Diffusion maps, reduction coordinates, and low dimensional representation of stochastic systems
read more
Citations
Stochastic Processes in Physics and Chemistry
Enhancing Important Fluctuations: Rare Events and Metadynamics from a Conceptual Viewpoint
destiny: diffusion maps for large-scale single-cell data in R.
Data-Driven Sparse Sensor Placement for Reconstruction: Demonstrating the Benefits of Exploiting Known Patterns
Determination of reaction coordinates via locally scaled diffusion map
References
Normalized cuts and image segmentation
Normalized cuts and image segmentation
Stochastic processes in physics and chemistry
Methods of Mathematical Physics
Related Papers (5)
A global geometric framework for nonlinear dimensionality reduction.
Frequently Asked Questions (8)
Q2. What is the main reason for the use of diffusion maps?
Beyond the benefits of dimensional reduction, the projection of the system onto the diffusion map coordinates also allows systematic design of computational experiments, where biased simulations are initialized at chosen values of the diffusion map coordinates, thus allowing efficient exploration of the dynamics of the system in these coordinates.
Q3. What is the equilibration time of the modified system?
since ψ1 is a slow coordinate, the equilibration time of the modified system is still of the same order of magnitude as the fast relaxation time τR.
Q4. What is the way to compute the eigenfunctions of the FP operator?
Their computational approach is closely related to the transfer operator approach [46], which also computes an approximation to the eigenfunctions of the FP operator [27], and to Perron cluster analysis [18, 19].
Q5. How do the authors compute the diffusion map?
By computing the diffusion map, this simply amounts to searching for a rotation angle θ which makes the variable w cos(θ) + z sin(θ) as closest to oneto-one with ψ1 as possible.
Q6. How long would it take to find the other metastable states?
In this case, starting from xL a direct simulation would require an extremely long time to exit this well and find the other metastable states.
Q7. What is the way to solve the problem of non-reversible diffusions?
The authors also note that [27] in fact considered the more general case of non-reversible diffusions and proved that the backward eigenfunctions can be used to partition the space into metastable states in this case as well.
Q8. What is the main difference between the two approaches?
The main differences in their work is that the proposed dimensionality reduction is intrinsically related to the dynamics, and has provably good properties in approximating long-term behavior of the system.