Jan 16, 2018 - Recoil energy measured by light, charge or phonons. Direct detection. Accelerating the Search for Dark Ma
Introduction into direct and indirect Dark Matter searches and their challenges
Interpreting extragalactic backgrounds via angular cross-correlations
Marco Regis
(Torino)
Two big challenges “… the moment of truth has come for WIMPs: either we will discover them in the
next five to ten years, or we will witness the inevitable decline of WIMP paradigm.”
G.Bertone, Nature, 2010
The period we have been witnessing is very relevant for WIMPs. What are the lessons we learned?
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Two big challenges “… the moment of truth has come for WIMPs: either we will discover them in the
next five to ten years, or we will witness the inevitable decline of WIMP paradigm.”
G.Bertone, Nature, 2010
The period we have been witnessing is very relevant for WIMPs. What are the lessons we learned?
1) We do not know what to search for 2) There is no (totally) clean observable
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Two big challenges We do not know what to search for From modified gravity
(i.e. DM does not exist) … ... to non-particle/baryonic DM (e.g. primordial black holes) ...
ALL
potentially VIABLE
… to very
specialized
DM models
Credit: M. Weiss/CfA
There is no (totally)
Credit: Simon Swordy (U. Chicago), NASA
clean observable
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Two big challenges We do not know what to search for Unknown fundamental (microscopic) nature not only impacts the
type/strength of non-gravitational signal but implies also macroscopic uncertainties:
- Clustering (especially at small-scales)
→
Buckley&Peter 2017
- DM phase-space distribution in the Galaxy
see talks by Nicolao Fornengo, David Harvey and Andrea de Simone
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Two big challenges There is no (totally) clean observable Direct/indirect detection target cosmo signals, where there are many other players, “backgrounds” are typically poorly known
J. Monroe Credit: Simon Swordy (U. Chicago), NASA
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Direct
detection
Direct detection Scattering of WIMPs off target atomic nuclei Recoil energy measured by light, charge or phonons.
CDMS image
Undagoitia&Rauch 2017
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Characteristic WIMP signatures
Annual modulation: DAMA/LIBRA, ANAIS, KIMS, DM-Ice, PICO-LON, SABRE
Diurnal modulation: DAMA with larger mass could likely access it
Directionality: Nuclear emulsion (NEWS), Anysotropic crystals
(ADAMO), Liquid Ar TPC, Negative Ion Time Expansion Chamber (NITEC), Carbon nanotubes, DRIFT, MIMAC, DMTPC, NEWAGE,
D3,
…
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
DAMA/LIBRA NaI(Tl) scintillation detectors located at LN Gran Sasso
DAMA/LIBRA
o tw e s c e n i d s ca de
annual modulation
Detection at ~ 9 C.L. PROPER MODULATION FEATURES: cosine like, T~1 year, t0 ~ 2nd of June
NO MODULATION above 6 keV and in the 2-6 keV multiple-hits residual rate. Not easy to reconcile it with other sources of background.
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
CoGeNT Coherent Germanium Neutrino Technology
2 n i
(Soudan Underground Laboratory, Minnesota) Ge-detectors
0 1 0
Aalseth+ 2010
Large background contamination, but low threshold
2 n i
1 1 0
Aalseth+ 2012
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
CDMS Cryogenic Dark Matter Search experiment
(Soudan Underground Laboratory, Minnesota)
Ge-Si detectors measuring phonons and ionization
2 n i
9 0 0
Ahmed+ Science2010
2009: 2 events on Ge (expected background = 0.8 events, P=23%) 2013: 3 events on Si (expected background = 0.4 events, P=5.4%)
Agnese+ PRL2013
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Direct detection - status Undagoitia&Rauch 2017
Under the hypothesis of contact-type spin-independent interactions
Roszkowski+ 2017
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Direct detection Beyond the easiest case (contact scalar interactions)
Full set of effective operators for DM – nucleus scattering Relevant for GeV-TeV DM
Fitzpatrick et al., JCAP 1302 (2013) 004
Interactions on (bound) electrons
+ interferences
Relevant for keV – MeV DM
DM phase space distribution in the Galaxy (esp. the high-v tail) - simulations - observations (GAIA)
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Indirect
detection
Indirect detection Annihilations (or decays) of DM particles in astrophysical objects
generate fluxes of “standard” detectable particles.
Energy of the process set by the DM mass → WIMPs ~ GeV-TeV
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Indirect detection channels
WIMPs are a primary source of
high-energy charged cosmic-rays, Hillas 2006
gamma-rays and
neutrinos
NASA image
Other wavelengths for
candidates different from WIMPs (e.g. X-rays for sterile neutrinos)
Plus radiative emission in the radio, X and
gamma bands from WIMP induced e+-e
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Positron fraction pre-PAMELA AMS01 Collab. PLB2007
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Positron fraction with PAMELA Cirelli+2008
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Positron fraction after AMS-02 AMS02 Collab. PRL2014
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Positron fraction interpretations DiMauro+ 2014
Boudaud+ 2015
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Anti-matter excesses Evoli+,2015
DAMPE spectrum of e+ and e-
Yuan+,2017
AMS-02
antiprotons
Cuoco+,2016
Tentative anti-He events in AMS-02 Coogan&Profumo, 2017
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Galactic Center pre-Fermi LAT “… it is possible to successfully identify dark matter
annihilation radiation, even in the presence of significant astrophysical backgrounds.” Dodelson, Hooper, Serpico, PRD2008
MR&Ullio,2008
Crowdy place at other
wavelengths
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Galactic Center with Fermi LAT Daylan+ 2014
First analyses by
Fermi-LAT Collaboration, 2017
Goodenough&Hooper (2009) and Vitale+ (2009)
Calore+ 2014
de Boer+ (2016) → molecular clouds distribution
Yang & Aharonian (2016) and Macias+ (2016) → bi-lobed structure
Bartels+ (2017) → traces stellar mass in the boxy and nuclear bulge
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Galactic Center interpretations Dark Matter
Difficult to make strong statements Daylan+ 2014
Different explanations:
●
Source of CR derived from observed level of star formation
rate and supernova explosion at GC ●
Population of millisecond pulsars
(e.g., Gaggero+2015, Carlson+2016)
(e.g., Bartels+2016, Lee+2015)
Fermi-LAT Collaboration, 2017
Fermi-LAT Collaboration, 2017
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Dwarf spheroidal galaxies Many dSph discovered with SDSS and DES, and
likely many more to come from near-future surveys.
Low level of star formation Hernandez et al., 2000 Carina dSph Inferred from colourmagnitude diagram
Metal-poor stellar systems McConnachie, 2013
→ The non-thermal
emission related to star-formation is expected to be extremely faint.
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Dark matter in dSph McConnachie, 2013
Very large mass-to-light ratio
→ Mass budget largely dominated by DM
MAGIC+FermiLAT Collab., 2016
Very promising target for particle DM
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Prediction for dSph The detection of -ray emission from the direction of a dSph is guaranteed!
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Prediction for dSph The detection of -ray emission from the direction of a dSph is guaranteed! Plenty of background sources in the Universe!
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Prediction for dSph The detection of -ray emission from the direction of a dSph is guaranteed! Plenty of background sources in the Universe! Are we already witnessing this situation?
GeringerSameth+, 2015
DES+FermiLAT Collab., 2016
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Already observed in ReticulumII ?
Radio counterpart for possible ray excess
> -1
MR+, JCAP 2017
~ 1.5 → BL Lac?
S25 ~ 10-12 erg cm-2s-1
Ackermann et al. 2015
-rays:
Barolo Astroparticle Meeting, Barolo, 05/09/2017 Marco Regis
Lesson We have been discovering new properties of the “backgrounds” as we try to discover DM
→ “Backgrounds” are known up to an O(1) factor.
Supervised vs unsupervised algorithm?
Deep learning is mandatory, but is it enough?
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Statistical techniques The indirect Dark Matter signal is typically predicted to be feeble but extended → statistical correlations can be a powerful tool
Recently adopted techniques in gamma-rays (to infer unresolved sources): - 1-point statistics
(Malyshev&Hogg 2011,Zechlin+2015, Lee+2015, Feyereisen+2015, Zechlin,MR+2016, Lisanti+2016)
- Wavelet decomposition - Stacking
(Bartels+ 2015)
(Branchini,MR+2017, Lisanti+2017, ...)
- Information field theory
(D3PO, Selig+ 2013,2015)
- Brock-Dechert-Scheinkman statistic
(Baxter&Dodelson2011)
- Poisson ordered-pixel method (Campbell+ 2017)
- Multi-Scale Variance Stabilizing Transform (Schmitt+2010)
- Convolutional neural networks with mock images (Caron+2017) - 2-point auto-correlation
(Fermi-LAT 2012, Fornasa+2016, Fornasa,MR+ 2017)
- 2-point cross-correlation (MR+2012-2017, Ando+2014-2017, Shirasaki+2014-2017)
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Angular cross-correlations N O
SI S U C ISbelieve a certain Map where we D e h source tpopulation is present as y bunresolved component D E P WIMP DM in the Fermi-LAT map) O (e.g. O C S
Tracer of the sources we are looking for: - Catalog of the same sources but at different wavelength
- Gravitational tracer (e.g. lensing or galaxy/cluster catalog)
→ CCFab(|-) → Clab correlation in
physical space
correlation in
harmonic space
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
Tomography Cuoco,MR+, ApJS 2015
MR+, PRL 2015
DM peaks at low z, whilst
TOMOGRAPHIC APPROACH
astrophysical sources peak at z > 0.5
Accelerating the Search for Dark Matter with Machine Learning, Leiden, 16/01/2018 Marco Regis
3D Power Spectrum Typically obtained from Simulations or Halo model Cuoco+, ApJS 2015
Camera, Fornasa, Fornengo, MR, ApJ 2013
It is (roughly speaking) mapped in the multipole range 100 < l < 1000
European Physical Society Conference on High Energy Physics 2017, Venice, 06/07/2017 Marco Regis
Angular cross-correlation outcomes Cuoco,MR+, ApJS 2015
→ Competitive technique
→ Significant degeneracies are still present in current analyses
European Physical Society Conference on High Energy Physics 2017, Venice, 06/07/2017 Marco Regis
How to combine different pieces? - Redshift distribution - Angular spectrum - Energy spectrum - Temporal dependence?
→ Machine Learning can help!
European Physical Society Conference on High Energy Physics 2017, Venice, 06/07/2017 Marco Regis