Takens' theorem (1981) shows how lagged variables of a single time series can be used as proxy variables to reconstruct an attractor for an underlying dynamic process. State space reconstruction (SSR) from single time series has been a powerful approach for the analysis of the complex, non-linear systems that appear ubiquitous in the natural and human world. The main shortcoming of these methods is the phenomenological nature of attractor reconstructions. Moreover, applied studies show that these single time series reconstructions can often be improved ad hoc by including multiple dynamically coupled time series in the reconstructions, to provide a more mechanistic model. Here we provide three analytical proofs that add to the growing literature to generalize Takens' work and that demonstrate how multiple time series can be used in attractor reconstructions. These expanded results (Takens' theorem is a special case) apply to a wide variety of natural systems having parallel time series observations for variables believed to be related to the same dynamic manifold. The potential information leverage provided by multiple embeddings created from different combinations of variables (and their lags) can pave the way for new applied techniques to exploit the time-limited, but parallel observations of natural systems, such as coupled ecological systems, geophysical systems, and financial systems. This paper aims to justify and help open this potential growth area for SSR applications in the natural sciences.
Citation: Deyle ER, Sugihara G (2011) Generalized Theorems for Nonlinear State Space Reconstruction. PLoS ONE6(3): e18295. https://doi.org/10.1371/journal.pone.0018295
Editor: Matej Oresic, Governmental Technical Research Centre of Finland, Finland
Received: December 1, 2010; Accepted: February 24, 2011; Published: March 31, 2011
Copyright: © 2011 Deyle, Sugihara. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by National Science Foundation (NSF) DEB1020372, NSF-NOAA CAMEO program (a partnership between the NSF and NOAA) NA08OAR4320894/CAMEO, EPA-STAR Fellowship, Sugihara Family Trust, Deutsche Bank-Jameson Complexity Studies Fund, and the McQuown Chair in Natural Sciences, University of California San Diego. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
A growing realization in many natural sciences is that simple idealized notions of linearly decomposable, fixed equilibrium systems often do not accord with reality. Rather, empirical measurements on ecosystems, metabolic systems, financial networks, and the like suggest a more complex, but potentially more information-rich paradigm at work –. Despite a long history of linear methods development in the engineering sciences, natural systems are generally not well described as sums of independent frequencies that can be sensibly decomposed, analyzed as non-interacting, and reassembled (e.g. Fourier or spectral analysis) in the style of traditional reductionism , . Rather, quantitative measurements show many systems to be fundamentally non-equilibrium and unstable, in a manner more consistent with nonlinear (state dependent) dynamics occurring on a strange attractor manifold , where relationships between state variables cannot be studied independently of the overall system state –. This emergent comprehensive view may help explain why many natural systems, such a those mentioned above, appear so difficult to understand and predict. Mirage correlations are commonplace in nonlinear systems where the manifold may contain trajectories that can temporarily exhibit positive correlations between variables for surprisingly long time periods (and in some regions of the state space) and can subsequently and rapidly exhibit negative correlations or no relationship in other time periods (and other regions of ). This transient property of apparent non-stationarity in correlations is one of the confounding phenomena faced by traditional linear models that require continual refitting and exhibit little or no predictive power.
In this paper, we present two general theorems that addresses the problem of characterizing the coupled dynamics of nonlinear systems using time series observations on a manifold . A special case of this theorem, attributed originally to Takens , provided the first sketch of a mathematical proof for reconstructing a diffeomorphic shadow manifold using lags of a single time series as coordinate axes. The basic idea, that was earlier demonstrated by Packard, Crutchfield, Farmer, and Shaw  and Crutchfield , is that under generic conditions, a shadow manifold can be created using time-lagged observations of based on a single observation function (Cartesian coordinate variable) that is a smooth and smoothly invertible mapping with . Subsequently, Sauer, Yorke, and Casdagli  provided a definitive proof and an explicit extension of Takens' theorem to fractal sets; their theorems are also more powerful than the original theorem, as they show embeddings are not just generic in the sense of being open and dense in the set of all mappings, but in fact almost every mapping in the sense of prevalence  is an embedding (see  for an in-depth explanation of the advantages of “prevalence” over “generic”). The theorem was also extended by Stark, Broomhead, Davies, and Huke ,  and Stark  to include certain classes of stochastic systems. Practical methods for reconstruction have also been explored, particularly to address the presence of noise in real data (e.g. , ). Casdagli et al.  give a thorough treatment of such techniques based on transformations of univariate maps, showing how optimal noise reduction can be achieved. These very important prior results all focused on reconstruction from a single time series; however, as proven below, they can be extended to the more practically significant case where multiple observation functions are used to generate .
Here we prove the more general case of multivariate embeddings (embeddings using multiple time series and lags thereof), and show how time series information can be leveraged if multiple time series and their lags are used to construct embeddings of . These theorems pave the way for more extensive use of state space reconstruction methods in practical applications where long time series may not be available, so that multiple diffeomorphic embeddings may be created in factorial fashion to more fully exploit the coupled non-redundant information that can be extracted from multiple time series (multiple observation functions of dynamics on a manifold) to create predictive shadow manifolds . The use of multiple time series allows the possibility of noise reduction that exceeds the limitations of univariate reconstructions in the presence of noise .
The possibility of extending Takens' theorem to allow lags of multiple observation functions was mentioned in Remark 2.9 from , but was not explicitly proven. The remark was also restricted to mappings strictly formed from consecutive lags, which is not the only possibility that needs to be considered in the multivariate case. Given the potential importance of multivariate reconstructions, we believe a full proof is required—in particular, one that extends the generalization to non-consecutive lags. We show how Takens' theorem is a special case of our more general Theorem 2 (below) and by following the structure of Takens' original proof we clarify the logic and highlight the restrictions and special cases (non-generic cases) that can arise in its application to real world systems. We then give explicit proof of a stronger version of Remark 2.9 from Sauer et al.  that allows non-consecutive lags. This third theorem is stronger than the first two in the sense that it shows embeddings are prevalent and not just generic. For those less familiar, we begin with a brief overview of some basic terms and concepts used in our proofs.
Some Basic Concepts of Embedding Theory
Consider the classic Lorenz attractor  shown in Figure 1(a), consisting of trajectories in three-dimensional space that together define a butterfly shaped surface or manifold. For simplicity, a manifold can be thought of as a generalized, n-dimensional surface embedded in some higher dimensional space, where the dimension of the manifold may be fractal (as is the case for the Lorenz attractor). More generally, an embedding is a multivariate transformation of a manifold that resolves all trajectories on the original manifold without crossings. That is, an embedding is globally in that it resolves all singularities in trajectories that define the manifold (singularities are points on the manifold where trajectories cross so that future paths are not uniquely determined).
The Lorenz attractor  is shown with three shadow manifolds created from lag-coordinate transformations. The typical parameters were used: , , and , giving the three coupled equations as , , and . The solution was computed using a fourth order Runge-Kutta method with a time step of , and the time lag used to create the shadow manifolds was . (A) The trajectory shown in the , , and coordinates of the original system reveals a two-lobed manifold. (B) A univariate transformation using time lags of the -coordinate, , preserves this two-lobed structure (and other topological properties), verifying Takens' theorem. (C) A univariate transformation using time lags of the -coordinate, , does not preserve the two-lobed structure. Local neighborhoods of the original attractor are, however, preserved. Thus, though this mapping violates a genericity assumption of the original theorem and is not an embedding, it is an immersion of the original manifold. (D) A multivariate transformation using both the - and -coordinates, , fulfills the assumptions of Theorems 2 and 7. As predicted, it also preserves the two-lobed structure of the Lorenz and is a valid embedding.
An immersion is a local embedding that may not preserve the global topology of a manifold. Rather an immersion preserves the topology of every local neighborhood of the original manifold, so that each point of the tangent space of the immersed manifold has the same dimensionality as the true manifold. Thus, an immersion is a mapping that is between any given “piece” of the true manifold and the immersed manifold. However, this condition does not guarantee that the global topology is preserved. This is illustrated in Figure 1(c), where two different pieces of the original manifold are mapped to the same piece of the immersed manifold, producing an immersion that is not an embedding. Immersions are nonetheless a useful conceptual stepping stone for constructing proofs about embeddings, since all embeddings are necessarily immersions.
The Lorenz attractor, Figure 1(a), provides an excellent example to illustrate both of these concepts. Consider two different multivariate functions that transform the original manifold, and where is a small time lag as in Takens' theorem. Both of these functions map points on the true manifold to points on a shadow manifold, shown in Figures 1(b) and 1(c). Examining these shadow manifolds, it is evident that both are immersions of the Lorenz attractor, because zooming in on a particular piece of either will reveal that the tangent spaces have the same dimensionality as the original. However, only Figure 1(b) is an embedding that successfully reproduces the two lobes of the butterfly. The reconstruction in Figure 1(c), based on lags of the -coordinate, fails to do so, because the two fixed points of the original attractor have the same -coordinate; they are mapped to the same point on the shadow manifold, so the two lobes are stacked on top of each other. This singularity is a consequence of a special, non-generic symmetry in the Lorenz system that violates an assumption of Takens' theorem. Figure 1(d) shows an embedding based on lags of both - and -coordinates and is an example of the generalized mappings addressed in this paper.
Two Theorems in the Style of Takens: The Generic Case
Let be a compact manifold of dimension . A dynamical system is a diffeophorism defining the trajectories or “flow” on for discrete time or a vector field on for continuous time. Takens  proved generically that given and , a smooth observation function can be used to construct an embedding of in dimensions under the transformation where . Here the components correspond to time-lagged observations of the dynamics on defined by . Notice that such mappings involve a single distinct observation function (i.e. a single time series), and represent a small subset in the larger set of all possible mappings that could, for example, involve multiple time series and their lags.
Takens explicitly refers only to the unlagged as an observation function, but in its most general sense an observation function is any . Thus, the functions , corresponding to the lags of the time series are technically observation functions as well. This bears mention, because in the more general case of mappings , the observation functions making up the components of are not all derived from a single time series, but can be various lags of multiple time-series. To treat these cases, it is necessary to acknowledge that these are all observation functions, and we will refer to distinct time series as “unlagged” observation functions.
For a mapping in the larger set of all mappings , consider the case with component functions which are multiple unlagged observation functions of (i.e. multiple time series). Again, an observation function is any function that assigns a real number to each point on the manifold . For a mapping , we can think of in terms of its component functions, which correspond to the coordinates in . These component functions may all be lags of a single distinct observation function tracking a dynamical system, as in the case of Takens, or they may be multiple observation functions, as in the case of Whitney, or they may be lags of multiple observation functions, as in Theorems 2 and 7 below.
The question arises whether general multivariate mappings form legitimate embeddings. Here we present two theorems: one that demonstrates that maps created from distinct observation functions are generically embeddings and another that shows that maps created from lags of multiple observation functions are also generically embeddings. Both of these theorems generalize Takens' theorem for which the component functions only involve a single observation function.
It follows from Whitney  that generically is an embedding. Note, however, that Whitney's work does not apply to the specific subsets of involving fixed lagged relationships as discussed by Takens for reconstructing attractor manifolds for dynamic systems. That is, Whitney's theorem is generic and does not address these specific subsets of which have “measure zero” (e.g. in the sense of “shy” defined in ). To tackle this problem, we look to the proof of Takens and see that it can be readily generalized to the other subsets of , including the case of generic .
Recall that, for a compact manifold, a mapping that is an immersion and injective is also necessarily an embedding. Thus, Takens' general approach was to first show that (i) immersions are dense in the set of mappings , then that (ii) there is a dense set of mappings within this set of immersions. Since the set of embeddings is open in the set of all possible mappings, Takens concludes that mappings in are generically embeddings. The critical word here is “generically,” meaning there can be exceptions (and as explained in , the set of such exceptions doesn't necessarily have zero measure).
To demonstrate both (i) and (ii), Takens argues that even when the property of interest (e.g. the property) does not hold for some particular mapping, by making an arbitrarily small perturbation, it is possible to find a nearby mapping for which that property holds. The key to the theorem and also to adapting it to other sets of mappings is finding how to make these perturbations. The proof is most straightforward for the general case involving distinct observation functions (each a distinct time series) because it is possible to perturb the component functions of independently. Thus we begin with this proof to add clarity to the more powerful main theorem 2 involving lags of multiple observation functions.
Consider a compact, -dimensional manifold and a set of observation functions , where smoothly; by “smooth” we mean at least . Then it is a generic property of all possible that the mapping defined as is an embedding.
Consider an arbitrary set of observation functions on . We define a corresponding mapping by letting each of these observation functions be one of the component functions of . Now, recall that an immersion is a map with a derivative that is globally injective, i.e. . We denote the total derivative of a function as . If the derivative is evaluated at a particular point in the domain of , we will write , and if is a matrix, then we denote the derivative at a particular point and along a particular tangent vector as .
For any point , we can perturb the co-vectors independently by perturbing individual . By making infinitesimal perturbations at points for which , we can get a set of observables arbitrarily close to such that for all —i.e., is an immersion. Since the set of immersions is open in the set of all mappings, there is a neighborhood around this such that every is an immersion.
Since immersions are local embeddings, we can find a such that on the manifold, implies . Here we depart from Takens' notation and let denote infinitesimal separations between two points on the manifold to avoid confusion with the later defined which is used to perturb the observable; is any fixed metric on . In fact for this fixed , there is a subset such that for any in , the associated map is an immersion, and implies that .
Next, we show that we can find a globally arbitrarily close to . To do this, we construct a finite collection of subsets such that the are open subsets of , the collection covers , and for every . Then, we take a partition of unity corresponding to these , so that we can vary the value of any by an infinitesimal amount without altering the value of for .
We now consider the mapping defined as . We define the set as , so that (by our choice of ), the mapping is necessarily injective on the complement of in . Furthermore, note that the intersection of with the diagonal of gives the set of points , and therefore is equivalent to injective. Our task, then, is to perturb the manifold using the and so that it does not intersect the diagonal manifold .
At each we know that , so and cannot belong to the same . Consequently, varying an or only alters the value of at either or (respectively). In the tangent space , then, the direction of the infinitesimal changes given by the and are all linearly independent (indeed orthogonal) and as such span . Since the tangent spaces of and are at most and dimensional, respectively, we can construct a vector from a linear combination of and that lies outside of both and . Therefore, an infinitesimal perturbation corresponding to this linear combination will move the sub-manifolds and away from each other at the point without creating a new intersection at another point. By keeping the size of these perturbations sufficiently small, we ensure that we stay confined to , so that is still an immersion. This is a more transparent statement of the transversality argument used in the Takens proof (1981).
Thus, we have shown that for any arbitrary set of observables , we can find a set of observables arbitrarily close to such that is an embedding—i.e., there is a dense set of observables such that is an embedding. The set of embeddings is open in the set of all mappings, so this set is dense and open, meaning that the embedding property is generic over all mappings.
When mappings are confined to fixed lag relationships, Takens showed it is valid to independently perturb each component of at a given point of the domain by perturbing the unlagged observation function, , in the other parts of the domain corresponding to neighborhoods of the lagged states , , etc. This ensures that the perturbations to maintain the structure of the lag relationships and that we have not inadvertently left the subset of interest. As we now show, this allows the above result to be easily extended to families of maps having component functions that are the lags of multiple observation functions. This is the relevant case for many practical examples where lags of multiple time series (multiple variables or observation functions) are required to achieve a mechanistic reconstruction of (e.g. ). It also allows information on to be leveraged when the time series are short, as is the case in many physical and biological problems , .
Before starting the proof, however, we must clarify exactly what the “subsets of interest” are. We define these sets as follows. First, we say is a lag of the observable if we can write for positive . We consider the lags in the positive time direction only to simplify notation in the proof, noting that the results apply equally to negative lags. Let be the subset of for which , is an unlagged observable, i.e. is not a lag of another . We begin with the “unlagged” observation functions, , or observation functions that are not a lag of another observable in . Now define a set for each that contains and any other observation function in which is a lag of it. That is, is the set of that are lags of given as , where the lags are distinct for fixed . This choice of and determine a subset containing all choices of observables which obey the correct lag relationships under a dynamical system . Note that each element of can be identified by the dynamical system and the . We denote such an element, then, as .
Consider a diffeomorphism on some compact manifold of dimension , along with observation functions , smoothly; by “smooth” we mean at least . Restrict the to have the lag relationships corresponding to a collection of sets and lags under the dynamical system , and impose the following generic ,  properties on :
- The set of periodic points with period has finitely many points,
- The eigenvalues of at each in a compact neighborhood are distinct and not equal to 1.
The proof of this theorem closely follows the logic of the previous proof and the original argument of Takens . As noted above, any perturbations to via its component functions must remain within (the set of observables having the desired lag relationships under prescribed by the and the ). Here we must also deal with points of that are fixed points or periodic under the dynamical system , i.e. the points for which there exists a such that (including the fixed point case, ). The above proof shows that the mapping is generically an immersion because the co-vectors can be independently perturbed. This is also true for non-periodic points where there are fixed lag relationships between some observables, as we can perturb in the neighborhood of and thus perturb without affecting in the neighborhood of .
Note that periodic points can exist such that the period or some integer multiple of it, , is the fixed time lag between two observables belonging to the same . Let be a compact neighborhood of all such points. For , the vectors and cannot necessarily be perturbed independently. Nonetheless, while for such a point, it is not generally true that . By assumption, for each , the eigenvalues of the are distinct and not equal to 1. Thus, by the chain rule, it is clear that and are linearly independent. As noted above, all the other can be perturbed independently, so we can find a set of observables arbitrarily near in for which is an immersion on . Note that because the set of immersions is open, there is an open neighborhood in around this for which every set of observables in that neighborhood gives an immersion.
We must also satisfy injective. The proof above relied on the ability to independently perturb the manifold at any point by an infinitesimal amount in any coordinate direction. For a periodic point on with perioid and two observables related as and , it is impossible to independently perturb locally in the coordinate or , as you also perturb or . By assumption, the set has a finite number of elements. For such a generic and any set , any neighborhood of the will contain a set of observables for which the unlagged observation functions take distinct values at each point in .
We first perturb the to find an open neighborhood of observables which give immersions when restricted to the set . We then further perturb the observables to find within this neighborhood a set of observables for which is also injective and therefore an embedding (on ). Since embeddings are dense in the space of all mappings, there is a neighborhood such that for all , the map is an embedding.
We now show that we can find a such that is an embedding on all of . We first note that at points , the vectors can be perturbed independently, so we can find for which is an immersion. Because an immersion is a local embedding, there is a such that for , implies that . Since the set of immersions is open in the set of possible mappings, there is a neighborhood such that for any , the corresponding mapping is an immersion. Thus, for the same as above, implies .
Now we need to show that there is a such that is also injective on . As noted in the first proof, this is equivalent to for the mapping defined as . If and are both in or , we already know that . Thus we restrict ourselves to the set .
To perturb the manifold away from at points of intersection, , we must be able to find variations for which the tangent vector is linearly independent from the tangent vectors and and lies outside of . In the first proof, it was obvious that each component of could be perturbed independently. Now we must be more careful. We do this by first creating a collection of open subsets of , , with the following properties:
- The cover the closure of .
- For each and , the diameter of is less than .
- For all choices of , the set intersects with for at most one .
- For and such that for some , , and , no two of belong to the same .
Take a partition of unity corresponding to this . Because of the way we constructed the , we can vary the value of each at any point by an infinitesimal amount without altering the value of the other in the neighborhood of . We make this explicit as follows. To perturb the , we take for corresponding to . To perturb the other ( for some ), we perturb for corresponding to . Consider the perturbations, , which are independent shifts at in distinct . In , we note that each corresponding tangent vector lies outside of . Note the together with any basis of form a linearly independent set of vectors. Since the dimension of is at most , there must be a linear combination of the that lies outside of both and , which can be used to perturb away from . By keeping variations in the sufficiently small, we can find a set of such that and (where now corresponds to the map). This pair gives a mapping that is both an immersion and injective, and thus is an embedding. Because was an arbitrarily small neighborhood of any point in , this means embeddings are dense in , and the set of embeddings is open in the set of mappings. Thus, the map given by is generically an embedding.
Just as Takens extends the original result for discrete time to dynamical systems in continuous time, we can extend our result as follows:
Consider a smooth vector field on some compact manifold along with observables , smoothly; by “smooth” we mean at least . Define as the flow on . Suppose we restrict the to have the lag relationships corresponding to a collection of sets and lags under the discrete dynamical system , where is a constant. We impose the following generic properties on :
- For points such that , the eigenvalues of are distinct and not equal to 1.
- No periodic integral curve of has integer period .
A Theorem in the Style of Sauer et al.: The Prevalent Case
We now give an explicit proof of Remark 2.9 from  using the framework constructed in their original paper, but we extend the language to cover reconstructions using non-consecutive lags (from multiple time series). The proof uses Lemma 4.1, 4.6, and 4.11 from  to show that mappings and immersions are prevalent in the space , just as Sauer et al. use Lemma 4.6 to prove Theorem 3.3, and Lemmas 4.1 and 4.11 to prove Theorem 3.5. These lemmas are now stated (for the proofs, see their original paper).
(Originally part 2 of 4.1) Let and be positive integers, distinct points in , in , and in . Then there exists a polynomial in variables of degree at most such that for , .
(Originally 4.6) Let be a compact subset of . Let be Lipschitz maps. For each integer , let be the set of pairs in for which the matrix has rank , and let lower boxdim . Define . If for all integers , then for outside a measure zero subset of , the map is .
(Originally 4.11) Let be a compact subset of a smooth manifold embedding in . Let be a set of smooth maps from an open neighborhood of to . For each positive integer , let be the subset of the unit tangent bundle such that the matrix has rank , and let lower boxdim . Define . If for all integers , then for almost every , the map is an immersion on .
To apply these lemmas, it is necessary to restrict the dimension of the sets of periodic orbits—that is, the sets for . For the case of consecutive lags, Sauer et al. state sufficient conditions to be boxdim . A sufficient condition for non-consecutive lags is a bit more complicated. Define the constants such that for at least one and . Also, define . A sufficient condition on the is .
Let be a diffeomorphism on an open subset of , and let be a compact subset of , . Let be a collection of sets and a set of lag relationships as above, such that . Assume that for every positive integer , the set of periodic points of period satisfies , and that for each point of , the Jacobian has distinct eigenvalues. Then, for almost every set of observation functions satisfying the given lag relationships, the map is an embedding on .
Without loss of generality, assume we have ordered the components of with and all its lags first, then and its lags, etc. That is,To show prevalence, we find a suitable probe space (see ). The infinite dimensional space for the univariate theorem is the observation functions , smoothly. For maps constructed from multiple lags, this becomes the sets of unlagged observation functions. Sauer et al. take the probe space for the univariate theorem to be any set of polynomials in variables which include all such polynomials up to degree . It is now necessary to have a set of polynomials for each of the . Thus, we take the probe space for this theorem to be the Cartesian product of copies of .
Let be a basis for . We want to show that for almost all choices of coefficients , the map defined by the observation functions is an embedding. We first demonstrate that almost every is , proceeding as in the proof of Theorem 4.3 in .
To sensibly apply Lemma 5, we adopt the following convention: think of as a perturbation of , which is the summed effect of perturbations on each separately. For each pair , and , there is a map which is for if and otherwise. The components of are either or of the form . Consequently, , which matches the structure Lemma 5.
We now check that the rank of the matrix satisfies the conditions of Lemma 5 for each pair of distinct . Note that to avoid confusion with the previous section of this paper and Takens' original work, we continue to use row vectors to describe the transformations . However, Sauer et al.  prefer column vectors, so it is necessary to use of transposes in several instances. Thus, we haveNote that is a block diagonal matrix, and so it has rank equal to the sum of the rank of the blocks. Each of the blocks can be rewritten as the product of two matrices, and , where the entries of are values of a single polynomial and the entries of are each one of . Note, there are multiple possible choices for and that give the same .
Case 1: First consider and that do not both lie in a periodic orbit of integer period less than . We specify so that the first rows, where is the size of the set , correspond to the , and the next correspond to the . is onto, so the rank of is just the sum of the ranks of the . For this case, contains a copy of , and thus will have rank . The entire matrix will thus have rank , which satisfies the conditions of Lemma 5.
Case 2: Now consider and in separate periodic orbits with periods and such that and . will have fewer rows corresponding to the for some (there will also be a reduction in the number of rows associated with ). In this case, will still contain the column space of and thus . Again the are onto, and so the rank of is the rank of .
The dimension of the set of all pairs and is . By the conditions placed on the size of the , we can conclude that , and thus that Lemma 5 applies to this case as well.
Case 3: Finally we consider and in the same -periodic orbit, . Now the matrix becomes more complicated, since some of the pertaining to may be equal to pertaining to . Consequently, the are no longer guaranteed to contain the column space of the identity. Each does contain the column space of an dimensional matrix with along the upper diagonal and a single off the diagonal in each column. Using elementary operations, it is possible to make the first columns of upper diagonal for some integer . Thus, the rank of each is at least and the entire matrix has .
The dimension of the set of all such and is just . By the imposed conditions, , and Lemma 5 applies.
Now we want show that almost every is an immersion. We check that the matrixhas full rank and thus satisfies the conditions of Lemma 6 for each in the tangent bundle . Note that this is a block diagonal matrix with blocks, so it is sufficient to show that the columns of the th block span the subspace for . We consider two cases.
Case 1: Consider first the subset of that are not periodic with period . The entries of each block are of the form . Since is a diffeomorphism and , we know that . Furthermore, the are distinct points. Examining Lemma 4, it is clear that the columns span . The dimension of is at most , so we may apply Lemma 6.
Case 2: Now consider the subset of that are periodic with period . By the conditions of the theorem, has distinct eigenvalues from . Therefore, . Furthermore, the relationship depends on , and again referencing Lemma 4, it is clear that the columns span . The dimension of is certainly less than , so we can safely apply Lemma 6.
Theorem 7 can be extended to continuous dynamical systems (smooth vector fields on a manifold) by letting the flow of be in the statement of the theorem.
Theorem 1 and the more general result presented in Theorem 2 (and its corollary) were given proofs intended to follow those presented by Takens. The original “transversality” argument, however, has been replaced with what we reckon is a simpler and more direct argument. These clarify how perturbations to the observation functions can be constructed and highlight why dimensions are necessary to have mappings that are generically embeddings. Theorem 7 is similar to Theorem 2, but takes advantage of the more powerful framework, built around the notion of prevalence, established by Sauer et al. . It also provides more specific conditions on the periodic orbits than Theorem 2 and thus can be applied to certain non-generic situations that Takens' original framework would exclude. Namely, the set of periodic points need not be finite (as required in Takens' original theorem and Theorem 2), so long as the dimensionality does not exceed the bounds stated in Theorem 7. Theorem 7 is an extension of Remark 2.9 in , which we explicitly proved by determining a sufficient restriction for the periodic orbits when the lags composing aren't necessarily consecutive.
This work also develops a language to describe a wider family of cases for reconstructing state space manifolds from multiple observational time series to encourage wider applicability of SSR in the natural sciences. For example, these results can be extended to another special case of interest for reconstructions using time derivatives , when multiple observation functions are available. The argument for this case is analogous to that used by Takens  for the case when all the derivatives are from a single observation function. Furthermore, these theorems validate heuristic work using spatial lag reconstructions and mixed spatial and temporal lag reconstructions to study spatially coupled dynamics .
More importantly, in terms of future applications, Theorems 2 and 7 set the stage for practical reconstruction of state space manifolds from multiple observation functions. This is significant in answering objections to single variable state space reconstruction (SSR) concerning the excessive phenomenology of lagged-coordinate embeddings . These two theorems provide proof of principle for modeling attempts of nonlinear dynamics in the natural sciences involving multiple time series (e.g. ), and lays bare the rather non-restrictive assumptions required in such applications for building mechanistic models from multiple time series variables. Moreover, it gives support to the notion of using multiple embeddings as a potentially efficient way of extracting information from time series data of limited length, but where there are potentially many simultaneous observations of dynamics on the same attractor manifold. By reducing correlations in noise between the reconstructed coordinates, these techniques should allow reconstructions to exceed the limitations placed on univariate methods , as heuristic examples have already suggested . The potential information leverage provided by multiple embeddings possible from novel combinations of variables (and their lags) can pave the way for a plethora of new applied techniques to exploit the time-limited, but parallel observations of nature . This paper is intended to complement the existing literature on SSR and help promote this potential growth area in the natural sciences.
We wish to thank Hao Ye, James Crutchfield, John Melack, Donald DeAngelis, Simon Levin, J. Doyne Farmer, Martin Casdagli, Tim Sauer, Sarah Glaser, Chih-hao Hsieh, Stephen Munch and Charles Peretti, Michael Fogarty, Alec MacCall, Andrew Rosenberg, Les Kaufman, and Irit Altman for helpful comments and editorial advice.
Conceived and designed the experiments: ERD GS. Performed the experiments: ERD GS. Analyzed the data: ERD GS. Contributed reagents/materials/analysis tools: ERD GS. Wrote the paper: ERD GS.
- 1. Casdagli M (1989) Nonlinear prediction of chaotic time series. Physica D 35: 335–356.M. Casdagli1989Nonlinear prediction of chaotic time series.Physica D35335356
- 2. Crutchfield JP (1979) Prediction and stability in classical mechanics. JP Crutchfield1979Prediction and stability in classical mechanics.Senior thesis in physics and mathematics, University of California, Santa Cruz. Senior thesis in physics and mathematics, University of California, Santa Cruz.
- 3. Farmer JD, Sidorowich JJ (1987) Predicting chaotic time series. Physical Review Letters 59: 845–848.JD FarmerJJ Sidorowich1987Predicting chaotic time series.Physical Review Letters59845848
- 4. Jenouvrier S, Weimerskirch H, Barbraud C, Park YH, Cazelles B (2005) Evidence of a shift in the cyclicity of antarctic seabird dynamics linked to climate. Proceedings of the Royal Society of London, Series B 272: 887–895.S. JenouvrierH. WeimerskirchC. BarbraudYH ParkB. Cazelles2005Evidence of a shift in the cyclicity of antarctic seabird dynamics linked to climate.Proceedings of the Royal Society of London, Series B272887895
- 5. Lo TT, Hsu HH (2010) Change in the dominant decadal patterns and the late 1980s abrupt warming in the extratropical northern hemisphere. Atmospheric Science Letters 11: 210–215.TT LoHH Hsu2010Change in the dominant decadal patterns and the late 1980s abrupt warming in the extratropical northern hemisphere.Atmospheric Science Letters11210215
- 6. Planque B, Batten SD (2000) Calanus finmarchicus in the north atlantic: the year of Calanus in the context of interdecadal change. ICES Journal of Marine Science 57: 1528–1535.B. PlanqueSD Batten2000Calanus finmarchicus in the north atlantic: the year of Calanus in the context of interdecadal change.ICES Journal of Marine Science5715281535
- 7. Ramanathan A, Wang C, Schreiber SL (2002) Perturbational profiling of a cell-line model of tumorigenesis by using metabolic measurements. Proceedings of the National Academy of Sciences of the United States of America 102: 5992–5997.A. RamanathanC. WangSL Schreiber2002Perturbational profiling of a cell-line model of tumorigenesis by using metabolic measurements.Proceedings of the National Academy of Sciences of the United States of America10259925997
- 8. Ruelle D (1989) Chaotic evolution and strange attractors: the statistical analysis of time series for deterministic nonlinear systems. D. Ruelle1989Chaotic evolution and strange attractors: the statistical analysis of time series for deterministic nonlinear systems.Cambridge University Press, Cambridge. Cambridge University Press, Cambridge.
- 9. Scheffer M, Carpenter SR (2003) Catastrophic regime shifts in ecosystems: linking theory to observation. Trends in Ecology & Evolution 18: 648–656.M. SchefferSR Carpenter2003Catastrophic regime shifts in ecosystems: linking theory to observation.Trends in Ecology & Evolution18648656
- 10. Seth AK, Izhikevich E, Reeke GN, Edelman GM (2006) Theories and measures of consciousness: an extended framework. Proceedings of the National Academy of Sciences of the United States of America 103: 10799–10804.AK SethE. IzhikevichGN ReekeGM Edelman2006Theories and measures of consciousness: an extended framework.Proceedings of the National Academy of Sciences of the United States of America1031079910804
- 11. Soramaki K, Bech ML, Arnold J, Glass RJ, Beyeler WE (2007) The topology of interbank payment flows. Physica A 379: 317–333.K. SoramakiML BechJ. ArnoldRJ GlassWE Beyeler2007The topology of interbank payment flows.Physica A379317333
- 12. Takens F (1981) Detecting strange attractors in turbulence. In: Rand DA, Young LS, editors. Symposium on Dynamical Systems and Turbulence. Berlin: Springer-Verlag. pp. 366–381.F. Takens1981Detecting strange attractors in turbulence.DA RandLS YoungSymposium on Dynamical Systems and TurbulenceBerlinSpringer-Verlag, volume 898 of Lecture Notes in Mathematics. 366381 , volume 898 of Lecture Notes in Mathematics.
- 13. Wagner BK, Kitami T, Gilbert TJ, Peck D, Ramanathan A, et al. (2008) Large-scale chemical dissection of mitochondrial function. Nature Biotechnology 26: 343–351.BK WagnerT. KitamiTJ GilbertD. PeckA. Ramanathan2008Large-scale chemical dissection of mitochondrial function.Nature Biotechnology26343351
- 14. May RM, Levin SA, Sugihara G (2008) Complex systems: Ecology for bankers. Nature 451: 893–895.RM MaySA LevinG. Sugihara2008Complex systems: Ecology for bankers.Nature451893895
- 15. Sugihara G (2010) On early warning signs. G. Sugihara2010On early warning signs.Seed Magazine Global Reset 2010. Seed Magazine Global Reset 2010.
- 16. Sugihara G (2010) Nature is nonlinear. Kyoto Journal 75: 56–57.G. Sugihara2010Nature is nonlinear.Kyoto Journal755657
- 17. Brock WA, Sayers CL (1988) Is the business cycle characterized by deterministic chaos? Journal of Monetary Economics 22: 71–90.WA BrockCL Sayers1988Is the business cycle characterized by deterministic chaos?Journal of Monetary Economics227190
- 18. Brock WA, Malliaris AG (1989) Differential Equations, Stability and Chaos in Dynamic Economics, volume 27 of Advanced Textbooks in Economics. North-Holland. WA BrockAG Malliaris1989Differential Equations, Stability and Chaos in Dynamic Economics, volume 27 of Advanced Textbooks in Economics.North-Holland
- 19. Brock WA, Hsieh DA, LeBaron BD (1991) Nonlinear Dynamics, Chaos, and Instability: Statistical Theory and Economic Evidence. WA BrockDA HsiehBD LeBaron1991Nonlinear Dynamics, Chaos, and Instability: Statistical Theory and Economic Evidence.MIT Press, Cambridge, Massachusettes. MIT Press, Cambridge, Massachusettes.
- 20. Dixon PA, Milicich MJ, Sugihara G (1999) Episodic fluctuations in larval supply. Science 283: 1528–1530.PA DixonMJ MilicichG. Sugihara1999Episodic fluctuations in larval supply.Science28315281530
- 21. Hsieh CH, Glaser SM, Lucas AJ, Sugihara G (2005) Distinguishing random environmental fluctuations from ecological catastrophes for the North Pacific Ocean. Nature 435: 336–340.CH HsiehSM GlaserAJ LucasG. Sugihara2005Distinguishing random environmental fluctuations from ecological catastrophes for the North Pacific Ocean.Nature435336340
- 22. Hsieh CH, Anderson C, Sugihara G (2008) Extending nonlinear analysis to short ecological time series. American Naturalist 171: 71–80.CH HsiehC. AndersonG. Sugihara2008Extending nonlinear analysis to short ecological time series.American Naturalist1717180
- 23. Rodo X, Pascual M, Fuchs G, Faruque ASG (2002) ENSO and cholera: a nonstationary link related to climate change? Proceedings of the National Academy of Sciences of the United States of America 99: 12901–12906.X. RodoM. PascualG. FuchsASG Faruque2002ENSO and cholera: a nonstationary link related to climate change?Proceedings of the National Academy of Sciences of the United States of America991290112906
- 24. Schaffer WM (1984) Stretching and folding in lynx fur returns: evidence for a strange attractor in nature? American Naturalist 124: 789–820.WM Schaffer1984Stretching and folding in lynx fur returns: evidence for a strange attractor in nature?American Naturalist124789820
- 25. Sugihara G, Grenfell B, May RM (1990) Distinguishing error from chaos in ecological time series. Philosophical Transactions of the Royal Society of London, Series B: Biological Science 330: 235–250.G. SugiharaB. GrenfellRM May1990Distinguishing error from chaos in ecological time series.Philosophical Transactions of the Royal Society of London, Series B: Biological Science330235250
- 26. Sugihara G, May RM (1990) Nonlinear forecasting as a way of distinguishing chaos from measurement error in a data series. Nature 344: 734–741.G. SugiharaRM May1990Nonlinear forecasting as a way of distinguishing chaos from measurement error in a data series.Nature344734741
- 27. Sugihara G (1994) Nonlinear forecasting for the classification of natural time series. Philosophical Transactions of the Royal Society of London, Series A 348: 477–495.G. Sugihara1994Nonlinear forecasting for the classification of natural time series.Philosophical Transactions of the Royal Society of London, Series A348477495
- 28. Packard NH, Crutchfield JP, Farmer JD, Shaw RS (1980) Geomtery from a time series. Physical Review Letters 45: 712–716.NH PackardJP CrutchfieldJD FarmerRS Shaw1980Geomtery from a time series.Physical Review Letters45712716
- 29. Sauer T, Yorke JA, Casdagli M (1991) Embedology. Journal of Statistical Physics 65: 579–616.T. SauerJA YorkeM. Casdagli1991Embedology.Journal of Statistical Physics65579616
- 30. Hunt BR, Sauer T, Yorke JA (1992) Prevalence: a translation-invariant “almost every” on infinite-dimensional spaces. Bulletin of the American Mathematical Society 27: 217–238.BR HuntT. SauerJA Yorke1992Prevalence: a translation-invariant “almost every” on infinite-dimensional spaces.Bulletin of the American Mathematical Society27217238
- 31. Stark J, Broomhead DS, Davies ME, Huke JP (1997) Takens embedding theorems for forced and stochastic systems. Nonlinear Analysis: Theory, Methods, and Applications 30: 5303–5314.J. StarkDS BroomheadME DaviesJP Huke1997Takens embedding theorems for forced and stochastic systems.Nonlinear Analysis: Theory, Methods, and Applications3053035314
- 32. Stark J, Broomhead DS, Davies ME, Huke JP (2003) Delay embeddings for forced systems: II. Stochastic forcing. Journal of Nonlinear Science 13: 519–577.J. StarkDS BroomheadME DaviesJP Huke2003Delay embeddings for forced systems: II. Stochastic forcing.Journal of Nonlinear Science13519577
- 33. Stark J (1999) Delay embeddings for forced systems: I. Deterministic forcing. Journal of Nonlinear Science 9: 255–332.J. Stark1999Delay embeddings for forced systems: I. Deterministic forcing.Journal of Nonlinear Science9255332
- 34. Gibson J, Farmer JD, Casdagli M, Eubank S (1992) An analytic approach to practical state space reconstruction. Physica D 57: 1–30.J. GibsonJD FarmerM. CasdagliS. Eubank1992An analytic approach to practical state space reconstruction.Physica D57130
- 35. Casdagli M, Eubank S, Farmer JD, Gibson J (1991) State space reconstruction in the presence of noise. Physica D 41: 52–98.M. CasdagliS. EubankJD FarmerJ. Gibson1991State space reconstruction in the presence of noise.Physica D415298
- 36. Ye H, Sugihara G, Deyle ER (2011) Leverging information from multiple time series with ensemble state space reconstructions. H. YeG. SugiharaER Deyle2011Leverging information from multiple time series with ensemble state space reconstructions.In preparation. In preparation.
- 37. Lorenz EN (1963) Deterministic nonperiodic flow. Journal of Atmospheric Science 20: 130–141.EN Lorenz1963Deterministic nonperiodic flow.Journal of Atmospheric Science20130141
- 38. Whitney H (1936) Differentiable manifolds. Annals of Mathematics 37: 645–680.H. Whitney1936Differentiable manifolds.Annals of Mathematics37645680
- 39. Huke JP (1993) Embedding nonlinear dynamical systems, a guide to Takens' theorem. JP Huke1993Embedding nonlinear dynamical systems, a guide to Takens' theorem.Internal report, DRA, Malvern, UK. Internal report, DRA, Malvern, UK.
- 40. Crutchfield JP, McNamara BS (1987) Equations of motion from a data series. Complex Systems 1: 417.JP CrutchfieldBS McNamara1987Equations of motion from a data series.Complex Systems1417
- 41. Ørstavik S, Stark J (1998) Reconstruction and cross-prediction in coupled map lattices using spatio-temporal embedding techniques. Physics Letters A 247: 145–160.S. ØrstavikJ. Stark1998Reconstruction and cross-prediction in coupled map lattices using spatio-temporal embedding techniques.Physics Letters A247145160