23.3.2  Finding approximate solutions of equations involving one variable

The fsolve or nSolve command can solve equations or systems of equations. This section will discuss solving equations; systems will be discussed in the next section.

The cfsolve command is the complex version of fsolve, with the same arguments. The only difference is that cfsolve gives numeric solutions over the complex numbers, even if Xcas is not in complex mode (see Section 2.5.5). fsolve will return complex roots, but only in complex mode.

fsolve solves numeric equations of the form:

Unlike solve (Section 9.3.6) or proot (Section 23.3.4), it is not limited to polynomial equations.

• fsolve takes one mandatory argument and three optional arguments:
  • eqn, an equation involving one variable.
  • Optionally, var, the variable (by default x).
  • Optionally, init, an initial approximation or range.
  • Optionally, method, the name of an iterative method to be used by the GSL solver. The possible values are:

    bisection_solver.

    This algorithm of dichotomy is the simplest but also generally the slowest. It encloses the zero of a function on an interval. Each iteration cuts the interval into two parts, the middle point value is calculated. The function sign at this point gives you the half-interval on which the next iteration will be performed.

    brent_solver.

    The Brent method interpolates of f at three points, finds the intersection of the interpolation with the x axis, computes the sign of f at this point and chooses the interval where the sign changes. It is generally faster than bisection.

    falsepos_solver.

    The “false position” algorithm is an iterative algorithm based on linear interpolation: it computes the value of f at the intersection of the line (a,f(a)), (b,f(b)) with the x axis. This value gives us the part of the interval containing the root, and on which a new iteration is performed. The convergence is linear but generally faster than bisection.

    newton_solver.

    This is the classic Newton method. The algorithm starts at an initial value x₀, then finds the intersection x₁ of the tangent at x₀ to the graph of f, with the x axis, the next iteration is done with x₁ instead of x₀. The x_i sequence is defined by

    x0=x0,    xn+1=xn−

    f(xn) f′(xn)

    .

    If the Newton method converges, then the convergence is quadratic for roots of multiplicity 1.

    secant_solver.

    The secant method is a simplified version of Newton’s method. The computation of x₁ is done using Newton’s method, but then the computation of f′(x_n), n>1 is done approximately. This method is used when the computation of the derivative is expensive:

    xi+1=xi−

    f(xi) f′est

    ,   f′est=

    f(xi)−f(xi−1) xi−xi−1

    .

    The convergence for roots of multiplicity 1 is of order (1+√5)/2 ≈ 1.62… .

    steffenson_solver.

    The Steffenson method is generally the fastest method. It combines Newton’s method with a “delta-two” Aitken acceleration: with Newton’s method, you get the sequence x_i and the convergence acceleration gives the Steffenson sequence

    Ri=xi−

    (xi+1−xi)2  (xi+2−2 xi+1+xi)

    .

• fsolve(eqn ⟨,var,init,method ⟩) returns an approximate solution to the equation eqn.

Examples

Input in real mode:

Input in complex mode:

Input in any mode: