Scribd Logo
Upload

Welcome to Scribd!

  • 2 views

    Continuity of The Roots of A Polynomial

    Uploaded by

    neeru572010

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Continuity of The Roots of A Polynomial

    Uploaded by

    neeru572010
    2 views3 pages

    Original Title

    Continuity_of_the_Roots_of_a_Polynomial

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    2 views3 pages

    Continuity of The Roots of A Polynomial

    Uploaded by

    neeru572010

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 3

    Continuity of the Roots of a Polynomial

    S. Kumaresan
    School of Math. and Stat.
    University of Hyderabad
    Hyderabad 500046
    kumaresa@gmail.com

    We give three proofs of the result which says that the roots of a polynomial depend
    “continuously” on the coefficients of the polynomial. The first proof uses Rouche’s theorem.
    The second proof is quite elementary. The third (if written) will justify what is in the quotes
    and is a highbrow proof.
    p
    Y
    Pn
    Theorem 1. Let f (z) := k=0 ak zk = an (z − zj )mj , an 6= 0 and g(z) := (a0 + ε0 )z +
    j=1
    · · · + (an−1 + εn−1 )z n−1 + an z n . Let 0 < rk ≤ minj6=k |zk − zj |. Then there exists an ε > 0
    such that if |εi | ≤ ε, then g has precisely mk zeros in B(zk , rk ) =: Bk .
    Pn−1 i P i
    P on Bk , hi := g−f , h(z) = i=0 εi z satisfies |h(z)| ≤ εi (|zi | + ri ) ≤ Mk ε,
    Proof. Note that
    where Mk := [|zi | + ri ] . But, on ∂Bk , we have
    Y Y Y
    |f (z)| = |an | |z − zj |mj = |an ||z − zk |mk |z − zj |mj ≥ |an |rkmk (|zj − zk | − rk )mj .
    j j6=k j6=k

    Call the right hand side of the last inequality as δk . So, if we choose
     
    δk
    ε < min ,1 ≤ k ≤ n ,
    Mk

    we then have |h(z)| < |f (z)| on ∂Bk . This means by Rouche’s theorem that f and f + h = g
    have the same number of zeroes in Bk . By our choice of rk , the only zeroes of f in Bk is zk
    with multiplicity mk . Hence the result.

    Theorem 2. Let
    Y
    f (z) = z n + an−1 z n−1 + · · · + a0 = (z − λi )
    Y
    g(z) = z n + αn−1 z n−1 + · · · + α0 = (z − µi ).

    Let λ be a root of f with multiplicity m and ε > 0 be given. Then for |ai − αi | sufficiently
    small for each i, g has at least m roots within an ε-distance of λ.

    1
    Proof. Suppose not. Then there exists a sequence {fk } of polynomials which converge to f
    such that fk has fewer than m roots in B(λ, ε). Since the coefficients {aki }k converge, the set
    (k) (k)
    {(a0 , . . . , an−1 | k ∈ N} is bounded in Cn .
    (k) (k) (k) (k)
    Let fk (z) = (x − λ1 ) · · · (x − λn ). Then {(λP
    1 ) . . . λn ) | k ∈ N} is a bounded subset of
    Cn . (For, if λ is a root of f , then |λ| ≤ max{1, ni=1 |ai |}.) It, therefore, has a convergent
    subsequence. Without loss of generality, assume λki → µi for each i.
    (k) (k)
    Now, since λki → µi , fk (z) = (z − λ1 ) · · · (z − λN ) converge to h(z) = (z − µ1 ) · · · (z −
    µn ). But then, by uniqueness of limits, h(z) = f (z). Hence, m of µi ’s must equal λ, a
    contradiction.

    Remark 3. By repeated application of the theorem to all distinct roots λi with multiplicity
    mi , we see that in the statement we can conclude g must have precisely m roots within an
    ε-distance of λ.

    Third Proof

    We assume that all polynomials P of degree n are normalized so that P (z) = z n +


    a1 z n−1 + · · · + an . We identify the polynomial P with the vector (a1 , . . . Q
    , an ) ∈ Cn . By the
    fundamental theorem of algebra we can factorize P as follows: P (z) = nj=1 (z − ξj ) for a
    finite set of elements ξj ∈ C. We also know that the coefficients aj are symmetric polynomials
    of the roots ξj : X
    ak = σk (ξ1 , . . . , ξn ) = ξj1 · · · ξjk
    where the sum is over all possible k subsets {j1 , . . . , jk } of {1, . . . , n}. Define σ : C → C be
    the map σ(ξ) := (σ1 (ξ), . . . , σn (ξ)). Then σ is continuous and σ is onto, by the fundamental
    theorem of algebra. However, σ is not one-one.
    Let Sn , the symmetric group on n symbols act on Cn by µ(z1 , . . . , zn ) := (zµ(1) , . . . , zµ(n) ).
    This induces an equivalence relation ∼ so that the equivalence classes are the orbits under
    the group action. Let Cn / ∼ be the quotient space and π : Cn → Cn / ∼ be the quotient map.
    Now let σ̃ be the unique map σ̃ : Cn / ∼→ Cn such that σ̃ ◦ π = σ. Using the standard results
    from the theory of quotient spaces, one shows easily that σ̃ is a continuous bijection. (At this
    point, it is very tempting to think of the result which says that a bijective continuous map of
    a compact space onto a Hausdorff space is a homeomorphism. See Remark 6 at the end.

    Proposition 4. The map σ̃ : Cn / ∼→ Cn is a homeomorphism.

    Proof. We define a metric on Cn / ∼ which induces the quotient topology as follows:

    d(π(z), π(w)) := min{|z 0 − w0 | : z 0 ∈ π(z), w0 ∈ π(w)}.

    Note that
    d(π(z), π(w)) := min{|z − w0 | : w0 ∈ π(w)}.
    Given z, v, w ∈ Cn , choose v 0 ∈ π(v) so that d(π(z), π(v)) = |z − v 0 |. For each w0 ∈ π(w), we
    then have |z − w0 | ≤ |z − v 0 | + |v 0 − w0 |. Hence,

    2
    d(π(z), π(w)) ≤ min{|z − v 0 | + |v 0 − w0 | : w0 ∈ π(w)}
    = d(π(z), π(v)) + d(π(v), π(w)).

    It follows that π : Cn → (Cn / ∼, d) is continuous.


    We now claim that d induces the quotient topology. Since π : C → (Cn / ∼, d) is a
    continuous bijection, it is enough to show that π is open. Let U ⊂ C be open. Let [α] ∈
    π(U ). We assume without loss of generality that α ∈ U . Then there exists r > 0 such that
    B(α, r) ⊂ U . We claim that B([α], r) ⊂ π(U ). Let [β] ∈ B([α], r). We can find β ∈ [β] such
    that d([α], [β]) = d(α, β). The result follows form this. (Alternative proof of the Claim: For
    any set A ⊂ Cn , we have π −1 (π(A)) is open (closed) if A is open (closed). Thus π is an open
    and closed map. Hence the claim.)
    Let B(0, R) denote the open ball in Cn / ∼. We claim that σ̃ is a homeomorphism of
    B(0, R) onto its image. It is enough to show that σ̃ is closed. Let K ⊂ B(0, R) be closed. Then
    π −1 (K) is a closed and bounded subset of Cn and hence is compact. Thus, σ̃(K) = σ(π −1 (K))
    is compact and hence closed in Cn .
    We complete the proof by showing that σ̃ : Cn / ∼→ Cn is open. Let U be open subset of
    Cn / ∼ and x ∈ U . Choose ε > 0 and R > 0 such that B(x, ε) ⊂ U and B(x, ε) ⊂ B(0, R).
    Since σ̃ is open on B(0, R), it follows that σ̃(x) lies in the interior of σ̃(B(x, ε)). Since x was
    arbitrary, this completes the proof.

    Theorem 5. Suppose
    k
    Y
    P (z) = z n + a1 z n−1 + · · · + an = (z − ξj )mj
    j=1

    for distinct ξ1 , . . . , ξk . Let ε > 0 be given such that for i 6= j, we B(ξi , ε) ∩ B(ξj , ε) = ∅. Then
    there exists δ > 0 so that b ∈ B(a, δ) implies that the polynomial

    Q(z) := z n + b1 z n−1 + · · · + bn

    has exactly mj roots (counted with multiplicity) in B(ξj , ε).

    Proof. Let τ : Cn → Cn / ∼ be inverse of σ̃. Given a ∈ Cn , the coefficients of P , by the last


    proposition, there exists a δ > 0 such that τ (B(a, δ)) ⊂ B(τ (a), ε). Spell this out in terms of
    the metrics on Cn and Cn / ∼ to arrive at the theorem.

    Remark 6. It is tempting to introduce some projective notions to “simplify” the proof of


    the proposition conceptually.

    You might also like