## On the distribution of spacings between zeros of the zeta function (1987)

Venue: | MATH. COMP |

Citations: | 83 - 9 self |

### BibTeX

@ARTICLE{Odlyzko87onthe,

author = {A. M. Odlyzko},

title = {On the distribution of spacings between zeros of the zeta function},

journal = {MATH. COMP},

year = {1987},

volume = {48},

pages = {273--308}

}

### Years of Citing Articles

### OpenURL

### Abstract

A numerical study of the distribution of spacings between zeros of the Riemann zeta function is presented. It is based on values for the first 10 5 zeros and for zeros number 10 12 + 1 to 10 12 + 10 5 that are accurate to within ± 10 − 8, and which were calculated on the Cray-1 and Cray X-MP computers. This study tests the Montgomery pair correlation conjecture as well as some further conjectures that predict that the zeros of the zeta function behave similarly to eigenvalues of random hermitian matrices. Matrices of this type are used in modeling energy levels in physics, and many statistical properties of their eigenvalues are known. The agreement between actual statistics for zeros of the zeta function and conjectured results is generally good, and improves at larger heights. Several initially unexpected phenomena were found in the data and some were explained by

### Citations

667 |
Random Matrices
- Mehta
- 1991
(Show Context)
Citation Context ...ir pair correlations becomes 1 − ( ( sin πu)/(πu) ) 2 . (The pair correlations of the GOE and GSE are different.) For more information about the GUE and the other ensembles, the reader is referred to =-=[2, 4, 11, 41, 54, 59, 63]-=-. The possible connection between zeros of the zeta function and eigenvalues of random matrices is of interest in number theory because of the Hilbert and Po ´lya conjectures [56, 59] which say that t... |

521 |
The Advanced Theory of Statistics
- KENDALL, STUART
- 1967
(Show Context)
Citation Context ...h the GUE predictions. One can obtain a better quantitative measure of the degree of fit between the observed and the predicted distributions using the histogram data (e.g., one can use the χ 2 -test =-=[47]-=- or the asymptotic results of [24]). In our case, since we know the predicted distribution quite well (at least numerically), we will use the quantile-quantile (q − q) plot to detect deviations betwee... |

234 |
Riemann’s Zeta Function
- Edwards
- 1974
(Show Context)
Citation Context ...hat has been observed so far appears to be S(t) = 2. 3136 . . . for some t near γ n , n = 1 , 333 , 195 , 692 [R]. (Large values of S(t) are associated to violations of Gram’s and Rosser’s rules, see =-=[10, 23, 49, 59]-=-, which explains why some statistics about them are available.) Aside from (2.6) and Selberg’s estimate (2.16), several other estimates have been proved. For example, if S 1 (t) = ∫ 0 t S(u) du , then... |

197 |
On Epstein’s zeta-function
- Titchmarsh
- 1934
(Show Context)
Citation Context ... has even conjectured [58] that (2.14) represents the right rate of growth of S(t). For statistical studies of the kind we are undertaking, though, the typical size of S(t) is more important. Selberg =-=[66, 68]-=- has shown that for all k ε Z + , ∫ 0 T 2k S(t) dt ∼ _ ________ ( 2k) ! k T ( log log T) as T → ∞ . (2.16) 2k k! ( 2π) This means that a typical value of S(t) is on the order of ( log log t) 1/2 , an ... |

184 |
Statistical theory of energy levels of complex systems
- Dyson
- 1962
(Show Context)
Citation Context ...ctions do not satisfy the RH in general. We briefly discuss large δ n. By [15], π log p( 0 ,t) ∼ − 8 2 _ __ 2 t as t → ∞ (2.25) (a result that had been derived earlier but less rigorously by F. Dyson =-=[22]-=-). Hence the GUE prediction is that max δ n ∼ π N + 1 ≤ n ≤ N + M − 1 ( 8 log M) 1/2 as N,M → ∞ with M not too small compared with N, and this implies Montgomery’s conjecture [58; Eq. (11)] that max (... |

134 |
The Riemann Zeta-function
- Ivić
- 1985
(Show Context)
Citation Context ...ates in that case are much easier. (Values of γ n for 1 ≤ n ≤ 2000 were taken from the 105 decimal place values computed in [60].) The computations described here relied on the Riemann-Siegel formula =-=[23, 43, 68]-=-,s- 26 - which requires on the order of t 1/2 operation to compute ζ( 1/2 + it). A new method has recently been invented [61] which enables one to compute on the order of T 1/2 values of ζ( 1/2 + it) ... |

61 |
The pair correlation of zeros of the zeta function
- Montgomery
- 1973
(Show Context)
Citation Context ...ed from studying problems harder than the RH if the RH itself could not be proved. A major step towards a detailed study of the distribution of zeros of the zeta function was made by H. L. Montgomery =-=[56, 57]-=-. Under the assumption of the RH, he showed that if we defines− 1 F(α , T) = 2π(T log T) - 2 - Σ 0 < γ ≤ T 0 < γ ′ ≤ T T iα (γ − γ ′ ) 4 + (γ − γ ′ ) 2 _ ___________ 4 (1.1) for α and T real, T ≥ 2, w... |

44 |
les “formules explicites” de la théorie des nombres premiers
- Weil, Sur
- 1952
(Show Context)
Citation Context ...s that connect zeros of the zeta function and prime numbers. Some of them, such as the exact formula for π(x) in terms of the zeros, or the more general ‘‘explicit formulas’’ of Guinand [38] and Weil =-=[72]-=- are well known. A related formula was proved by Landau [48]. (See [37] for a somewhat stronger version.) It says, under the assumption of the RH (which we assume for convenience, although Landau prov... |

38 |
The differences between consecutive primes
- Heath-Brown
- 1979
(Show Context)
Citation Context ...eros of many Dirichlet Lfunctions, then the analogue of Montgomery’s conjecture (1.3) is true for 1 ≤ α ≤ 2. For other results about Montgomery’s pair correlation conjecture and its consequences, see =-=[30, 31, 34, 35, 36, 42, 44, 59]-=-. Overall, though, there is very little solid theoretical evidence in favor of even the Montgomery pair correlation conjecture (1.4), and essentially none for the speculative conjectures discussed abo... |

31 |
Contributions to the theory of the Riemann zeta-function
- Selberg
- 1946
(Show Context)
Citation Context ... has even conjectured [58] that (2.14) represents the right rate of growth of S(t). For statistical studies of the kind we are undertaking, though, the typical size of S(t) is more important. Selberg =-=[66, 68]-=- has shown that for all k ε Z + , ∫ 0 T 2k S(t) dt ∼ _ ________ ( 2k) ! k T ( log log T) as T → ∞ . (2.16) 2k k! ( 2π) This means that a typical value of S(t) is on the order of ( log log t) 1/2 , an ... |

29 |
On the zeros of the Riemann zeta function in the critical strip
- Lune, Riele, et al.
- 1986
(Show Context)
Citation Context ...long time, and many unsuccessful attempts have been made to either prove or disprove it by analytic methods. A series of large computations have also been made, culminating in the recent verification =-=[52]-=- that the RH holds for the first 1. 5 × 10 9 zeros, an effort that involved over a thousand hours on a modern supercomputer. A proof of the RH would lead to tremendous improvements in estimates for va... |

28 |
On pair correlations of zeros and primes in short intervals, Analytic number theory and Diophantine problems
- Goldston, Montgomery
- 1984
(Show Context)
Citation Context ...eros of many Dirichlet Lfunctions, then the analogue of Montgomery’s conjecture (1.3) is true for 1 ≤ α ≤ 2. For other results about Montgomery’s pair correlation conjecture and its consequences, see =-=[30, 31, 34, 35, 36, 42, 44, 59]-=-. Overall, though, there is very little solid theoretical evidence in favor of even the Montgomery pair correlation conjecture (1.4), and essentially none for the speculative conjectures discussed abo... |

28 | te Riele, Disproof of the Mertens conjecture
- Odlyzko, J
- 1985
(Show Context)
Citation Context ...s discovered by an application of a method for constructing values of t for which ⎪Z(t)⎪is large that was based on the Lovasz lattice basis reduction algorithm in a manner somewhat similar to that of =-=[60]-=-. Several other values of t < ∼ 10 11 with ⎪Z(t)⎪large were constructed that way, and a substantial fraction of them corresponded to violations of Rosser’s rule, although all the violations discovered... |

24 |
On the distribution of primes in short intervals, Mathematika 23
- Gallagher
- 1976
(Show Context)
Citation Context ...er number theoretic functions. The number of primes in short intervals, for example, is observed experimentally and is conjectured theoreticallys- 3 - to be distributed like a Poisson random variable =-=[29]-=-. The conjecture (1.4) suggests some further topics for investigation. In the language of mathematical physics, this conjecture says that 1 − ( ( sin πu)/(πu) ) 2 is the pair correlation function of z... |

14 |
Extreme values of the Riemann zeta function
- MONTGOMERY
- 1977
(Show Context)
Citation Context ... The true rate of growth is likely to be much smaller, though. It has been known [68; Ch. 14.12] for a long time that on the RH, S(t) = Ω ± ( ( log t) 1/2 − ε ) as t → ∞ for any ε > 0, and Montgomery =-=[58]-=- has shown more recently that It is thought likely that S(t) = Ω ± ( ( log t) 1/2 ( log log t) − 1/2 ) as t → ∞ . (2.14) S(t) = O( ( log t) 1/2 + ε ) as t → ∞ (2.15) for every ε > 0 (cf. [49]) and Mon... |

13 | A test of a computer’s floating-point arithmetic unit
- Schryer
- 1981
(Show Context)
Citation Context ...g point arithmetic unit, but instead it is sufficient to test whether a particular machine obeys the specifications of the model. It is of course impractical to test all possible cases. N. L. Schryer =-=[64, 65]-=- has implemented a very extensive set of tests designed to reveal deviations from the rules of the model and has used it to test many computers. Schryer’s tests are based on the fact that there tend t... |

11 |
On the distribution of the zeros of the Riemann zeta function in short intervals
- Fujii
- 1975
(Show Context)
Citation Context ...gnitude is probably again around ( log t) 1/2 . Therefore there is a lot of cancellation in the local behavior of S(t). Many results about the distribution of values of S(t + h) − S(t) are also known =-=[25, 26, 27, 32, 33, 44, 69]-=-. For example, Theorem 13.2 of [69] says that if 1/2 < η ≤ 1, then for T η < H ≤ T, 0 < h < 1, T + H ∫ T ⎧ ⎨S(t + h) − S(t) � ⎫ ⎬ ⎭ 2k dt = HA k ⎧ ⎨log ( 2 + h log T) � ⎫ ⎬ ⎭ ⎧ + O⎪H c ⎪ � k k k ⎧ ⎨k ... |

11 |
A summation formula in the theory of prime numbers
- Guinand
- 1948
(Show Context)
Citation Context ...in the formulas that connect zeros of the zeta function and prime numbers. Some of them, such as the exact formula for π(x) in terms of the zeros, or the more general ‘‘explicit formulas’’ of Guinand =-=[38]-=- and Weil [72] are well known. A related formula was proved by Landau [48]. (See [37] for a somewhat stronger version.) It says, under the assumption of the RH (which we assume for convenience, althou... |

11 | Über Riemanns Nachlass zur analytischen Zahlentheorie. Quellen und Studien zur Geschichte de Mathematik, Astromie und Physik - Siegel - 1933 |

10 |
Primes and zeros in short intervals
- Gallagher, Mueller
- 1978
(Show Context)
Citation Context ...eros of many Dirichlet Lfunctions, then the analogue of Montgomery’s conjecture (1.3) is true for 1 ≤ α ≤ 2. For other results about Montgomery’s pair correlation conjecture and its consequences, see =-=[30, 31, 34, 35, 36, 42, 44, 59]-=-. Overall, though, there is very little solid theoretical evidence in favor of even the Montgomery pair correlation conjecture (1.4), and essentially none for the speculative conjectures discussed abo... |

9 |
Gaps between primes, and the pair correlation of zeros of the zeta-function
- Heath-Brown
- 1982
(Show Context)
Citation Context |

9 |
Distribution of the zeros of the Riemann zeta function
- MONTGOMERY
- 1974
(Show Context)
Citation Context ...ed from studying problems harder than the RH if the RH itself could not be proved. A major step towards a detailed study of the distribution of zeros of the zeta function was made by H. L. Montgomery =-=[56, 57]-=-. Under the assumption of the RH, he showed that if we defines− 1 F(α , T) = 2π(T log T) - 2 - Σ 0 < γ ≤ T 0 < γ ′ ≤ T T iα (γ − γ ′ ) 4 + (γ − γ ′ ) 2 _ ___________ 4 (1.1) for α and T real, T ≥ 2, w... |

8 |
On the Riemann zeta function, mean value theorems and the distribution of
- Ghosh
- 1983
(Show Context)
Citation Context ...gnitude is probably again around ( log t) 1/2 . Therefore there is a lot of cancellation in the local behavior of S(t). Many results about the distribution of values of S(t + h) − S(t) are also known =-=[25, 26, 27, 32, 33, 44, 69]-=-. For example, Theorem 13.2 of [69] says that if 1/2 < η ≤ 1, then for T η < H ≤ T, 0 < h < 1, T + H ∫ T ⎧ ⎨S(t + h) − S(t) � ⎫ ⎬ ⎭ 2k dt = HA k ⎧ ⎨log ( 2 + h log T) � ⎫ ⎬ ⎭ ⎧ + O⎪H c ⎪ � k k k ⎧ ⎨k ... |

8 |
Bohigas: Fluctuation Properties of Nuclear Energy Levels: Do Theory and Experiment
- Haq, Pandey, et al.
- 1982
(Show Context)
Citation Context ...ir pair correlations becomes 1 − ( ( sin πu)/(πu) ) 2 . (The pair correlations of the GOE and GSE are different.) For more information about the GUE and the other ensembles, the reader is referred to =-=[2, 4, 11, 41, 54, 59, 63]-=-. The possible connection between zeros of the zeta function and eigenvalues of random matrices is of interest in number theory because of the Hilbert and Po ´lya conjectures [56, 59] which say that t... |

7 |
Distribution Theorems of L-functions
- JOYNER
- 1986
(Show Context)
Citation Context |

7 |
The probabilities for several consecutive eigenvalues of a random matrix
- Mehta, Cloizeaux, et al.
- 1972
(Show Context)
Citation Context ...449 , 114. Thus these values provide upper bounds for the minimal δ n in the intervals investigated by those authors, and it is likely that they are the minimal values. . . .s- 19 - Since it is known =-=[54, 55]-=- that p( 0 ,t) has the Taylor series expansion near t = 0 given by π p( 0 ,t) = 3 2 _ __ 2 2π t − 45 4 _ ___ 4 π t + 315 6 _ ___ 6 t +... , (2.23) (for comparison, we note that p( 1 , t) = π 6 t 7 /40... |

6 |
Asymptotische Entwicklungen der Dirichletschen L-Reihen
- Deuring
- 1967
(Show Context)
Citation Context ...ture would imply some unusual behavior among the primes. It would also be desirable to obtain numerical data for zeros of Dirichlet L-functions, using the generalization of the Riemann-Siegel formula =-=[20, 21]-=- or the new method of [61]. Some data are available for zeros of Epstein zeta functions [8], and they don’t obey the GUE predictions, as they have many small spacings between consecutive zeros. This i... |

6 |
On the histogram as a density estimator
- Freedman, Diaconis
- 1981
(Show Context)
Citation Context ...ain a better quantitative measure of the degree of fit between the observed and the predicted distributions using the histogram data (e.g., one can use the χ 2 -test [47] or the asymptotic results of =-=[24]-=-). In our case, since we know the predicted distribution quite well (at least numerically), we will use the quantile-quantile (q − q) plot to detect deviations between the empirical and theoretical di... |

6 |
correlation of zeros of the zeta function
- Gallagher
- 1985
(Show Context)
Citation Context |

6 |
Zeros of the Riemann zeta function: Conjectures and computations, manuscript in preparation
- Odlyzko
(Show Context)
Citation Context ...ir pair correlations becomes 1 − ( ( sin πu)/(πu) ) 2 . (The pair correlations of the GOE and GSE are different.) For more information about the GUE and the other ensembles, the reader is referred to =-=[2, 4, 11, 41, 54, 59, 63]-=-. The possible connection between zeros of the zeta function and eigenvalues of random matrices is of interest in number theory because of the Hilbert and Po ´lya conjectures [56, 59] which say that t... |

5 | Stochastic behavior in quantum scattering - Gutzwiller - 1983 |

5 | Turing, A method for the calculation of the zeta-function - M - 1943 |

4 |
An approximate functional equation for Dirichlet L-functions
- Davies
- 1965
(Show Context)
Citation Context ...ture would imply some unusual behavior among the primes. It would also be desirable to obtain numerical data for zeros of Dirichlet L-functions, using the generalization of the Riemann-Siegel formula =-=[20, 21]-=- or the new method of [61]. Some data are available for zeros of Epstein zeta functions [8], and they don’t obey the GUE predictions, as they have many small spacings between consecutive zeros. This i... |

4 |
On the zeros of Dirichlet L-functions
- Fujii
- 1976
(Show Context)
Citation Context ...gnitude is probably again around ( log t) 1/2 . Therefore there is a lot of cancellation in the local behavior of S(t). Many results about the distribution of values of S(t + h) − S(t) are also known =-=[25, 26, 27, 32, 33, 44, 69]-=-. For example, Theorem 13.2 of [69] says that if 1/2 < η ≤ 1, then for T η < H ≤ T, 0 < h < 1, T + H ∫ T ⎧ ⎨S(t + h) − S(t) � ⎫ ⎬ ⎭ 2k dt = HA k ⎧ ⎨log ( 2 + h log T) � ⎫ ⎬ ⎭ ⎧ + O⎪H c ⎪ � k k k ⎧ ⎨k ... |

4 |
Neue Herleitung und explicite Restabschatzung der Riemann-Siegel-Formel
- Gabcke
- 1979
(Show Context)
Citation Context ... 1/2 in a total of O(T 1/2 + ε ) operations for all ε > 0 (i.e., O(T ε ) per value), but this method is quite involved and it has yet to be implemented. and We first recall the Riemann-Siegel formula =-=[10, 23, 28, 43, 59, 68]-=-. Let θ(t) = arg [π − it /2 Γ( 1/4 + it /2 ) ] , Z(t) = exp (i θ (t) ) ζ( 1/2 + it) . Then Z(t) is real for real t, and any sign change of Z(t) corresponds to a zero of ζ(s) on the critical line. We d... |

4 |
Ozluk, Pair correlation of zeros of Dirichlet L-functions
- E
- 1982
(Show Context)
Citation Context ...values might apply at least approximately to the zeros of the zeta function. Since Montgomery’s proof of (1.2) (which depends on the RH) was published, several other results have been obtained. Ozluk =-=[62]-=- has considered a q-analogue of Montgomery’s method and showed, roughly speaking, that if one considers a function similar to the F(α , T) of (1.1), but where one sums over zeros of many Dirichlet Lfu... |

3 |
On Riemann’s zeta function — Sign changes of S(t
- Ghosh
- 1981
(Show Context)
Citation Context |

3 |
On the distribution of the zeros of the Riemann zeta function
- Lehman
- 1970
(Show Context)
Citation Context ...ntgomery [58] has shown more recently that It is thought likely that S(t) = Ω ± ( ( log t) 1/2 ( log log t) − 1/2 ) as t → ∞ . (2.14) S(t) = O( ( log t) 1/2 + ε ) as t → ∞ (2.15) for every ε > 0 (cf. =-=[49]-=-) and Montgomery has even conjectured [58] that (2.14) represents the right rate of growth of S(t). For statistical studies of the kind we are undertaking, though, the typical size of S(t) is more imp... |

3 |
Odlyzko and A. Scho . . nhage, Fast algorithms for multiple evaluations of the Riemann zeta function
- M
(Show Context)
Citation Context ... × 10 11 , and given the available computational resources and the method used it was about as high as one could go. (A more efficient algorithm for evaluating the zeta function was invented recently =-=[61]-=-, but it is quite complicated and has not yet been implemented.) The computations used on the order 20 hours on the Cray X-MP, and the capabilities of this modern supercomputer were essential for the ... |

3 |
A Fortran computer program for calculating the linear prolate functions
- Buren
- 1976
(Show Context)
Citation Context ...), which is where one would expect numerical differentiation routines to be least accurate. The f n (x) are essentially the linear prolate functions. They were computed using the program of Van Buren =-=[71]-=-, with slight modifications introduced by this author and S. P. Lloyd. In Van Buren’s notation, the λ n defined above become λ n (πt /2 ), and the f n (x) become (λ n (πt /2 ) ) − 1/2 ψ n (πt /2 ,x). ... |

3 |
Riele, Optimization of a program for the verification of the Riemann hypothesis, Supercomputer 5
- Winter, te
- 1985
(Show Context)
Citation Context ...be used for zeros γ n with n not much larger than 3 . 10 12 due to the limitations on memory storage. In terms of speed, on the Cray X-MP it ran at about two thirds of the speed of the latest version =-=[73]-=- of the Cyber-205 program (written partiallys- 39 - in assembly language) that was used for the verification of the RH for the first 1. 5 × 10 9 zeros [52], since on average it required about 3. 5 . 1... |

2 |
On the zeros of Dirichlet L-functions. II (With corrections to ‘‘On the zeros of Dirichlet L-function
- Fujii
- 1981
(Show Context)
Citation Context |

2 |
Prime numbers and the pair correlation of zeros of the zeta function
- Goldston
- 1985
(Show Context)
Citation Context |

2 |
A formula of Landau and mean values of ζ(s
- Gonek
(Show Context)
Citation Context ...em, such as the exact formula for π(x) in terms of the zeros, or the more general ‘‘explicit formulas’’ of Guinand [38] and Weil [72] are well known. A related formula was proved by Landau [48]. (See =-=[37]-=- for a somewhat stronger version.) It says, under the assumption of the RH (which we assume for convenience, although Landau proved an unconditional result), that for any y > 0 as N → ∞,sn = 1 ΣN e i ... |

2 |
ber die Nullstellen der Zetafunktion
- Landau, U
- 1911
(Show Context)
Citation Context ... Some of them, such as the exact formula for π(x) in terms of the zeros, or the more general ‘‘explicit formulas’’ of Guinand [38] and Weil [72] are well known. A related formula was proved by Landau =-=[48]-=-. (See [37] for a somewhat stronger version.) It says, under the assumption of the RH (which we assume for convenience, although Landau proved an unconditional result), that for any y > 0 as N → ∞,sn ... |

2 |
de Lune, Some observations concerning the zero-curves of the real and imaginary parts of Riemann’s zeta function
- van
- 1983
(Show Context)
Citation Context ...ich would enable one to explore the behavior of Z(t) near such special points where it is large.) Other methods of constructing values of t with interesting behavior of Z(t) are presented in [46] and =-=[50]-=-. Van de Lune [50] has found several values of t for which ⎪Z(t)⎪ is very large. Thanks to D. Hejhal, the National Science Foundation, and the Minnesota Supercomputer Institute, this author was able t... |

2 |
Rigorous high speed separation of zeros of Riemann’s zeta function
- Lune, Riele, et al.
- 1981
(Show Context)
Citation Context ...ore accurate. The error estimates in it are not completely rigorous, as the error bounds assumed for the manufacturer’s cosine routine are based on statistical sampling. The main program described in =-=[51, 52, 73]-=- used the manufacturer’s dp cosine routines only to compute a few values to medium accuracy. These values were then used inside the program to obtain sp values of the cosine by linear interpolation. T... |

2 |
The Distribution of the Values
- Tsang
- 1984
(Show Context)
Citation Context |

1 |
On the Dyson-Montgomery hypothesis
- Joyner
(Show Context)
Citation Context ...this implies Montgomery’s conjecture [58; Eq. (11)] that max (γ n + 1 − γ n ) ∼ 8 ( 2 log T) γ n < T − 1/2 . (2.26) ⎪S(t)⎪ = O( ( log t) 1/2 ( log log t) − 1/2 ) as well as another argument of Joyner =-=[45]-=- suggest that max (γ n + 1 − γ n ) = O( ( log T) γ n < T − 1/2 ( log log T) − 1/2 ) , (2.27)s- 21 - which contradicts (2.26). Because of the slow growth of ( log T) 1/2 and of ( log log T) 1/2 , our d... |

1 |
Einige Ausnahmen zur Rosserschen Regel in der Theorie der Riemannschen Zetafunktion, Computing 27
- Karkoschka, Werner
- 1981
(Show Context)
Citation Context ...uture, which would enable one to explore the behavior of Z(t) near such special points where it is large.) Other methods of constructing values of t with interesting behavior of Z(t) are presented in =-=[46]-=- and [50]. Van de Lune [50] has found several values of t for which ⎪Z(t)⎪ is very large. Thanks to D. Hejhal, the National Science Foundation, and the Minnesota Supercomputer Institute, this author w... |