mathjs 7.2.0 → 7.5.1

package/src/function/algebra/solver/lusolve.js

@@ -53,67 +53,57 @@ export const createLusolve = /* #__PURE__ */ factory(name, dependencies, ({ type
   return typed(name, {
 
     'Array, Array | Matrix': function (a, b) {
-      // convert a to matrix
       a = matrix(a)
-      // matrix lup decomposition
       const d = lup(a)
-      // solve
       const x = _lusolve(d.L, d.U, d.p, null, b)
-      // convert result to array
       return x.valueOf()
     },
 
     'DenseMatrix, Array | Matrix': function (a, b) {
-      // matrix lup decomposition
       const d = lup(a)
-      // solve
       return _lusolve(d.L, d.U, d.p, null, b)
     },
 
     'SparseMatrix, Array | Matrix': function (a, b) {
-      // matrix lup decomposition
       const d = lup(a)
-      // solve
       return _lusolve(d.L, d.U, d.p, null, b)
     },
 
     'SparseMatrix, Array | Matrix, number, number': function (a, b, order, threshold) {
-      // matrix lu decomposition
       const d = slu(a, order, threshold)
-      // solve
       return _lusolve(d.L, d.U, d.p, d.q, b)
     },
 
     'Object, Array | Matrix': function (d, b) {
-      // solve
       return _lusolve(d.L, d.U, d.p, d.q, b)
     }
   })
 
   function _toMatrix (a) {
-    // check it is a matrix
     if (isMatrix(a)) { return a }
-    // check array
     if (isArray(a)) { return matrix(a) }
-    // throw
     throw new TypeError('Invalid Matrix LU decomposition')
   }
 
   function _lusolve (l, u, p, q, b) {
-    // verify L, U, P
+    // verify decomposition
     l = _toMatrix(l)
     u = _toMatrix(u)
-    // validate matrix and vector
-    b = solveValidation(l, b, false)
+
     // apply row permutations if needed (b is a DenseMatrix)
-    if (p) { b._data = csIpvec(p, b._data) }
+    if (p) {
+      b = solveValidation(l, b, true)
+      b._data = csIpvec(p, b._data)
+    }
+
     // use forward substitution to resolve L * y = b
     const y = lsolve(l, b)
     // use backward substitution to resolve U * x = y
     const x = usolve(u, y)
+
     // apply column permutations if needed (x is a DenseMatrix)
     if (q) { x._data = csIpvec(q, x._data) }
-    // return solution
+
     return x
   }
 })

package/src/function/algebra/solver/usolve.js

@@ -16,7 +16,7 @@ export const createUsolve = /* #__PURE__ */ factory(name, dependencies, ({ typed
   const solveValidation = createSolveValidation({ DenseMatrix })
 
   /**
-   * Solves the linear equation system by backward substitution. Matrix must be an upper triangular matrix.
+   * Finds one solution of a linear equation system by backward substitution. Matrix must be an upper triangular matrix. Throws an error if there's no solution.
    *
    * `U * x = b`
    *
@@ -32,7 +32,7 @@ export const createUsolve = /* #__PURE__ */ factory(name, dependencies, ({ typed
    *
    * See also:
    *
-   *    lup, slu, usolve, lusolve
+   *    usolveAll, lup, slu, usolve, lusolve
    *
    * @param {Matrix, Array} U       A N x N matrix or array (U)
    * @param {Matrix, Array} b       A column vector with the b values
@@ -42,67 +42,64 @@ export const createUsolve = /* #__PURE__ */ factory(name, dependencies, ({ typed
   return typed(name, {
 
     'SparseMatrix, Array | Matrix': function (m, b) {
-      // process matrix
       return _sparseBackwardSubstitution(m, b)
     },
 
     'DenseMatrix, Array | Matrix': function (m, b) {
-      // process matrix
       return _denseBackwardSubstitution(m, b)
     },
 
     'Array, Array | Matrix': function (a, b) {
-      // create dense matrix from array
       const m = matrix(a)
-      // use matrix implementation
       const r = _denseBackwardSubstitution(m, b)
-      // result
       return r.valueOf()
     }
   })
 
   function _denseBackwardSubstitution (m, b) {
-    // validate matrix and vector, return copy of column vector b
+    // make b into a column vector
     b = solveValidation(m, b, true)
-    // column vector data
+
     const bdata = b._data
-    // rows & columns
+
     const rows = m._size[0]
     const columns = m._size[1]
+
     // result
     const x = []
-    // arrays
-    const data = m._data
-    // backward solve m * x = b, loop columns (backwards)
+
+    const mdata = m._data
+    // loop columns backwards
     for (let j = columns - 1; j >= 0; j--) {
       // b[j]
       const bj = bdata[j][0] || 0
       // x[j]
       let xj
-      // backward substitution (outer product) avoids inner looping when bj === 0
+
       if (!equalScalar(bj, 0)) {
-        // value @ [j, j]
-        const vjj = data[j][j]
-        // check vjj
+        // value at [j, j]
+        const vjj = mdata[j][j]
+
         if (equalScalar(vjj, 0)) {
           // system cannot be solved
           throw new Error('Linear system cannot be solved since matrix is singular')
         }
-        // calculate xj
+
         xj = divideScalar(bj, vjj)
+
         // loop rows
         for (let i = j - 1; i >= 0; i--) {
           // update copy of b
-          bdata[i] = [subtract(bdata[i][0] || 0, multiplyScalar(xj, data[i][j]))]
+          bdata[i] = [subtract(bdata[i][0] || 0, multiplyScalar(xj, mdata[i][j]))]
         }
       } else {
-        // zero value @ j
+        // zero value at j
         xj = 0
       }
       // update x
       x[j] = [xj]
     }
-    // return column vector
+
     return new DenseMatrix({
       data: x,
       size: [rows, 1]
@@ -110,71 +107,71 @@ export const createUsolve = /* #__PURE__ */ factory(name, dependencies, ({ typed
   }
 
   function _sparseBackwardSubstitution (m, b) {
-    // validate matrix and vector, return copy of column vector b
+    // make b into a column vector
     b = solveValidation(m, b, true)
-    // column vector data
+
     const bdata = b._data
-    // rows & columns
+
     const rows = m._size[0]
     const columns = m._size[1]
-    // matrix arrays
+
     const values = m._values
     const index = m._index
     const ptr = m._ptr
-    // vars
-    let i, k
+
     // result
     const x = []
-    // backward solve m * x = b, loop columns (backwards)
+
+    // loop columns backwards
     for (let j = columns - 1; j >= 0; j--) {
-      // b[j]
       const bj = bdata[j][0] || 0
-      // backward substitution (outer product) avoids inner looping when bj === 0
+
       if (!equalScalar(bj, 0)) {
-        // value @ [j, j]
+        // non-degenerate row, find solution
+
         let vjj = 0
+
         // upper triangular matrix values & index (column j)
-        const jvalues = []
-        const jindex = []
+        const jValues = []
+        const jIndices = []
+
         // first & last indeces in column
-        const f = ptr[j]
-        let l = ptr[j + 1]
-        // values in column, find value @ [j, j], loop backwards
-        for (k = l - 1; k >= f; k--) {
-          // row
-          i = index[k]
-          // check row
+        const firstIndex = ptr[j]
+        const lastIndex = ptr[j + 1]
+
+        // values in column, find value at [j, j], loop backwards
+        for (let k = lastIndex - 1; k >= firstIndex; k--) {
+          const i = index[k]
+
+          // check row (rows are not sorted!)
           if (i === j) {
-            // update vjj
             vjj = values[k]
           } else if (i < j) {
             // store upper triangular
-            jvalues.push(values[k])
-            jindex.push(i)
+            jValues.push(values[k])
+            jIndices.push(i)
           }
         }
-        // at this point we must have a value @ [j, j]
+
+        // at this point we must have a value in vjj
         if (equalScalar(vjj, 0)) {
-          // system cannot be solved, there is no value @ [j, j]
           throw new Error('Linear system cannot be solved since matrix is singular')
         }
-        // calculate xj
+
         const xj = divideScalar(bj, vjj)
-        // loop upper triangular
-        for (k = 0, l = jindex.length; k < l; k++) {
-          // row
-          i = jindex[k]
-          // update copy of b
-          bdata[i] = [subtract(bdata[i][0], multiplyScalar(xj, jvalues[k]))]
+
+        for (let k = 0, lastIndex = jIndices.length; k < lastIndex; k++) {
+          const i = jIndices[k]
+          bdata[i] = [subtract(bdata[i][0], multiplyScalar(xj, jValues[k]))]
         }
-        // update x
+
         x[j] = [xj]
       } else {
-        // update x
+        // degenerate row, we can choose any value
         x[j] = [0]
       }
     }
-    // return vector
+
     return new DenseMatrix({
       data: x,
       size: [rows, 1]

package/src/function/algebra/solver/usolveAll.js

@@ -0,0 +1,199 @@
+import { factory } from '../../../utils/factory'
+import { createSolveValidation } from './utils/solveValidation'
+
+const name = 'usolveAll'
+const dependencies = [
+  'typed',
+  'matrix',
+  'divideScalar',
+  'multiplyScalar',
+  'subtract',
+  'equalScalar',
+  'DenseMatrix'
+]
+
+export const createUsolveAll = /* #__PURE__ */ factory(name, dependencies, ({ typed, matrix, divideScalar, multiplyScalar, subtract, equalScalar, DenseMatrix }) => {
+  const solveValidation = createSolveValidation({ DenseMatrix })
+
+  /**
+   * Finds all solutions of a linear equation system by backward substitution. Matrix must be an upper triangular matrix.
+   *
+   * `U * x = b`
+   *
+   * Syntax:
+   *
+   *    math.usolve(U, b)
+   *
+   * Examples:
+   *
+   *    const a = [[-2, 3], [2, 1]]
+   *    const b = [11, 9]
+   *    const x = usolve(a, b)  // [ [[8], [9]] ]
+   *
+   * See also:
+   *
+   *    usolve, lup, slu, usolve, lusolve
+   *
+   * @param {Matrix, Array} U       A N x N matrix or array (U)
+   * @param {Matrix, Array} b       A column vector with the b values
+   *
+   * @return {DenseMatrix[] | Array[]}  An array of affine-independent column vectors (x) that solve the linear system
+   */
+  return typed(name, {
+
+    'SparseMatrix, Array | Matrix': function (m, b) {
+      return _sparseBackwardSubstitution(m, b)
+    },
+
+    'DenseMatrix, Array | Matrix': function (m, b) {
+      return _denseBackwardSubstitution(m, b)
+    },
+
+    'Array, Array | Matrix': function (a, b) {
+      const m = matrix(a)
+      const R = _denseBackwardSubstitution(m, b)
+      return R.map(r => r.valueOf())
+    }
+  })
+
+  function _denseBackwardSubstitution (m, b_) {
+    // the algorithm is derived from
+    // https://www.overleaf.com/project/5e6c87c554a3190001a3fc93
+
+    // array of right-hand sides
+    const B = [solveValidation(m, b_, true)._data.map(e => e[0])]
+
+    const M = m._data
+    const rows = m._size[0]
+    const columns = m._size[1]
+
+    // loop columns backwards
+    for (let i = columns - 1; i >= 0; i--) {
+      let L = B.length
+
+      // loop right-hand sides
+      for (let k = 0; k < L; k++) {
+        const b = B[k]
+
+        if (!equalScalar(M[i][i], 0)) {
+          // non-singular row
+
+          b[i] = divideScalar(b[i], M[i][i])
+
+          for (let j = i - 1; j >= 0; j--) {
+            // b[j] -= b[i] * M[j,i]
+            b[j] = subtract(b[j], multiplyScalar(b[i], M[j][i]))
+          }
+        } else if (!equalScalar(b[i], 0)) {
+          // singular row, nonzero RHS
+
+          if (k === 0) {
+            // There is no valid solution
+            return []
+          } else {
+            // This RHS is invalid but other solutions may still exist
+            B.splice(k, 1)
+            k -= 1
+            L -= 1
+          }
+        } else if (k === 0) {
+          // singular row, RHS is zero
+
+          const bNew = [...b]
+          bNew[i] = 1
+
+          for (let j = i - 1; j >= 0; j--) {
+            bNew[j] = subtract(bNew[j], M[j][i])
+          }
+
+          B.push(bNew)
+        }
+      }
+    }
+
+    return B.map(x => new DenseMatrix({ data: x.map(e => [e]), size: [rows, 1] }))
+  }
+
+  function _sparseBackwardSubstitution (m, b_) {
+    // array of right-hand sides
+    const B = [solveValidation(m, b_, true)._data.map(e => e[0])]
+
+    const rows = m._size[0]
+    const columns = m._size[1]
+
+    const values = m._values
+    const index = m._index
+    const ptr = m._ptr
+
+    // loop columns backwards
+    for (let i = columns - 1; i >= 0; i--) {
+      let L = B.length
+
+      // loop right-hand sides
+      for (let k = 0; k < L; k++) {
+        const b = B[k]
+
+        // values & indices (column i)
+        const iValues = []
+        const iIndices = []
+
+        // first & last indeces in column
+        const firstIndex = ptr[i]
+        const lastIndex = ptr[i + 1]
+
+        // find the value at [i, i]
+        let Mii = 0
+        for (let j = lastIndex - 1; j >= firstIndex; j--) {
+          const J = index[j]
+          // check row
+          if (J === i) {
+            Mii = values[j]
+          } else if (J < i) {
+            // store upper triangular
+            iValues.push(values[j])
+            iIndices.push(J)
+          }
+        }
+
+        if (!equalScalar(Mii, 0)) {
+          // non-singular row
+
+          b[i] = divideScalar(b[i], Mii)
+
+          // loop upper triangular
+          for (let j = 0, lastIndex = iIndices.length; j < lastIndex; j++) {
+            const J = iIndices[j]
+            b[J] = subtract(b[J], multiplyScalar(b[i], iValues[j]))
+          }
+        } else if (!equalScalar(b[i], 0)) {
+          // singular row, nonzero RHS
+
+          if (k === 0) {
+            // There is no valid solution
+            return []
+          } else {
+            // This RHS is invalid but other solutions may still exist
+            B.splice(k, 1)
+            k -= 1
+            L -= 1
+          }
+        } else if (k === 0) {
+          // singular row, RHS is zero
+
+          const bNew = [...b]
+          bNew[i] = 1
+
+          // loop upper triangular
+          for (let j = 0, lastIndex = iIndices.length; j < lastIndex; j++) {
+            const J = iIndices[j]
+            bNew[J] = subtract(bNew[J], iValues[j])
+          }
+
+          B.push(bNew)
+        }
+      }
+    }
+
+    return B.map(x => new DenseMatrix({ data: x.map(e => [e]), size: [rows, 1] }))
+  }
+})