Classification of multistationarity for mass action networks with one-dimensional stoichiometric subspace

We may assume that i_0=1. By separating the first row in \Gamma we write in block matrix form

so \Gamma r(y)=0 if and only if {\bar\Gamma}{\bar r}({\bar y})=0.

For the nondegeneracy part we use (?). We can write Dr(x)=D_{r(x)}R^tD_{1/x}, where R denotes the reactant matrix having the source complexes of \cal N as columns, and D_\eta denotes the diagonal matrix with entries of vector \eta on the diagonal. If v=(v_1,\ldots, v_n)\in\text{im }\Gamma is a nonzero vector, then v_1=0 and \bar v=(v_2,\ldots, v_n)\in \text{im }{\bar\Gamma}, \bar v\neq 0. Conversely, if {\bar v}\in\text{im }{\bar\Gamma} is nonzero, then v=(0,{\bar v})\in \text{im }{\Gamma} is nonzero. Using block matrix forms we have

so \Gamma Dr(x)|_{x=y}v=0 if and only if {\bar\Gamma}D{{\bar r}({\bar x})}_{{\bar x}= {\bar y}}{\bar v}=0, therefore y is nondegenerate if and only if {\bar y} is.

We have f(\eta,\kappa;t)v=F(\eta+tv) and f'(\eta,\kappa;t)v=DF|_{\eta+tv} v. Then f'(\eta,\kappa;t_0)=0 if and only if v\in\ker(DF|_{\eta+t_0v}).

If the polynomial f(\eta,\kappa;t) has an infinite number of roots it must be identically zero.

f'(\eta,\kappa;t_0)=p'(t_0)q(t_0)+p(t_0)q'(t_0)=p(t_0)q'(t_0), so f'(\eta,\kappa;t_0)=0 if and only if q'(t_0)=0.

1. We have

and the conclusion follows from proposition 2.6 and lemma 3.3.

2. Similar to 1.

If \cal N has the capacity for MPE (MPNE) then {\cal N}_I does by corollary 2.7 and Proposition 3.4(1). If {\cal N}_I has the capacity for MPE (MPNE) we let \eta'+t_1v_I, \eta'+t_2v_I be distinct positive equilibria (nondegenerate equilibria) of {\cal N}_I and for each j\in J choose \eta_j large enough so that \eta_j+t_1v_j>0, \eta_j+t_2v_j>0. Then \eta+t_1v, \eta+t_2v distinct are positive equilibria (nondegenerate equilibria) of ({\cal N},\kappa) by corollary 2.7 and Proposition 3.4(2).

We have v_iv_j(a_i-b_i)(a_j-b_j)<0. Assume that there exists s\notin\{i,j\} such that a_s\neq b_s. Then v_iv_s(a_i-b_i)(a_s-b_s)<0 or v_jv_s(a_j-b_j)(a_s-b_s)<0 and therefore {\cal N}_{\{i,s\}}\in {\cal Z} or {\cal N}_{\{j,s\}}\in {\cal Z}, which is a contradiction.

The statement is equivalent if we replace \cal N by its essential projection, so we may assume that \cal N is an essential network. Clearly, if {\cal N}\in{\cal S}_2^{nz} then \cal N does not have 1-alt^c subnetworks of three reactions. Conversely, suppose \{a\to a', b\to b', c\to c', d\to d'\} is a subnetwork of \cal N that is in 1-alt^c , that a\to a', c\to c' form the (\gets,\to) pattern on {\cal N}_{\{i\}} with c_i<a_i and b\to b', d\to d' form the (\to,\gets) pattern on {\cal N}_{\{j\}} with d_j<b_j. If i=j it is easy to check that either \{a\to a', b\to b', c\to c', d\to d'\}\in2-alt (a contradiction: 1-alt^c networks have 1-alt^c subnetworks composed of three reactions) or that {\cal N}_{\{i\}} has the arrow diagram (\leftarrow\!\!\bullet\!\!\rightarrow,\leftarrow\!\!\bullet\!\!\rightarrow). Suppose i\neq j, and suppose v_i=v_j=1 (the case v_i=-v_j=1 is similar). If a_i>b_i then \{a\to a', b\to b', d\to d'\}\in1-alt^c , contradiction. If a_i<b_i then {\cal N}_{\{i\}}\in2-alt , contradiction. Therefore a_i=b_i. Likewise, c_i=d_i.

Therefore all 1D projections of \{a\to a', b\to b', c\to c', d\to d'\} have the arrow diagram (\leftarrow\!\!\bullet\!\!\rightarrow,\leftarrow\!\!\bullet\!\!\rightarrow). \cal N cannot more source complexes on any 1D projection, as that would mean {\cal N}\in2-alt .

Suppose for two reactions e\to e', f\to f' we have e_1=f_1. We show that e=f. Suppose e_k\neq f_k for some k\in S. If e\to e' and f\to f' have opposite directions, then they form one of the (\to,\gets) and (\gets, \to) patterns in {\cal N}_{\{k\}}. We can pick a reaction g\to g' with g_1\neq f_1 such that \{e\to e', f\to f', g\to g'\}\in1-alt^c , contradiction, so e_k=f_k. Since e\to e' and f\to f' were arbitrary, it follows that all source complexes that have the same projection on \{1\} are equal. Therefore {\cal N} contains exactly two source complexes a=b, c=d and {\cal N}=\{a\to a', b\to b', c\to c', d\to d'\}. If for some i,j\in S v_iv_j(c_i-a_i)(c_j-a_j)<0 then \{a\to a', c\to c'\} is a zigzag in {\cal N}_{\{i,j\}}, contradiction. Therefore {\cal N}\in{\cal S}_2^{nz}.

If {\cal N}\in \cal D then each 1D projection of \cal N has two source complexes, contradicting {\cal N}\in2-alt . Therefore \cal N has the capacity for MPNE by Theorem 4.1.

Assume without loss of generality that {\cal N}=\{a\rightleftharpoons b, c\to c'\} is an essential network. Suppose {\cal N} has the capacity for MPNE. Then \cal N must have rank 1 [JoshiShiu17], discussed above. By Theorem 4.1, \cal N must be 1-alt complete, which implies that c\notin\{a,b\}. It is clear that {\cal N}\notin {\cal L}\cup{\cal S}_2^z (for these classes reaction vectors cannot be parallel to the line of source complexes) and that {\cal N}\notin {\cal S}_2^{nz} (that requires four reactions). Moreover, {\cal N}\notin\cal C, since then it would have arrow diagram in Figure ?(i),(ii), contradiction. The other implication follows from Corollary 4.3.

We have

and the conclusion follows from Proposition 3.1.

(i) Let p=a_1-b_1=b_2-a_2<0, q=c_2-b_2>0 and let \kappa_1, \kappa_2, \kappa_3 denote the rate constants of a\to a', b\to b', c\to c'. Then

Let

Note that \eta+t_1v, \eta+t_2v are positive equilibria of ({\cal N},\kappa) if and only if g(t_1)=g(t_2)>0 (this will be the value of \kappa_2), and moreover they are nondegenerate if and only of g'(t_1), g'(t_2) are nonzero (lemma 3.3).

We have

Set \eta=(2,4) and \kappa_1=q, \kappa_3=-p2^{-p-2q}, so that g'(0)=0. We calculate

Since g(0)>0 Lemma 5.3 applies, and we may pick t_1\neq t_2 close to 0 such that g(t_1)=g(t_2)>0 and g'(t_1), g'(t_2) are both nonzero. We set \kappa_2=g(t_1) and lemma 3.3 implies that \eta+t_1v, \eta+t_2v are compatible nondegenerate equilibria of ({\cal N},\kappa).

(ii) Let p=a_1-b_1=b_2-a_2<0, q=c_2-b_2<0. The formula for f and g are the same as in (i). Setting \eta=(4,2) \kappa_1=-q, \kappa_3=-p2^{p-2q} we obtain g(0)=0 and g''(0)=(-p-q-2)2^{p-4}. Therefore if p+q\neq -2 Lemma 5.3 applies as above and \cal N has the capacity for MPNE. If p+q=-2 then p=-1 and q=-1, i.e. {\cal N}\in{\cal C} with corner c, Figure ?(i).

(iii) Let p=a_1-b_1>0 and q=c_2-b_2<0. We have

and let g(t)=\kappa_1 (\eta_1+t)^p+\kappa_3 (\eta_2+t)^q. Set \eta=(1,1) and k_1=-q, k_3=p. Then g(0)>0, g'(0)=0 and g''(0)=pq(q-p)<0. Therefore Lemma \ref{lem:ginj} applies as above and {\cal N} has the capacity for MPNE.

(iv) Let p=a_1-b_1<0 and q=c_2-b_2<0; We have

and let g(t)=\kappa_1 (\eta_1+t)^p+\kappa_3 (\eta_2-t)^q. Set \eta=(1,1) and k_1=-q, k_3=-p. Then g(0)>0, g'(0)=0 and g''(0)=-pq(q+p-2)>0. Therefore Lemma \ref{lem:ginj} applies as above and {\cal N} has the capacity for MPNE.

(v) Let p=a_1-b_1<0 and q=c_2-b_2<0. We have p=a_1-b_1>0 and q=c_2-b_2>0; f and g are as in (iv). Set \eta=(1,1) and k_1=q, k_3=p. Then g(0)>0, g'(0)=0 and g''(0)=-pq(q+p-2). If p+q-2\neq 0 then {\cal N} has the capacity for MPNE by Lemma \ref{lem:ginj}. If p+q=2 then p=1 and q=1 and \{a\to a', b\to b', c\to c'\}\in{\cal C} with corner b, see Figure ?(ii).

If \cal N has a zigzag of slope different from -1, then \cal N has the capacity for MPNE by Theorem ?, ?. Suppose {\cal N}=\{a\to a',b\to b', c\to c'\} and that \{a\to a',b\to b'\} form a zigzag of slope -1 like in Figure \ref{fig:s2z}(iii). Suppose c\to c' has the same direction as a\to a' (we may permute coordinates and change the directions of all arrows to arrive at this case). Set v=(1,1). If c_1>b_1 or c_2<b_2 then {\cal N} is 2-alternating, and therefore has the capacity for MPNE by corollary 4.3. We assume c_1\le b_1 and c_2\ge b_2. If c_1<b_1 and c_2>b_2 then \{b\to b', c\to c'\} form a zigzag of slope different from -1, and \cal N has the capacity for MPNE by Theorem ?, ?. The remaining cases have the arrow diagrams in Figure ?(i),(ii) and the conclusion follows from lemma 5.4.

Let {\cal N}=\{a\to a', b\to b', c\to c'\} with directions of reactions given by \lambda_1=\lambda_3=1 for a\to a', c\to c' and \lambda_2=-1 for b\to b'.

If v=(1,1) then up to changing directions of all reactions the 1D projections are as follows:

If c_1<b_1 or a_2>b_2 then {\cal N} is 2-alternating and has the capacity for MPNE. If c_1>b_1 then \{b\to b', c\to c'\} is a zigzag, contradiction; so c_1=b_1. Likewise, a_2=b_2. Let p=a_1-b_1>0 and q=c_2-b_2<0. The aroow diagram of \cal N is that of Figure ?(iii). By lemma 5.4 \cal N has the capacity for MPNE.

If v=(1,-1) then up to changing directions of all reactions the 1D projections are either

or

First consider case (i) in the diagram. If c_1>b_1 or a_2>b_2 then {\cal N} is 2-alternating and has the capacity for MPNE. If c_1<b_1 then \{b\to b', c\to c'\} is a zigzag, contradiction; so c_1=b_1. Likewise, a_2=b_2. The arrow diagram of \cal N is that of Figure ?(iv), and \cal N has the capacity for MPNE by lemma 5.4.

For case (ii) it is argued similarly that c_1=b_1 and a_2=b_2. This corresponds to the arrow diagram in Figure ?(v) and by lemma 5.4 either \cal N has the capacity for MPNE of {\cal N}\in{\cal C}.

Suppose \cal N does not have the capacity for MPNE.

Let {a\to a',b\to b'} form a zigzag (necessarily of slope -1) like the one in Figure \ref{fig:s2z}(iii). Let c\to c' be a different reaction of \cal N. By Lemma 5.5 either a,b,c are collinear or {a\to a',b\to b', c\to c'}\in{\cal C}, in other words c=(a_1,b_2). Suppose we have both types, i.e. c\to c' and d\to d' are such that {a\to a',b\to b', c\to c'}\in{\cal C} and a,b,d are collinear. Let \kappa_1,\kappa_2,\kappa_3,\kappa_4 denote the rate constants of a\to a', b\to b', c\to c', d\to d'. We have d_1+d_2=b_1+b_2, so d_2-c_2=1-(d_1-c_1)=1-p. We have

where g(t)=-\kappa_1(\eta_2+t)+\kappa_2(\eta_1+t) -\kappa_4\lambda_4(\eta_1+t)^p(\eta_2+t)^{1-p} and \lambda_4=1 or \lambda_4=-1. Note that p=0 or p=1 give d=a or d=b, so we have p<0 or p>1. Suppose p>1 (the other case is similar). If \lambda_4=1, we set \eta=(2,4) and we let \kappa_1=2^{-p-1}(p-1), \kappa_2=2^{-p-1}(3p+1), \kappa_4=1 and we get g(0)=2^{-p+2}p, g'(0)=0 and g''(0)=-2^{-p-1}p(p-1)\neq 0. If \lambda_4<0 we set \eta=(4,2), \kappa_1=2^p, \kappa_2=2^{p-1}p and \kappa_4=1 and we have g(0)=2^{p+1}p, g'(0)=0 and g''(0)=2^{p-3}p(p-1)\neq 0. In both cases Lemma 5.3 applies as above and ({\cal N},\kappa) has multiple nondegenerate equilibria, contradiction.

Therefore either {\cal N}\in{\cal C} or {\cal N} has all source complexes on a line of slope -1. In the latter case if {\cal N}\notin {\cal S}_2^z\cup{\cal L} then \cal N is 2-alternating and has the capacity for MPNE, contradiction.

Suppose the subnetwork \{a\to a', b\to b', c\to c'\} of \cal N is 1-alt^c with a\to a', c\to c' having the same direction. By Lemma 5.6 \cal N has the capacity for MPNE unless the arrow diagram of \{a\to a', b\to b', c\to c'\} is that in Figure ?(ii). Assume that’s the case, and let d\to d' be a reaction of \cal N with d\notin\{a,b,c\}.

First assume d\to d' has the same direction as a\to a'. If d_1>b_1 then \{a\to a', b\to b', d\to d'\}\in1-alt^c \setminus{\cal Z} and since d\neq c by Lemma 5.6 \cal N has the capacity for MPNE. If d_1<b_1 then \{b\to b', c\to c', d\to d'\}\in1-alt^c \setminus{\cal Z} and by Lemma 5.6 \cal N has the capacity for MPNE. Likewise, \cal N has the capacity for MPNE if d_2\neq b_2. Since d\neq b we conclude that \cal N has MNPE.

Assume d\to d' has the direction of b\to b'. If d_1>c_1 or d_2>a_2 then {\cal N}\in2-alt has capacity for MPNE. Assume d_1\le c_1 and d_2\le a_2. If d_1<a_1 and d_2<a_2 then \{a\to a', d\to d'\} is a zigzag, contradiction. Likewise, if d_1<c_1 and d_2<c_2 \{c\to c', d\to d'\} is a zigzag, contradiction. We therefore have d_1\in\{a_1,c_1\}, d_2\in\{a_2, c_2\}. Since d is different from a,b,c we have d=(c_1, a_2). Then \{a\to a', d\to d', c\to c'\}\in1-alt^c \setminus{\cal C}, and by lemma 5.6 \cal N has the capacity for MPNE.

Suppose {\cal N}_{\{1,2\}}\in {\cal S}_2^{z}\cup \cal L, and assume that reactions a\to a' and b\to b' of \cal N form a zigzag in {\cal N}_{\{1,2\}}. Since \cal N does not have the capacity for MPNE it follows from corollary 4.9 that \{1,2\} is the only projection of \{a\to a', b\to b'\} with zigzag, and therefore we have a_k=b_k for all k\in S\setminus \{1,2\} by lemma 3.12. Any reaction c\to c' with c\neq a and having the same direction as b\to b' can replace b\to b' in the argument above, and therefore a_k=c_k for all k\in S\setminus \{1,2\}. In the same way c_k=b_k for all k\in S\setminus \{1,2\} when c\to c' has the same direction as a\to a' and c\neq b. Therefore all 1D projections other than on \{1\}, \{2\} have a single source complex, which contradicts the fact that \cal N is an essential network if n>3. Therefore n=2 and {\cal N}={\cal N}_{\{1,2\}}.

First we consider networks on three species, n=3. Let {\cal N}=\{a\to a',b\to b', c\to c'\} and suppose that {\cal N}_{\{1,2\}} forms one of the arrow diagram in Figure ?.

Case 1: {\cal N}_{\{1,2\}}\in{\cal Z}, Figure ?(i). If a_3\neq c_3 then \{a\to a', b\to b'\} is a zigzag on the \{1,2\} or \{1,3\} projections; this subnetwork of \cal N also forms a zigzag on the \{1,2\} projection; by corollary 4.9 \cal N has the capacity for MPNE, contradiction. Therefore a_3=b_3. If a_3=b_3=c_3 then since \{a\to a', b\to b'\} are arbitrary representatives of all reactions with the same projections on \{1,2\} it follows that {\cal N}_{\{k\}} has a single source complex. But \cal N is essential, contradiction.

We assume that c_3\neq a_3=b_3 and suppose a\to a' has positive direction in {\cal N}_{\{3\}}, i.e. a'_3>a_3 (the case of the opposite direction is treated in the same way). If c_3<a_3 then {\cal N}_{\{1,3\}} has the arrow diagram in Figure ?(i), and so {\cal N}\in1-alt^c \setminus {\cal D}, contradicting the lack of capacity for MPNE of \cal N.

Therefore a_3=b_3<c_3, and it follows that {\cal N}_{\{1,3\}} has the arrow diagram in Figure ?(ii). But {\cal N}_{\{1,3\}}\in {\cal D} so we must have {\cal N}\in {\cal C}, and c_k-a_k=1. Letting \gamma=(a_1,b_2,a_1) we have a-\gamma=(0,1,0), b-\gamma=(1,0,0) and c-\gamma =(0,1,0). Therefore {\cal N}\in {\cal C}.

Case 2: {\cal N}_{\{1,2\}}\in{\cal C}\setminus{\cal Z}, Figure ?(ii). If \cal N has a 2D projection in {\cal Z}\cap {\cal C} then Case 1 implies that {\cal N}\in {\cal C}. We show that otherwise we get contradiction. Suppose no 2D projection of {\cal N} is in (\cal C)\cup{\cal Z}. We have

and therefore

in particular (c_3-b_3)(b_3-a_3)\ge 0, i.e. either (c_3-b_3)(b_3-a_3)>0 or {\cal N}_{\{3\}}\in2-alt . The latter contradicts the lack of capacity for MPNE of \cal N.

Assume a_3=c_3 (the other case is similar). If a_3=b_3=c_3 then as above {\cal N}_{\{3\}} contains a single source complex and \cal C is not essential, contradiction. If a_3=c_3\neq b_3 then either \{b\to b', c\to c'\} is a zigzag on the \{1,3\} projection, or \{a\to a', b\to b'\} is a zigzag on the \{2,3\} projection. The projections of a, b, c on \{1,3\} are pairwise distinct: for example a_3\neq b_3, b_3\neq c_3, and c_1\neq a_1. Moreover, they are not collinear. Since {\cal N}\in{\cal D} it follows that {\cal N}_{\{1,3\}}\in {\cal C}\cap{\cal Z}, contradicting our assumption.

Now we consider n\ge 4.

Assume by permuting coordinates if necessary that {\cal N}_{\{1,2\}}\in{\cal C} has one of the arrow diagrams in Figure ?. For each k\in S, {\cal N}_{\{k\}} contains at least two source complexes (\cal N is essential). Let \gamma_k denote the minimum of all source complexes of {\cal N}_{\{k\}}. For any k\ge 3, {\cal N}_{\{1,2,k\}}\in{\cal D} is an essential 1-alt complete network on three species, and so it must be in \cal C. It follows that for any complex d of \cal N and for any k\in S d_k=\gamma_k or d_k=\gamma_k+1. We check that the latter cannot hold on more than one coordinate k. Suppose k,l are different and d_k=\gamma_k, d_l=\gamma_l. Each 1D projection of \cal N contains at least one of the (\to,\gets) or (\gets, \to) patterns, and so either {\cal N}_{\{1,k,l\}} is 1-alt complete or {\cal N}_{\{2,k,l\}} is. In either case, that 3D projection of \cal N must be in \cal C by hypothesis and therefore the projection of d-\gamma on \{1,k,l\} has only one nonzero entry, contradiction. Therefore for any source complex d of \cal N we have d-\gamma\in\{e_1,\ldots, e_n\} and {\cal N}\in{\cal C}.

Let {\cal N}_{\{1,2\}}=\{a\to a', b\to b', c\to c', d\to d'\} have one of the arrow diagrams in Figure ?. Let k\neq 1,2. If a_k\neq b_k then {\cal N}_{\{1,k\}}\in1-alt^c and it contains at least three source complexes: (a_1,a_k),(a_1,b_k),(c_1,c_k) are distinct and moreover not collinear, which implies that {\cal N}_{\{1,k\}}\in{\cal C}, and by Proposition 5.10 {\cal N}\in{\cal C}. However, \cal C all 2D projections of \cal C networks have three source complexes, contradicting {\cal N}_{\{1,2\}}\in{\cal S}_2^{nz}. Likewise, if c_k\neq d_k we get a contradiction.

It remains to consider the case a_k=b_k and c_k=d_k. If a_k=b_k=c_k=d_k then since a\to a', b\to b', c\to c', d\to d' were arbitrary representatives of reactions with the same projections on \{1,2\}, it follows that {\cal N}_{\{k\}} has a single source complex, contradiction. We must have n=2 and {\cal N}={\cal N}_{\{1,2\}}\in{\cal S}_2^{nz}. If on the other hand a_k=b_k\neq c_k=d_k for all k then {\cal N} has two source complexes a=b and c=d and either v_1v_k(c_1-a_1)(c_k-a_k) has the same sign for all k\neq 1 (i.e. {\cal N}\in {\cal S}_2^{nz}) or for some k\neq 1,2 the projection {\cal N}_{\{1,k\}} is a zigzag. In the latter case, since {\cal N}_{\{1,k\}}\in\cal D we must have {\cal N}={\cal N}_{\{1,k\}}\in{\cal S}_2^z, contradiction.