Profile cover photo
Profile photo
Fredrik Johansson
158 followers
158 followers
About
Posts

Post has attachment
http://fredrikj.net/blog/2018/07/interesting-interference-integrals/

Diffraction catastrophe integrals describe complex wave interference phenomena. This post is about a simple test implementation in Arb. Scroll to the end for a note on T. Pearcey's remarkable hand-drawn plots from 1946!
Add a comment...

Post has attachment

Post has attachment

Post has attachment
http://fredrikj.net/blog/2018/06/the-arb-matrix-reloaded/

Basic linear algebra in Arb is getting a significant upgrade in Arb 2.14 which is due to be released within a few weeks. The main improvements are more accurate linear solving and faster matrix multiplication.
Add a comment...

Post has attachment

Post has attachment
Spektrum der Wissenschaft (the German edition of Scientific American) recently asked me if they could use one of my renders of the j-function for an illustration. The article is now available in the May issue: https://www.spektrum.de/magazin/mondschein-verbindung-fuer-eine-paria-gruppe/1555490
Add a comment...

Post has attachment
Add a comment...

Post has shared content
+Fredrik Johansson has a new paper out

/Computing Stieltjes constants using complex integration/
https://arxiv.org/abs/1804.01679
The Stieltjes constants γ_n are the coefficients appearing in the Laurent series of the Riemann zeta function at s=1. We give a simple and efficient method to compute a p-bit approximation of γ_n with rigorous error bounds. Starting from an integral representation due to Blagouchine, we shift the contour to eliminate cancellation. The integral is then evaluated numerically in ball arithmetic using the Petras algorithm, with the use of a Taylor expansion for bounds near the saddle point. This appears to be the first algorithm for Stieltjes constants with uniformly low complexity with respect to both n and p. An implementation is provided in the Arb library. We can, for example, compute γ_n to 1000 digits in a minute for any n≤10100.
Add a comment...

LCZero (http://lczero.org/) is an open source neural network chess engine trained from scratch in the spirit of AlphaZero. The training is done as a distributed computing project. It seems to be making slow but steady progress, and could be competitive with top chess engines in a few months. It would be nice if more people with fast GPUs volunteered to speed up the process. (Caution: the progress graph on the front page shows Elo rating measured by self-play games, which is not calibrated against human players.)

You can play against the most recent version of the network at http://play.lczero.org/. It's currently at a level where it's fun to play against as a total chess novice. I played it a few times on fast mode and won two games when I didn't blunder (one game below). Just so I can tell people "I once beat LCZero" in the future when this will sound much more impressive.

1. e4 h5
2. Nc3 c6
3. Nf3 d5
4. d3 dxe4
5. Nxe4 Nd7
6. Be2 e5
7. O-O Be7
8. d4 exd4
9. Nxd4 Ngf6
10. Nxf6+ Nxf6
11. Bg5 O-O
12. Re1 c5
13. Nf3 Bd7
14. Qd2 Ne4
15. Bxe7 Nxd2
16. Bxd8 Raxd8
17. Nxd2 Rfe8
18. Nb3 Bc6
19. Nxc5 Rc8
20. Bxh5 Rxe1+
21. Rxe1 g6
22. Bg4 Rc7
23. b4 Bd5
24. a3 g5
25. c3 b6
26. Nd7 Bb3
27. Nf6+ Kg7
28. Nh5+ Kg6
29. h3 Rxc3
30. Re7 f5
31. Bf3 a6
32. Rg7+ Kh6
33. Ra7 Kg6
34. Rxa6 Rc2
35. Rxb6+ Kf7
36. Rb7+ Kg6
37. Rb5 Rb2
38. g4 Ba2
39. gxf5+ Kf7
40. Bd5+ Ke7
41. f6+ Kd6
42. f7 Ke7
43. Rb7+ Kf8
44. Bxa2 Rxa2
45. Ra7 Rc2
46. Nf6 Rc1+
47. Kg2 Rb1
48. Rb7 Rd1
49. Nh7+ Kg7
50. Nxg5 Kf8
51. h4 Rd2
52. h5 Rd4
53. h6 Rxb4
54. Rxb4 Ke7
55. Re4+ Kf8
56. Re8#
Add a comment...

Post has attachment
Wait while more posts are being loaded