Bryan Ostdiek, Ph.D.
Theoretical particle physicist
Harvard University



  • Postdoctoral Fellow at Harvard University
  • Postdoctoral Fellow at the University of Oregon
  • Ph.D. from the University of Notre Dame
  • B.S. from Montana State University
  • Interests

    I am interested in finding dark matter in our universe. This comes in the form of LHC collider searches and using information from satellites and telescopes.



    Besides studying the universe, I also enjoy camping and hiking. I grew up with the mountains in the picture above out my back door.


    My Curriculum Vitae

    The presence of a light "stop squark" (see lower pannel) can affect the measurement of the mass of the top quark. [1909.09670]

    The decay of a light stop squark looks very similar to the decay of a top quark.
          With this, stops could bias the measurement of the top mass. This plot shows the regions
          in the ''real'' top mass vs the stop mass. The red regions would lead to a measured value
          within the uncertainty of ATLAS's combined measurement. The blue region actually leads to a
          better fit than the Standard Model alone.

    We use Gaia data to fild old stars which have accreted onto the Milky Way, which can help us learn about the distribution of dark matter in our galaxy. [1907.06652] [1907.07190] [1907.07681]

    We used deep learning to classify which stars fell onto the galaxy
          vs those which were born in situ. In the process, we found a stellar stream which may be the remnant of a disrupted
          dwarf galaxy merging with the Milky Way. The middle pannels show the velocity distribution of the stars which fell onto the galaxy,
          while the lower pannels are for the disk stars.

    Dark Mesons could make up the dark matter in the Universe. Current LHC searches have surprisingly low sensitivity. [1809.10184]

    Dark Mesons would decay similar to a heavy Higgs, with some model dependence on the preference for gauge bosons or fermions.
          With no current direct search, we recast many other searches designed for other models and the sensitivity to these models is very low.

    Diving into machine learning to discover what is important for the machine to accurately classify. We "plane" away the information. [1709.10106]

    Computers are great at finding patterns to classify signal and background.
          However, using this for physics, we want to know what it is learning.
          Through 'planing' we are able to strip the machine of the ability to find a certain pattern, which helps to see how important it is.

    Searching for supersymmetric particles is one of the main goals of the LHC. There are excellent bounds on "stop squarks" except for a small splinter. [1804.00111]

    The "stealth stop" is notoriously hard to search for. This happens when the stop is close in mass to the top quark.
          Besides difficult searches, the region of parameter space is also non-trivial to simulate, and we showcase some of the reasons, as well as providing a pedagogical guide.

    Machine learning for the LHC often uses simulated data to train the classifier. "Weak supervision" allows for training directly on real collider data. [1706.09451]

    When training directly on real data, we do not have access to "truth-level" labels. This paper
          shows that with weak supervision, the actual labels used by the network are unimportant.

    Collider signals of dark photons are an exciting probe for new gauge forces and are characterized by events with boosted lepton jets. [1612.00026]

    Electron lepton-jets are difficult to search for because of large backgrounds. One method is to use the muon chambers, but much parameter
          space does not have particles that live long enough to reach.

    R-parity conserving supersymmetry yields a dark matter candidate. Future experiments can cover most of for the "well-tempered neutralino." [1510.03460]

    A combination of 2σ exclusions from future indirect (CTA and HAWC),
          direct (XENON1T and LZ), and collider searches (charged tracks and compressed events at 100 TeV) are shown over the surface of thermal relic neutralinos.


    Are you interested in high energy physics or machine learning? Send me a message: