1: \begin{abstract}

2: Arclength continuation and branch switching are enormously successful algorithms

3: for the computation of bifurcation diagrams. Nevertheless, their combination

4: suffers from three significant disadvantages. The first is that they attempt to

5: compute only the part of the diagram that is continuously connected to the

6: initial data; disconnected branches are overlooked. The second is that the

7: subproblems required (typically determinant calculation and nullspace

8: construction) are expensive and hard to scale to very large discretizations. The

9: third is that they can miss connected branches associated with nonsimple

10: bifurcations, such as when an eigenvalue of even multiplicity crosses the

11: origin. Without expert knowledge or lucky guesses, these techniques alone can

12: paint an incomplete picture of the dynamics of a system.

13: 

14: In this paper we propose a new algorithm for computing bifurcation diagrams,

15: called deflated continuation, that is capable of overcoming all three of these

16: disadvantages. The algorithm combines classical continuation with a deflation

17: technique that elegantly eliminates known branches from consideration, allowing

18: the discovery of disconnected branches with Newton's method. Deflated

19: continuation does not rely on any device for detecting bifurcations and does not

20: involve computing eigendecompositions; all subproblems required in deflated

21: continuation can be solved efficiently if a good preconditioner is available

22: for the underlying nonlinear problem.

23: We prove sufficient conditions for the convergence of Newton's method

24: to multiple solutions from the same initial guess, providing insight into which

25: unknown branches will be discovered. We illustrate the success of the method on

26: several examples where standard techniques fail.

27: 

28: \end{abstract}

29: