Eric Hung-Lin Liu
ehliu@mit.edu
Term Address:
97 Moore St, Apt #3
Cambridge, MA 02139
919-357-7695
Education
Home Address:
313 Heidinger Drive
Cary, NC 27511-5669
919-481-2478
Massachusetts Institute of Technology
Cambridge, MA
Candidate for Bachelor of Science degrees in Aerospace Engineering and General Mathematics, June 2008.
Coursework includes: Unified Engineering, Real Analysis I, Numerical Analysis, Complex Variables,
Probabilistic Systems, Aerodynamics, Numerical Methods in PDEs, Adv. Fluid Mechanics, Structure and
Interpretation of Computer Programs, Numerical Linear Algebra, Optimization Methods. GPA: 5.0/5.0
North Carolina State University
Raleigh, NC
Attended supplemental math classes between August 2002 and December 2003: Calculus III, Applied
Differential Equations I, and Linear Algebra
Experience
MIT Aerospace Computational Design Lab: ProjectX
Cambridge, MA
Undergraduate Researcher
Summer 2007 to Fall 2007
Investigated a ProjectX issue with failed convergence in the presence of elements with multiple boundary
conditions. Identified situations under which the problem occurs. Attempted both an numerical and
analytic analysis of the problem; work is ongoing. Additionally worked with another undergraduate to
demonstrate and document the viability of using time-varying linearization (for sensitivity analysis) of the
primal and adjoint problems via a simple example—van der Pol oscillator. Method applies to nonlinear
systems with unstable linear modes, contrasts with previous approaches involving complex, stronger linear
solvers. Efforts resulted in a research paper that is in the final stages of preparation for publication.
Undergraduate Researcher
Summer 2005 to Spring 2006
Worked with a team of 10 graduate students developing a new computational fluids package; implemented
iterative linear solvers (including Gauss-Seidel and GMRES) with preconditioning on large, sparse matrices
arising from problems of fluid dynamics. Preconditioned GMRES yielded a 200 to 400% decrease in the
run-time of the sparse-solver while proving to be most robust iterative scheme. Aided in updates and
revisions to current coding and documentation standards to improve clarity and ease of use.
MIT Department of Aeronautics and Astronautics
Cambridge, MA
Undergraduate Teaching Assistant
Fall 2006 to Present
Held 5 weekly office hours with graduate TAs to assist students with homework assignments and lab
projects. Responded to student emails and engaged in weekly 1-on-1 tutoring sessions. Worked in the
2006/2007 school year in Unified Engineering, the department’s fundamental sophomore course.
Additionally, graded weekly problem sets for Aerodynamics (Fall 2007), working closely with the professor
to monitor students’ strengths, weaknesses, and conceptual gaps. Held 2 weekly office hours with the
graduate TA and another undergraduate to aid students with homework and projects; also available by
appointment.
The MathWorks: Embedded MATLAB
Cambridge, MA
Software Developer
Summer 2006
Developed numerical library functions (~20) for the EML team. Project goal is to create a MATLAB to C
compiler that supported as much of MATLAB’s numerical functionality as possible while decreasing the
execution time. Tasks ranged from special functions (using rational approximation) to linear algebra
routines (e.g. schur, hessenberg, rcond) to making previous code more efficient in terms of operation
counts and especially array accesses. Helped design an upcoming interface to the optimized BLAS and
LAPACK for the EML libraries. Uncovered bugs and worked with other employees to create fixes.
Skills
Programming Languages: C/C++ (with basic MPI), MATLAB, Ada95, FORTRAN77, Scheme, and Java
Operating Systems: Windows XP, 2000, NT 4.0, 9.x, MIT-Athena, Red Hat Linux, Ubuntu
Computing: MS Word, PowerPoint, Excel; Adobe Acrobat; Maple; LaTeX; Knowledgeable in PC hardware
installation and troubleshooting.
Languages: Fluent in Chinese (Mandarin)
Additional Relevant Experiences:
Investigating the Orr-Sommerfeld Equation
Spring 2007
In my junior year, I researched and wrote on the stability of Poiseuille flow under small perturbations as a term
project in a class on fluid mechanics. I was curious to see why turbulent transition is predicted at Re = 5772 by the
well-accepted theory of small perturbations but actually observed around Re ≈ 2000. The work was fundamentally
an investigation of transient growth in non-normal operators—specifically the Orr-Sommerfeld Operator. I
implemented a differential equation solver based on (Chebyshev) Pseudospectral Methods, replicating numerous
existing results and observing how classical, eigenvalue-based asymptotic analysis does not apply in the case of nonnormal operators. Transient effects, which can be characterized through the singular value decomposition, can
induce sufficiently large growth factors such that nonlinear effects take over and the system “blows up” in a way not
predicted by asymptotic analysis.
Local Flow Sensing and Separation Control in Simulated Flapping Flight
Fall 2007
Working with one other undergraduate and a faculty advisor (Professor Mark Drela) I investigated a theory on how
bats may utilize the hairs on their wings for flow sensing. There is evidence that bats use their wing hairs to aid in
flight (i.e. performance worsens when the hairs are shed). We designed an experiment to evaluate the idea that the
hairs alert bats to flow separation. That is, in attached flow, the hair will experience a certain bending moment; if the
flow separates, the bending moment will decrease. Bats may use this information to alter their wing geometry to
prevent flow separation. We constructed a thin airfoil mated to a mechanical flapping mechanism; the device was
mounted in a wind tunnel. Preston tubes measuring the average total pressure in the boundary layer served as our
“hairs.” Acting on the pressure data, we implemented a PD controller to drive the pitch of the wing. With our
experimental set up, we were able to reduce the maximum angle of attack by more than 50% compared to the
uncontrolled case. Unfortunately there was still a separation bubble, but we believe that lag induced by smalldiameter tubing was a significant detractor to the performance of the control law.
Research Interests:
As indicated in my “Summary of Research” (reproduced below) posted on the CSGF website, my research interests
lie predominately in the field of Finite Element Methods in a higher order (p), Discontinuous-Galerkin framework.
Naturally I am also concerned with the numerous issues of unstructured mesh generation and adaptation commonly
associated with DG implementations. I am working on developing DG methods, heading toward increased
automation (e.g. through output-based adaptation) and decreased error with particular application to aerospace
problems.
Reproduced from my “Summary of Research”:
Over the past few decades, Computational Fluid Dynamics (CFD) models have become a critical part of most aircraft
design efforts. Unfortunately the tools in use today are not sufficiently accurate; they cannot answer modern
engineering questions with a reasonable level of confidence. For example, a recent AIAA survey of various CFD
solvers found a 1e-3 variation in their drag coefficient predictions, a level of error which translates to the loss of as
many as 100 passengers on a long range transport aircraft. Part of the issue is that many current codes were designed
to run on 1990s supercomputers–computing power which is available in today“s commodity clusters.
I plan to work on the development of new high fidelity CFD routines with increased levels of automation. In
particular, I am interested in Finite Element Methods in a higher order, Discontinuous-Galerkin (DG) framework.
Among other advantages, higher order methods allow for increased resolution of flow phenomena and faster solution
convergence. One drawback of DG is that there is redundancy in the degrees of freedom since we allow
discontinuities at element boundaries. Nonetheless, DG offers significant algorithmic advantages, increased stability,
and high parallelizability, making the extra cost arguably worthwhile. Given the amount of computing power
available to us, higher order DG methods are potentially one way of working toward more accurate CFD models.