Millisecond Pulsar Discovery via Gamma-Ray Pulsa ons
Holger J. Pletsch (for AEI Hannover Group and LAT Collabora on)
Max Planck Institute for Gravitational Physics (Albert Einstein Institute) © SDO/AIA (Sun), AEI
H J Pletsch Fermi Symposium 2012, Monterey, CA 1 Execu ve summary
• Millisecond pulsars (MSPs): - old neutron stars, spun up by accre ng ma er from companion star, - reach high rota on rates of several hundreds of Hertz. • Previously: ALL such MSPs ("recycled" pulsars) discovered by spin-modulated radio emission. DOI: 10.1126/science.1229054
• Compu ng-intensive blind search in Fermi-LAT data using advanced methods (with par al constraints from op cal data): ⇒ Discovery of PSR J1311-3430 - First MSP found via gamma-ray pulsa ons! - Extremely compact binary: Shortest orbital period! - Clarifies nature of decade-long enigma.
© NASA (Pulsar), NASA/ESA, M.J. Jee and H. Ford (Johns Hopkins University) (Hubble Field), AEI/Milde Marke ng Science Communica on H J Pletsch 2 Background
30 years ago: First MSP detected in radio observa ons. • Backer et al., Nature 1982 • Fermi LAT confirmed that many radio-detected MSPs also pulsate in gamma-rays: Abdo et al., Science 2009 → Gamma-ray pulsa ons revealed
−9 only by rota on parameters 10 Radio discovered 1E13 G 1E33 erg/s Gamma−ray discovered 1E39 erg/s 1E36 erg/s obtained from radio telescopes. −11 e.g., Smith et al., A&A 2008 10 Part of a binary system Young &
1E12 G normal pulsars )
− 1 −13 Successful blind searches for 10 1E30 erg/s • 1 kyr
10 kyr gamma-ray pulsars in LAT data: −15 Abdo et al., Science 2009 10 - 24 (with 1 year) Saz Parkinson et al. ApJ 2010 100 kyr 1 Gyr
1 Myr 10 Gyr - 10 (with 3 years) HJP et al. ApJ & ApJL 2012 −17 10 1E11 G → Large frac on of young pulsars 10 Myr 100 Gyr Period derivative (s s 100 Myr −19 → Most are radio-quiet Ray et al. 2012 10 1E10 G
−21 1E8 G MSPs 1E9 G 10 No MSP previously found in −3 −2 −1 0 1 • 10 10 10 10 10 blind search of gamma-ray data. Spin period (s) H J Pletsch 3 Computa onal Challenge
• Blind-search problem for gamma-ray pulsars: - Computa onally demanding since pulsar parameters unknown a priori → Must be explicitly searched on dense grid (for isolated pulsars: spin frequency, frequency deriva ve & sky posi on) Brady et al., 1998 → For mul ple-year observa on mes: Chandler et al., 2001 Atwood et al., 2006 number of grid points tremendously large Ziegler et al., 2008 HJP et al., 2012 • Blind searches for MSPs vastly more difficult: - Must scan up to higher spin frequencies [to and beyond 716Hz] - Plus, most MSPs are in binaries: → Addi onally unknown orbital parameters → Increases computa onal complexity by orders of magnitude - Blind binary-MSP searches in LAT data were hitherto virtually unfeasible
H J Pletsch 4 Target source: 2FGL J1311.7-3429 (formerly uniden fied) First seen by EGRET. • (e.g., 2EG J1314-3430) • Most significant (43σ) uniden fied LAT source in 2FGL. • Good pulsar candidate: - low flux variability - very curved spectrum
op cal 93-minute modula on • Crucial: In search for op cal counterparts, Romani (2012) iden fied quasi-sinusoidal op cal flux modula on. (Romani ApJL 2012, Kataoka et al. ApJ 2012) → Conjecture: "black widow" pulsar binary - MSP irradiates companion star - Hea ng one side of companion, explains op cal brightness varia on
H J Pletsch 5 The search space
• "Black widow" pulsar interpreta on: Pulsar irradia ng companion - Op cal varia on associated with period of circular orbit. - Confines sky posi on. → These constraints made blind binary-MSP search feasible. BUT: s ll enormous computa onal © AEI challenge, because uncertain es on orbital parameters . by far larger than required for pulsar detec on (and f, f unknown)
→ Pulsar search parameter space le 5-dimensional: 1. Spin frequency: 0 < f < 1400 Hz . 2. Its rate of change: -5x10-13 Hz/s < f < 0 3. Orbital period: Porb = 5626.0 ± 0.1 s 4. Time of ascending: Tasc = 56009.131 ± 0.012 MJD 5. Projected semi-major axis: 0 < x < 0.1 lt-s
H J Pletsch 6 Hierarchical search strategy
Hierarchical, 3-staged search scheme: · Goal: discarding unpromising regions in parameter space as early as possible. · Previously enabled detec on of 10 young (non-MSP) pulsars in blind searches. HJP et al. ApJ & ApJL 2012
Total data set (~4 years) Most 12-day window ... compute 1. Semi-coherent stage: cycles - Sliding coherence window, extending Atwood et al., ApJL 2006; HJP, PRD 2011 spent at summing coherent Fourier power; Zooming in this - Coarse graining: Least sensi ve, but most efficient to scan en re search space stage. - Incorporates photon weights (à la Kerr, ApJ 2011) - Uses heterodyning to process in bands of f using FFT
"zooming in" 2. Coherent follow-up:
Dr. Timo Mappes, www.musop n.com. - For every semi-coherent candidate compute fully coherent Fourier power over en re data set, on significantly refined grid.
3. Including higher signal harmonics (H-test): de Jager et al. 1989 - Typically pulse profile non-sinusoidal, also Fourier power at harmonics. - For every coherent candidate sum power (over en re data) from harmonically related frequencies, using a further refined grid.
H J Pletsch 7 Key novelty: metric search grids
Brady et al. PRD 1998
• Based on concepts from gravita onal-wave searches. Metric • Define a distance metric on search parameter space: approxima on - Measure of expected frac onal loss in S/N for given signal at nearby grid point. - Local Taylor-expansion of frac onal loss around
signal loca on to 2nd order gives the metric. Frac onal loss in S/N Brady et al., PRD 1998; HJP & Allen, PRL 2009; HJP, PRD 2010 Parameter offset • Problem: Metric orbital components explicitly . depend on parameters (unlike metric in f and f ) → Simple la ce would either over- or undercover • Solu on: New grid construc on algorithm to u lize metric - First place orbital grid points at random. (fast MC integra on using metric provides total number of grid points required) - Then move those that are too close or too far apart by barycentric shi s towards op mal coverage. - Designed to never lose > 30% in S/N for any signal. Fehrmann & HJP, in prep H J Pletsch 8 Searching 4 years of LAT data
• Input LAT data for the search: - 1437-day me span; LAT photons within 15° around target.
• Compu ng done on the ATLAS cluster (6780 CPU-cores) - Analyzed full spin-frequency range in bands of 128 Hz each (via heterodyning): → accommodates memory limita ons → adapt orbital search to each band (points increase as f3) - For example, a band near 700Hz Number of grid points: . f: 108, f: 102, orb: 107 → total: 1017
ATLAS compu ng facility, Hannover
H J Pletsch 9 The PSR J1311-3430 system (1/2)
Photon probability weight 250 > 0.1 GeV • Following the discovery: → pulsar ming to precisely measure 200 the system parameters ( ) 150
100 Weighted counts 50 Weighted pulse profile
80 < 0.3 GeV 60 40 20 W. counts 0.3 − 1 GeV 90 60 30 W. counts 70 1 − 3 GeV 50 30
W. counts 10 3 − 7 GeV 12 9 6 3 W. counts 5 > 7 GeV 3
W. c. 1 0 0.5 1 1.5 2 Pulse phase
H J Pletsch 10 The PSR J1311-3430 system (2/2)
• Rota onal ephemeris also constrains companion mass: mc > 0.0083 MSun (~8 MJupiter)
[for mp = 1.35 MSun, i=90°] 15 c
14 m
13 i=90° 12 1.35 Solar masses)