**Background: The Descartes Rule of Sweeps and the Descartes Signature**The

*Descartes Rule of Signs* states, for a polynomial with non-zero real coefficients, that the number of positive roots is no greater than the number of sign changes in the coefficient sequence. The

*Rule of Sweeps* generalizes this result to cover polynomials with non-real coefficients, and it gives rise to the

*Descartes signature,* a function that conveys information about potential roots in any direction of the complex plane.

Imagine a needle with one end anchored at the origin and the other end initially pointing in the direction of the trailing coefficient of a polynomial,

, with (non-zero, but not-necessarily-real) complex coefficients. (The trailing coefficient is the coefficient of the term of least degree.) Let the needle sweep—always counter-clockwise—to point in the direction of the second, third, etc. coefficients, "stalling in place" when successive coefficients lie in the same direction, and tracing out a "sweep spiral". The

*positive sweep* of

is the total angle swept by the needle; the

*negative sweep* of

is the total angle swept by a clockwise-spinning needle (or, equivalently, a counter-clockwise-spinning needle that takes the coefficients in reverse order); these values are usually different, and

*the sweep*is defined to be the smaller of the two. The sweep plays the role of the sign-change tally from the Descartes Rule of Signs, providing a bound on the number of positive roots of

:

*The Descartes Rule of Sweeps.* The maximum number of positive roots of

is

, or 1 less if a line through the origin meets the leading and trailing coefficients but misses at least one other.

Note that the first part of the rule of sweeps covers the Descartes rule of signs, since each coefficient sign change contributes

to a polynomial's sweep.

To study the roots in a particular direction of the complex plane, define an auxiliary polynomial

, where

is the degree of the smallest-degree term of

whose coefficients are obtained by "fanning out" the coefficients of

by multiples of

. If

is a positive root of

, then

is a root of

; thus, applying the rule of sweeps to

gives a bound on the number of roots of

that lie in direction

Treating

as a parameter ranging from

to

, the

*Descartes Signature*—

—is the function that gives that bound, and thus its graph conveys a comprehensive summary of the potential number of roots of

in all directions. Writing

,

, and

for the positive, negative, and minimal sweep of

gives the following result:

The maximum number of roots of

in direction

is given by

, where

, or 1 less if etc., etc., etc.

*Note*: Because the Descartes signature depends only on the *directions* of a polynomial's coefficients, and not on their *magnitudes*, an infinite family of polynomials share the same signature. For instance, has the same Descartes signature as

, for any positive numbers

,

, and

. As a result, a polynomial's signature rarely reveals unambiguous information about its roots; nevertheless, as this Demonstration shows, the signature can be helpful in ruling out entire intervals of

values that cannot serve as directions of roots.

For more information: See "The Descartes Rule of Sweeps and the Descartes Signature" [

online PDF].

**The Demonstration: The Display Modes***sweep spirals* mode shows the coefficients defined by the c

*oefficient angles* settings, and the positive (blue) and negative (red) sweep spirals joining them. When the a

*uxiliary angle* is non-zero, the display illustrates the "fanning out" process that yields the coefficients and spirals of

. The top of the display records the color-coded values of the sweep functions, which give the total angles swept out by the spirals (blue and red), and the minimum of these numbers (purple).

*sweep graphs* mode shows the graphs of

(blue) and

(red), revealing these to be piecewise linear functions with predictable jumps: the jumps occur when a coefficient of

"catches up with" its predecessor as they fan out; the size of the jump is

times the number of instances of this catching-up that occur.

*Descartes signature* mode shows the graph of

(purple) and

(black and green). The green dots indicate when the "or 1 less …" aspect of the rule of sweeps has come into play. The top of the display records the value of the Descartes signature function at the current auxiliary angle; this value turns green for "or 1 less ..." values. (See

*Implementation Caveat* below.)

**The Demonstration: The Controls***sweep spirals / sweep graphs / Descartes signature*: This control selects the display mode (see above).

*coefficient angles*: Define a polynomial by clicking the checkboxes for the powers to include, and setting the sliders for the direction of the coefficients. (As with the rule of signs, the rule of sweeps ignores the size of the coefficients.)

*auxiliary angle (**)*: In s

*weep spirals* mode, this value fans out the polynomial coefficients, with the

-th power coefficient rotating through the angle

. In

*sweep graphs* and

*Descartes signature* modes, the value controls a vertical indicator of the

parameter.

**The Demonstration: Things to Try***Polynomials of the form* . The Descartes signature clearly identifies

symmetrically-spaced angles (corresponding to the "

-th roots of unity") as the only viable directions for roots.

*Polynomials of the form* . Such a polynomial's roots are also roots of unity (except 1 is missing). The Descartes signature is less successful at highlighting them than the preceding example, but the graph is interesting.

The conditions defining the various jumps in the graph involve comparing combinations of numbers against π. This requires a certain amount of computational accuracy, which is undermined by the "play" in the sliders controlling the coefficient and auxiliary angles. The user should therefore not be surprised by errors in the displayed value of

at points of discontinuity.

Likewise, the user should be encouraged to validate the (non-)appearance of "green dots" in

*Descartes signature* mode. This can be done by moving the auxiliary angle indicator to a suspect location and then to switch back to

*sweep spirals* mode to verify that the coefficients are (or are reasonably close to being) in their proper configuration.

Note that, for the "or 1 less ..." condition to be considered at all, the polynomial sweep must be a multiple of π (an easy situation to spot, as this happens at the endpoints of the steps in the signature graph); generally speaking, this is enough. For the sweep to be a multiple of π and yet for the "or 1 less ..." condition to

*fail*, all of the coefficients must be collinear with the origin—for instance, they could be all real—which is fairly unusual.