This is the rst part of a series consisting of two papers. In this paper we de ne the category of partial di erential equations. Special cases of morphisms from an object (equation) are symmetries of the equation and reductions of the equation by a symmetry groups, but there are many other morphisms. We develop a special-purpose language for description and study of the internal structure of this category. We are mostly interested in a subcategory that arises from second order parabolic equations on arbitrary manifolds. In the second part of the series we study the internal structure of this subcategory in detail.
/ / M 0 / / N 0
0 1(x) !
If F : ( ; E) ! ( 0; E0) is a morphism in PDE 0, then F de nes a bijection between the set of all solutions of E0 and the set of all F -projectable solutions of E.
In Section 2 we de ne a bigger category PDE , whose objects are pairs (N; E) with N a smooth manifold
and E a subset of the bundle J mk(N ) of k-jets of m-dimensional submanifolds of N , and whose morphisms from
(N; E) to (N 0; E0) are maps N ! N 0 satisfying some analogue of conditions (
The category PDE generalizes the notion of symmetry group in two directions: 1. Automorphisms group of an object (N; E) in PDE is the symmetry group of the equation E. 2. For a symmetry group G of E the natural projection N ! N=G de nes the morphism (N; E) ! (N=G; E=G) in PDE . Here E=G is the equation describing G-invariant solutions of E.
Note that morphisms of PDE go beyond morphisms of this kind.
In Sections 2 and 6 we discuss the relations between our approach to the factorization of PDE and the other approaches.
Then we discuss the possibility of the introduction of a certain structure in PDE formed by a lattice of subcategories. These subcategories may be obtained by restricting to equations of speci c kind (for example, elliptic, parabolic, hyperbolic, linear, quasilinear equations etc.) or to the morphisms of speci c kind (for example, morphisms respecting the projection of N on a base manifold M as in PDE 0) or both. When we interested in solutions of some equation it is useful to look for its quotient objects because every quotient object gives us a class of solutions of the original equation. It may happen that the position of an object in the lattice gives information on the morphisms from the object and/or on the kind of the simplest representatives of quotient objects. In Section 4 we develop a special-purpose language for description and study of such situations. We introduce a number of partial orders on the class of all subcategories of xed category and depict these orders by various arrows (see Table 1 and Fig. 3). For instance, we say that a subcategory C1 is closed in a category C and depict C / C1 if every morphism in C with source from C1 is a morphism in C1; we say that C1 is plentiful in C and depict C / C1 if for every A 2 ObC1 and for every quotient object of A in C there exists a representative of this quotient object in C1; and so on.
We use this language in the second part of this series [9] for a detail study of the full subcategory PE of PDE that arises from second order parabolic equations posed on arbitrary manifolds, but we hope that our approach based on category theory may be useful for other types of PDE as well. An object of PE is an equation for an unknown function u(t; x), x 2 X having the form ut = Pi;j bij (t; x; u)uij +Pi;j cij (t; x; u)uiuj +Pi bi(t; x; u)ui + q(t; x; u) in local coordinates xi on X, where X is a smooth manifold. We prove that every morphism in PE is of the form (t; x; u) 7! (t0(t); x0 (t; x) ; u0(t; x; u)) (Theorem 1). Particularly, PE appears to be a subcategory of PDE 0.
The use of the structure of PE developed in the second paper [9] is illustrated there on the example of the
reaction-di usion equation
ut = a(u) ( u +
ru) + q(x; u);
x 2 X; t 2 R;
(
Let M , K be smooth manifolds. A system E of k-th order partial di erential equations for a function u : M ! K is given as a system of equations l(x; u; : : : ; u(k)) = 0 involving x, u and the derivatives of u with respect to x up to order k, where x = (x1; : : : ; xm) are local coordinates on M and u = (u1; : : : ; uj ) are local coordinates on K. Further we will use the words \partial di erential equation", \PDE" or \equation" instead of \a partial di erential equation or a system of partial di erential equations" for short.
Recall some things about jets and related notions. The k-jet of a smooth function u : M ! K at a point x 2 M is the equivalence class of smooth functions M ! K whose value and partial derivatives up to k-th order at x coincide with the ones of u. All k-jets of all smooth functions M ! K form the smooth manifold J k(M; K), and the natural projection k : J k(M; K) ! J 0(M; K) = M K de nes a smooth vector bundle over M K, which is called k-jet bundle. For every function u : M ! K, its k-th prolongation jk(u) : M ! J k(M; K) is naturally de ned. k-th order PDE for functions acting from M to K can be considered as a subset E of J k(M; K); solutions of E are functions u : M ! K such that the image of jk(u) is contained in E.
In more general situation we have a smooth ber bundle : N ! M instead of a projection M K ! M , and sections s : M ! N instead of functions u : M ! K. Denote by the space of smooth sections of ; recall that a section of is a map s : M ! N such that s is the identity. De nitions of the k-jet bundle k : J k( ) ! J 0( ) = N and of the k-th prolongation jk : ! k : J k( ) ! M are the same as those for functions. Let E be a subset of J k( ); then E can be considered as a k-th order partial di erential equation for sections of , that is s 2 is a solution of E if the image of jk(s) is contained in E.
Let : N ! M , 0 : N 0 ! M 0 be smooth ber bundles. Let F : ! 0 be a smooth bundle morphism with the additional property that F is a di eomorphism on the bers (see Fig. 1). F induces the map F : 0 ! ; denote by F its image. We say that a section of is F -projectable if it is contained in F . If F is surjective, then F is injective, so it de nes the map F# : F ! 0. If additionally F is submersive, then it de nes the map F k : JFk ( ) ! J k( 0), where JFk ( ) = F J k( 0) is the bundle of k-jets of F -projectable sections of (see Fig. 2). Recall that a map F is called a submersion if dF : TxN ! TF (x)N 0 is surjective at each point x 2 N .
It turns out that the language of category theory is very convenient for our study of PDE. Recall that a category C consists of a collection of objects ObC, a collection of morphisms (or arrows) HomC and four operations. The rst two operations associate with each morphism F of C its source and its target, both of which are objects of C. The remaining two operations are an operation that associates with each object C of C an identity morphism idC 2 HomC and an operation of composition that associates to any pair (F; G) of morphisms of C such that the source of F coincides with the target of G another morphism F G, their composite. These operations should satisfy some natural axioms [3, sec. I.1].
De nition 2. PDE 0 is the category whose objects are pairs smooth viber bundle, k 2 N, and morphisms from an object
0 : N 0 ! M 0; E0 J k( 0) are smooth bundle morphisms F : ! F -projection of E.
If F : ( ; E) ! ( 0; E0) is a morphism in PDE 0, then F E0 and the set of all F -projectable solutions of E. : N ! M; E J k( ) with being a
: N ! M; E J k( ) to an object 0 admitted by E such that E0 is the de nes a bijection between the set of all solutions of 2
In this section we de ne the category PDE of partial di erential equations, whose objects are pairs (N; E) such that N is a smooth manifold and E is a subset of the bundle J mk(N ) of k-jets of m-dimensional submanifolds of N .
Let N be a Cr-smooth manifold, 0 < m < dim N . The jet bundle k : J mk(N ) ! N is a ber bundle with the ber J mk(N ) x over x 2 N , where J mk(N ) x is the set of equivalence classes of smooth m-dimensional submanifolds L of N passing through x under the equivalence relation of k-th order contact in x.
The k-jet of a k-smooth m-dimensional submanifold L over x 2 L is the equivalence class from J mk(N ) x determined by L. Thus we have the prolongation map jk : L ! J mk(N ) taking each point x 2 L to the k-jet of L over x (so it is the section of the ber bundle J mk(N ) restricted to L N ). For every k > l 0 the natural projection k;l : J mk(N ) ! J ml(N ) maps the k-jet of L to the l-jet of L over x for every m-dimensional submanifold L of N and every x 2 L.
For a submanifold L of N the di erential of the prolongation map jk : L ! J mk(N ) takes the tangent bundle T L to the tangent bundle T J mk(N ). The closure of the union of the images of T L in T J mk(N ) when L runs over all m-dimensional submanifolds of N is the vector subbundle of T J mk(N ); it is called the Cartan distribution on J mk(N ).
Let E be a submanifold of J k( ), : N ! M , m = dim M . The graph of a section is an m-dimensional submanifold of N , so J k( ) is an open subspace of J mk(N ) and E could be considered as a submanifold of J mk(N ). The extended version of E is de ned as the closure of E in J mk(N ) [4, p. 222]. Since we don't plan to consider in nitesimal properties of E in contrast to the Lie group analysis of PDE, we would consider any subset E of J mk(N ) as a partial di erential equations. By de nition, solutions of such an equation are smooth m-dimensional non-vertical integral manifolds of the Cartan distribution on J mk(N ) that are contained in E. Note that for any m-dimensional submanifold L of N its prolongation jk(L) is a non-vertical integral manifold of the Cartan distribution on J mk(N ). Therefore, if E J mk(N ) is obtained from a traditional PDE as it was described above, and if L is the graph of a section u of , then L is a solution of E in the above sense if and only if u is a solution of the corresponding traditional PDE in the traditional sense. In addition there is allowed the possibility of both multi-valued solutions and solutions with in nite derivatives (see [4, sec. 3.5] for the details). Wherever we write concrete equation in the traditional form below we mean the extended version of this equations, that is the closure of the corresponding set in J mk(N ).
Now let us introduce some auxiliary notation.
Let F : N ! N 0 be a map. We say that L N is F -projected if L = F 1(F (L)). Note that if F is a surjective submersion and L is an F -projectable submanifold of N , then L0 = F (L) is a submanifold of N 0.
Let N , N 0 be Cr-smooth manifolds, 0 < m < dim N . Let F : N ! N 0 be a surjective submersion of smoothness class Cs, k s r.
De nition 3. F -projectable jet bundle J mk;F (N ) is the submanifold of J mk(N ) consisting of k-jets of all mdimensional F -projectable submanifolds of N .
We write JFk (N ) instead of J mk;F (N ) if the value of m is clear from context.
There is natural isomorphism between the bundles J mk;F (N ) and F J mk0 (N 0) over N , where F J mk0 (N 0) = J mk0 (N 0) N0 N is the pullback of J mk0 (N 0) by F , dim N m = dim N 0 m0. Therefore we can lift F to the map F k : J mk;F (N ) ! J mk0 (N 0) by the following natural way:
1) Suppose # 2 J mk;F (N ). Take an arbitrary F -projectable submanifold L of N such that the k-th prolongation of L pass through # (that is the k-jet of L over k(#)) is #.
2) Assign to # the point #0 2 J mk0 (N 0), where #0 is the k-jet of the submanifold L0 = F (L) N 0 over F k (#). objects of PDE are pairs (N; E), where N is a smooth manifold, E is a subset of J mk (N ) for some integer k; m 1; morphisms of PDE with a source A = (N; E) are all surjective submersions F : N ! N 0 admitted by E; target of such morphism is (N 0; E0) where E0 is F -projection of E; the identity morphism from A is the identity mapping of N , and the composition of morphisms is the composition of appropriate maps.
If F : (N; E) ! (N 0; E0) is a morphism in PDE , then F de nes a bijection between the set of all solutions of E0 and the set of all F k-projectable solutions of E.
In [5] the author de ned the following notion of a map admitted by a pair of equations: a map F : N ! N 0 is admitted by an ordered pair of equations (A; A0), A = (N; E), A0 = (N 0; E0) if for any L0 N 0 the following two conditions are equivalent:
L0 is the graph of a solution of E0,
F 1(L0) is the graph of a solution of E.
However, we are not happy with this de nition; in particular, because it deals only with global solutions of E. Therefore, here we de ne the notion of map admitted by an equation in terms of (locally de ned) jet bundles. Remark 1. Let A = (N; E) be an object of PDE . Then its automorphism group Aut(A) is the symmetry group for the equation E.
Remark 2. Suppose G is a subgroup of the symmetry group of E such that N=G is a smooth manifold. Then the natural projection N ! N=G de nes the morphism (N; E) ! (N=G; E=G) in PDE . Here E=G is the equation describing G-invariant solutions of E.
Therefore, reduction of E by subgroups of Aut(A) de nes a part of nontrivial morphisms from A. But the class of all morphisms from A is signi cantly richer than the class of morphisms arising from reduction by subgroups of Aut(A). Let Sol(A) be the set of all solutions of A, that is of all smooth m-dimensional nonvertical integral manifolds of the Cartan distribution on J mk(N ) that are contained in E. In general, the subset F (Sol(A0)) = F 1(L0) : L0 2 Sol(A0) Sol(A) of solutions of A arising from a morphism F : A ! A0 can not be represented as a set of solutions that are invariant under some subgroup of Aut(A). In particular, F (Sol(A0)) can be the set of G-invariant solutions, where G is a transformation group that is not necessarily a symmetry group of E. Moreover, for a morphism F : A ! A0 it may occur that for every nontrivial di eomorphism g of N there is an element in F (Sol(A0)) that is not g-invariant. More detailed discussion is given in Section 6; see also [6, sections 9, 10], [7, sections 7{9].
Our approach is conceptually close to the approach developed in [1] that deals with control systems. If we set aside the control part and look at this approach relative to ordinary di erential equations, then we get the category of ordinary di erential equations, whose objects are ODE systems of the form x_ = , x 2 X, where X is a manifold equipped with a vector eld , and morphism from a system A to a system A0 is a smooth map F from the phase space X of A to the phase space X0 of A0 that projected to 0. In other words, F is a morphism if it transforms solutions (phase trajectories) of A to solutions of A0: F (Sol(A)) = Sol(A0).
By contrast, we deal with pullbacks of the solutions of the quotient equation A0 to the solutions of the original equation A. In our approach, the number of dependent variables in the reduced PDE remains the same, while the number of independent variables is not increased. Thus, in the approach proposed, the quotient object notion is an analogue of the sub-object notion (in terminology of [1, sec. 5.1]) with respect to the information about the solutions of the given equation; however, it is similar to the quotient object notion with respect to interrelations between the given and reduced equations.
Note also that the above described category of ODE from [1] is isomorphic to a certain subcategory of PDE . Namely, consider the following subcategory PDE 1 of PDE : objects of PDE 1 are pairs (N; E), where N = X unknown function u : X ! R, 2 T X; morphisms of PDE 1 are morphisms of PDE of the form (x; u) 7! (x0(x); u).
R, E is a rst order linear PDE of the form L u = 1 for One can easily see that the category of ODE from [1] is isomorphic to PDE 1: the object L u = 1 corresponds to the object x_ = , and the morphism (x; u) 7! (x0(x); u) corresponds to the morphism x 7! x0(x).
The category of di erential equations was also de ned in [2, sec. 6.1.3] in a di erent way: objects are in nitedimensional manifolds endowed with integrable nite-dimensional distribution (particularly, in nite prolongations of di erential equations), and morphisms are smooth maps such that image of the distribution is contained in the distribution on the image, similarly to morphisms in [1]. Thus, the category of di erential equations de ned in [2] is quite di erent from the category PDE de ned here; one should keep it in mind in order to avoid confusion. The factorization of PDE A by a symmetry group described in [2] is PDE A0 on the quotient space describing images of all solutions of A at the projection to the quotient space: F (Sol(A)) = Sol(A0). In that approach, every factorization of A provides a part of the information about all the solutions of A. In our approach, factorization of A is such an equation A0 that the pullbacks of its solutions are solutions of A: F (Sol(A0)) Sol(A); so that from every factorization, we obtain the full information about a certain set of the solutions of the given equation.
The following two propositions are simple corollaries of our de nitions.
Proposition 2. Suppose (N; E), (N 0; E0), (N 00; E00) are objects of PDE , F : N ! N 0 is a morphism from (N; E) to (N 0; E0) in PDE , G : N 0 ! N 00 is surjective submersion. Then the following two conditions are equivalent (see Fig. 4):
G is a morphism from (N 0; E0) to (N 00; E00), GF is a morphism from (N; E) to (N 00; E00).
E \ J GkF (N )
J GkF (N ) JFk (N ) J k(N ) k N
F k F
(GF )k ' ' E0 \ J Gk(N 0) ' ' ' '
J Gk(N 0) J k(N 0) k0
Note that the Cartan distribution Ck(N ) on J mk (N ) restricted to J mk;F (N ) coincides with the lifting F k Cmk0 (N 0) of the Cartan distribution on J mk0 (N 0), m0 = m dim N + dim N 0. Taking this into account and using the analogy with higher symmetry group, we replace J mk;F (N ) to arbitrary submanifold of J mk(N ). Thus we obtain the category PDE ext with the same objects as PDE and extended set of morphisms involving transformations of jets. (This category will not be used in the rest of the paper.) De nition 6. An extended category of partial di erential equations PDE ext is de ned as follows: objects of PDE ext are pairs (N; E), where N is a smooth manifold, E is a subset of J mk (N ) for some integer k; m 1; morphisms of PDE ext from A = (N; E J mk(N )) to A0 = (N 0; E0 J mk00 (N 0)) are all pairs ; F~ such that is a smooth submanifold of J mk(N ), F~ : ! J mk00 (N 0) is a surjective submersion, the Cartan distribution on J mk (N ) restricted to coincides with the lifting F~ Cmk00 (N 0) of the Cartan distribution on J mk00 (N 0), and E \ = F~ 1(E0) (see right diagram on Fig. 3); the identity morphism from A is = J mk(N ); F~ = idN ; composition of J mk(N ); F~ : ! J mk00 (N 0) F~ 1( 0); F~0 F~ . and
For each integral manifold of the Cartan distribution on E0, its inverse image is an integral manifold of the Cartan distribution on E. Therefore, for each solution of E0, its pullback is a solution of E.
PDE is embedded to PDE ext by the following natural way: to the morphisms F : N ! N 0 of PDE from the equation of k-th order we assign the morphisms ; F~ of PDE ext such that = J mk;F (N ), F~ = F k. 4
We start with a review of some basic de nitions of category theory [3, sec. II.6]. Given a category C and an object A of C, one may construct the category (A # C) of objects under A (this is the particular case of the comma category): objects of (A # C) are morphisms of C with source A, and morphisms of (A # C) from one such object F : A ! B to another F 0 : A ! B0 are morphisms G : B ! B0 of C such that F 0 = G F .
Suppose C is a subcategory of PDE , A is an object of C. Then the category (A # C) of objects under A describes collection of quotient equations for A and their interconnection in the framework of C.
To each morphism F : A ! B with source A (that is to each object of the comma category (A # C)) assign the set F (Sol(B)) Sol(A) of such solutions of A that \projected" onto underlying space of B (space of dependent and independent variables). We can identify such morphisms that generated the same sets of solutions of A, that is identify isomorphic objects of the comma category (A # C).
Describe the situation more explicitly. An equivalence class of epimorphisms with source A is called a quotient object of A, where two epimorphisms F : A ! B and F 0 : A ! B0 are equivalent if F 0 = I F for some isomorphism I : B ! B0 [3, sec. V.7]. If F : A ! B and F 0 : A ! B0 are equivalent, then they lead to the same subsets of the solutions of A: F (Sol(B)) = F 0 (Sol(B0)). So if we interested only in the sets of the solutions of A, then all representatives of the same quotient object have the same rights.
Therefore, the following problems naturally arise: to study all morphisms with given source, to choose a \simplest" representative from every equivalence class, or to choose representative with the simplest target (that is the simplest quotient equation).
In order to deal with these problems we develop in this paper a special-purpose language.
Let us introduce a number of partial orders on the class of all categories to describe arising situations. First of all, we de ne a few types of subcategories.
De nition 7. Suppose C is a category, C1 is a subcategory of C.
C1 is called a wide subcategory of C if all objects of C are objects of C1.
C1 is called a full subcategory of C if every morphism in C with source and target from C1 is a morphism in C1.
We say that C1 is full under isomorphisms in C if every isomorphism in C with source and target from C1 is an isomorphism in C1.
We say that C1 is closed in C if every morphism in C with source from C1 is a morphism in C1. (Note that every subcategory that is closed in C is full in C.) We say that C1 is closed under isomorphisms in C if every isomorphism in C with source from C1 is an isomorphism in C1.
We say that C1 is dense in C if every object of C is isomorphic in C to an object of C1.
We say that C1 is plentiful in C if for every morphism F : A ! B in C, A 2 ObC1 , there exists an isomorphism I : B ! C in C such that I F 2 HomC1 (in other words, for every quotient object of A in C there exists a representative of this quotient object in C1). Such morphism I F we call C1-canonical for F . We say that C1 is fully dense (fully plentiful) in C if C1 is a full subcategory of C and C1 is dense (plentiful) in C.
The rst two parts of this de nition are standard notions of category theory, while the notions of the other parts are introduced here for the sake of description of the structure of PDE .
Remark 3. Using the notion of \the category of objects under A", we can de ne the notions of closed subcategory and plentiful subcategory by the following way:
Remark 4. C1 is fully dense in C if and only if the embedding functor C1 ! C de nes an equivalence of these categories.
Choose some category U , which is big enough to contain all needful for us categories as it's subcategories. For our purposes U = PDE is su cient.
De ne the category U , whose objects are subcategories C of U , and a collection HomU (C1; C2) of morphisms from C1 to C2 is a one-element set if C2 is subcategory of C1 and empty otherwise, so an arrow from C1 to C2 in U means that C2 is the subcategory of C1. Let U= be the discrete wide subcategory of U , that is objects of U= are all subcategories C of U , and the only morphisms are identities, so C1 and C2 are connected by arrow in U= only if C1 = C2.
De nition 8. Suppose C1, C2 are subcategories of U . A subcategory of U , whose objects are objects of C1 and C2 simultaneously, and whose morphisms are morphisms of C1 and C2 simultaneously, is called an intersection of C1 and C2 and is denoted by C1 \ C2. In other words, C1 \ C2 is the bered sum of C1 and C2 in U .
The following proposition is obvious: Proposition 3. Suppose C1 is closed in C, and C2 is (full/closed/dense/plentiful ) subcategory of C; then C1 \ C2 is closed in C2 and is (full/closed/dense/plentiful ) subcategory of C1.
Now we introduce some graphic designations for various types of subcategories of U (see Table 1). These designations will be used, particularly, for the representation of the structure of the category of parabolic equations described below.
We shall use the term \meta-category" both for the category U and for its subcategories de ned below to avoid confusion between U and \ordinary" categories which are objects of U ; and we shall use Gothic script for meta-categories except U . One may view these meta-categories as a partial orders on the class of all subcategories of U ; we prefer category terminology here since this allows us to use category constructions for the interrelations between various partial orders.
W F
Wide
Full FI Full under isomorphisms C CI D P
Close Close under isomorphisms Dense
Plentiful
Let us de ne wide subcategories W, F, FI, C, CI, D, and P of meta-category U . Objects of them are categories, while arrows from C1 to C2 have a di erent meaning: in the meta-category W it means that C2 is wide subcategory of C1, in the meta-category F it means that C2 is full subcategory of C1, in the meta-category FI it means that C2 is full under isomorphisms in C1, in the meta-category C it means that C2 is closed subcategory of C1, in the meta-category CI it means that C2 is closed under isomorphisms in C1, in the meta-category D it means that C2 is dense subcategory of C1, CI
FP
PD
W U≥ P
D FD
PW U =
U = a in the meta-category P it means that C2 is plentiful subcategory of C1,
We shall denote the intersections of these meta-categories by the concatenations of appropriate letters, for example: FD = F \ D. The following proposition is obvious.
Proposition 4. FI \ P = F \ P; CI \ P = C; F \ P \ D = F \ D:
Interrelations between \basic" meta-categories W, F, FI, C, CI, D, P and their intersections (\composed" meta-categories) are represented on Fig. 5(a). Here an arrow means the predicate \to be subcategory of"; we shall call it the \meta-arrow". For example, meta-arrow from D to W means that W is a subcategory of D. In the language of \ordinary" categories this meta-arrow means that the statement \C2 is wide in C1" implies that C2 is dense in C1. Everywhere on Fig. 5(a) a pair of meta-arrows with the same target means that this meta-category (target of these meta-arrows) is the intersection of two \top" meta-categories (sources of these meta-arrows). For example, FD = FP \ PD.
On Fig. 5(b) the same scheme is represented as on Fig. 5(a), but the letter names are replaced by the arrows of various types. b
Instead of investigation of all or the simplest morphisms with the given source, we want to introduce a certain structure in PDE , so that the position of an object in it gives an information about the morphisms from the object and about the kind of the simplest representatives of equivalence classes of the morphisms. In the second part of this series [9] we describe such a structure for the category of parabolic equations, choosing some subcategories of PE connected by the arrows from Fig. 5(b). Then we use this structure to describe the morphisms from nonlinear reaction-di usion equation. 5
Let us consider the class P (X; T; ) of di erential operators on a connected smooth manifold X, which depend additionally on a parameter t (\time"), locally having the form
Lu =
X bij (t; x; u)uij +
X cij (t; x; u)uiuj +
X bi(t; x; u)ui + q(t; x; u); i;j x 2 X; t 2 T;
i u 2 ; in some neighborhood of each point, in some (and then arbitrary) local coordinates xi on X. Here subscript i denotes partial derivative with respect to xi, quadratic form bij = bji is positive de nite, and cij = cji. Both T and may be bounded, semi-bounded, or unbounded open intervals of R.
De nition 9. The category PE of parabolic equations of the second order is the full subcategory of PDE , whose objects are pairs A = (N; E), N = T X such that X is a connected smooth manifold, T and are open intervals, and E is an equation of the form ut = Lu, L 2 P (X; T; ) (more exactly, E is the extended version of the equation ut = Lu, that is a closed submanifold of Jn2+1(T X ), n = dim X).
Example 1. Let k(x), x 2 R3 f0g be a spherical harmonic of the k-th order. Then the map (t; x; u) 7! (t; jxj ; u / k(x) ) de nes the morphism in the category PE from the object A corresponding to equation ut = u and X = R3 f0g, T = = R, to the object A0 corresponding to equation u0t0 = u0x0x0 k (k + 1) x0 2u0 and X0 = R+, T 0 = 0 = R. One may assign to the set Sol(A0) of all solutions of the quotient equation the set F (Sol(A0)) of such solutions of the original equation that may be written in the form u = k(x)u0 (t; jxj). Example 2. The following example shows that not every endomorphism in PE is an automorphism. Consider object A, for which X = S1 = R mod 1, T = = R, E : ut = uxx. Then the morphism from A to A de ned by the map (t; x; u) 7! (4t; 2x; u) has no inverse.
(t; x; u) 7! (t0(t); x0 (t; x) ; u0(t; x; u)) ;
with submersive t0(t), x0(t; x), and u0 (t; x; u). Isomorphisms in PE are exactly di eomorphisms of the form (
X cijfifjfu + X bififu2 i;j i qfu3: Suppose F : A ! A0 is a morphism in PE, N 0 = X0
T 0
0, and E0 is de ned by the equation u0 = X Bi0j0 (t0; x0; u0) u0i0j0 + X Ci0j0 (t0; x0; u0) u0i0 u0j0 + X Bi0 (t0; x0; u0) u0i0 + Q (t0; x0; u0) : i0;j0 i0;j0 i0 Consider the extended analog of the last equation: f 0t0 f 02u0 = X Bi0j0 f 0i0j0 f 0u0
2 i0;j0
f 0i0u0 f 0j0 + f 0j0u0 f 0i0 f 0u0 + f 0i0 f 0j0 f 0u0u0 X Ci0j0 f 0i0 f 0j0 f 0u0 + X Bi0 f 0i0 f 0u0
2 i0;j0 i0
3
Qf 0u0 ;
(
Recall that F : (t; x; u) 7! (t0; x0; u0) is a morphism in PE if and only if for each point # 2 N and for each
submanifold L0 of N 0, F (#) 2 L0, the following two conditions are equivalent:
the 2-jet of L0 at the point F (#) satis es (
In other words, the conditions \f 0 is solution of (
2. In the obtained identity substitute the combinations of the derivatives of f 0 for @f 0=@t0 by formula (
3. Solve the system 1 = 0; : : : ; s = 0 of partial di erential equations for a map F .
Let us realize this procedure. Note that we shall not write out function completely. Instead we consider only some of its coe cients and then use the obtained information about F in order to simplify step by step.
First note that the derivatives of the forth order arise only in term @2f 0=@t02 when we ful ll the step 2 of the above procedure. Write this term before the nal realization of step 2 for the sake of simplicity: = X bij t0it0j fu2 t0it0ufj fu
2 t0j t0ufifu + t0ufifj i;j = X bij (t0ifu
t0ufi) (t0j fu i;j t0i (f 0t0 t0u + f 0x0 x0u + f 0u0 u0u) = t0u (f 0t0 t0i + f 0x0 x0i + f 0u0 u0i) (here and below we use the notation and so on). Hence we obtain the following system of equations: One of the following three conditions holds: 1. t0u = 0, t0x = 0; 2. t0u = 0, t0x 6= 0; 3. t0u 6= 0.
In the second case, u0u = x0u = 0. Taking into account the equality t0u = 0, we obtain a desired contradiction to the assumption that F is a submersion.
In the third case, we get from (
X cij !i!j +
X bi!i
q : i;j i Denote the expression in square brackets by (t; x; u). Then ft = fu. Expressing derivatives of f in terms of derivatives of f 0, we obtain t0t = t0u, x0t = x0u, u0t = u0u. Consider the eld of hyperplanes that kill the 1-form dt0 in the tangent bundle T M (recall that t0u 6= 0, so dt0 is non-degenerated). The di erential of F vanishes on these hyperplanes because du0 ^ dt0 = dx0i0 ^ dt0 = 0. Therefore rank(dF ) 1. Since dim N 0 3, F can not be submersive, which contradicts the de nition of an admitted map.
Finally, we see that only the rst case is possible. Hence t0 is a function of t, and f 0t0 may appear only in the representation of ft. Let us look at the terms of containing (f 0u0 ) 2: =
X i0;j0;k0;l0
t0tx0iu0 x0ju0 B0k0l0 f 0i0 f 0j0 f 0k0 f 0l0 f 0u0u0 (f 0u0 ) 2 + : : : : Substitution of any covector ! = Pi0 !i0 dx0i0 2
(T X0) to the expression
X i0;j0;k0;l0 t0tx0iu0 x0ju0 B0k0l0 !i0 !j0 !k0 !l0 = t0t
X x0iu0 !i0 2 i0
X B0k0l0 !k0 !l0 k0;l0 into account that F is submersive, we obtain t0t 6= 0. Therefore Pi0 x0iu0 !i0 = 0 for any !, that is x0u implies x0 = x0 (t; x), t0 = t0 (t), which completes the proof.
As Remark 2 shows, our de nition of morphism in PDE is a generalization of the reduction by a symmetry group. So we can obtain sets of solutions more general than the sets of group-invariant solutions provided by the group analysis of PDE (though our approach is more laborious owing to the non-linearity of the system of PDE describing a morphisms). Let us illustrate this by an example of a primitive morphism. De nition 10. A morphism F : A ! B of a category C is called a reducible in C if there exist non-invertible morphisms G : A ! C, H : C ! B in C such that F = H G. Otherwise, a morphism is called primitive in C.
Note that the reduction of PDE by a symmetry group de nes a primitive morphism if and only if this group has no proper subgroups The reduction by any symmetry group that is not a discrete cyclic group of prime order may be always represented as a superposition of two nontrivial reductions, so the corresponding morphism is a superposition of two non-invertible morphisms and therefore is reducible. In particular, this situation takes place for any nontrivial connected Lie group.
However, the situation for morphisms is completely di erent. Even a morphism that decreases the number of independent variables by 2 or more may be primitive; below we present an example of such a morphism. In contrast, in the Lie group analysis we always have one-parameter subgroups of a symmetry group, so the morphism corresponding to a symmetry group is always reducible.
Example 3. Consider the following morphism F : A ! B in PE :
A is the heat equation ut = a(u) u posed on X = f(x; y; z; w) : z < wg
R4 equipped with the metric 01
0 0 0
2, a 2= Aexp [ Adeg. In the coordinate form, A looks as a 1(u)ut = uxx + uyy 2 uyz 2 uyw + 1 +
2 uzz B is the heat equation a 1(u)ut = uxx + uyy posed on Y = f(x; y)g = R2 equipped with Euclidean metric.
The morphism F is de ned by the map (t; (x; y; z; w); u) 7! (t; (x; y); u).
This morphism decreases the number of independent variables by 2 and nevertheless is primitive in PE .
Additional examples of morphisms that are not de ned by any symmetry group of the given PDE, and also a detailed investigation of the case dim Y = dim X 1, may be found in [6, sections 9, 10] and [7, sections 7{9]. Acknowledgements. This work was partially supported by the RFBR grants 15-01-02352 and 15-51-06001 (Russia). I am grateful to Vladimir Rubtsov and Yuri Zarhin for useful remarks.