nLab real quadratic function

Contents

Context

Algebra

Analysis

Contents

Definition

A real quadratic function is a quadratic function in the real numbers. Equivalently, it is a solution to the third-order linear homogeneous ordinary differential equation

d 3fdx 3=0\frac{d^3 f}{d x^3} = 0

with initial conditions

f(0)=cf(0) = c
dfdx(0)=b\frac{d f}{d x}(0) = b
d 2fdx 2(0)=2a\frac{d^2 f}{d x^2}(0) = 2 a

Properties

Extrema

Given a real quadratic function f(x)ax 2+bx+cf(x) \coloneqq a x^2 + b x + c, if |a|>0\vert a \vert \gt 0, after completing the square, the quadratic function may be written as

f(x)=a(x+b2a) 2b 24ac4a f(x) = a \left(x + \frac{b}{2a}\right)^2 - \frac{b^2 - 4 a c}{4a}

The extremum occurs at

x=b2ax = -\frac{b}{2a}

and the extremum is a minimum if a>0a \gt 0 and a maximum if a<0a \lt 0. The value of ff at the extremum is

f(x)=4acb 24af(x) = \frac{4 a c - b^2}{4a}

When a>0a \gt 0, the real quadratic function is said to be concave up, while when a<0a \lt 0, the real quadratic function is said to be concave down.

If a=0a = 0, the real quadratic function is degenerate?; it becomes a real affine function.

In constructive mathematics

In classical mathematics, the law of trichotomy holds in the real numbers, so the three cases above cover every real number. However, in constructive mathematics, trichotomy does not hold in the real numbers, and as a result, there exists real quadratic functions f:f:\mathbb{R} \to \mathbb{R} such that one cannot decide whether ff is concave up, concave down, or degenerate. Similarly, there exists real quadratic functions f:f:\mathbb{R} \to \mathbb{R} such that one cannot decide whether the extremum is a minimum, is a maximum, or does not exist due to the degeneracy of ff.

Exact zeroes

Partial inverse functions

Given a real quadratic function f:f:\mathbb{R} \to \mathbb{R}, f(x)ax 2+bx+cf(x) \coloneqq a x^2 + b x + c for real numbers aa \in \mathbb{R}, bb \in \mathbb{R}, cc \in \mathbb{R} such that |a|>0\vert a \vert \gt 0, since ff is continuously differentiable if |a|>0\vert a \vert \gt 0, due to the inverse function theorem, there are two branches of the partial inverse function of ff, gg and hh, such that if a>0a \gt 0, then gg and hh have domain

(cb 24a,)\left(c - \frac{b^2}{4a}, \infty\right)

and if a<0a \lt 0, gg and hh have domain

(,cb 24a)\left(-\infty,c - \frac{b^2}{4a}\right)

In both cases, gg and hh are defined as solutions to the first-order nonlinear ordinary differential equation

(2aF+b)dFdx=1(2 a F + b) \frac{d F}{d x} = 1

with the initial conditions for the principal branch gg are

g((a+b+c)+(cb 24a))=1b2ag\left((a + b + c) + \left(c - \frac{b^2}{4a}\right)\right) = 1 - \frac{b}{2a}

and the initial conditions for the reflected branch hh are

h((ab+c)+(cb 24a))=1b2ah\left((a - b + c) + \left(c - \frac{b^2}{4a}\right)\right) = -1 - \frac{b}{2a}

As a result, gg and hh are automatically pointwise continuously differentiable, and for every xx in the domain of gg and hh,

g(x)>b2ag(x) \gt -\frac{b}{2a}
h(x)<b2ah(x) \lt -\frac{b}{2a}

Since this differential equation is separable, one could use separation of variables to see that gg and hh satisfy the functional equations ag(x) 2+bg(x)+c=xa g(x)^2 + b g(x) + c = x and ah(x) 2+bh(x)+c=xa h(x)^2 + b h(x) + c = x.

If the discriminant Δ=b 24ac\Delta = b^2 - 4 a c is greater than zero Δ>0\Delta \gt 0, then the quadratic function ff has two zeroes apart from each other. The larger of the two zero of f(x)f(x) is at g(0)g(0) and the smaller zero of f(x)f(x) is at h(0)h(0). If the discriminant is less than or equal to zero Δ0\Delta \leq 0, then zero isn’t in the domain of either gg and hh.

Square roots

If a=1a = 1, b=0b = 0, and c=0c = 0, then the real quadratic function above is the square function f(x)x 2f(x) \coloneqq x^2. The principal branch of the continuous partial inverse function of the square function is called the continuous principal square root sqrt cont:(0,)\mathrm{sqrt}_\mathrm{cont}:(0, \infty) \to \mathbb{R}, and is the solution to the differential equation

(2sqrt cont)dsqrt contdx=1(2 \mathrm{sqrt}_\mathrm{cont}) \frac{d \mathrm{sqrt}_\mathrm{cont}}{d x} = 1

with initial condition sqrt cont(1)=1\mathrm{sqrt}_\mathrm{cont}(1) = 1. The reflected branch of the continuous partial inverse function of the square function is simply the negation of the principal square root sqrt cont:(0,)-\mathrm{sqrt}_\mathrm{cont}:(0, \infty) \to \mathbb{R}.

Quadratic formula

For any quadratic function ff, the principal branch of its inverse has a canonical definition in terms of the continuous principal square root:

g(x)=b+sqrt cont(4ax+b 24ac)2ag(x) = \frac{-b + \mathrm{sqrt}_\mathrm{cont}(4ax + b^2 - 4ac)}{2a}

and the reflected branch has a canonical definition in terms of the continuous principal square root as well:

h(x)=bsqrt cont(4ax+b 24ac)2ah(x) = \frac{-b - \mathrm{sqrt}_\mathrm{cont}(4ax + b^2 - 4ac)}{2a}

The continuous quadratic formula is the evaluation of the two branches at zero, if zero is in the domain of gg and hh:

g(0)=b+sqrt cont(b 24ac)2ag(0) = \frac{-b + \mathrm{sqrt}_\mathrm{cont}(b^2 - 4ac)}{2a}
h(0)=bsqrt cont(b 24ac)2ah(0) = \frac{-b - \mathrm{sqrt}_\mathrm{cont}(b^2 - 4ac)}{2a}

The continuous quadratic formula is only valid for real quadratic functions with positive discriminant, unlike the quadratic formula in the discrete field of quadratic irrational numbers, which is valid for quadratic functions with rational coefficients and a non-negative discriminant. But there’s an analytic reason for why the quadratic formula should be restricted to real quadratic functions with positive discriminant: for any real quadratic function ff with a non-positive discriminant Δ0\Delta \leq 0, there is no interval on the real numbers such that the conditions required for the intermediate value theorem are fulfilled: if a>0a \gt 0, then f(x)0f(x) \geq 0 for all real numbers xx, and if a<0a \lt 0, then f(x)0f(x) \leq 0 for all real numbers xx.

Negative and zero discriminant

As stated above, when a real quadratic function ff has a non-positive discriminant, Δ0\Delta \leq 0, if a>0a \gt 0, then f(x)0f(x) \geq 0 for all real numbers xx, and if a<0a \lt 0, then f(x)0f(x) \leq 0 for all real numbers xx. When ff has a negative discriminant, Δ<0\Delta \lt 0, then ff is provably always apart from zero: if a>0a \gt 0, then f(x)>0f(x) \gt 0 for all real numbers xx, and if a<0a \lt 0, then f(x)<0f(x) \lt 0 for all real numbers xx. As a result, ff has no zeroes in the real numbers if the determinant of ff is negative.

Now, suppose that ff has zero discriminant. This means that b 24ac=0b^2 - 4ac = 0, and after completing the square, the resulting function is equal to

f(x)=a(x+b2a) 2f(x) = a \left(x + \frac{b}{2a}\right)^2

By definition of an ordered field, there are no multiplicative nilpotent elements in the real numbers whose absolute value is greater than zero, and thus every nilpotent element is equal to zero. As a result,

f(x)=a(x+b2a) 2=0f(x) = a \left(x + \frac{b}{2a}\right)^2 = 0
(x+b2a) 2=0\left(x + \frac{b}{2a}\right)^2 = 0
x+b2a=0x + \frac{b}{2a} = 0
x=b2ax = -\frac{b}{2a}

There is only one zero, which occurs at the extremum of the real quadratic function.

In constructive mathematics

In classical mathematics, the law of trichotomy holds in the real numbers, so the three cases above cover every real number. However, in constructive mathematics, trichotomy does not hold in the real numbers, and as a result, there exists real quadratic functions f:f:\mathbb{R} \to \mathbb{R} such that one cannot decide whether the discriminant of ff is positive, negative, or zero. As a result, there exist real quadratic functions where one cannot decide the number of zeroes the function has. Furthermore, one cannot prove that a real quadratic function with positive determinant has exactly two zeroes.

Newton’s method

Given a real quadratic function f:f:\mathbb{R} \to \mathbb{R}, f(x)ax 2+bx+cf(x) \coloneqq a x^2 + b x + c for real numbers aa \in \mathbb{R}, bb \in \mathbb{R}, cc \in \mathbb{R}, where |a|>0\vert a \vert \gt 0, suppose we want to approximate a zero of ff. There is an algorithm called Newton's method which would allow us to do so. The real quadratic function has discriminant Δ=b 24ac\Delta = b^2 - 4 a c and has 2 zeroes if Δ>0\Delta \gt 0, 1 zero at the extremum if Δ=0\Delta = 0, or no zeroes of Δ<0\Delta \lt 0. (In constructive mathematics, there are real quadratic functions where it isn’t possible to determine the number of zeroes the real quadratic function has.)

Assume that Δ>0\Delta \gt 0. Select a real number x 0x_0 \in \mathbb{R} as the initial guess, where 2ax 0+b>02 a x_0 + b \gt 0 or 2ax 0+b<02 a x_0 + b \lt 0, and the next guess would be defined as

x i+1=x iax i 2+bx i+c2ax i+bx_{i+1} = x_i - \frac{a x_i^2 + b x_i + c}{2 a x_i + b}

Newton’s method isn’t valid when 2ax 0+b=02 a x_0 + b = 0 because of division by zero. When 2ax 0+b>02 a x_0 + b \gt 0, Newton’s method converges towards the zero at x>b2ax \gt -\frac{b}{2a}, and when 2ax 0+b<02 a x_0 + b \lt 0, Newton’s method converges towards the zero x<b2ax \lt -\frac{b}{2a}.

See also

References

See also:

Last revised on August 21, 2024 at 01:52:45. See the history of this page for a list of all contributions to it.