (dist > tolerance) return false
+  }
+  return true
+}
+
+// Merge coplanar adjacent triangles in a flat non-indexed position array.
+// Takes 9 floats per triangle (3 vertices x 3 coords), groups adjacent coplanar
+// triangles via union-find, then re-triangulates each group with fan triangulation
+// to eliminate internal diagonal edges.
+export function merge_coplanar_triangles(
+  positions: Float32Array,
+  tolerance = 1e-4,
+): Float32Array {
+  const n_triangles = Math.floor(positions.length / 9)
+  if (n_triangles === 0) return new Float32Array(positions)
+
+  // === Step 1: Extract triangles and compute plane for each ===
+  type TriPlane = {
+    verts: [Vec3, Vec3, Vec3]
+    normal: Vec3
+    plane_d: number
+    degenerate: boolean
+  }
+  const tri_planes: TriPlane[] = []
+  for (let tri_idx = 0; tri_idx < n_triangles; tri_idx++) {
+    const base = tri_idx * 9
+    const va: Vec3 = [positions[base], positions[base + 1], positions[base + 2]]
+    const vb: Vec3 = [positions[base + 3], positions[base + 4], positions[base + 5]]
+    const vc: Vec3 = [positions[base + 6], positions[base + 7], positions[base + 8]]
+    const edge_ab: Vec3 = [vb[0] - va[0], vb[1] - va[1], vb[2] - va[2]]
+    const edge_ac: Vec3 = [vc[0] - va[0], vc[1] - va[1], vc[2] - va[2]]
+    const raw_normal = cross_3d(edge_ab, edge_ac)
+    const len = Math.hypot(raw_normal[0], raw_normal[1], raw_normal[2])
+    if (len < tolerance) {
+      tri_planes.push({
+        verts: [va, vb, vc],
+        normal: [0, 0, 0],
+        plane_d: 0,
+        degenerate: true,
+      })
+      continue
+    }
+    // Normalize and canonicalize: first non-zero component must be positive
+    let normal: Vec3 = [raw_normal[0] / len, raw_normal[1] / len, raw_normal[2] / len]
+    const CANON_EPS = 1e-12
+    const first_nonzero =
+      Math.abs(normal[0]) > CANON_EPS
+        ? normal[0]
+        : Math.abs(normal[1]) > CANON_EPS
+          ? normal[1]
+          : normal[2]
+    if (first_nonzero < 0) normal = [-normal[0], -normal[1], -normal[2]]
+    const plane_d = dot(normal, va)
+    tri_planes.push({ verts: [va, vb, vc], normal, plane_d, degenerate: false })
+  }
+
+  // === Step 2: Build adjacency via edge hash map ===
+  // Quantize vertex to integer grid for hashing (only used for equality, not coords)
+  const vert_key = (v: Vec3): string =>
+    `${Math.round(v[0] / tolerance)},${Math.round(v[1] / tolerance)},${Math.round(
+      v[2] / tolerance,
+    )}`
+  const edge_key = (va: Vec3, vb: Vec3): string => {
+    const ka = vert_key(va)
+    const kb = vert_key(vb)
+    return ka < kb ? `${ka}|${kb}` : `${kb}|${ka}`
+  }
+  // Map edge -> list of triangle indices sharing that edge
+  const edge_to_tris = new Map<string, number[]>()
+  for (let tri_idx = 0; tri_idx < n_triangles; tri_idx++) {
+    const { verts, degenerate } = tri_planes[tri_idx]
+    if (degenerate) continue
+    const edges = [
+      edge_key(verts[0], verts[1]),
+      edge_key(verts[1], verts[2]),
+      edge_key(verts[0], verts[2]),
+    ]
+    for (const ek of edges) {
+      const existing = edge_to_tris.get(ek)
+      if (existing) existing.push(tri_idx)
+      else edge_to_tris.set(ek, [tri_idx])
+    }
+  }
+
+  // === Step 3: Union-Find grouping of coplanar adjacent triangles ===
+  const parent = new Int32Array(n_triangles)
+  const rank = new Int32Array(n_triangles)
+  for (let idx = 0; idx < n_triangles; idx++) parent[idx] = idx
+  const find = (x: number): number => {
+    while (parent[x] !== x) {
+      parent[x] = parent[parent[x]] // path compression
+      x = parent[x]
+    }
+    return x
+  }
+  const union = (a: number, b: number): void => {
+    const ra = find(a),
+      rb = find(b)
+    if (ra === rb) return
+    if (rank[ra] < rank[rb]) parent[ra] = rb
+    else if (rank[ra] > rank[rb]) parent[rb] = ra
+    else {
+      parent[rb] = ra
+      rank[ra]++
+    }
+  }
+  for (const tri_list of edge_to_tris.values()) {
+    if (tri_list.length !== 2) continue
+    const [idx_a, idx_b] = tri_list
+    const pa = tri_planes[idx_a]
+    const pb = tri_planes[idx_b]
+    if (pa.degenerate || pb.degenerate) continue
+    // Check coplanarity: same canonical normal direction AND same plane distance
+    const normal_dot =
+      pa.normal[0] * pb.normal[0] + pa.normal[1] * pb.normal[1] + pa.normal[2] * pb.normal[2]
+    if (Math.abs(normal_dot) < 1 - tolerance) continue
+    if (Math.abs(pa.plane_d - pb.plane_d) > tolerance) continue
+    union(idx_a, idx_b)
+  }
+
+  // === Step 4: Collect groups ===
+  const groups = new Map<number, number[]>()
+  for (let idx = 0; idx < n_triangles; idx++) {
+    const root = find(idx)
+    const group = groups.get(root)
+    if (group) group.push(idx)
+    else groups.set(root, [idx])
+  }
+
+  // === Step 5: Merge each group and re-triangulate ===
+  const output: number[] = []
+  // Push a triangle's 3 vertices (9 floats) to the output
+  const emit_tri = (va: Vec3, vb: Vec3, vc: Vec3): void => {
+    output.push(va[0], va[1], va[2], vb[0], vb[1], vb[2], vc[0], vc[1], vc[2])
+  }
+  const emit_original = (members: number[]): void => {
+    for (const tri_idx of members) {
+      const { verts } = tri_planes[tri_idx]
+      emit_tri(verts[0], verts[1], verts[2])
+    }
+  }
+  for (const members of groups.values()) {
+    if (members.length === 1) {
+      emit_original(members)
+      continue
+    }
+
+    const { normal } = tri_planes[members[0]]
+    // Collect all unique vertices from the group
+    const seen_keys = new Map<string, Vec3>()
+    for (const tri_idx of members) {
+      for (const vert of tri_planes[tri_idx].verts) {
+        const key = vert_key(vert)
+        if (!seen_keys.has(key)) seen_keys.set(key, vert)
+      }
+    }
+    const unique_verts = [...seen_keys.values()]
+    if (unique_verts.length < 3) {
+      emit_original(members)
+      continue
+    }
+
+    // Project to 2D using in-plane basis
+    const [u_vec, v_vec] = compute_in_plane_basis(normal)
+    const pts_2d = unique_verts.map((vertex): Vec2 => [dot(u_vec, vertex), dot(v_vec, vertex)])
+
+    const hull = convex_hull_2d(pts_2d)
+    if (hull.length < 3) {
+      emit_original(members)
+      continue
+    }
+
+    // Map 2D hull vertices back to nearest 3D vertex
+    const hull_3d: Vec3[] = hull.map((pt) => {
+      let best_dist = Infinity
+      let best_idx = 0
+      for (let idx = 0; idx < pts_2d.length; idx++) {
+        const du = pts_2d[idx][0] - pt[0]
+        const dv = pts_2d[idx][1] - pt[1]
+        const dist = du * du + dv * dv
+        if (dist < best_dist) {
+          best_dist = dist
+          best_idx = idx
+        }
+      }
+      return unique_verts[best_idx]
+    })
+
+    // Fan-triangulate from hull vertex 0
+    for (let idx = 1; idx < hull_3d.length - 1; idx++) {
+      emit_tri(hull_3d[0], hull_3d[idx], hull_3d[idx + 1])
+    }
+  }
+
+  return new Float32Array(output)
+}
+
+// Compute axis-aligned bounding box of Vec3 vertices
+export function compute_bounding_box(vertices: Vec3[]): { min: Vec3; max: Vec3 } {
+  if (vertices.length === 0) {
+    return { min: [0, 0, 0], max: [0, 0, 0] }
+  }
+
+  const min: Vec3 = [...vertices[0]]
+  const max: Vec3 = [...vertices[0]]
+
+  for (const vert of vertices) {
+    if (vert[0] < min[0]) min[0] = vert[0]
+    if (vert[1] < min[1]) min[1] = vert[1]
+    if (vert[2] < min[2]) min[2] = vert[2]
+    if (vert[0] > max[0]) max[0] = vert[0]
+    if (vert[1] > max[1]) max[1] = vert[1]
+    if (vert[2] > max[2]) max[2] = vert[2]
+  }
+
+  return { min, max }
+}
+
+// Check if a matrix is square with dimension NxN
+export function is_square_matrix(matrix: unknown, dim: number): boolean {
+  if (!Array.isArray(matrix)) return false
+  if (matrix.length !== dim) return false
+  return matrix.every((row) => Array.isArray(row) && row.length === dim)
+}
+
+// --- 2D Geometry Utilities ---
+
+// Point-in-polygon test using ray casting algorithm
+// Returns true if point (x, y) is inside the polygon defined by vertices
+export function point_in_polygon(point_x: number, point_y: number, vertices: Vec2[]): boolean {
+  if (vertices.length < 3) return false
+  let [inside, prev_idx] = [false, vertices.length - 1]
+
+  for (let idx = 0; idx < vertices.length; idx++) {
+    const [x_i, y_i] = vertices[idx]
+    const [x_j, y_j] = vertices[prev_idx]
+
+    // Check if horizontal ray from point crosses this edge
+    if (y_i !== y_j && y_i > point_y !== y_j > point_y) {
+      const x_intersect = ((x_j - x_i) * (point_y - y_i)) / (y_j - y_i) + x_i
+      if (point_x < x_intersect) inside = !inside
+    }
+    prev_idx = idx
+  }
+
+  return inside
+}
+
+// Compute axis-aligned bounding box of 2D vertices
+export function compute_bounding_box_2d(vertices: Vec2[]): {
+  min: Vec2
+  max: Vec2
+  width: number
+  height: number
+} {
+  if (vertices.length === 0) {
+    return { min: [0, 0], max: [0, 0], width: 0, height: 0 }
+  }
+
+  let [min_x, min_y] = vertices[0]
+  let [max_x, max_y] = vertices[0]
+
+  for (const [x, y] of vertices) {
+    if (x < min_x) min_x = x
+    if (x > max_x) max_x = x
+    if (y < min_y) min_y = y
+    if (y > max_y) max_y = y
+  }
+
+  const width = max_x - min_x
+  const height = max_y - min_y
+  return { min: [min_x, min_y], max: [max_x, max_y], width, height }
+}
+
+// Calculate true geometric centroid of a polygon using shoelace formula
+// Falls back to vertex average for degenerate cases (< 3 vertices or zero area)
+export function polygon_centroid(vertices: Vec2[]): Vec2 {
+  if (vertices.length === 0) return [0, 0]
+  if (vertices.length < 3) {
+    const sum_x = vertices.reduce((acc, [x]) => acc + x, 0)
+    const sum_y = vertices.reduce((acc, [, y]) => acc + y, 0)
+    return [sum_x / vertices.length, sum_y / vertices.length]
+  }
+
+  let [signed_area, cx, cy] = [0, 0, 0]
+
+  for (let idx = 0; idx < vertices.length; idx++) {
+    const [x0, y0] = vertices[idx]
+    const [x1, y1] = vertices[(idx + 1) % vertices.length]
+    const cross = x0 * y1 - x1 * y0
+    signed_area += cross
+    cx += (x0 + x1) * cross
+    cy += (y0 + y1) * cross
+  }
+
+  signed_area *= 0.5
+
+  // Fall back to vertex average for degenerate polygons
+  if (Math.abs(signed_area) < EPS) {
+    const sum_x = vertices.reduce((acc, [x]) => acc + x, 0)
+    const sum_y = vertices.reduce((acc, [, y]) => acc + y, 0)
+    return [sum_x / vertices.length, sum_y / vertices.length]
+  }
+
+  const factor = 1 / (6 * signed_area)
+  return [cx * factor, cy * factor]
+}
+
+// Solve linear system Ax = b via LU decomposition with partial pivoting.
+// Returns null if the system is singular (no unique solution).
+// Fast-paths for 2x2 (Cramer's rule) and 3x3 (matrix inverse).
+export function solve_linear_system(
+  A: number[][], // NxN coefficient matrix
+  b: number[], // N-element right-hand side
+): number[] | null {
+  const n = A.length
+  if (n === 0 || b.length !== n || !A.every((row) => row.length === n)) return null
+
+  // 2x2 fast path via Cramer's rule
+  if (n === 2) {
+    const det = A[0][0] * A[1][1] - A[0][1] * A[1][0]
+    if (Math.abs(det) < EPS) return null
+    return [(b[0] * A[1][1] - b[1] * A[0][1]) / det, (A[0][0] * b[1] - A[1][0] * b[0]) / det]
+  }
+
+  // 3x3 fast path via matrix inverse
+  if (n === 3) {
+    const det = det_3x3(A as Matrix3x3)
+    if (Math.abs(det) < EPS) return null
+    const inv = matrix_inverse_3x3(A as Matrix3x3)
+    return mat3x3_vec3_multiply(inv, b as Vec3) as number[]
+  }
+
+  // General NxN: LU decomposition with partial pivoting + forward/back substitution
+  const lu = A.map((row) => [...row])
+  const perm = Array.from({ length: n }, (_, idx) => idx)
+
+  for (let col = 0; col < n; col++) {
+    // Find pivot
+    let [max_row, max_val] = [col, Math.abs(lu[col][col])]
+
+    for (let row = col + 1; row < n; row++) {
+      const val = Math.abs(lu[row][col])
+      if (val > max_val) [max_val, max_row] = [val, row]
+    }
+    if (max_val < EPS) return null // singular
+
+    // Swap rows
+    if (max_row !== col) {
+      ;[lu[col], lu[max_row]] = [lu[max_row], lu[col]]
+      ;[perm[col], perm[max_row]] = [perm[max_row], perm[col]]
+    }
+
+    // Eliminate below pivot
+    const pivot = lu[col][col]
+    for (let row = col + 1; row < n; row++) {
+      const factor = lu[row][col] / pivot
+      lu[row][col] = factor // store L factor in lower triangle
+      for (let k = col + 1; k < n; k++) {
+        lu[row][k] -= factor * lu[col][k]
+      }
+    }
+  }
+
+  // Apply permutation to b
+  const pb = perm.map((idx) => b[idx])
+
+  // Forward substitution (Ly = Pb)
+  for (let row = 1; row < n; row++) {
+    for (let col = 0; col < row; col++) {
+      pb[row] -= lu[row][col] * pb[col]
+    }
+  }
+
+  // Back substitution (Ux = y)
+  const x = Array.from<number>({ length: n }).fill(0)
+  for (let row = n - 1; row >= 0; row--) {
+    let sum = pb[row]
+    for (let col = row + 1; col < n; col++) {
+      sum -= lu[row][col] * x[col]
+    }
+    x[row] = sum / lu[row][row]
+  }
+
+  return x
+}
+
+export const cross_2d = (origin: Vec2, point_a: Vec2, point_b: Vec2): number =>
+  (point_a[0] - origin[0]) * (point_b[1] - origin[1]) -
+  (point_a[1] - origin[1]) * (point_b[0] - origin[0])
+
+// Full 2D convex hull via Andrew's monotone chain algorithm.
+// Returns vertices in counter-clockwise order.
+export function convex_hull_2d(points: Vec2[]): Vec2[] {
+  if (points.length < 3) return [...points]
+
+  const sorted = points.toSorted((a, b) => a[0] - b[0] || a[1] - b[1])
+
+  // Lower hull
+  const lower: Vec2[] = []
+  for (const pt of sorted) {
+    while (
+      lower.length >= 2 &&
+      cross_2d(lower[lower.length - 2], lower[lower.length - 1], pt) <= 0
+    ) {
+      lower.pop()
+    }
+    lower.push(pt)
+  }
+
+  // Upper hull
+  const upper: Vec2[] = []
+  for (let idx = sorted.length - 1; idx >= 0; idx--) {
+    const pt = sorted[idx]
+    while (
+      upper.length >= 2 &&
+      cross_2d(upper[upper.length - 2], upper[upper.length - 1], pt) <= 0
+    ) {
+      upper.pop()
+    }
+    upper.push(pt)
+  }
+
+  // Remove last point of each half (it's the first point of the other)
+  lower.pop()
+  upper.pop()
+
+  return [...lower, ...upper]
+}