DOCSLIB.ORG

Absolute Convergence: True Trees from Short Sequences

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Absolute Convergence: True Trees from Short Sequences Absolute Convergence: True Trees From Short Sequences Tandy Warnow∗ Bernard M.E. Moret† Katherine St. John‡ Abstract evolution if, for every model tree (i.e., rooted tree and Fast-converging methods for reconstructing phylogenetic the associated random variables) and every ε > 0, there trees require that the sequences characterizing the taxa be is a sequence length k (which depends on the method, of only polynomial length, a major asset in practice, since the model tree, and ε) such that the method recovers the real-life sequences are of bounded length. However, of the topology (the edges) of the model tree with probability half-dozen such methods proposed over the last few years, at least 1−ε, if it is given sequences of length at least k. only two fulfill this condition without requiring knowledge For many models (such as the Jukes-Cantor model [13], of typically unknown parameters, such as the evolutionary the simplest four-state model, as well as more complex rate(s) used in the model; this additional requirement models, such as the General Markov (GM) [19] model), severely limits the applicability of the methods. We say that even simple distance methods are easily established to methods that need such knowledge demonstrate relative fast be statistically consistent. convergence, since they rely upon an oracle. We focus on The sequence length required by a method is a sig- the class of methods that do not require such knowledge and nificant aspect of its performance since real data sets are thus demonstrate absolute fast convergence. We give a very of limited length. (Computational requirements are also general construction scheme that not only turns any relative important, but it is possible to wait longer or get more fast-converging method into an absolute fast-converging one, powerful machines, while it is not possible to get longer but also turns any statistically consistent method that sequences than exist in nature.) Consequently, experi- converges from sequences of length O(eO(diam(T ))) into an mental and analytical studies have attempted to bound absolute fast-converging method. the sequence lengths required by different phylogenetic methods. Methods that perform well (with respect to 1 Introduction topology estimation) from sequences of realistic lengths (bounded by at most a few thousand nucleotides) are Phylogenetic reconstruction methods build an evolu- very desirable, especially if the topological accuracy re- tionary tree from a collection of taxa given, for example, mains good when the rate of evolution and number of by molecular sequences. These methods are designed to taxa increase. recover the “true” evolutionary tree as often as possible. In an earlier paper [12], we defined the notion of fast Not all are guaranteed to do so with high probability convergence under the Jukes-Cantor model of evolution. under reasonable conditions; even those that offer this In this paper we introduce the concept of absolute guarantee vary considerably in their requirements. Un- fast convergence and extend this definition to more der some models of evolution, no method can be guaran- general models, such as the General Time Reversible teed to recover the true tree with high probability, due Markov model. A method is absolute fast-converging if to unidentifiability. Under other models, many methods it is fast-converging and does not need to know any will be able to recover the tree if given long enough se- of the parameters of the model in order to achieve quences. The latter methods are said to be statistically fast convergence. In the Jukes-Cantor model, such consistent under the model of evolution. Formally, a parameters might be the minimum (f) or maximum (g) method is statistically consistent for a specific model of expected number of times a site changes on any edge in the tree. A method can be fast converging only if it ∗Dept. of Computer Science, U. of Texas at Austin, [email protected]; supported by NSF grant 94-57800 and by is given knowledge of one or both of these values: such the David and Lucile Packard Foundation. a method is not absolute fast-converging and we say †Dept. of Computer Science, U. of New Mexico, instead that it is relative fast-converging. [email protected]; supported in part by NSF grant 00-81404. Only a few methods have been proved to be abso- ‡ Graduate Center and Dept. of Math. and Computer Science, lute fast-converging even under the simple Jukes-Cantor Lehman College, City U. of New York, [email protected]; supported in part by NSF grant 99-73874. model: two “DCM-boosted” quartet methods [12] and the Short-Quartet methods [8, 9]. Methods that have lution of a single site is modeled through the use of been proved relative fast-converging under the Jukes- “stochastic substitution matrices,” 4 × 4 matrices (one Cantor model include DCM-boosted neighbor-joining for each tree edge) in which every row sums to 1. A (DCM-NJ) [12], the Harmonic Greedy Triplets (HGT) stochastic model of how a single site evolves can thus method of Cs˝ur¨osand Kao [4], and a method of Cryan, have up to 12 free parameters. The simplest such model Goldberg, and Goldberg (CGG) [3]. These methods is the Jukes-Cantor model, with one free parameter, and are only relative fast-converging under the Jukes-Cantor the most complex is the General Markov model, with all model, rather than absolute fast-converging, because 12 parameters [19]. they require knowledge of f or g; without such knowl- edge, they have to guess the parameter (or, more pre- Definition 1. The GM model of single-site evolution cisely, they have to guess which of the trees they have is defined as follows. constructed is the correct tree). Their guessing strate- 1. The nucleotide in a random site at the root is drawn gies are not provably correct. Consequently, in absence from a known distribution, in which each nucleotide of knowledge about f and g, these methods are not even has positive probability. statistically consistent. In this paper, we describe a very simple algorithm, 2. The probability of each site substitution on an which we call Short Quartet Support (SQS). This al- edge e of the tree is given by a 4 × 4 stochastic gorithm selects the true tree (under the GM model) substitution matrix M(e) in which det(M(e)) is not with high probability from a collection of trees, given 0, 1, or −1. sequences of only polynomial length. Since SQS does This model is generally used in a context where not require knowledge of the model parameters, it can all sites evolve identically and independently (the iid be used to turn a relative fast-converging method into assumption), although sometimes a distribution of rates an absolute fast-converging method. Consequently, across sites is also given. In this paper, we use the GM a straightforward use of SQS produces absolute fast- model with iid site evolution. converging versions of the relative fast-converging meth- We denote a model tree in the GM model as a pair, ods DCM-Neighbor-Joining [12], HGT [4], and CGG [3]. (T, {Me : e ∈ E(T )}), or more simply as (T,M). We SQS can also be combined with the first phase of assume that the number of changes obeys a Poisson the Disk-Covering Method (DCM) [12] to produce a distribution. For each edge e ∈ E(T ), we define the technique we call DCM∗ for reducing the dependency ∗ weight of the edge λ(e) to be − log |det(Me)|. This of methods on sequence lengths. In particular, DCM allows us to define the matrix of leaf-to-leaf distances, turns methods that converge from sequence lengths that P {λij}, with λij = λ(e) and where Pij is the grow exponentially in the diameter (the longest path e∈Pij path in T between leaves i and j. Note that {λ } is a length in the tree) into methods that are absolute fast- ij symmetric matrix. It is a well-known fact that, given converging. Since the diameter of an n-leaf tree can √ the distance matrix {λ }, it is easy to recover the be as large as n − 1 and is typically Ω( n), this tech- ij underlying leaf-labelled tree T in polynomial time. nique turns methods that are statistically consistent, This general model of site evolution subsumes the but not even relative fast-converging, into absolute fast- great majority of other models examined in the phyloge- converging methods. Finally, SQS provides a very gen- netic literature, including the Hasegawa-Kishino-Yano eral framework within which absolute fast-converging (HKY) model, the Kimura 2-parameter model (K2P), methods can be developed. the Kimura 3-ST model (K3ST), the Jukes-Cantor We state our results in terms of the General Time model (JC), etc. These models are all special cases Reversible Markov model of evolution, which contains, of the General Markov model, because they place as a special case, the Jukes-Cantor model. restrictions on the form of the stochastic substitution matrices (see [14] for more information about stochastic 2 Basics models of evolution). Most distance-based methods 2.1 Stochastic models of DNA sequence evolu- are statistically consistent under the General Markov tion. A model of DNA sequence evolution must de- model, because statistically consistent methods exist scribe the probability distribution of the four states, for estimating the matrix {λij} above. (A method for A, C, T, G, at the root, the evolution of a random site estimating the matrix {λij} is statistically consistent (i.e., position within the DNA sequence) and how the if each of the distance estimates dij converges to the evolution differs across the sites.
Load more

Recommended publications
  • Ch. 15 Power Series, Taylor Series
    Ch. 15 Power Series, Taylor Series 서울대학교 조선해양공학과 서유택 2017.12 ※ 본 강의 자료는 이규열, 장범선, 노명일 교수님께서 만드신 자료를 바탕으로 일부 편집한 것입니다. Seoul National 1 Univ. 15.1 Sequences (수열), Series (급수), Convergence Tests (수렴판정) Sequences: Obtained by assigning to each positive integer n a number zn z . Term: zn z1, z 2, or z 1, z 2 , or briefly zn N . Real sequence (실수열): Sequence whose terms are real Convergence . Convergent sequence (수렴수열): Sequence that has a limit c limznn c or simply z c n . For every ε > 0, we can find N such that Convergent complex sequence |zn c | for all n N → all terms zn with n > N lie in the open disk of radius ε and center c. Divergent sequence (발산수열): Sequence that does not converge. Seoul National 2 Univ. 15.1 Sequences, Series, Convergence Tests Convergence . Convergent sequence: Sequence that has a limit c Ex. 1 Convergent and Divergent Sequences iin 11 Sequence i , , , , is convergent with limit 0. n 2 3 4 limznn c or simply z c n Sequence i n i , 1, i, 1, is divergent. n Sequence {zn} with zn = (1 + i ) is divergent. Seoul National 3 Univ. 15.1 Sequences, Series, Convergence Tests Theorem 1 Sequences of the Real and the Imaginary Parts . A sequence z1, z2, z3, … of complex numbers zn = xn + iyn converges to c = a + ib . if and only if the sequence of the real parts x1, x2, … converges to a . and the sequence of the imaginary parts y1, y2, … converges to b. Ex.
    [Show full text]
  • Topic 7 Notes 7 Taylor and Laurent Series
    Topic 7 Notes Jeremy Orloff 7 Taylor and Laurent series 7.1 Introduction We originally defined an analytic function as one where the derivative, defined as a limit of ratios, existed. We went on to prove Cauchy's theorem and Cauchy's integral formula. These revealed some deep properties of analytic functions, e.g. the existence of derivatives of all orders. Our goal in this topic is to express analytic functions as infinite power series. This will lead us to Taylor series. When a complex function has an isolated singularity at a point we will replace Taylor series by Laurent series. Not surprisingly we will derive these series from Cauchy's integral formula. Although we come to power series representations after exploring other properties of analytic functions, they will be one of our main tools in understanding and computing with analytic functions. 7.2 Geometric series Having a detailed understanding of geometric series will enable us to use Cauchy's integral formula to understand power series representations of analytic functions. We start with the definition: Definition. A finite geometric series has one of the following (all equivalent) forms. 2 3 n Sn = a(1 + r + r + r + ::: + r ) = a + ar + ar2 + ar3 + ::: + arn n X = arj j=0 n X = a rj j=0 The number r is called the ratio of the geometric series because it is the ratio of consecutive terms of the series. Theorem. The sum of a finite geometric series is given by a(1 − rn+1) S = a(1 + r + r2 + r3 + ::: + rn) = : (1) n 1 − r Proof.
    [Show full text]
  • 11.3-11.4 Integral and Comparison Tests
    11.3-11.4 Integral and Comparison Tests The Integral Test: Suppose a function f(x) is continuous, positive, and decreasing on [1; 1). Let an 1 P R 1 be defined by an = f(n). Then, the series an and the improper integral 1 f(x) dx either BOTH n=1 CONVERGE OR BOTH DIVERGE. Notes: • For the integral test, when we say that f must be decreasing, it is actually enough that f is EVENTUALLY ALWAYS DECREASING. In other words, as long as f is always decreasing after a certain point, the \decreasing" requirement is satisfied. • If the improper integral converges to a value A, this does NOT mean the sum of the series is A. Why? The integral of a function will give us all the area under a continuous curve, while the series is a sum of distinct, separate terms. • The index and interval do not always need to start with 1. Examples: Determine whether the following series converge or diverge. 1 n2 • X n2 + 9 n=1 1 2 • X n2 + 9 n=3 1 1 n • X n2 + 1 n=1 1 ln n • X n n=2 Z 1 1 p-series: We saw in Section 8.9 that the integral p dx converges if p > 1 and diverges if p ≤ 1. So, by 1 x 1 1 the Integral Test, the p-series X converges if p > 1 and diverges if p ≤ 1. np n=1 Notes: 1 1 • When p = 1, the series X is called the harmonic series. n n=1 • Any constant multiple of a convergent p-series is also convergent.
    [Show full text]
  • The Ratio Test, Integral Test, and Absolute Convergence
    MATH 1D, WEEK 3 { THE RATIO TEST, INTEGRAL TEST, AND ABSOLUTE CONVERGENCE INSTRUCTOR: PADRAIC BARTLETT Abstract. These are the lecture notes from week 3 of Ma1d, the Caltech mathematics course on sequences and series. 1. Homework 1 data • HW average: 91%. • Comments: none in particular { people seemed to be pretty capable with this material. 2. The Ratio Test So: thus far, we've developed a few tools for determining the convergence or divergence of series. Specifically, the main tools we developed last week were the two comparison tests, which gave us a number of tools for determining whether a series converged by comparing it to other, known series. However, sometimes this P 1 P n isn't enough; simply comparing things to n and r usually can only get us so far. Hence, the creation of the following test: Theorem 2.1. If fang is a sequence of positive numbers such that a lim n+1 = r; n!1 an for some real number r, then P1 • n=1 an converges if r < 1, while P1 • n=1 an diverges if r > 1. If r = 1, this test is inconclusive and tells us nothing. Proof. The basic idea motivating this theorem is the following: if the ratios an+1 an eventually approach some value r, then this series eventually \looks like" the geo- metric series P rn; consequently, it should converge whenever r < 1 and diverge whenever r > 1. an+1 To make the above concrete: suppose first that r > 1. Then, because limn!1 = an r;, we know that there is some N 2 N such that 8n ≥ N, a n+1 > 1 an )an+1 > an: But this means that the an's are eventually an increasing sequence starting at some value aN > 0; thus, we have that limn!1 an 6= 0 and thus that the sum P1 n=1 an cannot converge.
    [Show full text]
  • 1 Convergence Tests
    Lecture 24Section 11.4 Absolute and Conditional Convergence; Alternating Series Jiwen He 1 Convergence Tests Basic Series that Converge or Diverge Basic Series that Converge X Geometric series: xk, if |x| < 1 X 1 p-series: , if p > 1 kp Basic Series that Diverge X Any series ak for which lim ak 6= 0 k→∞ X 1 p-series: , if p ≤ 1 kp Convergence Tests (1) Basic Test for Convergence P Keep in Mind that, if ak 9 0, then the series ak diverges; therefore there is no reason to apply any special convergence test. P k P k k Examples 1. x with |x| ≥ 1 (e.g, (−1) ) diverge since x 9 0. [1ex] k X k X 1 diverges since k → 1 6= 0. [1ex] 1 − diverges since k + 1 k+1 k 1 k −1 ak = 1 − k → e 6= 0. Convergence Tests (2) Comparison Tests Rational terms are most easily handled by basic comparison or limit comparison with p-series P 1/kp Basic Comparison Test 1 X 1 X 1 X k3 converges by comparison with converges 2k3 + 1 k3 k5 + 4k4 + 7 X 1 X 1 X 2 by comparison with converges by comparison with k2 k3 − k2 k3 X 1 X 1 X 1 diverges by comparison with diverges by 3k + 1 3(k + 1) ln(k + 6) X 1 comparison with k + 6 Limit Comparison Test X 1 X 1 X 3k2 + 2k + 1 converges by comparison with . diverges k3 − 1 √ k3 k3 + 1 X 3 X 5 k + 100 by comparison with √ √ converges by comparison with k 2k2 k − 9 k X 5 2k2 Convergence Tests (3) Root Test and Ratio Test The root test is used only if powers are involved.
    [Show full text]
  • Math 113 Lecture #30 §11.6: Absolute Convergence and the Ratio and Root Tests
    Math 113 Lecture #30 x11.6: Absolute Convergence and the Ratio and Root Tests Absolute Convergence. The Integral Test and the Comparison Tests apply to series with positive terms, and the Alternating Series Test applies to series the sign of whose terms alternate regularly. But how do we test the convergence or divergence of a series in which the terms are positive and negative without any regularity in the switching of the sign? The answer is to consider a related series in which all the terms are positive. P P Definitions. A series an is called absolute convergent if the series janj converges. P A series an that is convergent but not absolutely convergent is called conditionally convergent. Conditionally convergent series like alternating series 1 X (−1)n n n=1 are quite bizarre in that the order of the terms can be rearranged to sum to any real number. Absolutely convergent series like 1 X (−1)n n2 n=1 always sum to the same thing no matter how the order of the terms are rearranged. Here is another nice property of absolute convergent series. P P Theorem. If an is absolutely convergent, then an is convergent. Proof. Since either janj is either an or −an, then 0 ≤ an + janj ≤ 2janj: P P Assuming that an is absolutely convergent implies that 2janj is convergent. P The Comparison Test applies to show that (an + janj) is convergent too. Since an = an + janj − janj, then X X X an = an + janj − janj P which shows that an is the sum of two convergent series, and is convergent too.
    [Show full text]
  • Power Series, Taylor Series and Analytic Functions (Section 5.1)
    Power Series, Taylor Series and Analytic Functions (section 5.1) DEFINITION 1. A power series about x = x0 (or centered at x = x0), or just power series, is any series that can be written in the form 1 X n an(x − x0) ; n=0 where x0 and an are numbers. P1 n DEFINITION 2. A power series n=0 an(x − x0) is said to converge at a point x if the limit m X n lim an(x − x0) exists and finite. m!1 n=0 1 X n REMARK 3. A power series an(x − x0) always converges at x = x0. n=0 1 X EXAMPLE 4. For what x the power series xn = 1 + x + x2 + x3 + ::: converges ? n=0 m X xn = n=0 If jxj < 1 If jxj > 1 1 X n Absolute Convergence: The series an(x − x0) is said to converge absolutely at x if n=0 1 X n janj jx − x0j converges. n=0 If a series converges absolutely then it converges (but in general not vice versa). 1 X xn EXAMPLE 5. The series converges at x = −1, but it doesn't converges absolutely n n=1 there: 1 1 1 − + − ::: = ln 2 2 3 but the series of absolute values is the so-called harmonic series 1 1 1 + + + ::: 2 3 and it is divergent. 1 X n Fact: If the series an(x − x0) converges absolutely at x = x1 then it converges absolutely for n=0 all x such that jx − x0j < jx1 − x0j 1 This immediately implies the following: 1 X n THEOREM 6.
    [Show full text]
  • Half-Plane of Absolute Convergence
    221- Extremal values of Dirichlet L-functions in the half-plane of absolute convergence par JÖRN STEUDING RÉSUMÉ. On démontre que, pour tout 03B8 réel, il existe une infinité de s = 03C3 + it avec 03C3 ~ 1 + et t ~ +~ tel que log log t log log log log t Re {exp i 03B8) log L (s, } ~ log La démonstration est basée sur une version effective du théorème de Kronecker sur les approximations diophantiennes. ABSTRACT. We prove that for any real 03B8 there are infinitely many values of s = 03C3 + it with 03C3 ~ 1 + and t ~ +~ such that Re { exp ( i 03B8) log L (s, } ~ log The proof relies on an effective version of Kronecker’s approxima- tion theorem. 1. Extremal values Extremal values of the Riemann zeta-function in the half-plane of abso- lute convergence were first studied by H. Bohr and Landau [1]. Their results rely essentially on the diophantine approximation theorems of Dirichlet and Kronecker. Whereas everything easily extends to Dirichlet series with real coefficients of one sign (see [7], §9.32) the question of general Dirichlet se- ries is more delicate. In this paper we shall establish quantitative results for Dirichlet L-functions. Let q be a positive integer and let X be a Dirichlet character mod q. As usual, denote by s = Q + it with a, t E IEB, i2 = -1, a complex variable. Then the Dirichlet L-function associated to the character X is given by - . where the product is taken over all primes p; the Dirichlet series, and so the Euler product, converge absolutely in the half-plane 7 &#x3E; 1.
    [Show full text]
  • Generating Functions
    MATH 324 Summer 2012 Elementary Number Theory Notes on Generating Functions Generating Functions In this note we will discuss the ordinary generating function of a sequence of real numbers, and then find the generating function for some sequences, including the sequence of Mersenne numbers and the sequence of Fibonacci numbers. Definition. Given a sequence ak k≥0 of real numbers, the ordinary generating function, or generating f g function, of the sequence ak k≥0 is defined to be f g 1 k 2 G(x) = akx = a0 + a1x + a2x + : ( ) k=0 · · · ∗ X The sum is finite if the sequence is finite, and infinite if the sequence is infinite. In the latter case, we will think of ( ) as either a power series with x chosen so that the sum converges, or as a formal power series, where we ∗are not concerned with convergence. Example 1. Suppose that n is a fixed positive integer and n ak = k for k = 0; 1; : : : ; n: From the binomial theorem we have n n n n n n (1 + x) = + x + x2 + + x ; 0 1 2 · · · n n and the generating function for the sequence ak k is f g =0 n G(x) = (1 + x) : Note: The advantage of what we have done is that we have expressed G(x) in a very simple encoded form. If we know this encoded form for G(x); it is now easy to derive ak by decoding or expanding G(x) as a k power series and finding the coefficient of x : Often we will be able to find G(x) without knowing ak; and then be able to solve for ak by expanding G(x): Example 2.
    [Show full text]
  • Improper Integrals
    Improper Integrals R. C. Daileda By an improper integral we will mean an integral of the form Z 1 f(x) dx: (1) a The goal of this note is to carefully define, and then study the properties of, improper integrals. To this end, let a 2 R and let f be a function that is Riemann integrable on every finite subinterval of [a; 1). We then define Z 1 Z A f(x) dx = lim f(x)dx a A!1 a provided the limit on the right-hand side exists. In this case we say that the improper integral (1) converges and that its value is that of the limit. Otherwise we say (1) diverges. It turns out that the theory of improper integrals closely mirrors that of infinite series, and we will prove several results which can be viewed as direct analogues of results about series. For example, one can use the usual limit laws to immediately verify that improper integrals (when convergent) obey the usual linearity of definite integrals. Another more powerful result is the following alternate formulation of convergence. Theorem 1 (Cauchy Criterion). The improper integral (1) converges if and only if for every > 0 there is an M ≥ a so that for all A; B ≥ M we have Z B f(x) dx < . A Proof. Suppose that the improper integral converges to L. Let > 0. Using the definition of convergence, choose M ≥ a so large that if A ≥ M then Z A f(x) dx − L < : a 2 Then if B ≥ M as well we have Z B Z B Z A f(x) dx = f(x) dx − f(x)dx A a a Z B Z A = f(x) dx − L + L − f(x)dx a a Z B Z A ≤ f(x) dx − L + L − f(x)dx a a < + = .
    [Show full text]
  • Absolute Convergence and the Ratio and Root Tests
    Section 11.6: Absolute Convergence and the Ratio and Root Tests Absolute and Conditional Convergence: Absolute Convergence Test (ACT): If ∑|| converges, then ∑ also converges. Definitions: 1. ∑ is absolutely convergent or converges absolutely (CA) if ∑|| converges. 2. ∑ is conditionally convergent or converges conditionally (CC) if ∑ converges but ∑|| diverges. NOTE: After determining convergence by the Alternating Series Test (AST), then use Integral Test, Comparison Test (CT), or Limit Comparison Test (LCT) on ∑|| to determine absolute convergence (CA) or conditional convergence (CC). Examples: 1 1. 1 2 1 2. ⁄ 1 ln 3. sin 4. ⁄ 3 The Ratio Test: The Ratio Test (RT): Let ∑ be a series of nonzero terms and suppose | | lim . → || (i) If 1, the series converges absolutely. (ii) If 1, the series diverges. (iii) If 1, the test is inconclusive. NOTES: 1. For any type of series: positive, alternating, or other. 2. If 1, the test fails. You must use a different test. 3. If ∞, the series diverges ( ρ does not have to be finite for this test). 4. If 0, the series converges ( ρ can have a value of 0 in this test). 5. This test is most useful with series involving powers and factorials. Useful Facts for Factorials: 1. 2 ⋅ 4 ⋅ 6 ⋅ ⋯ ⋅ 2 2! 2 1! 2 1! 2. 1 ⋅ 3 ⋅ 5 ⋅ ⋯ ⋅ 2 1 2⋅4⋅6⋅⋯⋅2 2! Examples: 13 1. ! 4 12⋅4⋅6⋅⋯⋅2 2. 4 The Root Test: The Root Test (RoT): Let ∑ be a series of nonzero terms and suppose lim || . → (i) If 1, the series converges absolutely. (ii) If 1, the series diverges.
    [Show full text]
  • Changing the Order of Integration
    CHANGING THE ORDER OF INTEGRATION E. R. LOVE (Received 15 April 1969) Communicated by E. Strzelecki Repeated improper Riemann integrals arise in a variety of contexts, and the validity of changing the order of integration is often in question. Fubini's theorem ensures the equality of two repeated Lebesgue integrals when one of them is ab- solutely convergent. For many years I have assumed that an analogous test is ap- plicable to repeated improper i?-integrals, since they will be absolutely convergent and therefore in agreement with the corresponding L-integrals. There is a fallacy in this reasoning. The assumed absolute convergence of one of the repeated improper i?-integrals may not ensure absolute convergence of the other. It certainly ensures absolute convergence and equality of both the corresponding repeated ^-integrals; but there may be no connection between these and the other repeated improper i?-integral if this is only conditionally convergent. The first purpose of this note is to demonstrate by an example that this pos- sibility must be taken seriously. Example 1 shows that reversal of order in an ab- solutely convergent repeated improper i?-integral can destroy absolute conver- gence while preserving convergence; and this even for a continuous integrand. The other purpose is to give theorems justifying change of order in repeated improper i?-integrals. In Theorems 1,2, 3 respectively it is assumed that both, one, neither repeated integral is absolutely convergent. Theorem 2 justifies my former erring ways, after Example 1 has shown that absolute convergence of both repeated integrals cannot be proved from that of one.
    [Show full text]