Bourbaki

§ 1. Comparison of functions on a filtered set

Let E be a set filtered by a filter with base F\mathfrak{F} (Gen. Top., I, p. 58); in this chapter we shall consider functions whose domain of definition is a subset of E belonging to the filter base F\mathfrak{F} (the subset depending on the function) and which take their values either in the field of real numbers R\mathbf{R}, or, more generally, in a normed vector space over a valued field (Gen. Top., IX, p. 170).

In applications, E will most often be a subspace of the real space Rn\mathbf{R}^n, or the extended line R‾\overline{\mathbf{R}}, and F\mathfrak{F} will be the trace on E of the filter of neighbourhoods of a closure point of E, or else the filter of complements of relatively compact subsets of E ("neighbourhoods of the point at infinity").

It will not in general suffice to know that such a function tends to a given limit along F\mathfrak{F} to be able to treat all the problems of "passage to the limit along F\mathfrak{F}" for expressions formed from this function.

For example, when the real variable x tends to +∞+\infty the three functions xx, x2x^2 and x\sqrt{x} all tend to +∞+\infty, but, of the expressions

(x+1)2−x2,(x+1)−x,x+1−x, (x+1)^2 - x^2, \quad (x+1) - x, \quad \sqrt{x+1} - \sqrt{x},

the first tends to +∞+\infty, the second to 1, the third to 0.

It is thus important to know not just the limit value of a function along F\mathfrak{F} (when this limit exists) but also the "manner" in which the function approaches its limit; in other words, one is led to classify the set of functions which tend to the same limit.

1. Comparison relations: I. Weak relations

In what follows we denote by V a normed vector space over a valued field K, and by H(F,V)\mathscr{H}(\mathfrak{F}, V) the set of functions with values in V, each of which is defined on a subset of E belonging to the filter base F\mathfrak{F}. The relations which we shall define between such functions have a local character relative to the filter with base F\mathfrak{F}: let us clarify what we mean by this. If f and g are two functions from H(F,V)\mathscr{H}(\mathfrak{F}, V), recall that the relation "there is a set Z∈FZ \in \mathfrak{F} such that f and g are defined and equal on Z" is an equivalence relation R∞R_\infty on H(F,V)\mathscr{H}(\mathfrak{F}, V) (Gen. Top.*, I, p. 66). This being so, we shall say that a relation S involving a function ff of H(F,V)\mathscr{H}(\mathfrak{F}, V) is of local character (along F\mathfrak{F}) relative to ff, if it is compatible (in ff) with the equivalence relation R∞R_\infty (Set Theory, II, p. 117); we know that if f~\tilde{f} is the germ of ff along F\mathfrak{F}, the equivalence class of ff modulo R∞R_\infty (an element of the quotient set H∞(F,V)=H(F,V)/R∞\mathscr{H}_\infty(\mathfrak{F}, V) = \mathscr{H}(\mathfrak{F}, V)/R_\infty), then one can derive from S, by passage to the quotient, a relation between f~\tilde{f} and the other arguments of S, and that, conversely, every relation of this nature defines a relation of local character relative to ff.

Example

If ff and gg are two functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}), the relation "there is an X∈FX \in \mathfrak{F} such that ff and gg are defined on XX and f(t)≤g(t)f(t) \leq g(t) for every t∈Xt \in X" is of local character relative to ff and gg. We denote by f~≤g~\tilde{f} \leq \tilde{g} the relation obtained by passing to the quotient (for ff and gg); we remark that if f~≤g~\tilde{f} \leq \tilde{g} then there exist a function f1∈f~f_1 \in \tilde{f} and a function g1∈g~g_1 \in \tilde{g}, defined on all of E, such that f1(t)≤g1(t)f_1(t) \leq g_1(t) for all t∈Et \in E.

Remarks.* 1) Let ViV_i (1≤i≤n1 \leq i \leq n) be nn normed vector spaces over K, and φ\varphi a function defined on V1×V2×⋯×VnV_1 \times V_2 \times \cdots \times V_n, with values in V; by passing to the quotient along R∞R_\infty the function φ\varphi defines a map from

H∞(F,V1)×⋯×H∞(F,Vn) \mathscr{H}_\infty(\mathfrak{F}, V_1) \times \cdots \times \mathscr{H}_\infty(\mathfrak{F}, V_n)
into H∞(F,V)\mathscr{H}_\infty(\mathfrak{F}, V), which one most often denotes by φ(f~1,…,f~n)\varphi(\tilde{f}_1, \ldots, \tilde{f}_n) (Gen. Top., I, p. 67). For example, taking φ\varphi to be the maps (x,y)↦x+y(x, y) \mapsto x + y and x↦xλx \mapsto x\lambda (λ∈K\lambda \in K), one thus defines, for any two germs f~,g~\tilde{f}, \tilde{g} in H∞(F,V)\mathscr{H}_\infty(\mathfrak{F}, V), the elements f~+g~\tilde{f} + \tilde{g} and f~λ\tilde{f}\lambda, and one verifies immediately that the rules of composition (f~,g~)↦f~+g~(\tilde{f}, \tilde{g}) \mapsto \tilde{f} + \tilde{g} and (λ,f~)↦f~λ(\lambda, \tilde{f}) \mapsto \tilde{f}\lambda define on H∞(F,V)\mathscr{H}_\infty(\mathfrak{F}, V) a vector space structure over the field K; in this space O~\tilde{O} is the class formed by functions equal to 0 on a set in F\mathfrak{F}, and −f~-\tilde{f} is the class formed by functions equal to −f-f on a set in F\mathfrak{F}. In the same way, if V is an algebra over K, one defines on H∞(F,V)\mathscr{H}_\infty(\mathfrak{F}, V) a second internal rule of composition (f~,g~)↦f~g~(\tilde{f}, \tilde{g}) \mapsto \tilde{f}\tilde{g} by taking φ(x,y)=xy\varphi(x, y) = xy; together with the two preceding rules this defines an algebra structure over K on H∞(F,V)\mathscr{H}_\infty(\mathfrak{F}, V); if V has a unit element e then H∞(F,V)\mathscr{H}_\infty(\mathfrak{F}, V) has as a unit element the class e~\tilde{e} formed by the functions equal to e on some set in F\mathfrak{F}; for f~\tilde{f} to be invertible in H∞(F,V)\mathscr{H}_\infty(\mathfrak{F}, V) it is necessary and sufficient that for some f∈f~f \in \tilde{f} there exists a Z∈FZ \in \mathfrak{F} such that f(t)f(t) is invertible in V for every t∈Zt \in Z (in which case this condition is satisfied by every function in the class f~\tilde{f}).

2) With the same notation, let ψ\psi be a map of a subset of ∏i=1nVi\prod_{i=1}^n V_i into V; we denote by ψ(f1,f2,…,fn)\psi(f_1, f_2, \ldots, f_n) the function equal to ψ(f1(t),f2(t),…,fn(t))\psi(f_1(t), f_2(t), \ldots, f_n(t)) at every point t∈Et \in E where the fi(t)f_i(t) are defined and where the point (fi(t))(f_i(t)) belongs to the set where ψ\psi is defined 1^1. For example, f+gf + g is the function equal to f(t)+g(t)f(t) + g(t) at every point t∈Et \in E where ff and gg are both defined. Observe that the map (f,g)↦f+g(f, g) \mapsto f + g is not a group law on H(F,V)\mathscr{H}(\mathfrak{F}, V), since if ff is not defined on all of E there is no function g∈H(F,V)g \in \mathscr{H}(\mathfrak{F}, V) such that f+g=0f + g = 0.

Definition 1

Given two real functions* f,gf, g belonging to H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}), which are ≥0\geq 0 on a set in F\mathfrak{F}, we say that ff is dominated by gg, or that gg dominates ff (along F\mathfrak{F}*), and write* f⪯gf \preceq g or g⪰fg \succeq f, if there are a set X∈FX \in \mathfrak{F} and a number k>0k > 0 such that f(t)≤k g(t)f(t) \leq k\ g(t) for every t∈Xt \in X (in other words, if there exists a k>0k > 0 such that f~≤k g~\tilde{f} \leq k\ \tilde{g})

1^1 In particular, in all that follows, given a function ff in H(F,V)\mathscr{H}(\mathfrak{F}, V) we shall denote by ∥f∥\|f\| the function t↦∥f(t)∥t \mapsto \|f(t)\| which belongs to H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) and is defined on the same set as ff: we expressly draw attention to the fact that in this chapter ∥f∥\|f\| is a function and not a number.

Given two normed vector spaces V1,V2V_1, V_2, and two functions f1,f2f_1, f_2 belonging to H(F,V1)\mathscr{H}(\mathfrak{F}, V_1) and H(F,V2)\mathscr{H}(\mathfrak{F}, V_2) respectively, one says that f1f_1 is dominated by f2f_2 (along F\mathfrak{F}), and writes f1≼f2f_1 \preccurlyeq f_2 or f2⪰f1f_2 \succeq f_1, if ∥f1∥≼∥f2∥\|f_1\| \preccurlyeq \|f_2\|.

The relation f1≼f2f_1 \preccurlyeq f_2 is clearly of local character in f1f_1 and f2f_2; it is thus equivalent to the relation f~1≼f~2\tilde{f}_1 \preccurlyeq \tilde{f}_2 derived by passing to the quotient. When ff and gg are real functions one must be careful not to confuse the relations f~≼g~\tilde{f} \preccurlyeq \tilde{g} and f~≤g~\tilde{f} \leq \tilde{g}.

Note that for every scalar λ≠0\lambda \neq 0 the relation f1≼f2λf_1 \preccurlyeq f_2 \lambda is equivalent to f1≼f2f_1 \preccurlyeq f_2. If f1≼f2f_1 \preccurlyeq f_2 there exists a set X∈FX \in \mathfrak{F} such that f1(x)=0f_1(x) = 0 for every point x∈Xx \in X where f2(x)=0f_2(x) = 0.

Examples. 1) The relation f≼1f \preccurlyeq 1 means that ff is bounded on a set of F\mathfrak{F}.

2) For every function ff of H(F,V)\mathscr{H}(\mathfrak{F}, V), and every scalar λ≠0\lambda \neq 0, one has f≼fλf \preccurlyeq f \lambda. 3) When xx tends to +∞+\infty one has sin⁡2x≼sin⁡x\sin^2 x \preccurlyeq \sin x. 4) When (x,y)(x, y) tends to (0,0)(0, 0) in R2\mathbf{R}^2 one has

xy≼x2+y2. xy \preccurlyeq x^2 + y^2.

The following propositions are immediate consequences of def. 1:

Proposition 1

If* f,g,hf, g, h are three functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) then the relations f≼gf \preccurlyeq g and g≼hg \preccurlyeq h imply f≼hf \preccurlyeq h.

Proposition 2

Let* f1,f2f_1, f_2 be two functions in H(F,V)\mathscr{H}(\mathfrak{F}, V) and gg a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}). The relations f1≼gf_1 \preccurlyeq g and f2≼gf_2 \preccurlyeq g imply f1+f2≼gf_1 + f_2 \preccurlyeq g.

Further:

Proposition 3

Let* V1,V2,VV_1, V_2, V be three normed spaces over the same valued field, and (x,y)↦[x.y](x, y) \mapsto [x.y] a bilinear map from V1×V2V_1 \times V_2 into VV. If f1f_1 and f2f_2 are functions in H(F,V1)\mathscr{H}(\mathfrak{F}, V_1) and H(F,V2)\mathscr{H}(\mathfrak{F}, V_2) respectively, and g1,g2g_1, g_2 are two functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) such that f1≼g1f_1 \preccurlyeq g_1 and f2≼g2f_2 \preccurlyeq g_2 then [f1.f2]≼g1g2[f_1.f_2] \preccurlyeq g_1 g_2.

Indeed (Gen. Top., IX, p. 173, th. 1) there exists a number a>0a > 0 such that

∣[f1.f2]∣≤a∥f1∥∥f2∥. |[f_1.f_2]| \leq a \|f_1\| \|f_2\|.

Corollary

If* VV is a normed algebra, if f1,f2f_1, f_2 are two functions in H(F,V)\mathscr{H}(\mathfrak{F}, V), and g1,g2g_1, g_2 are two functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}), then the relations f1≼g1,f2≼g2f_1 \preccurlyeq g_1, f_2 \preccurlyeq g_2 imply f1f2≼g1g2f_1 f_2 \preccurlyeq g_1 g_2.

The relation f≼gf \preccurlyeq g between functions in H(F,V)\mathscr{H}(\mathfrak{F}, V) is transitive by prop. 1; since it is reflexive the relation "*f \preccurlyeq g and g \preccurlyeq f*" is an equivalence relation on H(F,V)\mathscr{H}(\mathfrak{F}, V) (Set Theory, II, p. 113).

Definition 2

Given two functions* f,gf, g of H(F,V)\mathscr{H}(\mathfrak{F}, V) we say that ff and gg are similar (along F\mathfrak{F}), and write f≃gf \simeq g, if f≼gf \preccurlyeq g and g≼fg \preccurlyeq f.

For every scalar λ≠0\lambda \neq 0 the relation f≃gf \simeq g is equivalent to f≃gλf \simeq g \lambda. It implies the existence of a set X∈FX \in \mathfrak{F} such that that the subset of XX of points where f(x)=0f(x) = 0 is identical with the subset of XX of points where g(x)=0g(x) = 0.

Example 1

For a real function f∈H(F,R)f \in \mathscr{H}(\mathfrak{F}, \mathbf{R}) the relation f≃1f \simeq 1 means that there are two numbers a>0, b>0a > 0,\ b > 0 such that a≤∣f(x)∣≤ba \leq |f(x)| \leq b on a set in F\mathfrak{F}, or that the function log⁡∣f∣\log |f| is bounded on a set in F\mathfrak{F}: one then says that ff is logarithmically bounded on a set in F\mathfrak{F}.

Example 2

Let VV be a normed space over a non-discrete valued field KK, and let f(x)=a0xn+a1xn−1+⋯+anf(x) = a_0 x^n + a_1 x^{n-1} + \cdots + a_n be a polynomial in the variable x∈Kx \in K, with coefficients in VV, such that a0≠0a_0 \neq 0. For every vector b≠0b \neq 0 one has f(x)≃bxnf(x) \simeq b x^n as ∣x∣|x| tends to +∞+\infty.

Example 3

We have seen that sin⁡2x≼sin⁡x\sin^2 x \preccurlyeq \sin x as xx tends to +∞+\infty, but we do not have sin⁡2x≃sin⁡x\sin^2 x \simeq \sin x, even though these functions vanish at the same points.

Example 4

One has x2+xy+y2≃x2+y2x^2 + xy + y^2 \simeq x^2 + y^2 when (x, y)(x,\ y) tends to (0, 0)(0,\ 0) in R2\mathbf{R}^2, but not xy≃x2+y2xy \simeq x^2 + y^2.

It follows immediately from prop. 3 of V, p. 213, that if f1, f2, g1, g2f_1,\ f_2,\ g_1,\ g_2 are functions in H(F,K)\mathscr{H}(\mathfrak{F}, K) (KK being any valued field) the relations f1≃g1f_1 \simeq g_1 and f2≃g2f_2 \simeq g_2 imply that f1f2≃g1g2f_1 f_2 \simeq g_1 g_2.

We remark that in contrast the relations f1≃g1f_1 \simeq g_1 and f2≃g2f_2 \simeq g_2 do not imply that f1+f2≃g1+g2f_1 + f_2 \simeq g_1 + g_2, as is shown by the example f1(x)=g1(x)=x2, f2(x)=−(x2+x), g2(x)=−(x2−1)f_1(x) = g_1(x) = x^2,\ f_2(x) = -(x^2 + x),\ g_2(x) = -(x^2 - 1), as the real variable xx tends to +∞+\infty.

The comparison relations f≼g, f≃gf \preccurlyeq g,\ f \simeq g are said to be weak. We say that two functions f, gf,\ g from H(F,V)\mathscr{H}(\mathfrak{F}, V) are weakly comparable if they satisfy one (at least) of the relations f≼g, g≼ff \preccurlyeq g,\ g \preccurlyeq f.

Remark 1

Two functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) need not be weakly comparable, as is shown by the example of the functions 1 and xsin⁡xx \sin x as xx tends to +∞+\infty.

Remark 2

Denote by R0R_0 the relation f≃gf \simeq g on H(F,V)\mathscr{H}(\mathfrak{F}, V), and by H0(F,V)\mathscr{H}_0(\mathfrak{F}, V) the quotient set H(F,V)/R0\mathscr{H}(\mathfrak{F}, V)/R_0; note that the relation RR implies R0R_0. Passing to the quotient the relation f≼gf \preccurlyeq g gives, by prop. 1 of V, p. 213, an order relation on H0(F,V)\mathscr{H}_0(\mathfrak{F}, V) (Set Theory, III, p. 134); the preceding example shows that H0(F,V)\mathscr{H}_0(\mathfrak{F}, V) is not totally ordered by this relation.

2. Comparison relations: II. Strong relations

Definition 3

Given two real functions* f, gf,\ g belonging to H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}), which are ≥0\geq 0 on a set in F\mathfrak{F}, one says that ff is negligible relative to gg, or that gg is preponderant over ff (along F\mathfrak{F}), and one writes f≪gf \ll g or g≫fg \gg f if, for every ε>0\varepsilon > 0, there exists a set X∈FX \in \mathfrak{F} such that f(t)≤εg(t)f(t) \leq \varepsilon g(t) for every t∈Xt \in X.

Given two normed spaces V1, V2V_1,\ V_2 and two functions f1, f2f_1,\ f_2 belonging to H(F,V1)\mathscr{H}(\mathfrak{F}, V_1) and H(F,V2)\mathscr{H}(\mathfrak{F}, V_2) respectively, one says that f1f_1 is negligible relative to f2f_2 (along F\mathfrak{F}), and one writes f1≪f2f_1 \ll f_2 or f2≫f1f_2 \gg f_1, if ∥f1∥≪∥f2∥\|f_1\| \ll \|f_2\|.

For every scalar λ≠0\lambda \neq 0 the relation f1≪f2λf_1 \ll f_2 \lambda is equivalent to f1≪f2f_1 \ll f_2. The relation f1≪f2f_1 \ll f_2 implies f1≼f2f_1 \preccurlyeq f_2 but is not equivalent to it.

Note that the relation f1≼f2f_1 \preccurlyeq f_2 by no means implies the relation "f1≪f2f_1 \ll f_2 or f1≍f2f_1 \asymp f_2": one has sin⁡x≼1\sin x \preccurlyeq 1 as xx tends to +∞+\infty, but neither of the relations sin⁡x≪1\sin x \ll 1 or sin⁡x≍1\sin x \asymp 1 is true.

Examples. 1) The relation f≪1f \ll 1 means that ff tends to 0 along F\mathfrak{F}.

2) When α\alpha and β\beta are two real numbers such that α<β\alpha < \beta one has xα≪xβx^\alpha \ll x^\beta as xx tends to +∞+\infty. Similarly, when mm and nn are two rational integers such that m<nm < n one has zm≪znz^m \ll z^n as the complex number zz tends to ∞\infty.

3) As xx tends to +∞+\infty one has xn≪exx^n \ll e^x for every integer nn (III, p. 105).

4) In R2\mathbf{R}^2 one has, as (x,y)(x, y) tends to (0,0)(0, 0),

x2+y2≪∣x∣+∣y∣. x^2 + y^2 \ll |x| + |y|.

The following propositions can be deduced immediately from def. 3:

Proposition 4

If f,g,hf, g, h are three functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) then the relations f≼gf \preccurlyeq g and g≪hg \ll h (resp. f≪gf \ll g and g≼hg \preccurlyeq h) imply f≪hf \ll h.

Proposition 5

Let f1,f2f_1, f_2 be two functions in H(F,V)\mathscr{H}(\mathfrak{F}, V) and gg a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}). Then the relations f1≪gf_1 \ll g and f2≪gf_2 \ll g imply f1+f2≪gf_1 + f_2 \ll g.

Moreover, the same argument as in prop. 3 of V, p. 213 shows that:

Proposition 6

With the notation of prop. 3, the relations f1≪g1f_1 \ll g_1 and f2≼g2f_2 \preccurlyeq g_2 (resp. f1≼g1f_1 \preccurlyeq g_1 and f2≪g2f_2 \ll g_2) imply [f1.f2]≪g1g2[f_1.f_2] \ll g_1g_2.

Prop. 4 shows that the relation f≪gf \ll g between functions in H(F,V)\mathscr{H}(\mathfrak{F}, V) is transitive; but it is not reflexive: to be precise, the relation f≪ff \ll f implies that ff vanishes on a set in F\mathfrak{F} (in other words, ff is equivalent to 0 modulo R∞R_\infty); indeed, for an ε\varepsilon such that 0<ε<10 < \varepsilon < 1 there exists an X∈FX \in \mathfrak{F} such that ∥f(x)∥≤ε∥f(x)∥\|f(x)\| \leq \varepsilon \|f(x)\| for every x∈Xx \in X, which is possible only if f(x)=0f(x) = 0 for every x∈Xx \in X. It follows that the relation "f \ll g and g \ll f" is transitive and symmetric, but not reflexive: so it is not an equivalence relation (it implies that there is a set X∈FX \in \mathfrak{F} such that f(x)=g(x)=0f(x) = g(x) = 0 for every x∈Xx \in X).

Proposition 7

If ff and gg are two functions in H(F,V)\mathscr{H}(\mathfrak{F}, V) then the relation f−g≪ff - g \ll f is equivalent to f−g≪gf - g \ll g.

Indeed, f−g≪ff - g \ll f implies that for every ε>0\varepsilon > 0 there exists an X∈FX \in \mathfrak{F} such that ∥f(x)−g(x)∥≤ε∥f(x)∥\|f(x) - g(x)\| \leq \varepsilon \|f(x)\| for every x∈Xx \in X. But then (1−ε)∥f(x)∥≤∥g(x)∥(1 - \varepsilon) \|f(x)\| \leq \|g(x)\|, and consequently f≼gf \preccurlyeq g, whence (V, p. 215, prop. 4) f−g≪gf - g \ll g.

Corollary

The relation f−g≪ff - g \ll f is an equivalence relation on H(F,V)\mathscr{H}(\mathfrak{F}, V).

Indeed, if f−g≪ff - g \ll f and g−h≪gg - h \ll g then f−g≪gf - g \ll g, whence (V, p. 215, prop. 5) f−h≪gf - h \ll g, and consequently, since g⪯fg \preceq f, we have f−h≪ff - h \ll f, which shows that this relation is transitive; it is symmetric by prop. 7, and is clearly reflexive, whence the corollary.

Definition 4

Given two functions* f,gf, g of H(F,V)\mathscr{H}(\mathfrak{F}, V) one says that ff and gg are equivalent (along F\mathfrak{F}), and writes f∼gf \sim g, if f−g≪ff - g \ll f.

The relation f∼gf \sim g implies f≍gf \asymp g but is not equivalent to it.

Example 1

If aa is a constant function, ≠0\neq 0 on EE, the relation f∼af \sim a means that ff tends to aa along F\mathfrak{F}.

Example 2

Let VV be a normed space over a non-discrete valued field KK, and let f(x)=a0xn+a1xn−1+⋯+anf(x) = a_0 x^n + a_1 x^{n-1} + \cdots + a_n be a polynomial in the variable x∈Kx \in K, with coefficients in VV, such that a0≠0a_0 \neq 0. Then f(x)∼a0xnf(x) \sim a_0 x^n as ∣x∣|x| tends to +∞+\infty.

Example 3

When the real number xx tends to +∞+\infty one has (1+1x)sin⁡x∼sin⁡x\left( 1 + \frac{1}{x} \right) \sin x \sim \sin x.

Example 4

As the complex variable zz tends to 0 one has ez−1∼ze^z - 1 \sim z. More generally, if VV is a normed space over a valued field KK, and ff is a function defined on a neighbourhood of x0∈Kx_0 \in K, with values in VV, and admitting a derivative f′(x0)≠0f'(x_0) \neq 0 at the point x0x_0, then, as xx approaches x0x_0, we have f(x)−f(x0)∼f′(x0)(x−x0)f(x) - f(x_0) \sim f'(x_0)(x - x_0) (I, p. 3, def. 1).

Example 5

As (x,y)(x, y) approaches (0,0)(0, 0) in R2\mathbf{R}^2 one has

sin⁡2x+sin⁡2y∼x2+y2. \sqrt{\sin^2 x + \sin^2 y} \sim \sqrt{x^2 + y^2}.

Example 6

Let f(x,y)f(x, y) be a polynomial with real coefficients in two real variables x,yx, y, not having a constant term. If, as xx approaches 0 while remaining >0> 0, there is a function φ(x)\varphi(x), continuous on an interval [0,a][0, a], and such that φ(0)=0\varphi(0) = 0 and f(x,φ(x))=0f(x, \varphi(x)) = 0 for 0≤x≤a0 \leq x \leq a, one can show that there are a rational number rr and a real number λ≠0\lambda \neq 0 such that φ(x)∼λxr\varphi(x) \sim \lambda x^r (V, p. 259, exerc. 3).

Example 7

For every x>0x > 0 let π(x)\pi(x) denote the number of prime numbers which are ≤x\leq x; it has been proved that π(x)∼x/log⁡x\pi(x) \sim x / \log x 2^2 as xx tends to +∞+\infty.

Remark

Note that the relation f∼gf \sim g by no means implies that the difference f−gf - g tends to 0 along F\mathfrak{F}; this difference can even be unbounded, as is shown by the example x2+x∼x2x^2 + x \sim x^2 as xx tends to +∞+\infty.

Proposition 8

Let* KK be a valued field and f1,f2,g1,g2f_1, f_2, g_1, g_2 four functions in H(F,K)\mathscr{H}(\mathfrak{F}, K); then the relations f1∼g1f_1 \sim g_1 and f2∼g2f_2 \sim g_2 imply f1f2∼g1g2f_1 f_2 \sim g_1 g_2.

Indeed, we have f1f2−g1g2=f1(f2−g2)+(f1−g1)g2f_1 f_2 - g_1 g_2 = f_1 (f_2 - g_2) + (f_1 - g_1) g_2; since f1⪯g1f_1 \preceq g_1, f1−g1≪g1f_1 - g_1 \ll g_1 and f2−g2≪g2f_2 - g_2 \ll g_2, we have f1f2−g1g2≪g1g2f_1 f_2 - g_1 g_2 \ll g_1 g_2 (V, p. 215, prop. 5 and 6).

In contrast, we have given an example in V, p. 214 where one has f1=g1f_1 = g_1, f2∼g2f_2 \sim g_2 and yet the relation f1+f2≍g1+g2f_1 + f_2 \asymp g_1 + g_2 does not hold (so neither a fortiori does

f1+f2∼g1+g2). f_1 + f_2 \sim g_1 + g_2).
2 See, for example, A. E. INGHAM, The distribution of prime numbers (Cambridge Tracts, n 30), Cambridge University Press, 1932.

The comparison relations f≪g,f∼gf \ll g, f \sim g are called strong relations. Two functions f,gf, g from H(F,V)\mathscr{H}(\mathfrak{F}, V) are called comparable (or strongly comparable when one wants to avoid any possible confusion) if they satisfy one of the three relations: f≪g,f≫gf \ll g, f \gg g, or "there exists a λ≠0\lambda \neq 0 such that f∼gλf \sim g\lambda".

Remark 1

Two functions can be weakly comparable though not strongly comparable, for example the functions 1 and sin⁡x\sin x as xx tends to +∞+\infty.

Remark 2

In the definitions of the comparison relations f1≼f2f_1 \preccurlyeq f_2 and f1≪f2f_1 \ll f_2 the norms on the spaces V1,V2V_1, V_2 where f1f_1 and f2f_2 respectively take their values, are involved only apparently; in reality only the topologies on V1V_1 and V2V_2 are involved, for the relations f1≼f2f_1 \preccurlyeq f_2 and f1≪f2f_1 \ll f_2 are replaced by equivalent relations when one replaces the norm on V1V_1 or V2V_2 by an equivalent norm (\emph{Gen. Top.}, IX, p. 170, def. 7).

3. Change of variable

Let φ\varphi be a map from the set E′E' into EE such that φ−1(F)\varphi^{-1}(\mathfrak{F}) is a filter base on E′E'. It is clear that if f1,f2f_1, f_2 are functions in H(F,V1)\mathscr{H}(\mathfrak{F}, V_1) and H(F,V2)\mathscr{H}(\mathfrak{F}, V_2) respectively, then f1∘φ,f2∘φf_1 \circ \varphi, f_2 \circ \varphi belong to H(φ−1(F),V1)\mathscr{H}(\varphi^{-1}(\mathfrak{F}), V_1) and H(φ−1(F),V2)\mathscr{H}(\varphi^{-1}(\mathfrak{F}), V_2) respectively, and that the relation f1≼f2f_1 \preccurlyeq f_2 (resp. f1≪f2f_1 \ll f_2) is equivalent to f1∘φ≼f2∘φf_1 \circ \varphi \preccurlyeq f_2 \circ \varphi (resp. f1∘φ≪f2∘φf_1 \circ \varphi \ll f_2 \circ \varphi).

4. Comparison relations between strictly positive functions

Let gg be a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) which is strictly positive on a set in F\mathfrak{F}. Comparison relations featuring gg can be formulated in another way: the relation f≼gf \preccurlyeq g is equivalent to saying that ∥f∥/g\|f\|/g (which is defined on a set in F\mathfrak{F}) is bounded on a set in F\mathfrak{F}; the relation f≪gf \ll g is equivalent to saying that ∥f∥/g\|f\|/g tends to 0 along F\mathfrak{F}. If ff is a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) the relation f≍gf \asymp g means that f/gf/g is logarithmically bounded on a set in F\mathfrak{F}, and the relation f∼gf \sim g that f/gf/g tends to 1 along F\mathfrak{F}. If ff is a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) and positive on a set in F\mathfrak{F}, to say that ff and gg are comparable implies that f/gf/g tends to a limit (finite or equal to +∞+\infty) along F\mathfrak{F}.

Proposition 9

Let ff and gg be two functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}), strictly positive on a set in F\mathfrak{F}. For ff and gg to be comparable it is necessary and sufficient that for every number t⩾0t \geqslant 0, with at most one exception, the function f−tgf - tg has constant sign 3^3 on a set in F\mathfrak{F}.

The condition is necessary. Indeed, if f≪gf \ll g one has f−tg∼−tgf - tg \sim -tg except for t=0t = 0, so f−tgf - tg is strictly negative on a set in F\mathfrak{F}, except for t=0t = 0; if f≫gf \gg g then f−tgf - tg is strictly positive on a set in F\mathfrak{F} for all tt; finally, if f∼kgf \sim kg (kk constant >0> 0), then f−tg∼(k−t)gf - t g \sim (k - t)g except for t=kt = k, so, except perhaps for t=kt = k, f−tgf - t g has the sign of k−tk - t on a set in F\mathfrak{F}.

The condition is sufficient. Indeed, suppose that the ratio f/gf/g has two distinct cluster points α<β\alpha < \beta along F\mathfrak{F}. For every number tt such that α<t<β\alpha < t < \beta there then exist in every set X∈FX \in \mathfrak{F} two points x1,x2x_1, x_2 such that f(x1)/g(x1)<tf(x_1)/g(x_1) < t and f(x2)/g(x2)>tf(x_2)/g(x_2) > t; thus f(x)−tg(x)f(x) - t g(x) does not have constant sign on XX; we have arrived at a conclusion incompatible with the hypothesis. It follows that f/gf/g can have only one cluster value (finite or infinite) along the filter with base F\mathfrak{F}, and consequently, (Gen. Top., I, p. 85, corollary) has this value as its limit along F\mathfrak{F}.

Proposition 10

Let ff and gg be two functions in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) which are strictly positive on a set in F\mathfrak{F}; for every α\alpha real and ≠0\neq 0 the relation f≍gf \asymp g (resp. f∼gf \sim g) is equivalent to fα≍gαf^\alpha \asymp g^\alpha (resp. fα∼gαf^\alpha \sim g^\alpha); if α>0\alpha > 0 the relation f≼gf \preccurlyeq g (resp. f≪gf \ll g) is equivalent to fα≼gαf^\alpha \preccurlyeq g^\alpha (resp. fα≪gαf^\alpha \ll g^\alpha); if α<0\alpha < 0 it is equivalent to fα≽gαf^\alpha \succcurlyeq g^\alpha (resp. fα≻gαf^\alpha \succ g^\alpha).

The proofs are immediate.

One notes that in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) the set I′I' of functions strictly positive on a set in F\mathfrak{F} is such that Γ/R∞\Gamma/\mathbf{R}_\infty is a multiplicative group Γ∞\Gamma_\infty in H∞(F,R)\mathscr{H}_\infty(\mathfrak{F}, \mathbf{R}); Γ/R0\Gamma/\mathbf{R}_0 is identical to the quotient group of Γ∞\Gamma_\infty by the subgroup of classes (mod. R∞\mathbf{R}_\infty) of logarithmically bounded functions in Γ\Gamma; on Γ/R0\Gamma/\mathbf{R}_0 the order relation deduced from the relation f≼gf \preccurlyeq g by passage to the quotient is compatible with the group structure of Γ/R0\Gamma/\mathbf{R}_0 and thus makes it an ordered group.

Proposition 11

Let gg be a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) such that lim⁡Fg=+∞\lim_{\mathfrak{F}} g = +\infty; the relation f≪gf \ll g implies ef≪ege^f \ll e^g; the relation f∼gf \sim g implies log⁡f∼log⁡g\log f \sim \log g.

Indeed, if f≪gf \ll g then f−g=g(fg−1)f - g = g \left( \frac{f}{g} - 1 \right) tends to −∞-\infty along F\mathfrak{F}. Similarly, if f∼gf \sim g one has log⁡f=log⁡g+log⁡fg\log f = \log g + \log \frac{f}{g}, so log⁡f−log⁡g\log f - \log g tends to 0, and the same for log⁡flog⁡g−1=log⁡f−log⁡glog⁡g\frac{\log f}{\log g} - 1 = \frac{\log f - \log g}{\log g}.

On the other hand, note that the relation f∼gf \sim g does not imply ef∼ege^f \sim e^g, nor even that ef≍ege^f \asymp e^g, as is shown by the example where f(x)=x2, g(x)=x3+xf(x) = x^2, \ g(x) = x^3 + x as xx tends to +∞+\infty; similarly, the relation f≪gf \ll g does not imply log⁡f≪log⁡g\log f \ll \log g, as is shown by the example f(x)=x, g(x)=x2f(x) = x, \ g(x) = x^2 as xx tends to +∞+\infty.

Definition 5

Let gg be a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}), strictly positive on a set in F\mathfrak{F}, and such that lim⁡Fg=0\lim_{\mathfrak{F}} g = 0 or lim⁡Fg=+∞\lim_{\mathfrak{F}} g = +\infty. One says that a function f∈H(F,R)f \in \mathscr{H}(\mathfrak{F}, \mathbf{R}) is of order ρ\rho (finite or infinite) relative to gg if lim⁡Flog⁡(∣f∣)/log⁡g=ρ\lim_{\mathfrak{F}} \log(|f|)/\log g = \rho.

Note that if ff is of order ρ\rho relative to gg then ff is of order −ρ-\rho relative to 1/g1/g; one therefore need treat only the case where g(x)g(x) tends to +∞+\infty along F\mathfrak{F}.

Proposition 12

Let g be a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}) such that lim⁡Fg=+∞\lim_{\mathfrak{F}} g = +\infty; let f be a function in H(F,R)\mathscr{H}(\mathfrak{F}, \mathbf{R}).

a) For f to be of order +∞+\infty relative to g it is necessary and sufficient that f≫gαf \gg g^\alpha for every α≥0\alpha \geq 0.

b) For f to be of order −∞-\infty relative to g it is necessary and sufficient that f≪g−αf \ll g^{-\alpha} for every α>0\alpha > 0.

c) For f to be of finite order equal to ρ\rho relative to g it is necessary and sufficient that, for every ε>0\varepsilon > 0, one has gρ−ε≪f≪gρ+εg^{\rho-\varepsilon} \ll f \ll g^{\rho+\varepsilon}.

Let us prove c) for example. If the order of f relative to g is ρ\rho then for every ε>0\varepsilon > 0 there exists a set M∈FM \in \mathfrak{F} such that, for every x∈Mx \in M, we have

(ρ−ε2)log⁡g(x)≤log⁡∣f(x)∣≤(ρ+ε2)log⁡g(x), \left( \rho - \frac{\varepsilon}{2} \right) \log g(x) \leq \log |f(x)| \leq \left( \rho + \frac{\varepsilon}{2} \right) \log g(x),

or (g(x))ρ−ε2≤∣f(x)∣≤(g(x))ρ+ε2(g(x))^{\rho - \frac{\varepsilon}{2}} \leq |f(x)| \leq (g(x))^{\rho + \frac{\varepsilon}{2}}; since lim⁡Fg=+∞\lim_{\mathfrak{F}} g = +\infty one thus has gρ−ε≪f≪gρ+εg^{\rho-\varepsilon} \ll f \ll g^{\rho+\varepsilon} for every ε>0\varepsilon > 0; the converse is immediate. The proofs of a) and b) are similar.

Note that if f is of finite order ρ\rho relative to g then fg−ρfg^{-\rho} is of order 0 relative to g, and conversely; if f1f_1 (resp. f2f_2) is of order ρ1\rho_1 (resp. ρ2\rho_2) relative to g, and if ρ1+ρ2\rho_1 + \rho_2 is defined, then f1f2f_1 f_2 is of order ρ1+ρ2\rho_1 + \rho_2 relative to g.

Remark 1

Observe that though f may be of finite order ρ\rho relative to g nevertheless the ratio f/gρf/g^\rho need not tend to a limit; for example, every logarithmically bounded function is of order 0 relative to g, but need not have a limit along F\mathfrak{F}.

Remark 2

A function defined on a set in F\mathfrak{F} need not have a determinate order (finite or not) relative to g, for the functions having a determinate order relative to g are comparable to all the powers of g with at most one exception. Now f need not have this property, as one sees from the example g(x)=x, f(x)=1+x2sin⁡2xg(x) = x,\ f(x) = 1 + x^2 \sin^2 x (as x tends to +∞+\infty). In this example f is comparable to gαg^\alpha for α<0\alpha < 0 and α>2\alpha > 2; if one takes f(x)=exsin⁡2x+e−xcos⁡2xf(x) = e^x \sin^2 x + e^{-x} \cos^2 x then f is not comparable to any power (positive or negative) of g.

5. Notation

Given a real function f∈H(F,R)f \in \mathscr{H}(\mathfrak{F}, \mathbf{R}) it is often convenient in a formula to write O(f)O(f) for a function dominated by f, and o(f)o(f) for a negligible function relative to f. When, in a proof, there feature several functions dominated by the same function f (resp. negligible relative to f) we denote them by O1(f), O2(f)O_1(f),\ O_2(f), etc. (resp. o1(f), o2(f)o_1(f),\ o_2(f), etc.).

Many authors write O(f)O(f) (resp. o(f)o(f)) indiscriminately for all functions in a proof that are dominated by f (resp. negligible relative to f), an abuse of language which may risk confusion.

With this notation we can express props. 1, 2, 3 (V, p. 213) as follows: if g=O1(f)g = O_1(f) and h=O(g)h = O(g) then h=O2(f)h = O_2(f); one can write

∑i=1nλiOi(f)=On+1(f)(λi scalars)(1) \sum_{i=1}^{n} \lambda_i O_i(f) = O_{n+1}(f) \quad (\lambda_i \text{ scalars}) \tag{1}
O(f) O(g)=O(fg).(2) O(f) \, O(g) = O(fg). \tag{2}

Similarly, prop. 4 (V, p. 215) shows that if g=O1(f)g = O_1(f) and h=o(g)h = o(g) (resp. g=o1(f)g = o_1(f) and h=O(g)h = O(g)) then h=o2(f)h = o_2(f), and props. 5 and 6 (V, p. 5) can be expressed in the form

∑i=1nλioi(f)=on+1(f)(λi scalars)(3) \sum_{i=1}^{n} \lambda_i o_i(f) = o_{n+1}(f) \quad (\lambda_i \text{ scalars}) \tag{3}
o(f) O(g)=o(fg).(4) o(f) \, O(g) = o(fg). \tag{4}

The relation f∼gf \sim g is equivalent to f=g+o(g)f = g + o(g). The notation O(1)O(1) (resp. o(1)o(1)) denotes a function bounded on a set in F\mathfrak{F} (resp. a function tending to 0 along F\mathfrak{F}).

Exercises

See the exercises for § 1.

Transcribed from the printed volume. Not translated.