跳转到内容

A2.1 Vieta and Symmetric Information

此内容尚不支持你的语言。

Start with a quadratic whose roots are α\alpha and β\beta:

x2sx+p=0.x^2-sx+p=0.

Vieta gives

α+β=s,αβ=p.\alpha+\beta=s,\qquad \alpha\beta=p.

If we swap the names α\alpha and β\beta, these two expressions do not change. The coefficients can see

α+β,αβ.\alpha+\beta,\qquad \alpha\beta.

They cannot see which root was called α\alpha and which was called β\beta.

This is the first important point:

Vieta is not only a shortcut. It tells us which information about the roots is naturally visible from the coefficients.


Let

P(x)=i=1n(xri)P(x)=\prod_{i=1}^{n}(x-r_i)

be a monic polynomial with roots

r1,r2,,rn.r_1,r_2,\ldots,r_n.

When we expand the product,

P(x)=xne1xn1+e2xn2e3xn3++(1)nen,P(x)=x^n-e_1x^{n-1}+e_2x^{n-2}-e_3x^{n-3}+\cdots+(-1)^ne_n,

where

e1=iri,e_1=\sum_i r_i, e2=i<jrirj,e_2=\sum_{i<j}r_ir_j, e3=i<j<krirjrk,e_3=\sum_{i<j<k}r_ir_jr_k,

and so on, down to

en=r1r2rn.e_n=r_1r_2\cdots r_n.

These are called the elementary symmetric polynomials in the roots.

The word “elementary” means basic, not easy. These are the building blocks for many other expressions that are unchanged when the roots are renamed.

For a cubic,

(xu)(xv)(xw)=x3(u+v+w)x2+(uv+uw+vw)xuvw.(x-u)(x-v)(x-w)=x^3-(u+v+w)x^2+(uv+uw+vw)x-uvw.

So if

x3px2+qxr=0,x^3-px^2+qx-r=0,

then

e1=p,e2=q,e3=r.e_1=p,\qquad e_2=q,\qquad e_3=r.

For a quartic,

x4Ax3+Bx2Cx+D=0x^4-Ax^3+Bx^2-Cx+D=0

has

e1=A,e2=B,e3=C,e4=D.e_1=A,\qquad e_2=B,\qquad e_3=C,\qquad e_4=D.

An expression in the roots is symmetric if it is unchanged whenever the roots are renamed.

For roots u,v,wu,v,w, the expression

u2+v2+w2u^2+v^2+w^2

is symmetric.

The expression

u2v+v2w+w2uu^2v+v^2w+w^2u

is not symmetric. If we swap uu and vv, it becomes

v2u+u2w+w2v,v^2u+u^2w+w^2v,

which is usually different.

The test is:

If the roots are renamed, does the expression always keep the same value?

Coefficients naturally control symmetric information. They cannot by themselves identify an expression that depends on a preferred naming of the roots.


Let u,v,wu,v,w be the roots of

x3px2+qxr=0.x^3-px^2+qx-r=0.

Then

e1=p,e2=q,e3=r.e_1=p,\qquad e_2=q,\qquad e_3=r.

For example,

(u+v+w)2=u2+v2+w2+2(uv+uw+vw).(u+v+w)^2=u^2+v^2+w^2+2(uv+uw+vw).

Therefore

u2+v2+w2=e122e2=p22q.u^2+v^2+w^2=e_1^2-2e_2=p^2-2q.

Another useful expression is

(u+v)(v+w)(w+u).(u+v)(v+w)(w+u).

Expanding in a controlled way,

(u+v)(v+w)(w+u)=(u+v+w)(uv+uw+vw)uvw.(u+v)(v+w)(w+u)=(u+v+w)(uv+uw+vw)-uvw.

So

(u+v)(v+w)(w+u)=e1e2e3=pqr.(u+v)(v+w)(w+u)=e_1e_2-e_3=pq-r.

The calculation is useful, but the principle is more important:

If an expression is symmetric, first try to rewrite it using e1,e2,,ene_1,e_2,\ldots,e_n.


We will use the following theorem as a guiding principle.

Symmetric polynomial theorem, working version.
Every symmetric polynomial in r1,,rnr_1,\ldots,r_n can be written as a polynomial in

e1,e2,,en.e_1,e_2,\ldots,e_n.

For this course, we do not need the full proof. We do need to know how to use the idea in examples.

Boundary note: the word polynomial matters. Expressions involving division, square roots, or other non-polynomial operations need extra care.


A very important family of symmetric expressions is the power sums:

pk=r1k+r2k++rnk.p_k=r_1^k+r_2^k+\cdots+r_n^k.

The first few identities are

p1=e1,p_1=e_1, p2=e1p12e2,p_2=e_1p_1-2e_2, p3=e1p2e2p1+3e3.p_3=e_1p_2-e_2p_1+3e_3.

So

p2=e122e2,p_2=e_1^2-2e_2,

and

p3=e1(e122e2)e2e1+3e3=e133e1e2+3e3.p_3=e_1(e_1^2-2e_2)-e_2e_1+3e_3 =e_1^3-3e_1e_2+3e_3.

Power sums look different from Vieta expressions, but they are still symmetric. That is why they can be controlled by the same elementary data.


Let a,b,ca,b,c be real numbers such that

a+b+c=0,a+b+c=0,

and

(1+ax)(1+bx)(1+cx)=1+qx2+rx3(1+ax)(1+bx)(1+cx)=1+qx^2+rx^3

for all real xx.

Expanding the left hand side gives

1+(a+b+c)x+(ab+bc+ca)x2+abcx3.1+(a+b+c)x+(ab+bc+ca)x^2+abcx^3.

Since a+b+c=0a+b+c=0,

q=ab+bc+ca,r=abc.q=ab+bc+ca,\qquad r=abc.

Now define

Sn=an+bn+cnn.S_n=\frac{a^n+b^n+c^n}{n}.

The problem asks for the coefficient of xnx^n in

ln(1+qx2+rx3).\ln(1+qx^2+rx^3).

The key observation is

1+qx2+rx3=(1+ax)(1+bx)(1+cx).1+qx^2+rx^3=(1+ax)(1+bx)(1+cx).

So, as a power series about x=0x=0,

ln(1+qx2+rx3)=ln(1+ax)+ln(1+bx)+ln(1+cx).\ln(1+qx^2+rx^3) =\ln(1+ax)+\ln(1+bx)+\ln(1+cx).

Using

ln(1+t)=tt22+t33t44+,\ln(1+t)=t-\frac{t^2}{2}+\frac{t^3}{3}-\frac{t^4}{4}+\cdots,

we get

ln(1+ax)+ln(1+bx)+ln(1+cx)=n1(1)n+1an+bn+cnnxn.\ln(1+ax)+\ln(1+bx)+\ln(1+cx) =\sum_{n\ge1}(-1)^{n+1}\frac{a^n+b^n+c^n}{n}x^n.

Therefore the coefficient of xnx^n is

(1)n+1Sn.\boxed{(-1)^{n+1}S_n.}

This is not just a logarithm expansion. The logarithm turns a product into a sum, and the coefficients of that sum are power sums of the roots.


In the anchor problem, good writing should make the following visible:

  1. define SnS_n before using it;
  2. use a+b+c=0a+b+c=0 to remove the coefficient of xx;
  3. state why 1+qx2+rx31+qx^2+rx^3 factors as (1+ax)(1+bx)(1+cx)(1+ax)(1+bx)(1+cx);
  4. say that the logarithm is being expanded as a power series about x=0x=0;
  5. compare coefficients of xnx^n.

A compressed answer that only says “by expansion” hides too much.