EXISTENCE, UNIQUENESS, AND COMPUTATION OF THE RESPONSE OF AN ARRAY OF ELASTOPLASTIC SPRINGS TO PERTURBATIONS: A COMPLEMENTARITY APPROACH 1 Ivan Gudoshnikov [email protected] Department of Mathematical Sciences, The University of Texas at Dallas 1. Introduction In section 3 we consider a mechanical system which is a lattice of m elastoplasctic bars, but some of the bars have rigid controlled pistons, attached to their endpoints (see figure 1). Also there are external forces, applied to the nodes(joints) of the lattice. We assume, that we know lengths of pistons and external forces for each moment of time t ∈ [0, T ]. Figure 1. An example of the lattice with pistons. There are 13 elastoplastic bars and 3 rigid controlled pistons As suggested in [1, 6a], we describe state of each elastoplastic bar by two real values: elastic enlongation ek and plastic enlongation pk , so the sum ek + pk is total enlongation of the bar with respect to some fixed zero length (which corresponds to the origin in space of enlongations Rm ), see figure 2. The vectors e and p of, respectively, elastic and plastic enlongations of all bars describe a configuration of the enitre system at any moment of time. Figure 2. Elastic and plastic enlongations of a bar. We are looking for the functions e(t) and p(t), such that at any moment of time the resultant forces at all nodes are zero (it is called quasistatic evolution of the system). Besides this, the system of bars have to be geometrically valid, i.e. the total enlongation e + p has to be such that a) it is physically possible for the bars to stay connected, b) length of those bars, which have an attached piston, has to be the same as the length of the piston. The condition a) may be written as an inclusion of e + p into a secon-order hypersurface in Rm , but under assumption of small enlongation we may linearize it (similarly to the Finite Element Method) and obtain a linear constraint (which we call geometric constraint), of the type e(t) + p(t) ∈ Im D, 1 The research is partially supported by National Science Foundation grant CMMI-1436856 1 Ivan Gusoshnikov The response of elastoplastic springs to perturbations where D is a linear operator. The condition b) produces affine constraint (which we call additional constraint) e(t) + p(t) ∈ U + g(t), (1.1) m where U ⊂ R is a hyperplane and the vector g represents changing lengths of the pistons. Clearly, the intersection of th geometric constraint and the additional constraint also has the form (1.1) with some different U and g. Physically, a resultant force at a node is a composition of a stresses of all bars, connected to the node, external force, applied to the node, and reactions of the pistons. The stresses of bars are generated by the elastic enlongations according to the Hooke’s law (so only this part of a total enlongation of a bar generates stress). We collect stresses of all bars into a vector s(t), so the Hookes law becomes s(t) = Ae(t), (1.2) where A is a diagonal matrix of Hooke’s koefficients. To have quasistatic evolution it is clearly needed to have external forces such that they can be compensated by others mentioned forces. This implies that the external forces can be represented in form of a vector h(t) ∈ Rm , where each component hk is a magnitude of two forces, applied to both endpoints of a k−th bars in opposite directions (similarly to how the stresses are applied), see figure 3 for example. The resultant force vanishes at all nodes iff (1.3) ∃(rt ∈ U ⊥ )[h(t) + s(t) + rt = 0], where rt is reaction of the constraint. Figure 3. External forces, applied to the nodes, can be represented as h = (h1 , h2 , h3 ) ∈ R3 . + We also assume that a possible stress value of each bar is restricted by an interval [c− k , ck ]. The plastic component p is required to maintain such restriction. Assuming that the plastic enlongation of a bar changes only when its stress reaches the boundary, we can derive the following law of plasticity: (1.4) d p(t) ∈ NC (s(t)), dt 2 Ivan Gusoshnikov The response of elastoplastic springs to perturbations where NC (s(t)) is an outwards normal cone to the rectangular set C of all admissible vectors of stress values, taken at current stress point s(t). We also need initial conditions e0 , p0 , such that (1.1)-(1.4) are satisfied for t = 0. We call a lattice with pistons, described by (1.1)-(1.4) Physical Elastoplastic Model. The plasticity law (1.4) is a particular type of so-called resistance laws (see [1, Chapter 4]) and the same resistance law also corresponds to the Coloumb’s law of friction (see [1, 4e]). This gives us the opportunity to write another model, utilizing dry friction phenomena instead of plasticity. The new model consists of just two point in a vector space, connected by an elastic spring. The stress of the spring is denoted by s. The point p (we call it “plastic” component) experience dry friction, such that the force of friction always belongs to a closed convex set C. The other point x (the “visible” component) is locked in a translationally moving plane U + g(t) and submitted to an external force c(t) (see figure 4). For this model we also consider the quasistatic evolution problem. Then we show in the section 4, that the mathematical description of the Physical Elastoplastic Model will completely fit into this new model (h corresponds to c, e + p to x and others variables are the same up to a sign). f p c(t) s C −s x r U + g(t) Figure 4. Abstract elastoplastic problem. This model may be, in some sence, easier to imagine then the Physical Elastoplastic Model. But it also may be seen without any relation to the physical intuition, just as a purely abstract mathematical problem. It will be formulated in terms of abstract linear spaces, so we call it Abstract Elastoplastic Problem. In order to solve it we will make several assumptions and reformulate the problem in terms of a separable Hilbert space (in case of a Physical Elastoplastic Model we will still have it in terms of enlongations space Rm ). We do this by identifying the space of forces with the configurations space by the bijection A, so the stress s is identified with the elastic deviation e = x − p, external force c is identified with a vector A−1 c and so on. We introduce the new variables y := s − c − g, z := x − c − g. Then the Moreau Sweeping Process −y 0 ∈ N(C−c(t)−g(t))∩U ⊥ (y), y(0) = y0 . allows us to find the elastic part of solution. Further, the differential inclusion 0 z ∈ (NC (y + c + g) + y 0 ) ∩ U, z(0) = z0 . always has a solution, which is directly connected to the plastic part. Section 5 contains the proof of this method, which was also taken from [1] and [2] and filled with details. Notice, that the moving set in the Moreau Sweeping Process is a cross-section of the set C, translated on c + g, by the fixed orthogonal complement of the constraint plane (the ortogonality 3 Ivan Gusoshnikov The response of elastoplastic springs to perturbations here is understood in sence of a specific inner product). Recall, that the vectors c and g, may represent the applied external forces and lengths of the pistons respectively, so their change leads to the change of the cross-section, which sweeps the elastic part. Section 6 of the paper shows this process in detail by manually solving a simple concrete example of an Physical Elastoplastic Model. Also the analysis of the example highlights several general properties of such models. 2. Preliminaries 2.1. Normal cones and convex analysis. Definition 2.1. For a Hilbert space H with an inner product < ·, · > and a convex set C ⊂ H the support function of C is defined as ∗ δC : H → R ∪ {+∞} ∗ δC : x 7→ sup{< x, c >: c ∈ C} Definition 2.2. For a Hilbert space H with an inner product < ·, · >,a closed convex set C ⊂ H and y ∈ C the set NC (y) = {ξ ∈ H : ∀(c ∈ C)[< ξ, c − y >6 0]} is called an outwards normal cone to C at y. Remark 2.1. Sometimes we will use the concept of the normal cone in more general case, when < ·, · >: E × F → R is a bilinear form, defined on two linear spaces E, F , which not necessaraly satisfies all inner product axioms. Lemma 2.1. For any y, a ∈ H s.t. y + a ∈ C NC (y + a) = NC−a (y). Proof. NC (y + a) = {ξ ∈ H : ∀(c ∈ C)[< ξ, c − y − a >6 0]} = = {ξ ∈ H : ∀(c ∈ C − a)[< ξ, c − y >6 0]} = NC−a (y). Lemma 2.2. NC (y) = −N−C (−y) Proof. NC (y) = {ξ ∈ H : ∀(c ∈ C)[< ξ, c − y >6 0]} = {ξ ∈ H : ∀(c ∈ −C)[< ξ, −c − y >6 0]} = = {ξ ∈ H : ∀(c ∈ −C)[< ξ, (−y) − c >6 0]} = = −{ξ ∈ H : ∀(c ∈ −C)[< ξ, c − (−y) >6 0]} = −N−C (−y). Lemma 2.3. For closed convex sets C1 , C2 ⊂ H, s.t. int C1 ∩ C2 6= ∅ and y ∈ C1 ∩ C2 NC1 (y) + NC2 (y) = NC1 ∩C2 (y). 4 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Proof. Embedding NC1 (y) + NC2 (y) ⊂ NC1 ∩C2 (y): Put ξ1 ∈ NC1 (y), ξ2 ∈ NC2 (y), i.e. ∀(c ∈ C1 )[< ξ1 , c − y >6 0] hence ^ ∀(c ∈ C2 )[< ξ2 , c − y >6 0], h i ^ ∀(c ∈ C1 ∩ C2 ) (< ξ1 , c − y >6 0) (< ξ2 , c − y >6 0) which implies ∀(c ∈ C1 ∩ C2 )[< ξ1 , c − y > + < ξ2 , c − y >6 0] which is equivalent to ∀(c ∈ C1 ∩ C2 )[< ξ1 + ξ2 , c − y >6 0] i.e. ξ1 + ξ2 ∈ NC1 ∩C2 (y). For a proof of the reversed embedding see [2, Lemma 1(b)]. Lemma 2.4. For a Hilbert space H1 =H × R with the inner product < (x1 , ξ1 ), (x2 , ξ2 ) >1 =< x1 , x2 > +ξ1 ξ2 , for any ξ ∈ R we have NC (x) × {0} = NC×R ((x, ξ)). Proof. Let y ∈ NC (x), i.e. ∀(c ∈ C)[< y, c − y >6 0]. Observe, that for any c1 ∈ C, c2 ∈ R < (y, 0), (c1 , c2 ) − (x, ξ) >1 =< y, c1 − x > +0 · (c2 − ξ) =< y, c1 − x >6 0, so (y, 0) ∈ NC×R ((x, ξ)) . Now let (y1 , y2 ) ∈ NC×R ((x, ξ)), i.e. for all c1 ∈ C, c2 ∈ R < (y1 , y2 ), (c1 , c2 ) − (x, ξ) >1 6 0 < y1 , c1 − x > +y2 (c2 − ξ) 6 0 y2 (c2 − ξ) 6 − < y1 , c1 − x > . Two cases are possible: (a) y2 = 0, hence < y1 , c1 − x >6 0, so we have (y1 , y2 ) ∈ NC (x) × {0}. (b) y2 6= 0, < y1 , c1 − x > , y2 < y1 , c1 − x > + ξ. c2 6 − y2 But we can choose any c1 ∈ C and then choose < y1 , c1 − x > c2 > − + ξ, y2 since the right-hand side is just a number. So we have a contradiction and this case is not possible. 5 c2 − ξ 6 − Ivan Gusoshnikov The response of elastoplastic springs to perturbations Lemma 2.5. [2, Lemma 1(d)] If C ∈ H is nonempty, closed and convex, and ρ > 0 then the set Cρ := {x ∈ H : dist(x, H \ C) > ρ} is convex. Lemma 2.6. [2, Lemma 1(c)] If C ⊂ H is nonempty, closed and convex, ρ > 0 and x ∈ Cρ then ∗ ∀(y ∈ H) [ρ · kyk+ < y, x >6 δC (y)] . 2.2. Moreau sweeping process. Definition 2.3. ([2, Section 3.1]) Let H be a hilbert space, a map C : [0, T ] → 2H , and u0 ∈ C(0). We say that a function u : [0, 1] → H is a solution of the Moreau Sweeping Process ( −u0 (t) ∈ NC(t) (u(t)), u(0) = u0 ; iff u satisfies the following 4 properties: (1) (2) (3) (4) u(0) = u0 , for all t ∈ [0, T ] we have u(t) ∈ C(t), for almost all t ∈ (0, T ) there exists u0 (t), for almost all t ∈ (0, T ) we have −u0 ∈ NC(t) (u(t)). Theorem 2.1. ([1, Chapter 5]),[2, Section 3.1]) If C(t) : [0, T ] → 2H is absoluely continous in sence of Hausdorff distance and for every t ∈ [0, T ] the set C(t) is nonempty, closed and convex, then there exists a unique soluton u of the Moreau Sweeping Process. The function u is absolutely continous and u0 ∈ L2 ([0, T ]; H). If C(t) is Lipscitz-continous with a constant L in sence of Hausdorff distance, then then u is also Lipschitz-continous with L. 2.3. Measurable functions and multifunctions. Lemma 2.7. (see [6, Lemma 6.4.2]) Let X be a Hausdorff topological space and Y be a separable metric space. Then Borel σ-algebra, generated by product topology on X × Y is equal to product σ-algebra of Borel σ-algebras of X and Y : B(X × Y ) = B(X) ⊗ B(Y ) Definition 2.4. (see [4, Def. III.10], cf. [5, Def. 1.§3.1]) Let (T, S ) be a measurable space (i.e. S is a σ-algebra of subsets of T), X — separable metric space, Φ : T → {F ⊂ X : F − complete} ⊂ 2X then Φ is said to be S -measurable if T0 := {t ∈ T : Φ(t) = ∅} ∈ S and ∀(U ⊂ X : U − open) Φ−1 (U ) := {t ∈ T : Φ(t) ∩ U 6= ∅} ∈ S . 6 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Theorem 2.2. (see [4, Th. III.9]) Let (T, S ) be a measurable space, X — separable metric space, Φ : T → {F ⊂ X : F − complete} ⊂ 2X \ {∅} Φ is S -measurable iff it admits a pointwise dense sequence of strongly measurable selections: i h ∃({ϕi }i∈N : (ϕi : T → X) − S -measurable)∀(t ∈ T ) Φ(t) = {ϕi (t)}i∈N Theorem 2.3. (see [4, Prop. III.13]) If (T, S ) is a measurable space, X is a separable metric space, Φ : T → {F ⊂ X : F − complete} ⊂ 2X is S -measurable, then the graph of Φ satisfies G(Φ) := {(t, x) ∈ T × X : x ∈ Φ(t)} ∈ S ⊗ B(X). Theorem 2.4. (Projection theorem, see [5, Prop. 1.§3.6]) Let (Ω, A , ν) be a complete-measure space and (X, d) be a complete separable metric space(Polish space). Then for every A ∈ A ⊗ B(X) projΩ (A) := {ω ∈ Ω : ∃(x ∈ X)[(ω, x) ∈ A]} ∈ A . 2.4. Some properties of Bochner integral. Since we are dealing with absolutely continous functions and integration in a Hilbert space it may be useful to recall the following. For absolutely continous vector-valued functions and the Bochner integral the statement of the Fundamental Theorem of Lebesgue integral calculus (see [8, Th. 4.4.1] for real-valued version) is not generally true, see [9, Example is section XIII.4] for a counter-example. But in case of reflexive separable Banach space E an absolutely continous function f : [0, T ] → E has Bochner-integrable derivative f 0 , defined almost everywhere and Zb f (b) − f (a) = f 0 (s)ds a for a, b : 0 6 a 6 b 6 T , see [10, Th. 3.8.6], [11, Appendix]. Notice, that absolute continuity implies weak absolute continuity, since a continous linear functional is Lipschitz-continous with its norm and a composition of a Lipschitz-continous function and an absoluteny continous function is absolutely continous. In the opposite way, we always have a Bochner integral absolutely continous as a set function (see [10, Th. 3.7.11]). 3. Development of the Physical Elastoplastic Model 3.1. Configuration of the elastic model. For n ∈ N we consider a set of n points(nodes) A1 , A2 , ..An and let m pairs of them be connected by bars (where 1 6 m 6 21 n(n − 1)). We denote a bar, connecting nodes Aik and Ajk , by Bik jk or by Bk , where k = 1..m. We make a convention, that ik < jk for all k = 1..m and put K := {(ik , jk ) : k = 1..m}. Of course, any configuratiuon of the system may be described in terms of nodes coordinates, but in this text we assume, that movements of vertices Ai are small with respect to lengths of bars Bij . This allows us to linearize the relation (see further) between coordinates of the nodes and enlongations of bars Bk with respect to their lengths at a “zero” configuration, which we 7 Ivan Gusoshnikov The response of elastoplastic springs to perturbations artificially prefer. The choice of this zero state doesn’t play a role until we specify a relation between a configuration of the system and forces, applied to it. But, for example, the zero state may be s.t. all of the bars Bij experience no internal stress. We describe the system in terms of the following variables: • ξi ∈ R3 — coordinate vector of Ai at “zero” state, • xi ∈ R3 — small displacement vector of Ai from “zero” state, the total configuration of the system is denoted by the vector x = (xi )ni=1 ∈ X := R3n , • ξˆi ∈ R — coordinate vector of Ai , so ξˆi = ξi + xi , • θij = θk ∈ R — length of Bij (for i = ik , j = jk ) when the system is in zero state, i.e. when all xi = 0, • eij = ek ∈ R — enlongation of Bij (for i = ik , j = jk ) with respect to zero state, so length of Bij equals θij + eij when Ai , Aj are at ξˆi , ξˆj . The configuration of all bars is a m vector e = (ek )m k=1 ∈ E := R . The figure 5 shows how such system may look like. Figure 5. Configuration of an elastic physical model. 3.2. Linearization. In order to find a relation between x and e we consider a function ϕ : R3 × R3 → R, which gives a length of a bar θij + eij by the coordinates of its endpoints ξˆi , ξˆj : !1 3 2 2 X (3.1) ϕ(ξˆi , ξˆj ) = kξˆj − ξˆi k = ξˆjp − ξˆip , p=1 0 ϕ = ∂φ ∂φ ∂φ ∂φ ∂φ ∂φ , , , , , ∂ ξˆi1 ∂ ξˆi2 ∂ ξˆi3 ∂ ξˆj1 ∂ ξˆj2 ∂ ξˆj3 ! 1 = 2 2 21 3 P ξˆjk − ξˆik · k=1 · −2(ξˆj1 − ξˆi1 ), −2(ξˆj2 − ξˆi2 ), −2(ξˆj3 − ξˆi3 ), 2(ξˆj1 − ξˆi1 ), 2(ξˆj2 − ξˆi2 ), 2(ξˆj3 − ξˆi3 ) . We evaluate the derivative at the zero configuration ξi , ξj and put it as ϕ0 (ξi , ξj ) = (−αij1 , −αij2 , −αij3 , αij1 , αij2 , αij3 ) . 8 Ivan Gusoshnikov The response of elastoplastic springs to perturbations ξ −ξ Observe, that αij = kξjj −ξii k is a normalized vector along the bar Bij directed from ξi to ξj . Next, for ξi , ξj and small displacements xi ,xj we have by the definition of derivative: eij = ϕ(ξi + xi , ξj + xj ) − ϕ(ξi , ξj ) = ϕ0 (ξi , ξj ) · (xi , xj ) + o(k(xi , xj )k) = = −αij · xi + αij · xj + o(k(xi , xj )k) = αij · (xj − xi ) + o(k(xi , xj )k). So for small xi , xj we can assume (3.2) m e = (ek )m k=1 = (αik jk · (xjk − xik ))k=1 =: Dx, where D : R3n → Rm is a linear map. In the other words, for in case of small displacements rotations of the bars are negligible and don’t count for enlongations of the bars. Figures 6 and 7 can give an image of how the linearization works. Figure 6. Possible positions of a node Aj for xi = 0 and some fixed eij as described by the original relation (3.1)(solid) and as the linearized relation (3.2)(dashed). Figure 7. The real length of Bij and its linerized appriximation. The enlongation eij = αij · (xj − xi ) is the sum of lengths of the bold segments. 3.3. External and internal forces. Let each node Ai is submitted to a given external force yi ∈ R3 . Let the work done by force y = (yi )ni=1 ∈ Y for displacement x = (xi )ni=1 ∈ X of the system is given by the bilinear form ·, · : X × Y → R, (3.3) x, y = n X i=1 9 xi · yi . Ivan Gusoshnikov The response of elastoplastic springs to perturbations Here spaces X and Y are just two exemplars of (R3 )n , placed in seprating duality (see, for example, [3, II.40.§6.1]) by the form (3.3), and · is the usual inner product in R3 . Let each bar Bij generates force, applied at its endpoints Ai , Aj and directed along αij and −αij respectively. We call this force tension or stress. Since the magnitude of stress has to be the same for both endpoints, we can describe it as a number sij . Stress values for all bars form the linear space S := Rm 3 (sik jk )m k=1 = s. For small enlongations e ∈ E the work, done by s ∈ S is given by the bilinear form < ·, · >: E × S → R, < e, s >= − (3.4) m X eik jk sik jk . k=1 Here spaces E and S are two exemplars of Rm , placed in separating duality by the form (3.4). We have the minus sign in (3.4) since for both eij and sij positive (or negative) the corresponding displacement of endpoints of the bar has opposite to applied at the endponts stress direction, so the work, done by stress has to be negative in such case. In terms of the configuration spaces X and E we can write, that D:X→E Then substituting Dx to (3.4) we obtain a linear functional s ◦ D : x 7→< Dx, s >, (3.5) The following lemma shows, how to represent this functional in the form ·, ŷ , where ŷ is some element of Y , depending on s. In other words, we represent an operator, adjoint to D in terms of elements of Y and form (3.3). Lemma 3.1. Let i = 1..n be an index of a vector component of an element of Y , then X X (3.6) (D∗ s)i = sij αij − sji αji . j:(i,j)∈K j:(j,i)∈K defines a linear operator D∗ : S → Y such that ∀(x ∈ X, s ∈ S)[< Dx, s >= x, D∗ s ]. (3.7) Proof. Substituting definitions (3.3),(3.4) and (3.2) into (3.7) we obtain (3.8) − m X (αik jk · (xjk − xik )) sik jk = n X xi · (D∗ s)i i=1 k=1 n We fix an arbitrary i = 1..n, p = 1..3 and choose x = (xi0 p0 )3p0 =1 i0 =1 ∈ X such that the righthand side is equal to the real number (D∗ s)ip . Namely, all vectors components of x, s.t. i0 6= i are set to 0; vector xi is set to (δpp0 )3p0 =1 . Then in the left-hand side of (3.8) for jk = i we have X − αjip sji j:(j,i)∈K and for ik = i we have − X αijp (−1)sij j:(i,j)∈K 10 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Putting all together we obtain the numerical equality X (D∗ s)ip = αijp sij − j:(i,j)∈K X αjip sji , j:(j,i)∈K which may be written in vector form as (3.6). Lemma 3.2. D∗ = −DT . Proof. The definition (3.3) implies that for all x ∈ X, s ∈ S Dx · s = x, DT s , and by definition (3.4) we have −(Dx · s) =< Dx, s > . Thus x, −DT s = − x, DT s = x, D∗ s , Hence for all s ∈ S −DT s = D∗ s, −DT = D∗ . Lemma 3.3. Superposition of all forces vanishes iff y + D∗ s = 0. (3.9) Proof. Recall that the stress of a bar Bij , applied to a point Ai has a direction αij and magnitude sij and assume, that the superposition of tensions and external forces, applied to any Ai vanishes: X X yi + sij αij + sji (−αji ) = 0. j:(i,j)∈K j:(j,i)∈K We substitute (3.6) and conclude, that the equilibrium of forces is equivalent to y + D∗ s = 0. 3.4. Geometric constraint. Now we are interested in representing the model trough elements of spaces E and S only. If the load y : [0, T ] → Y is a known function of time, (3.9) implies that there is at least one function h(t) : [0, T ] → S, s.t. for all t ∈ [0, T ] (3.10) D∗ h(t) = y(t), i.e. h is a representation of external load in terms of the space S. It means that at any moment (3.11) y(t) ∈ D∗ S. This may be also interpreted as follows: external forces have to be such that they may be compensated by stresses of the bars to maintain the quasistatic character of the evolution. Next, notice that the equilibrium criterion (3.9) is equivalent to D∗ (h + s) = 0, which may be written as ∀(x ∈ X) [ x, D∗ (h + s) = 0] , 11 Ivan Gusoshnikov The response of elastoplastic springs to perturbations which is equivalent (by (3.7)) to ∀(x ∈ X) [< Dx, h + s >= 0] . (3.12) So h + s has to be orthogonal to the subspace of possible configurations DX in sence of the form (3.4), and this has the physical meaning: the system is at rest when the resultant force produces zero work. Moreover, if we want to consider evolution of the system in time in terms of enlongations, we have to watch not only the forces, but also the configuration of enlongations e ∈ E itself. It sould be consistent from the geometrical point of view, i.e. it should be realizable for at least one x ∈ X. In our linearized framework this condition can be written as the following inclusion: e ∈ DX, (3.13) which we call the geometric constraint. Notice, that (3.12) is also equivalent to h i ^ ∃(r ∈ F )∀(x ∈ X) (< Dx, r >= 0) (h + s + r = 0) , or ∃(r ∈ DX ⊥ )[h + s + r = 0], where the symbol ⊥ is understood in sense of the work form < ·, · >, but by the definition (3.4) the euclidean orthogonality is equivalent to ⊥. We call such r the reaction of the constraint (3.13). Example 3.1. Consider the system with three nodes at ξ1 = (0, 0, 0), ξ2 = (1, 0, 0), ξ3 = (0, 1, 0), connected by 3 bars. Then we have α12 = (1, 0, 0), α13 = (0, 1, 0), α23 1 1 = −√ , √ , 0 . 2 2 So (1, 0, 0) · (x21 − x11 , x22 − x12 , x23 − x13 ) x21 − x11 (0, 1, 0) · (x31 − x11 , x32 − x12 , x33 − x13 ) = x32 − x12 Dx = = 1 1 √1 √ √ (−x + x + x − x ) − 2 , 2 , 0 · (x31 − x21 , x32 − x22 , x33 − x23 ) 31 21 32 22 2 x11 x12 x13 −1 0 0 1 0 0 0 0 0 x21 0 0 0 1 0 x = 0 −1 0 0 22 0 0 0 √12 − √12 0 − √12 √12 0 x 23 x31 x32 x33 Observe, that by changing x11 ,x12 and x22 the one can independently adjust all three components of Dx to any desired value (see the figure 8), so DX = E. We also should calculate D∗ S to know admissible values of external load y. For the first time we use the definition (3.7) to find the matrix D∗ (but we could just take the negative transpose of D, according to the lemma 3.2): K = {(1, 2), (2, 3), (1, 3)}, 12 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 8. By moving nodes in such directions it is possible to adjust linerized distances between nodes independently. (D∗ s)1 = s12 α12 + s13 α13 = (s12 , s13 , 0) ∗ i = 2 : (D s)2 = s23 α23 − s12 α12 = − s√232 − s12 , s√232 , 0 i = 3 : (D∗ s)3 = −s13 α13 − s23 α23 = s√232 , − s√232 − s13 , 0 . i=1: We write h instead of s, following the formula (3.10): h12 1 y11 0 h13 y12 0 0 y13 − h√23 − h −1 12 y21 2 h 0 √23 = y = D∗ h = 22 2 y23 0 0 h y31 √23 0 2 y32 − h√23 − 0 h13 2 y33 0 0 0 1 0 0 0 0 0 −1 0 0 0 0 − √12 h11 √1 h . 13 2 0 h23 √1 2 1 − √2 0 Observe, that the matrix of D∗ is the negative transpose of D. Notice, that the image of D∗ is described by the following equation: y13 = 0, y 23 = 0, y33 = 0, y22 = y31 , y21 = −y11 − y22 , y32 = −y12 − y22 . It is reasonable, that the third components has to be zeroes, since all nodes lay in the xy−plane. The last pair of equations can be transformed to y11 + y21 + y22 = 0, y12 + y32 + y31 = 0; which mean a kind of symmetrical balance(see figure 9). Example 3.2. Consider the system with three nodes at ξ1 = (−1, 0, 0), ξ2 = (0, 0, 0), ξ3 = (1, 0, 0), 13 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 9. Admissible external forces, decomposed on directions. Same color means same magnitude. connected by 3 bars (figure 10). Figure 10. A system with three aligned bars. Observe, that α12 = α23 = α13 = (1, 0, 0), e12 α12 · (x2 − x1 ) x21 − x11 −1 0 0 1 e23 = Dx = α23 · (x3 − x2 ) = x31 − x21 = 0 0 0 −1 e13 α13 · (x3 − x1 ) x31 − x11 −1 0 0 0 Notice, that we have a relation e12 + e23 = e13 e12 + e23 − e13 = 0 < e, (1, 1, −1) >= 0 So we have the geometric constraint e ∈ DX = (1, 1, −1)⊥ . 14 x11 x12 x13 0 0 0 0 0 x21 0 0 1 0 0 x 22 0 0 1 0 0 x 23 x31 x32 x33 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 11. Space E = S = R3 with involved vectors and subspaces. Hence the reaction r and the superposition of stress and external force s + h = −r has to belong to the one-dimensional subspace V = span{(1, 1, −1)}, see figure 11. In other words, r12 = r23 = −r13 , i.e. s12 + h12 = s23 + h23 = −s13 − h13 . This conclusion coinsides with the observable physical requirement of having all forces compensating each other at each node. Let’s find admissible external forces. 1 0 1 0 0 0 0 0 0 −1 1 0 h12 h23 0 0 0 (3.14) D∗ h = 0 h13 0 0 0 −1 −1 0 0 0 0 0 0 As expected from the one-dimensional nature of the configuration, all second and third components of external forces have to be zero. So we are left with y11 , y21 , y31 , such that y11 = h12 + h13 , y21 = −h12 + h23 , y31 = −h23 − h13 . 15 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Fix arbitrary y11 , y31 and get from their formulae h12 = y11 − h13 , h23 = −y31 − h13 . By substituting this into the formula for y21 and cancelling h13 we finally obtain y21 = −y11 − y31 which means y11 + y21 + y31 = 0. 3.5. Additional constraints. The model also allows us to enforce additional artificial constraints of the form (3.15) e(t) ∈ U + g(t), where U is a fixed in time linear subspace of E and g : [0, T ] → E is a given function. Since (U + g(t)) ∩ DX may also be represented in a form (3.15) of a moving affine constraint with some different U and g, we will consider (3.15) as a general form of constraint on e, replacing (3.13). If we don’t have any additional constraint, we just put U = DX and g ≡ 0. Having additional constraints on the configuration gives more possibilities for forces h + s. Instead of (3.12) we put ∀(u ∈ U ⊂ DX)[< u, h + s >= 0], to have zero work produced by force on allowed displacements. In terms of reaction this may be written as (3.16) ∃(r ∈ U ⊥ )[h + s + r = 0]. Replacement of (3.12) by (3.16) requires a proper formal justification, which we will provide further as Lemma 3.4 for the following particular case of the constraint. A particular example of the affine moving constraint (3.15) may be imagined as a rigid telescopic piston, attached to two nodes along with elastic bar Bij . We suppose, that we can control, how the piston changes its length, so the length as a function of time is given by gij : [0, T ] → R. The function gij is a part of input data in our model. The following examples illustrate, how the telescopic pistons are interpreted as affine constraints in the space of enlongations E. Example 3.3. Consider the system from the example 3.1 with a rigid piston attached to the nodes A2 and A3 (figure 12). Figure 12. Three-nodes system with moving affine constraint. 16 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Let the length of the piston be given by g23 (t). Then we have the moving affine constraint in the form (3.15), where U = R × R × {0} and g(t) = (0, 0, g23 (t)). At a moment of time such constraint may look like figure 13. Figure 13. The moving constraint in the space E. Example 3.4. Consider the system from the example 3.2 with a rigid telescopic piston, attached to the nodes A1 and A3 , see figure 14. Figure 14. The system with three aligned bars and the constraint. For the original system from the example 3.2 we already had a geometric constraint of the plane DX = (1, 1, −1)⊥ (figure 11). Similarly to the exlample 3.3, the telescopic piston gives us the additional constraint of the form e ∈ Ua + ga (t), where Ua := {(e12 , e23 , 0) : e12 , e23 ∈ R}, ga (t) := (0, 0, g13 (t)), which looks like the one on the figure 13, except the third coordinate axis corresponds now to the bar B13 . 17 Ivan Gusoshnikov The response of elastoplastic springs to perturbations We have to find a linear subspace U ⊂ E and a function g : [0, T ] → E such that for all t ∈ [0, T ] U + g(t) = DX ∩ (Ua + ga (t)). From the obvious geometry we have U := span {(1, −1, 0)}, g(t) ∈ {(g12 , g23 , g13 (t)) : (g12 , g23 , g13 (t)) · (1, 1, −1) = 0}. Without loss of generality we may choose g(t) ⊥ U , i.e. g13 (t) g13 (t) g(t) := , , g13 (t) . 2 2 The figure 15 shows this process graphically. Figure 15. Intersection of the constraints. Lemma 3.4. If the system has pistons between nodes Aik , Ajk , where ik < jk , k = 1..p then the vanishing of forces criterion is (3.16), where U is intersection of the geometric constraint and constraints, induced by the pistons. Proof. Each piston can generate a reaction force, applied at its endpoints Aik and Ajk as vectors rk αik jk and −rk αik jk respectively for some scalars rk . Since the pistons are assumed to be perfectly rigid rk have no restrictions on magnitude. Each reaction can be also written as a vector rk∗ := (0, . . . , 0, rk , 0 . . . , 0) ∈ F, where rk has k-th position. Clearly rk∗ ⊥ Uk , where Uk := Rk−1 × {0} × Rm−k ⊂ E is a part of the constraint Uk + gk (t), generated by the k−th piston. Put r∗ := p X k=1 18 rk∗ . Ivan Gusoshnikov The response of elastoplastic springs to perturbations By applying the same argument as in the proof of the lemma 3.3 we can see, that resultant forces at all nodes vanish iff ! p X ∗ ∗ y+D s+ rk = y + D∗ (s + r∗ ) = 0. k=1 Similarly to what we had before, this means that D∗ (h + s + r∗ ) = 0, ∀(x ∈ X) [ x, D∗ (h + s + r∗ ) = 0] , ∀(x ∈ X) [< Dx, h + s + r∗ >= 0] , ∃(r ∈ DX ⊥ )[h + s + r + r∗ = 0], We put r ∗∗ ∗ ⊥ := r + r ∈ DX + p X Uk⊥ = DX ∩ k=1 U := DX ∩ p \ !⊥ Uk , k=1 p \ Uk . k=1 Thus we obtain (3.17) ∃(r∗∗ ∈ U ⊥ )[h + s + r∗∗ = 0] Notice, that U is the intersection of all constraint planes, which we have, so (3.17) is exactly (3.16). Remark 3.1. The moving constarints of the type (3.15) may be produced in a different way. We will briefly describe it here and refer the reader to [1, 3c] for better justification. Let some of the nodes Ai be locked in translationally moving smooth surfaces {x : ϕi (t, x) = 0}, in such way that a node can move inside the correspinding surface without any friction, but cannot leave it. Under the assumption of small displacements, we can replace the surfaces with their tangents, so the constraint on nodes becomes ∇ϕi (t, ξi ) · xi + ϕi (t, ξi ) = 0. Since the surfaces are moving translationally, the gradient ∇ϕ(t, ξi ) =: ∇ϕ(ξi ) does not depend on time, and we can write the constraint as xi ∈ Li + gi (t), 3 where Li is the linear subspace of R , orthogonal to ∇ϕi (ξi ) and gi : [0, T ] → R3 is such that ∇ϕi (ξi ) · (gi (t) − ξi ) = ϕi (t, ξi ), e.g. ϕi (t, ξi ) ∇ϕi (ξi ). k∇ϕi (ξi )k2 See figure 16 for illustration. Putting all of the inclusions together we have for some subspace L ⊂ X, and a function ψ : [0, T ] → X gi (t) = x ∈ L + ψ(t). Thus e = Dx ∈ DL + Lψ(t). Now put U := DL and g(t) := Lψ(t) to obtain (3.15). Notice that the one can model a large variety of mechanisms using this method. 19 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 16. Moving constraints with points Ai and Aj at some moment of time. The solid curves represent actual frictionless constraints, the dashed ones are used for the linearization procedure. 3.6. Split of e and the stress formula. In the previous sections we had established the constraints on a configuration and stress. Now we will define the dependence between them. We say that the stress of each bar is given by the Hooke’s law, but it is generated not by the entire enlongation, but only by its part, which we will call elastic enlongation. For each bar Bij we replace the single enlongation value eij with two variables: elastic enlongation of a bar, denoted again by eij and plastic enlongation pij . We will use the plastic enlongation later, while defining the law of plasticity. Length of a bar Bij is now the sum of three terms: θij + pij + eij ∈ R, where θij is again a length of the bar in zero configuration. The figure 17 illustrates two components of enlongation for each bar for a system, similar to example 3.3. y1 ξ1 x1 ξ3 α13 x3 y3 e23 p23 α12 g23 (t) ξ2 x2 α23 y2 Figure 17. A three-node example of elastoplastic system. The stress sij is produced by the elastic enlongation only, according to the Hooke’s law (3.18) sij = aij · eij where aij ∈ R is a fixed characteristic of the bar Bij , such that aij > 0. 20 Ivan Gusoshnikov The response of elastoplastic springs to perturbations To describe this in terms of the configuration spaces we put e = (eik jk )m k=1 ∈ E, p = (pik jk )m k=1 ∈ S, A : E → S, ai1 j1 0 ... 0 a ... i j 2 2 A= . .. .. . . . . 0 0 ... 0 0, .. . , aim jm and we obtain (3.19) s = Ae. In other words, to describe a bar as an elastic material we have to specify its relaxed length (s.t. the bar produces no stress) and the Hooke’s coefficient. In the model they are respectively θij + pij and aij . Notice, that the constraint (3.15) now applied to the total enlongation and should be written as follows: (3.20) e(t) + p(t) ∈ U + g(t). 3.7. Plasticity law. Now we have to explain the role of the plastic component p. We assume that there is a known initial value of p, and we will derive the plasticity law as a differential d inclusion of the derivalive dt p. First, we explain the plasticity law for a single bar Bij . Assign to the bar two fixed values: c− ij − + − + and c+ , s.t. c 6 0 6 c . So we want to have s ∈ [c , c ] during all evolution and change p ij ij ij ij ij ij ij accordingly to maintain this inclusion. At any moment of time there are three different possible scenarios: + (1) if sij (t) ∈ (c− ij , cij ) then there is no need to change pij , i.e. ṗij = 0. − (2) if sij (t) = cij < 0, then this corresponds to a maximal elastic compression of the bar. Indeed, due to (3.18) and aij > 0 a stress is negative iff elastic enlongation is negative. Geometrically, this may be seen on the figure 18. In this situation we say that p may decrease with any rate, i.e. ṗij 6 0. (3) Analogously, if sij (t) = c+ ij > 0, then this corresponds to a maximal elastic stretch of the bar and ṗ > 0. A possible scenario, which leads to change of pij is an increase of a total enlngation due to a moving constraint. The elastic enlongation will increase and pij will remain constant until the stress reaches the maximal value. Then the behavior of eij and pij will switch to the opposite. Notice, that all three cases can be summarizes in the following rule: (3.21) ṗij ∈ N[c− ,c+ ] (sij ), ij ij where the normal cone is understood in sence of an inner product in R (usual multiplication of numbers). The figure 19 illustrates this. Now we need to couple plasticity laws of all bars together. For ewach k = 1..m we rewrite (3.21) as (0, . . . , 0, ṗik jk , 0, . . . , 0) ∈ {0}k−1 × N[c− ,c+ ] (sij ) × {0}m−k , ij ij where the power is in the sence of Cartesian product. By summation over all k we obtain m X ṗ = (ṗi1 j1 , . . . , ṗim jm ) ∈ {0}k−1 × N[c− ,c+ ] (sij ) × {0}m−k . ij k=1 21 ij Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 18. A negative value of sij means an elastic compression of the bar. Figure 19. Normal cones, corresponding to all three cases. Notice, that written component-wisely, it implies the conjunction of all (3.21). Further applying lemma 2.4 to each term in the sum (m − 1) times and choosing the components of s as ξ from the lemma, we get the following: m X ṗ ∈ NRk−1 ×[c− ,c+ ]×Rm−k (s), ij ij k=1 where the normal cone is understood in sence of the usual dot product in Rm . Now put m \ + + − + − + m−k ] × . . . × [c , c ] = Rk−1 × [c− ⊂ S. , c C := [c− , c ] × [c ij , cij ] × R im jm im jm i2 j 2 i2 j 2 i1 j 1 i1 j 1 k=1 The figure 20 shows an example of C. Since int C 6= ∅, we apply lemma 2.3 and get m X NRk−1 ×[c− ,c+ ]×Rm−k (s) = NC (s). ij ij k=1 So we finally have the law of plasticity in the form ṗ ∈ NC (s). Recall, that the normal cone here is understood in sence of the usual dot product in Rm . We may rewrite it, using the normal cone in sence of the work form < ·, · >. By the definition of the normal cone and (3.4) we obtain (3.22) − ṗ ∈ NC (s). 22 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 20. C and s for the three-node example. At the end we shall say, that the set C here describes the plastic properties of the material, as long as the matrix A describes its elastic properties. 3.8. Formulation of the problem. Now we will put all parts of the model together. We consider evolution of the system on the time-interval [0, T ] with some initial state e0 , p0 ∈ E. The evolution is assumed to be quazistatic, i.e. at each node at each moment of time the resultant force is zero. External load y may be given as a function y : [0, T ] → Y , satisfying (3.11), but we find its representation h : [0, T ] → S, such that for all t ∈ [0, T ] we have D∗ h(t) = y(t). Putting all relations together we obtain the system, which we have to satisfy for all t ∈ [0, T ]: e(t) + p(t) ∈ U + g(t) — moving affine constraint (3.20), ∃(rt ∈ U ⊥ )[h(t) + s(t) + rt = 0] — quasistaticity criterion (3.16), — stress formula (3.19), s(t) = Ae(t) d − dt p(t) ∈ NC (s(t)) — law of plasticity (3.22). In addition, we have the initial condition e(0) = e0 p(0) = p0 . Recall, that U, g, h, A and e0 , p0 are given here, everything else we have to consider as unknowns. 3.9. Dependence on choice of h. Generally, the matrix D∗ is not one-to-one mapping, so it may happen, that for the same physical external force y(t) there are multiple h(t). Here we consider such case and show, that the choice of a different h doesn’t play any role. The proof is split on several lemmas since each of them may be useful by itself. Lemma 3.5. In the setting of the problem (3.20)-(3.22) let h1 and h2 be two representations of the force y(t). Then for all t ∈ [0, T ] h1 − h2 ∈ U ⊥ . 23 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Proof. By premise, for all t ∈ [0, T ] D∗ h1 = D∗ h2 = y Thus D∗ (h1 − h2 ) = 0, h1 − h2 ∈ Ker D∗ = DX ⊥ . Since U ⊥ = (DX ∩ Ua )⊥ = DX ⊥ + Ua⊥ , we have proven the statement of the lemma. Lemma 3.6. For any r∗ ∈ U ⊥ let (s, p) be a solution of the problem (3.20)-(3.22) with h(t) = h1 (t) + r∗. Then the problem (3.20)-(3.22) with h(t) = h1 (t) also has the same solution (s, p). Proof. The proof is trivial. Indeed, we have (3.16) as ∃(rt ∈ U ⊥ )[(h(t) + r∗ ) + s(t) + rt = 0]. For each t ∈ [0, T ] we put rt∗ = rt + r∗ , so, clearly, rt∗ ∈ U ⊥ and h(t) + s(t) + rt∗ = 0. Since the only changed terms are rt and h, and all other inclusions in the system don’t contain them, the same (p, s) is a solution of the system with h(t) = h1 (t). Theorem 3.1. Fix an external force y : [0, T ] → Y . The solution of the problem (3.20)(3.22) does not depend on the choice of different functions h, if for all t ∈ [0, T ] they all satisfy D∗ h(t) = y(t). Proof. Combine the lemmas 3.5 and 3.6. 4. Formulation of the Abstract Elastoplastic Problem 4.1. Formulation of the model. The system at any moment of time from [0, T ] is represented by two components, both from an abstract linear space of configurations U: • x — the visible (or exposed) component, submitted to external forces and a perfect affine bilateral constraint, i.e. (4.1) x ∈ L = U + g(t), where U is a closed linear subspace of U and g : [0, T ] → U is a given function called guiding or drivng. In terms of the physical interpetation, the point x is locked in a planar device. Inside the plane it can move freely without any friction, but it cannot leave the plane. • p — the hidden (or plastic) component. The difference x − p =: e ∈ U is called the elastic deviation. Assume that we also have an abstract linear space of forces F, and the bilinear form to calculate work, which places U and F in separating duality: (4.2) < ·, · >: U × F → R. The figure 4 illustrates all involved components, forces and constraints. Overall, there are five forces, acting in the system. Forces, applied to p: 24 Ivan Gusoshnikov The response of elastoplastic springs to perturbations • s — elastic force(produced by the spring), directed towards x, given by the formula s = A(e) = A(x − p), (4.3) where A:U →F is a given linear bijection, satisfying the following propreties: (4.4) ∀(u ∈ U \ {0})[< u, Au >> 0]. (4.5) ∀(u1 , u2 ∈ U)[< u1 , Au2 >=< u2 , Au1 >]. • f — the force of plastic resistance, satisfying the condition dp (4.6) ∈ NC (−f ) := {ξ ∈ U : ∀(c ∈ C)[< ξ, c − (−f ) >6 0]}, dt where C is given nonempty closed convex subset of F. Physically, the set C describes admissible values of a force of friction. This is a formulation of the law of friction, experienced by p. It is derived from the “Principle of Maximal Dissipation”, which says that a direction of a fiction force has to be s.t. its power is minimal over all possible directions(notice, that the power is negative due to the dissipative nature of a friction force). For details the reader may look at [1, 4.e] and [2, 3.2]. Notice, that lemma 2.2 and replacement C := −C allow us to rewrite the relation in the form −ṗ ∈ NC (f ) which is derived in [1, 4.e]. Forces, applied to x: • r — reaction of the constraint L. It should satisfy the condition (4.7) − r ∈ NL (x) := {ξ ∈ F : ∀(l ∈ L)[< l − x, ξ >6 0]}. Notice, that since L = U + g(t) and U is a linear space, we have NL (x) = L⊥ = U ⊥ = {ξ ∈ F : ∀(u ∈ U)[< ξ, u >= 0]}. • c — external force, called load, defined explicitely by a given function c : [0, T ] → F. • (−s) — elastic force towards p. Using quazistatic approach, i.e. assuming that the evolution is slow enough to neglect inertia, we obtain the two following equations, which physical meaning is vanishing of the superposition of the forces: (4.8) s + f = 0, (4.9) r + c − s = 0. Definition 4.1. To summarize the above, we say, that in the Abstract Elastoplastic problem we are given with: • driving function g : [0, T ] → U and a closed linear subspace U ⊂ U; • load function c : [0, T ] → F; • nonempty closed convex set C ⊂ F; • linear bijection A : U → F, s.t. (4.5) and (4.4) holds; • initial condition for x and p. We have to find: • x, p : [0, T ] → U, s.t. for all t ∈ [0, T ] the value of x(t) satisfies the constraint (4.1), • s, f, r : [0, T ] → F, satisfying formulae (4.3),(4.7) and equations (4.8),(4.9) for all t ∈ [0, T ], and (4.6) for almost all t ∈ [0, T ]. 25 Ivan Gusoshnikov The response of elastoplastic springs to perturbations 4.2. Connection with the physical model. In the following table we will show a possible substitution, which turns the Physical Elastoplastic Model into the Abstract Elastoplastic Problem: Abstract problem Physical elastoplastic model U E F S < ·, · > < ·, · > x total enlongation e + p p p e e s −s A −A C −C c h r r U U g g Several comments must be made here. (1) The relations (3.20) and (4.1) are identical. (2) We define s in the Abstract Problem as −s from the Physical Model. This turns (3.16) into (4.9) together with (4.7). (3) The substitution of s and A into the formula (3.19) leads to cancellation of the minus sign and (4.3). Clearly −A satisfies (4.5) as a diagonal matrix and (4.4) is true for the form (3.4). (4) The plasticity law (3.22) means ∀(c1 ∈ C)[< −ṗ, c1 − s >] 6 0 in terms of the Physical Model. Applying the change of variables, in terms of the Abstract Problem we get for all c1 ∈ C < −ṗ, −c1 + s >6 0 < ṗ, c1 − s >6 0 The relation (4.8) may be understood here as a definition of f , so we have < ṗ, c1 − (−f ) >6 0, and by the definition of a normal cone ṗ ∈ NC (−f ), which is precisely (4.6). 5. Solution of the Abstract Elastoplastic Problem through the Moreau Sweeping Process. This section contains a mathematical solution of the Abstract Elastoplastic Problem. Under certain assumptions a trajectory of the elastic component can be found through the Moreau Sweeping Process with a special moving set, and the trajectory of the plastic component can be calculated as an integral of a measurable selection from a special multifunction. 26 Ivan Gusoshnikov The response of elastoplastic springs to perturbations 5.1. Transition to a Hilbert space. At first step, we will get rid of the space F in our calculations, using the bijection A : U → F as an identification. Let H be a space of the same nature, as U. We define the bilinear form (·, ·) : H × H → R (5.1) (u1 , u2 ) =< u1 , Au2 > . This bilinear form defines the inner product in H and a pre-Hilber norm p |u|H = (u, u). Indeed, we can verify all axioms: (1) ∀(u1 , u2 ∈ H) [(u1 , u2 ) =< u1 , Au2 >=< u2 , Au1 >= (u2 , u1 )] by (4.5), (2) linearity of (·, ·) follows from linearity of A and the bilinear form < ·, · >, (3) ∀(u ∈ H)[(u, u) =< u, Au >> 0], (u, u) = 0 ⇔ u = 0 by (4.4), Assumption 0. H with the norm | · |H is complete and separable. Now we redefine all values, which we have in F, as themselves, but bijectively mapped to H: (5.2) c(t) := A−1 c(t), C := A−1 C, s(t) := A−1 s(t), f (t) := A−1 f (t), r(t) := A−1 r(t). Notice, that after this replacement an outwards normal cone in H, build up on the inner product (· , ·) according to the Definition 2.2 coincides with the sets, which we used in (4.6) and (4.7). Indeed, for arbitrary closed convex nonempty sets C1 ⊂ F, C2 ⊂ U and a ∈ F, b ∈ U we have ∀(c ∈ C1 )[< b, c − a >6 0] ⇐⇒ ∀(c ∈ C1 )[< b, A(A−1 c − A−1 a) >6 0] ⇐⇒ ⇐⇒ ∀(c ∈ A−1 C1 )[(b, c − A−1 a) 6 0] ⇐⇒ b ∈ NA−1 C1 (A−1 a), ∀(c ∈ C2 )[< c − b, a >6 0] ⇐⇒ ∀(c ∈ C2 )[< c − b, AA−1 a >6 0] ⇐⇒ ⇐⇒ ∀(c ∈ C2 )[(A−1 a, c − b) 6 0] ⇐⇒ A−1 a ∈ NC2 (b). Observe, that since U is linear subspace of U and L = U + g(t), for any x ∈ L the normal cone NL (x) is linear subspace of F, ortogonal to U . From the preceding argument we can see, that V := A−1 NL (x) is the ortogonal complement to U in the sence of (· , ·) Now we will rewrite the problem through elements of the space H only and get rid of several variabels: • (4.3) becomes s = x − p • ((4.9) and (4.7)) is equivalent to s − c ∈ V So we obtain the new problem, which is equivalent to the original one: find functions x, p, s : [0, T ] → H, satisfying for almost all t ∈ [0, T ]: (5.3) x∈U +g (5.4) s∈V +c (5.5) (5.6) x=p+s p0 ∈ N (s) C 27 Ivan Gusoshnikov The response of elastoplastic springs to perturbations and initial conditions x(0) = x0 , s(0) = s0 . 5.2. Assumptions. Now we introduce additional assumptions: Assumption 1. The given functions g and c are absolutely continous. Assumption 1a. The given functions g and c are Lipschitz-continous. Assumptions 1 and 1a are two possible variants. Remark 5.1. Since V the is ortogonal complement of U , we can represent g(t) = gU (t) + gV (t), where c(t) = cU (t) + cV (t), h i ^ ∀(t ∈ [0, T ]) (gU (t), cU (t) ∈ U ) (gV (t), cV (t) ∈ V ) But U and V are linear spaces, so x ∈ U + gU + gV = U + gV , s ∈ V + cU + cV = V + cU . and we can assume without loss of generality, that for any t ∈ [0, T ] g(t) ∈ V, (5.7) c(t) ∈ U. values of given g and c lay in V and U respectively. The equivalent statement for the Physical Elastoplastic Model was presented as Theorem 3.1. Assumption 2. The initial conditions x0 , s0 are s.t. x0 ∈ U + g(0), s0 ∈ (V + c(0)) ∩ C. Assumption 3. For any t ∈ [0, T ] (V + c(t)) ∩ int C 6= ∅ Notice, that (V + c(t)) ∩ C 6= ∅ is automatically required by (5.4) and (5.6) for the problem to be solvable (which means that the external load can be compensated by stress within given C and the system can develop quazistatically). The strengthened assumption with int C is called safe load hypothesis. Assumption 4. The set C is bounded. The figure 21 provides a two-dimensional(for simplicity) example of the problem (5.3) - (5.6) under the assumptions, described above. But the reader should remember, that a nontrivial Physical Elastoplastic Model necessarily has more than two dimensions. 28 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 21. A two-dimentional example of the problem (5.3) - (5.6). 5.3. Change of variables. We can rearrange (5.5) to x − s = p and by adding c − g to both parts obtain (x − g) + (c − s) = p + c − g, then (5.3) and (5.4) imply x − g = proj (p + c − g, U ) c − s = proj (p + c − g, V ) and by (5.7) we can continue: x − g = proj (p, U ) + proj (c, U ) − proj (g, U ) = proj (p, U ) + c − 0 c − s = proj (p, V ) + proj (c, V ) + proj (g, V ) = proj (p, V ) + 0 − g So we introduce new variables: (5.8) y := s − c − g = −proj (p, V ), (5.9) z := x − c − g = proj (p, U ) hence p = z − y, and we can reformulate the problem again as 0 z − y 0 ∈ NC (y + c + g), z ∈ U, (5.10) y∈V for almost all t ∈ [0, T ]. Remark 5.2. Notice, that by (5.8) two normal cones, mentioned in (5.6) and (5.10) are the same set: NC (y + c + g) = NC (s). 29 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Theorem 5.1. Under assumptions 0, 1 (resp. 1a), 3, 4 for y0 ∈ (C − c(0) − g(0)) ∩ V, z0 ∈ U there exist unique function y : [0, T ] → H and at least one function z : [0, T ] → H, s.t. (5.10) holds, y(0) = y0 , z(0) = z0 and y, z are absolutely continous (resp. Lipschitz-continous). 5.4. Proof of the theorem 5.1. We will split the proof on several following propositions. Step 5.1.1. Under the conditions of theorem 5.1 the functions y,z are solutions of (5.10) if and only if y is a solution of the Moreau Sweeping Process −y 0 ∈ N(C−c(t)−g(t))∩V (y), (5.11) y(0) = y0 . and for this y there exists z : [0, T ] → H, satisfying (5.10). Proof. By assumption 3 int C ∩ (V + c(t)) 6= ∅, i.e. h i ^ ∃(x ∈ H) (x ∈ int C) (x ∈ V + c(t)) so (x − c(t) ∈ int C − c(t)) ^ (x − c(t) ∈ V ) and since g(t) ∈ V (x − c(t) − g(t) ∈ int C − c(t) − g(t)) ^ (x − c(t) − g(t) ∈ V − g(t) = V ) so (int C − c(t) − g(t)) ∩ V = (int (C − c(t) − g(t))) ∩ V 6= ∅. Hence by lemmas 2.3 and 2.1 we obtain NC (y + c + g) + NV (y) = NC−c−g (y) + NV (y) = N(C−c−g)∩V (y). Notice, that for absolutely continous function z the condition z ∈ U is equivalent to −z 0 ∈ U = NV (y) for almost all t ∈ [0, 1] and z(0) ∈ U . So (5.12) − y 0 = z 0 − y 0 − z 0 ∈ NC (y + c + g) + NV (y) = N(C−c−g)∩V (y). The multimapping t 7→ (C − c − g) ∩ V is absolutely continous under assumptions 1, 3 and 4 (see [1, Section 5c]) and Lipschitz-continous under assumptions 1a, 3 and 4 (see [2, Lemma 6]). Step 5.1.2. From the step 5.1.1 we obtain well-posed Moreau Sweeping Process. Existence and uniqueness of its solution are guaranteed by theorem 2.1. Recall, that, by definition (5.8), finding y means finding the stress s an projection of plastic component p on V . The figure 22 shows the terms of the Moreau Sweeping Process for the configuration from the figure 21. But, we still have to find suitable function z, such that the pair (y, z) is a solution of (5.10). Observe, that all we need is a summable function z 0 , such that (5.13) z 0 ∈ (NC (y + c + g) + y 0 ) ∩ U. The following steps show that there is at least one such function, by constructing a summable function η = z 0 − y 0 first. 30 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 22. The moving set (black), the variable (y) and the normal cone (orange) for the Moreau Sweeping Process, generated by Abstract Elastoplastic Problem. Step 5.1.3. Put ∗ e Φ(t) := {w ∈ H : δC (w) = (w, y(t) + c(t) + g(t))}. The set e ∩ (U − y 0 (t)) Φ(t) := Φ(t) is nonempty for almost all t ∈ [0, T ]. Proof. By the inclusion (5.12) for almost any t ∈ [0, T ] we may decompose − y 0 (t) = w − u, (5.14) where u ∈ U and w ∈ NC (y(t) + c(t) + g(t)), so w ∈ U − y 0 (t) and ∀(ĉ ∈ C)[(w, ĉ − y(t) − c(t) − g(t)) 6 0] ∀(ĉ ∈ C)[(w, y(t) + c(t) + g(t)) > (w, ĉ)] ∗ (w, y(t) + c(t) + g(t)) > sup{(w, ĉ) : ĉ ∈ C} = δC (w) and also by (5.12) we can put y(t) + c(t) + g(t) =: ĉ ∈ C 31 Ivan Gusoshnikov The response of elastoplastic springs to perturbations so we obtain (w, y(t) + c(t) + g(t)) = δ ∗ (t). Step 5.1.4. There exists strongly measurable selection from Φ η : [0, T ] → H, i.e. for almost all t ∈ [0, T ] η(t) ∈ Φ(t) and the function kη(·)k : [0, T ] → R is Lebesgue-measurable. Proof. ∗ (1) Since C is closed, convex and nonempty, δC is lower semicontinous(see e.g. [1, Chapter 2]). Hence (see e.g.[7, Th. 3.3.1]) ∗ ∀(a ∈ R) [{w ∈ H : δC (w) > a} is open] , i.e. ∗ ∀(a ∈ R) [{w ∈ H : δC (w) > a} ∈ B(H)] . ∗ Therefore the function δC is B(H)-measurable ([7, 4.4]), and for the function ∗ δ̃C : [0, T ] × H → R, ∗ δ̃C : (t, w) 7→ δ ∗ (w), we have h i ∗ ∗ ∀(a ∈ R) {(t, w) ∈ [0, T ] × H : δ̃C (w) > a} = [0, T ] × {w ∈ H : δC (w) > a} ∈ B([0, T ]) ⊗ B(H) , ∗ so δ̃C is B([0, T ]) ⊗ B(H)-measurable. (2) Consider the function v : [0, T ] × H → R, v : (t, w) 7→ (w, y(t) + c(t) + g(t)). As composition of absolutely continous functions and inner product is continous mapping w.r. to the product topology on [0, T ] × H, and since (a, +∞] := {b ∈ R : b > a} is open for any a ∈ R, ∀(a ∈ R) v −1 ((a, +∞]) ∈ B([0, T ] × H) . Assumption 0 and lemma 2.7 imply that B([0, T ] × H) = B([0, T ]) ⊗ B(H), so v is B([0, T ]) ⊗ B(H)-measurable too. (3) By [7, Th. 4.4.5] ∗ e = {(t, w) : δ̃C G(Φ) (t, w) = v(t, w)} ∈ B([0, T ]) ⊗ B(H) ⊂ L ([0, T ]) ⊗ B(H). (4) Since y 0 ∈ L2 ([0, T ]; H), there exists L ([0, 1])-measurable function ψ : [0, T ] → R, s.t. ∀(t ∈ [0, T ]\Z) [y 0 (t) = ψ(t)] , where Z has measure zero. Since H is separable and U is closed, h i ∃ ({ai }i∈N ⊂ U ) {ai }i∈N = U . Therefore each function t 7→ (ai − ψ(t)) is L ([0, T ])-measurable and h i ∀(t ∈ [0, T ]) {ai − ψ(t)}i∈N = U − ψ(t) , 32 Ivan Gusoshnikov The response of elastoplastic springs to perturbations so by theorem 2.2 the multifunction t 7→ U − ψ(t) is L ([0, T ])-measurable, hence by theorem 2.3 its graph G(U − ψ) ∈ L ([0, T ]) ⊗ B(H). 0 (5) Denote by Z the measure-zero set of t ∈ [0, 1], where Φ(t) = ∅. Notice, that Z × H, Z 0 × H ∈ L ([0, T ]) ⊗ B(H) by the definition of product σ-algebra. It follows from the definition of σ − algebra, that e ∩ G(U − ψ)) ∪ (Z × H) ∪ (Z 0 × H) ∈ L ([0, T ]) ⊗ B(H) F := (G(Φ) The multifunction, defined by F as its graph is different from Φ only on the measure-zero set Z ∪ Z 0 and is L ([0, T ])-measurable. Indeed, for any open U ∈ H we have ([0, T ] × U ) ∩ F ∈ L ([0, T ]) ⊗ B(H) and by theorem 2.4 proj[0,1] (([0, T ] × U ) ∩ F ) ∈ L ([0, T ]). So by theorem 2.2 there exists a strongly measurable selection η : [0, T ] → H, s.t for almost all t ∈ [0, T ] η(t) ∈ Φ(t). Finally observe, that the numerical function kη(t)k is Lebesgue-measurable since H is separable(see [10, Th. 2.8.5, th. 3.5.2, th. 3.5.3]). Step 5.1.5. Observe, that for almost all t ∈ [0, T ] η(t) + y 0 (t) ∈ U (5.15) and for all ĉ ∈ C(t) (η(t), y(t) + c(t) + g(t)) = δ ∗ (η(t)) > (η(t), ĉ) so (η(t), ĉ − y(t) − c(t) − g(t)) 6 0 i.e (η(t) + y 0 (t)) − y 0 (t) = η(t) ∈ NC (y(t) + c(t) + g(t)), which, together with (5.15) means (cf. (5.13)) η(t) + y 0 (t) ∈ (NC (y(t) + c(t) + g(t)) + y 0 (t)) ∩ U. Step 5.1.6. There exists ρ > 0 and an absolutely continous function h : [0, T ] → H, s.t. for all t ∈ [0, T ] h(t) ∈ C ∩ (V + c(t)), (5.16) and the closed ball Bρ [h(t)] ⊂ C. (5.17) Proof. For any ρ > 0 define a set Cρ := {w ∈ H : dist (w, H \ C) > ρ} ⊂ C. First, we prove, that (5.18) ∃(ρ > 0)∀(t ∈ [0, T ]) [Cρ ∩ (V + c(t)) 6= ∅] . Assume the contrary, ∀(ρ > 0)∃(t ∈ [0, T ]) [Cρ ∩ (V + c(t)) = ∅] . 33 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Then take a sequence ρi → 0 and find the corresponding {ti } ⊂ [0, T ]. By compactness of [0, T ] we can choose a convergent to some t0 ∈ [0, T ] subsequence: tij → t0 . Notice that ρij → 0. Now take an arbitrary v ∈ V and observe, that v + c(tij ) ∈ / Cρ , i.e. dist(v + c(tij ), H \ C) < ρij Since for any A ∈ H the function dist(·, A) : H → R is continous, we have dist(v + c(t0 ), H \ C) = 0. This means, that v + c(t0 ) ∈ H \ C = H \ int C, so (V + c(t0 )) ∩ int C = ∅, which contradicts the assumption 3 (“safe load hypothesis”). Hence (5.18) is correct. Cρ is closed as continous inverse image of a closed set and convex by the lemma 2.5. Moreover, the multimapping t 7→ Cρ ∩ (V + c(t)) (5.19) is absolutely continous(again, by [1, Section 5c]), convex-valued and closed-valued. Take any h0 ∈ Cρ ∩ (V + c(0)) and put the function h to be the solution of the Moreau Sweeping Process with the moving set (5.19) and initial condition h0 . By the definition of Cρ the condition (5.17) is satisfied for all t ∈ [0, T ]. Step 5.1.7. The function kη(·)k : [0, T ] → R is Lebesgue-summable. Proof. By lemma 2.6 we obtain for all t ∈ [0, T ] ∗ ρ · kη(t)k + (η(t), h(t)) 6 δC (η(t)) Since η(t) ∈ Φ(t), ∗ ρ·kη(t)k = δC (η(t))−(η(t), h(t)) = (η(t), y(t)+c(t)+g(t))−(η(t), h(t)) = (η(t), y(t)+c(t)+g(t)−h(t)) = = (η(t) + y 0 (t), y(t) + c(t) + g(t) − h(t)) − (y 0 (t), y(t) + c(t) + g(t) − h(t)) By (5.16) we have h(t) ∈ V + c(t), i.e. h(t) − c(t) ∈ V , by step 5.1.1 we have y(t) ∈ V , and recall, that g(t) ∈ V (5.7). From the other side, we have (5.15), so since U ⊥ V we obtain (η(t) + y 0 (t), y(t) + c(t) + g(t) − h(t)) = 0. Hence (5.20) ρ · kη(t)k = −(y 0 (t), y(t) + c(t) + g(t) − h(t)) 6 ky 0 (t)k · ky(t) + c(t) + g(t) − h(t)k. The function t 7→ ky(t) + c(t) + g(t) + h(t)k is continous on the compact interval [0, T ], so let M be the maximum. Recall, that, by properties of Moreau Sweeping Process, y 0 (t) ∈ L2 ([0, T ], H) ⊂ L1 ([0, T ], H), so the Lebesgue integral ZT ZT kη(t)kdt 6 0 M 0 M ky (t)kdt = ρ ρ 0 ZT 0 34 ky 0 (t)kdt < ∞. Ivan Gusoshnikov The response of elastoplastic springs to perturbations Step 5.1.8. Finally we have proven, that η is strongly measurable and kη(·)k measurable and summable. So, by [10, Th. 3.7.4] the function η is Bochner-integrable, so we put Zt z(t) := z0 + (η(s) + y 0 (s))ds. 0 Observe, that step 5.1.5 shows, that such z satisfies conditions (5.10). By [10, Th. 3.7.11] the function z is absolutely continous. In case of Lipschitz-continous input the function y is Lipschitz-continous as a solution of Moreau Sweeping Process,(see step 5.1.1), hence its derivative y 0 is bounded by some constant Ly and from (5.20) we have for almost all t ∈ [0, T ] kz 0 (t)k 6 M T Ly . ρ Since z is absolutely continous as an integral, for t1 , t2 , s.t. 0 6 t1 6 t2 6 T we have t Zt2 Z 2 M T Ly 0 0 kz(t2 ) − z(t1 )k = · (t2 − t1 ), z (t)dt 6 kz (t)kdt 6 ρ t1 t1 so z is also Lipschitz-continous. This concludes the proof of the theorem 5.1. 6. Example 1 In this section we will develop and analyze an example of Physical Elastoplastic Model. The example is simple enough, so a physical intuition is applicable, but the example is still nontrivial in sence that it contains all essential components of the general model. Besides giving an illustration for the general model, the development of this example was motivated by the following questions: Question 6.1. Having a pair of input functions (c, g) of an Abstract Elastoplastic Problem and a response of the system (x, p), can we change both of the response components by alternating g only? Can we do it by alternating c only? Question 6.2. Does Question 6.1 make sence if the Abstract Elastoplastic Problem corresponds to some Physical Elastoplastic Model? In particular, is it possible for the plastic component of such system to change in response to external forces only, i.e. when g ≡ 0? It is quite easy to see, that we can, indeed, influence both components by changing g. But, since magnitudes of external forces are limited due to the quazistatic approach (moreover, by the safe load hypothesis), the question about c is less trivial. And, regarding a physical model, it may be hard to figure out immidiately, how elastoplastic bars may change ther plastic length, while external forces remain inside the elastic domain. 6.1. Example setup. Consider the mechanism, presented in example 3.4. Recall, that it consists from three aligned elastoplastic bars between the nodes at ξ1 = (−1, 0, 0), ξ2 = (0, 0, 0), ξ3 = (1, 0, 0) at the zero configuration. Also there is a controlled rigid piston between nodes A1 and A3 (see figure 14). Intersection of the geometric constraint and the moving constraint, induced by the piston gives us (see figure 15) the moving affine constraint of the type (3.20): e(t) + p(t) ∈ U + g(t), 35 Ivan Gusoshnikov The response of elastoplastic springs to perturbations where (6.1) U = span {(1, −1, 0)}, g(t) = g13 (t) g13 (t) , , g13 (t) . 2 2 in the space of enlongations E = R3 . We put the elastic boundaries for each bar as − + + − + c− 12 = c23 = −2, c12 = c23 = 2, c13 = −1, c13 = 1 and, for simplicity, their stiffness as a12 = a23 = a13 = 1. So, in terms of the Physical Elastoplastic Model, we have C = [−2, 2] × [−2, 2] × [−1, 1], A = I. Applying the change of variables from the table in the section 4.2 and the transition to a Hilbert space (5.2) we get the system of the form (5.3)-(5.6): x ∈ U + g, (5.3) s ∈ V + c, (5.4) x = p + s, (5.5) 0 p ∈ NC (s). (5.6) in terms of the space H = R3 with the usual dot product. Indeed, by substituting A = −I and the form (3.4) into (5.1) we obtain the usual dot product in a three-dimensional space, so Assumption 0 is clearly satisfied. The sign for s and C is changed to the opposite twicely. Notice, that Assumption 4 (boundness of C) is clearly satisfied here. The external load is now represented as (6.2) c(t) = −h(t). The terms U and g of the affine constraint for the space H are the same as (6.1), and V = U ⊥ = (1, −1, 0)⊥ = {s ∈ H : s12 = s23 }. Recall, that by the remark 5.1 we can take c(t) ∈ U, g(t) ∈ V without loss of generality. In our case the vector g(t) from (6.1) is already taken orthogonal to U . Inclusion c(t) ∈ U means that we can take a parameter function cpar : [0, T ] → R and put (6.3) c(t) = (cpar (t), −cpar (t), 0). The figure 23 illustrates the inclusions (5.3),(5.4) and (5.6). Observe from the figure 23, that Assumption 3 (the safe load hypothesis) means, that for all t ∈ [0, T ] the vector c(t) belongs to the interior of C, so the smaller rectangle (V + c(t)) ∩ C doesn’t degenerate to a segment. This means, that (6.4) ∀(t ∈ [0, T ]) [|cpar (t)| < 2] . 6.2. Physiscal meaning of the problem. Now we have the example problem formulated in terms of the enlongations space H. Before we will solve it, it seems useful to go back to the physical representation to check again the physical meaning of some elements of the problem. 36 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 23. The exposed component x belongs to the subspace U , shifted on the vector g (e.g. the black dot). The stress s has to be in the small rectangle, which is section of the set C by the subspace V , shifted on the green vector c. Derivative p0 must be a vector from a normal cone of the green frame box C, taken at s. 6.2.1. Applied external forces. Since we consider the external load from the subspace U only, i.e. it is in the form (6.3), where |cpar | < 2, we find the actual forces y ∈ Y , applied to the nodes A1 , A2 , A3 . To do this we substitute (6.3) to (6.2) and then submit h to the operator D∗ from (3.14): cpar (t) y(t) = −D∗ −cpar (t) , 0 y11 = −cpar , y21 = 2cpar , y31 = −cpar , y12 = y13 = y22 = y23 = y32 = y33 = 0. Observe, that these forces stretch one of the bars B12 , B23 , compress another, and the whole mechanism remains at the same place (see figure 24). Figure 24. Possible forces, applied to the nodes. 37 Ivan Gusoshnikov The response of elastoplastic springs to perturbations 6.2.2. Balance of forces. The inclusion (5.4), which right-hand part is shown on the figure 23 as the small rectangle, stands for the vanishing criterion of resultant forces at all nodes. We can verify this as follows. First, notice, that by (6.1) and (6.3) the inclusion (5.4) may be written here as (1, −1, 0) · s = (1, −1, 0) · (cpar , −cpar , 0), (6.5) s12 − s23 = 2cpar . From the other side, the balance of forces at A2 is s23 − s12 + y21 = 0, 2cpar = s12 − s23 , which coincides with (6.5). Assume, that a scalar r̂ is a reaction of the piston (which may be any number, since we assumed that the piston is perfectly rigid). This reaction is applied along α13 = (1, 0, 0) at A1 and along −α13 at A3 (similarly to stresses of elastoplastic bars). Then we can write the balance of forces at A1 and A3 as s12 + s13 + r̂ + y11 = 0, −s23 − s13 − r̂ + y13 = 0 respectively. We add the equations together to get rid of r̂ and get s12 − s23 − 2cpar = 0, which again coincides with (6.5). Remark 6.1. Notice, that when cpar = 0, i.e. the external force is zero, we have s ∈ V , which means that s12 = s23 , i.e. the bars B12 and B23 are stressed equally. 6.2.3. The moving affine constraint. Recall, that the inclusion (5.3) was obtained as an intersection of the geometric constraint and the constraint, generated by the rigid piston. Observe, that (5.3) together with (5.5) (which means, that the total enlongation is the sum of elastic and plastic ones) may be written as e + p − g = (τ, −τ, 0) where τ ∈ R is an arbitrary number. Coordinate-wisely: g e12 + p12 − 213 = τ, g13 e23 + p23 − 2 = τ, e13 + p13 − g13 = 0; e12 + p12 + e23 + p23 = g13 , e13 + p13 = g13 . Observe, that these are exactly properties of enlongations of bars in the mechanism on the figure 14. 6.3. Input functions and initial conditions. For simplicity, we put ċpar ≡ 1, cpar (0) = 0; i.e. cpar (t) = t, c(t) = (t, −t, 0). To satisfy (6.4) (i.e. Assumption 3) we put T such that 1 < T < 2. 38 Ivan Gusoshnikov The response of elastoplastic springs to perturbations For sake of Question 6.2 we put g ≡ 0, i.e. g13 ≡ 0. Such input functions are Lipschitz-continous, so Assumption 1a is satisfied. We also put the initial condtions as follows: x0 = 0, s0 = (1, 1, 0). Observe, that x0 ∈ U + 0 = U + g(0), s0 ∈ V + 0 = V + c(0), s0 ∈ C; so Assumption 2 is also satisfied. Remark 6.2. Notice, that Remark 6.1 applies here, i.e. the bars B12 and B23 have equal positive stress. But, since the visible configuration x of the system at the moment t = 0 coincides with the zero configuration, the positive elastic enlongations have to be compensated with nonzero plastic enlongations p(0) = x − s = (−1, −1, 0). (6.6) 6.4. Solution and the Moreau Sweeping Process. For now we have specified all elements of the problem and confirmed all assumptions, so we are ready to solve the problem. Recall, that to do this we perform the change of variables (5.8)-(5.9): y := s − c − g = s(t) − (t, −t, 0), z := x − c − g = x(t) − (t, −t, 0), y0 := s0 − 0 = (1, 1, 0), z0 := x0 − 0 = 0; to obtain the problem (5.10): 0 z − y 0 ∈ NC (y + c)), z ∈ U, y ∈V; for which we find the function s from the Moreau Sweeping Process (5.11): −y 0 ∈ N(C−c)∩V (y), y(0) = y0 . and the function z from the differential inclusion (5.13) 0 z ∈ (NC (y + c) + y 0 ) ∩ U, z(0) = z0 . In particular, for t ∈ [0, 1) we have the point y resting at y0 (see figure 25). Indeed, the plane V doesn’t change in time, and y0 remains in the interior of C until t = 1, so y is not pulled by any edge of the section (C − c(t)) ∩ V . Hence for t ∈ [0, 1) we have s(t) = y(t) + c(t) = (1, 1, 0) + (t, −t, 0) ∈ int C, NC−c (y) = NC (s) = {0}. 39 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 25. Points s (left) and y together with the Moreau Sweeping Process moving set (right) at moments t = 0, 0.5, 1. Thus ṗ|[0,1) ≡ 0 by (5.6), so, by (6.6) and (5.5), we have for t ∈ [0, 1) p(t) ≡ (−1, −1, 0), x(t) = s(t) + p(t) = (t, −t, 0), z(t) = x(t) − c(t) = 0. We also get by continuity y(1) = (1, 1, 0), s(1) = (2, 0, 0), p(1) = (−1, −1, 0), x(1) = (1, −1, 0), z(1) = 0. For t ∈ [1, T ] the point y is pulled by the face of C, which is a part of the plane y12 = 2 − t. The normal cone to this face is NC (s) = NC−c (y) = [0, +∞) × {0} × {0}. The figure 26 illustrates the process at this stage. Since y ∈ V we also have y23 = y12 = 2 − t. And, since both U = NV (y) and NC−c (y) in (5.12) belong to the plane y13 = 0, we have y(t) = (2 − t, 2 − t, 0). 40 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 26. Points s (left) and y (right) at moments t = 1, 1.6 in top orthographic projection. Thus s(t) = y(t) + c(t) = (2, 2 − 2t, 0), y 0 = (−1, −1, 0). Figure 27. Unique solution of the differential inclusion (5.13). Figure 27 shows, that the differential inclusion (5.13) implies z 0 = (1, −1, 0). Indeed, if we put w := (2, 0, 0) ∈ NC (y + c), then u := z 0 = y 0 + w = (1, −1, 0) ∈ U. 41 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Observe, that this coincides with the decomposition (5.14) in the Step 5.1.3 of the Theorem 5.1 proof. This decomposition may be illustrated here by the figure 28. Figure 28. Decomposition −y 0 = w − u; w ∈ NC (y + c + g), u ∈ U . Finally, we obtain for t ∈ [1, T ] z(t) = (t − 1, 1 − t, 0), x(t) = z(t) + c(t) = (2t − 1, 1 − 2t, 0), p(t) = x(t) − s(t) = (2t − 3, −1, 0). If we put all parts of the solution together we get the following piecewise-linear responce in terms of H: Function: y =s−c−g z =x−c−g s x p t ∈ [0, 1] (1, 1, 0) (0, 0, 0) (1 + t, 1 − t, 0) (t, −t, 0) (−1, −1, 0) t ∈ (1, T] (2 − t, 2 − t, 0) (t − 1, 1 − t, 0) (2, 2 − 2t, 0) (2t − 1, 1 − 2t, 0) (2t − 3, −1, 0) Remark 6.3. This result provides us the answer for the Questions 6.1 and 6.1. First, if we put both g and c to be constant-zero, then we clearly have trivial constant responce. Second, the calculated example of the Physical Elastoplastic Model shows, that we may force both of the components s and p to change even if g ≡ 0 without braking any of the assumptions, but this requires some bars to be stressed from the beginning. Third, putting c ≡ 0, we clearly may increase the magnitude of g beyond elastic boundaries of the bars (see figure 14), forcing them to change their elastic and plastic lengths. 6.5. Decomposition of the system. Observe from the figure 14 that the behavior of the bar B13 is independent from the behavior of the couple of bars B12 , B23 . The system can be considered as two independent systems as on the figure 29. This may also be seen from figure 25 or derived analytically as follows. We utilize the description from section 6.1, without specifying the functions cpar and g13 and exact values of elastic 42 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 29. Decomposition of the system from the example 1 boundaries. Put + − + 2 C1 := [c− 12 , c12 ] × [c23 , c23 ] ⊂ R , + C2 := [c− 13 , c13 ] ⊂ R, + − + − + C = [c− 12 , c12 ] × [c23 , c23 ] × [c13 , c13 ] = (C1 × R) ∩ (R × R × C2 ), so, similarly to the derivation of the plasticity law in section 3.7, by lemmas 2.3 and 2.4 NC−c(t)−g(t) (y) = NC1 −(c12 (t),c23 (t))−(g12 (t),g23 (t)) ((y12 , y23 ))×{0}+{0}×{0}×NC2 −c13 (t)−g13 (t) (y13 ) Put U1 := span {(1, −1)} ⊂ R2 , V1 := (1, −1)⊥ ⊂ R2 and notice, that by (6.1) we have U = U1 × {0}, so N(C−c(t)−g(t))∩V (y) = NC−c(t)−g(t) (y) + U = = (NC1 −(c12 (t),c23 (t))−(g12 (t),g23 (t)) ((y12 , y23 )) + U1 ) × {0} + {0} × {0} × NC2 −c13 (t)−g13 (t) (y13 ) = = N(C1 −(c12 (t),c23 (t))−(g12 (t),g23 (t)))∩V1 ((y12 , y23 )) × {0} + {0} × {0} × NC2 −c13 (t)−g13 (t) (y13 ). Notice, that the last equality relies on the safe load hypothesis, which is independent of the third component for this system due to (6.3) and (6.4). So the Moreau Sweeping Process (5.11) is decomposed on two independent parts: ( −(y12 , y23 )0 ∈ N(C1 −(c12 (t),c23 (t))−(g12 (t),g23 (t)))∩V1 ((y12 , y23 )), (6.7) 0 −y13 ∈ NC2 −c13 (t)−g13 (t) (y13 ). In its turn, the inclusion (5.13) is z 0 ∈ (NC−c(t)−g(t) (y) + y 0 ) ∩ U = 0 0 0 = (NC1 −(c12 (t),c23 (t))−(g12 (t),g23 (t)) ((y12 , y23 )) × {0} + {0} × {0} × NC2 −c13 (t)−g13 (t) (y13 ) + (y12 , y23 , y13 )) ∩ (U1 × {0}), which is equivalent to ( 0 0 (z12 , z23 )0 ∈ (NC1 −(c12 (t),c23 (t))−(g12 (t),g23 (t)) ((y12 , y23 )) + (y12 , y23 )) ∩ U1 , 0 z13 = 0. Observe, that the decomposition (6.7) consists from the two sweeping processes: the one is by a (possibly changing) interval on a plane, another one is by a fixed interval on a line. Next, we will describe another example, which has its sweeping process in R3 and cannot be decomposed further. 43 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 30. Example 2 7. Example 2 7.1. Model description. We consider the one-dimensional mechanical model, shown on the figure 30. It has 3 elastoplastic bars an 2 rigid controlled pistons. Similarly to section 6.5, the bars B13 and B24 , which could be placed along with the pistons, do not contribute to the behavior of the others bars. Their geometric constraints, composed with the additional constraints of the pistons will be enforced below directly as an additional constraints on the bars B12 , B23 , B34 . The lengths of the pistons are given as functions l13 (t), l24 (t). For simplicity we assume again, that all bars have stiffness equal to 1. Let the reference length of each bar be 1. Thus α = (1, 0, 0) and (omitting unused dimensions) we have x2 − x1 −1 Dx = x3 − x2 = 0 x4 − x3 0 1 −1 0 0 1 −1 x 0 1 x2 0 x3 . 1 x4 Clearly the matrix D has rank 3, so DX = R3 and there is no geometric constraint. Moreover, we can solve Dx = 0 x2 − x1 = 0, x3 − x2 = 0, x4 − x3 = 0; x1 = x2 = x3 = x4 , so (7.1) Ker D = span((1, 1, 1, 1)). Now let x be the vector of total enlongations: x := e + p = (x12 , x23 , x34 ) The additional constraints, produced by the pistons are ( x12 + x23 = l13 , x23 + x34 = l24 ( (1, 1, 0) · x = l13 , (0, 1, 1) · x = l24 44 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Thus x ∈ U + g, where U ⊥ span((1, 1, 0), (0, 1, 1)). Since (1,1,0) and (0,1,1) are linearly independent, U is 1-dimensional. Notice, that (1 − 1, 1) · (1, 1, 0) = (1 − 1, 1) · (0, 1, 1) = 0, so U = span((1, −1, 1)) . To find g, observe, that x − g ∈ U means ( (1, 1, 0) · (x − g) = 0, (0, 1, 1) · (x − g) = 0. Recall, that we also want g ⊥ U , so we have the following system: (1, 1, 0) · (x − (x − g)) = l13 , (0, 1, 1) · (x − (x − g)) = l24 , (1, −1, 1) · g = 0; (1, 1, 0) · g = l13 , (0, 1, 1) · g = l24 , (1, −1, 1) · g = 0; g12 + g23 = l13 , g23 + g34 = l24 , g12 − g23 + g34 = 0; g12 + g23 = l13 , g23 + g34 = l24 , 3g23 = l13 + l24 ; 1 g23 = 3 (l13 + l24 ), g12 = l13 − 31 (l13 + l24 ), g34 = l24 − 31 (l13 + l24 ); −1 2 1 1 g = (2l13 − l24 , l13 + l24 , −l13 + 2l24 ) = 1 l13 + 1 l24 . 3 3 2 −1 (7.2) Performing step of transition to Abstract Elastoplasic Problem, we get (meaning the notation of the Pysical Model to the left) A = −I, c = h, C = −C, and after transition to a Hilbert space we get −c = h, C = C and external forces, applied to the nodes are DT c = −DT (−c) = D∗ h. This means, that, as long as the safe load hypothesis is satisfied, we can apply forces from Im(DT ) = (Ker D)⊥ = (1, 1, 1, 1)⊥ by (7.1). This only means, that the sum of all external forces has to be zero, which is expectable requirement. But, as mentioned before, in order to produce different behaviors it makes sence to consider external forces from DT U only, i.e. we take them from the one-dimesional subspace. So for a parameter function cpar : [0, T ] → R we have external forces of the type −1 0 0 −cpar 1 −1 0 cpar −cpar = 2cpar . DT · (1, −1, 1) · cpar = 0 −2cpar 1 −1 cpar 0 0 1 cpar The figure 31 illustrates such forces for positive cpar . 45 Ivan Gusoshnikov The response of elastoplastic springs to perturbations Figure 31. Forces in example 2 The corresponding Moreau Sweeping Process (5.11) is ( −y 0 ∈ N(C−c−g)∩(1,−1,1)⊥ (y), y(0) = y0 . As shown on the figure 32, by taking different elastic boundaries of the bars we can get various shapes of the moving set, from a triangle to a hexagon. Figure 32. The moving sets under zero external force applied (h = 0) for C (the green boxes) equal (from left to right): [−1, 2.5] × [−2.5, 1] × [−1, 2.5], [−1, 2.5] × [−1, 1] × [−1, 2.5], [−1, 2.5] × [−1, 1] × [−1, 1], [−1, 1] × [−1, 1] × [−1, 1]. The vectors, denoted here by l13 and l24 , are directions from (7.2) for the corresponding coefficients. 46 Ivan Gusoshnikov The response of elastoplastic springs to perturbations References [1] Moreau J.-J. On unilateral constraints, friction and plasticity. New variational techniques in mathematical physics (Centro Internaz. Mat. Estivo (C.I.M.E.), II Ciclo, Bressanone, 1973), pp. 171-322. Edizioni Cremonese, Rome, 1974. [2] Kunze M., Monteiro Marques, M. D. P. An introduction to Moreau’s sweeping process. (English summary) Impacts in mechanical systems (Grenoble, 1999), 1-60, Lecture notes in phys., 551, Springer, Berlin, 2000. [3] Bourbaki N. Topological vector spaces: Chapters 1-5. Springer, 2003. [4] Castang C., Valadier M. Convex analysis and measurable mulifunctions. Lecture notes in mathematics, 580, Springer-Verlag Berlin, Heidelberg, 1977. [5] Deimling K. Multivalued differential equations. de Gruyter, Berlin, New York, 1992. [6] Bogachev V.I. Measure theory. Volume II. Springer-Verlag Berlin Heidelberg, 2007. [7] Lojasiewicz S. An introduction to the theory of real functions. John Wiley & sons, 1988. [8] Athreya Krishna B., Lahiri Soumendra N. Measure theory and probability theory. Springer, 2006. [9] Mikusinski J. The Bochner integral. Lehrbucher und Monographien aus dem Gebiete der exakten Wissenschaften, Mathematische Reihe, Band 55. Birkhauser Verlag, Basel-Stuttgart, 1978. [10] E. Hille, R.S. Philips, Functional Analysis and Semigroups, American Mathematical Society, Providence, R. I. 1957. [11] Komura Y. Nonlinear semi-groups in Hilbert spaces, J. Math. Soc. Japan, Vol. 19, No. 4, pp. 493-507, 1967. 47
© Copyright 2026 Paperzz