Bourbaki

1. Terms and relations

1. Signs and assemblies

The signs of a mathematical theory T\mathscr{T} [^1] are the following :

(1) The logical signs [^2] : □\square, τ\tau, ∨\vee, ¬\neg. (2) The letters.

By letters we mean upper and lower case Roman letters, with or without accents. Thus AA, A′A', A′′A'', A′′′A''', ... are letters. At any place in the text it is possible to introduce letters other than those which have appeared in previous arguments.

(3) The specific signs, which depend on the theory under consideration.

In the theory of sets we shall use only the following three specific signs : ==, ∈\in, ⊃\supset.

An assembly in T\mathscr{T} is a succession of signs of T\mathscr{T} written next to one another; certain signs, other than letters, may be joined in pairs by bars above the line, which are called links. \*For example, in the Theory of Sets, in which ∈\in is a specific sign,

τ∨¬∈□‾A′∈□‾A′′\overline{\overline{\tau \vee \neg \in \square} A' \in \square} A''

is an assembly.\*

The exclusive use of assemblies would lead to insuperable difficulties both for the printer and for the reader. For this reason current texts use abbreviating symbols (notably words of ordinary speech) which do not belong to formal mathematics. The introduction of such symbols is the object of definitions. Their use is not indispensable to the theory, and can often lead to confusion which only a certain familiarity with the subject will enable the reader to avoid.

Examples

Example 1 03P0

The assembly ∨1\vee 1 is represented by ⇒\Rightarrow.

Example 2 03P1

The following symbols represent assemblies (and very long ones at that) :

“3 and 4”

∅\varnothing
N\mathbf{N}
Z\mathbf{Z}

“the real line”

“the Γ\Gamma function”

f∘gf\circ g
π=2+3\pi=\sqrt{2}+\sqrt{3}
1∈21\in 2

“Every finite division ring is a field”

“The zeros of ζ(s)\zeta(s) other than −2-2, −4-4, −6-6, ... lie on the line R(s)=1/2R(s)=1/2.”

In general, the symbol used to represent an assembly contains all the letters which appear in the assembly. Nevertheless, this principle can sometimes be infringed without risk of confusion. *For example, “the completion of X” represents an assembly which contains the letter X, but which also contains the letter which represents the set of entourages of the uniform structure of X. On the other hand,

∫01f(x) dx\int_0^1 f(x)\,dx

represents an assembly in which the letter xx (and the letter dd) do not appear; and the assemblies represented by N\mathbf{N}, Z\mathbf{Z}, “the Γ\Gamma function” do not contain any letters.

A mathematical theory (or simply a theory) contains rules which allow us to assert that certain assemblies of signs are terms or relations of the theory, and other rules which allow us to assert that certain assemblies are theorems of the theory.

The description of these rules, which will appear in this chapter, does not belong to formal mathematics; the rules involve assemblies which are more or less undetermined, for example undetermined letters. To simplify the exposition it is convenient to denote such assemblies by less cumbersome symbols. We shall use, especially, combinations of signs (of a mathematical theory), bold-face italic letters (with or without indices or accents), and particular symbols, of which some examples will be given. Since our object is only to avoid circumlocutions (cf. note (*), § 3, no. 1, p. 28) we shall not enunciate strict general rules for the use of these symbols; the reader will be able to reconstruct without trouble the assembly in question, in each particular case. By abuse of language we shall often say that the symbols are assemblies, rather than that they denote assemblies : expressions such as “the assembly AA” or “the letter xx”, in the statements of the following rules, should therefore be replaced by “the assembly denoted by AA” or “the letter denoted by xx”.

Let AA and BB be assemblies. We shall denote by ABAB the assembly obtained by writing the assembly BB on the right of the assembly AA. We shall denote by ∨A¬B\vee A\neg B the assembly obtained by writing, from left to right, the sign ∨\vee, the assembly AA, the sign ¬\neg, the assembly BB. And so on.

Let AA be an assembly and let xx be a letter. We shall denote by τx(A)\tau_x(A) the assembly constructed as follows : form the assembly τA\tau A, link each occurrence of xx in AA to the τ\tau written on the left of AA, and then replace xx everywhere it occurs by the sign □\square. The assembly denoted by τx(A)\tau_x(A) therefore does not contain xx.

Example 03RY

The symbol τx(∈xy)\tau_x(\in xy) represents the assembly

τ∈□y. \tau\in\square y.

Let AA and BB be assemblies and let xx be a letter. The assembly obtained by replacing xx, wherever it occurs in AA, by the assembly BB is denoted by (B∣x) A(B\mathbin{|}x)\,A (read : BB replaces xx in AA). If xx does not appear in AA, then (B∣x) A(B\mathbin{|}x)\,A is identical with AA; in particular,

(B∣x) τx(A) (B\mathbin{|}x)\,\tau_x(A)

is identical with τx(A)\tau_x(A).

Example 03TO

If we replace xx by □\square wherever xx occurs in the assembly ∨∈xy=xx\vee\in xy=xx, we obtain the assembly ∨∈□y=□□\vee\in\square y=\square\square.

If AA is an assembly and we are interested particularly in a letter xx, or two distinct letters xx and yy (which may or may not appear in AA), we shall often write A{x}A\{x\} or A{x,y}A\{x,y\}. In this case we write A{B}A\{B\} instead of (B∣x) A(B\mathbin{|}x)\,A. We denote by A{B,C}A\{B,C\} the assembly obtained by simultaneously replacing xx by BB and yy by CC wherever they occur in AA (note that xx and yy may appear in BB and in CC); if x′x' and y′y' are distinct letters, other than xx and yy, which appear in neither AA, BB, nor CC, then A{B,C}A\{B,C\} is the same as (B∣x′)(C∣y′)(x′∣x)(y′∣y)A(B\mathbin{|}x')(C\mathbin{|}y')(x'\mathbin{|}x)(y'\mathbin{|}y)A.

Remark 03P2

When an abbreviating symbol Σ\Sigma is introduced, by means of a definition, to represent a certain assembly, the (usually tacit) convention is made of representing the assembly obtained by substituting an assembly BB for a letter xx in the original assembly, by the symbol obtained by the replacing the letter xx in Σ\Sigma by the assembly BB (or, more often, by an abbreviating symbol representing the assembly BB).

¶ *For example, having defined what assembly is represented by the symbol E⊗FE\otimes F, where EE and FF are letters — an assembly which, incidentally, contains other letters besides EE and FF — the symbol Z⊗FZ\otimes F can be used without further explanation.

¶ This rule can lead to confusions, which are avoided by the use of various typographical devices; the most common consists in replacing xx by (B)(B) in place of BB.

¶ *For example, M∩NM\cap N denotes an assembly containing the letter NN. If we substitute for NN the assembly represented by P∪QP\cup Q, we get an assembly denoted by M∩(P∪Q)M\cap(P\cup Q).

2. Criteria of substitution

Formal mathematics contains only explicitly written assemblies. Nevertheless, even with the use of abbreviating symbols, the development of mathematics strictly in accordance with this principle would lead to extremely long chains of reasoning. For this reason we shall establish criteria relating to indeterminate assemblies; each of these criteria will describe once for all the final result of a definite sequence of manipulations on these assemblies. These criteria are therefore not indispensable to the theory; their justification belongs to metamathematics.

The development of metamathematics itself requires, in practice, the use of abbreviating symbols, some of which have already been indicated. Most of these symbols are also used in mathematics.

We shall make use of the following criteria, called the criteria of substitution :

CS1. Let AA and BB be assemblies and let xx and x′x' be letters. If x′x' does not appear in AA, then (B∣x)A(B|x)A is identical with (B∣x′)(x′∣x)A(B|x')(x'|x)A.

CS2. Let AA, BB, and CC be assemblies and let xx and yy be distinct letters [^3]. If yy does not appear in BB, then (B∣x)(C∣y)A(B|x)(C|y)A is identical with

(C′∣y)(B∣x)A,(C'|y)(B|x)A,

where C′C' is the assembly (B∣x)C(B|x)C.

CS3. Let AA be an assembly and let xx and x′x' be letters. If x′x' does not appear in AA, then τx(A)\tau_x(A) is identical with τx′(A′)\tau_{x'}(A'), where A′A' is the assembly (x′∣x) A(x'|x)\,A.

CS4. Let AA and BB be assemblies and let xx and yy be distinct letters. If xx does not appear in BB, then (B∣y)τxA(B|y)\tau_xA is identical with τx(A′)\tau_x(A'), where A′A' is the assembly (B∣y) A(B|y)\,A.

CS5. Let AA, BB, CC be assemblies and let xx be a letter. The assemblies (C∣x)(¬A)(C|x)(\neg A), (C∣x)(∨AB)(C|x)(\vee AB), (C∣x)(⇒AB)(C|x)(\Rightarrow AB), (C∣x)(sAB)(C|x)(sAB) (where ss is a specific sign) are respectively identical with ¬A′\neg A', ∨A′B′\vee A'B', ⇒A′B′\Rightarrow A'B', sA′B′sA'B', where A′A', B′B' are respectively (C∣x) A(C|x)\,A, (C∣x) B(C|x)\,B.

As an example, let us indicate the principle of the verification of CS2. Compare the operation which takes us from AA to (B∣x)(C∣y) A(B|x)(C|y)\,A with the operation which takes us from AA to (C∣y)(B∣x) A(C|y)(B|x)\,A. In each operation, no sign which appears in AA and is distinct from xx and yy is altered. At every place where xx appears in AA, we have to substitute BB for xx in the first and in the second operation; this is clear in regard to the first operation, and in regard to the second it follows from the fact that yy does not appear in BB. Finally, at every place where yy appears in AA, the first operation consists in replacing CC for yy, then BB for xx at every place where xx appears in CC; but it is clear that this comes to the same thing as substituting for yy, wherever it appears in AA, the assembly (B∣x) C(B|x)\,C.

3. Formative constructions

Some of the specific signs of a theory are called relational, and the others are called substantific. With every specific sign is associated a natural number called its weight (which is practically always the number 2).

¶ An assembly is said to be of the first species if it begins with a τ\tau, or with a substantific sign, or if it consists of a single letter; otherwise it is of the second species.

¶ A formative construction in a theory T\mathscr{T} is a sequence of assemblies which has the following property : for each assembly AA of the sequence, one of the following conditions is satisfied :

(a) AA is a letter.

(b) There is in the sequence an assembly BB of the second species, preceding AA, such that AA is ¬B\neg B.

(c) There are two assemblies BB and CC of the second species (distinct or not), preceding AA, such that AA is ∨BC\vee BC.

(d) There is an assembly BB of the second species, preceding AA, and a letter xx such that AA is τx(B)\tau_x(B).

(e) There is a specific sign ss of weight nn [^4] in T\mathscr{T}, and nn assemblies A1,A2,…,AnA_1,A_2,\ldots,A_n of the first species, preceding AA, such that AA is sA1A2…AnsA_1A_2\ldots A_n.

¶ The assemblies of the first species (resp. of the second species) which appear in the formative constructions of T\mathscr{T} are called terms (resp. relations) in T\mathscr{T}.

Example. *In the theory of sets, in which ∈\in is a relational sign of weight 2, the following sequence of assemblies is a formative construction:*

AA′A′′∈AA′∈AA′′¬∈AA′∨¬∈AA′∈AA′′τ∨¬∈□A′∈□A′′ \begin{array}{c} A\\ A'\\ A''\\ \in AA'\\ \in AA''\\ \neg\in AA'\\ \vee\neg\in AA'\in AA'' \\[2pt] \tau\vee\neg\in\square A'\in\square A'' \end{array}

Hence the assembly given as an example in no. 1 is a term in the theory of sets.*

Remark. Intuitively, terms are assemblies which represent objects, and relations are assemblies which represent assertions which can be made about these objects. Condition (a) means that the letters represent objects. Condition (b) means that if BB is an assertion, then ¬B\neg B, called the negation of BB, is an assertion (which is read : not BB). Condition (c) means that if BB and CC are assertions, ∨BC\vee BC, which is called the disjunction of BB and CC, is an assertion (which is read : either BB or CC); thus ¬→BC\neg\to BC is an assertion (in words : “either not BB, or CC”, or “BB implies CC”). Condition (d) means that if BB is an assertion and xx a letter, then τx(B)\tau_x(B) is an object. Let us consider the assertion BB as expressing a property of the object XX; then, if there exists an object which has the property in question, τx(B)\tau_x(B) represents a distinguished object which has this property; if not, τx(B)\tau_x(B) represents an object about which nothing can be said. Finally, condition (e) means that if A1,A2,…,AnA_1,A_2,\ldots,A_n are objects, and if ss is a relational (resp. substantific) sign of weight nn, then sA1A2…AnsA_1A_2\ldots A_n is an assertion about the objects A1,…,AnA_1,\ldots,A_n (resp. an object depending on A1,…,AnA_1,\ldots,A_n).

Examples. The symbols ∅\varnothing, N\mathbf{N}, “the real line”, “the Γ\Gamma function”, f∘gf\circ g represent terms. The symbols π=2+3\pi=\sqrt{2}+\sqrt{3}, 1∈21\in2, “every finite division ring is a field”, “the zeros of ζ(s)\zeta(s) other than −2-2, −4-4, −6-6, ... lie on the line R(s)=1/2\mathfrak{R}(s)=1/2” represent relations. The symbol “3 and 4” represents neither a term nor a relation.

The initial sign of a relation is ∨\vee, ¬\neg, or a relational sign. The initial sign of a term is either τ\tau or a substantific sign, provided that the term does not consist of a single letter. The latter assertion follows from the fact that a term is an assembly of the first species. If AA is a relation, then AA features in a formative construction, is not a letter, and does not begin with τ\tau, so that three cases are possible : (1) AA is preceded by an assembly BB such that AA is ¬B\neg B; (2) AA is preceded by two assemblies BB and CC such that AA is ∨BC\vee BC; (3) AA is preceded by assemblies A1A_1, A2A_2, ... AnA_n such that AA is sA1A2...AnsA_1A_2 ... A_n, ss being a relational sign.

4. Formative criteria

CF1. If AA and BB are relations in a theory T\mathscr{T}, then ∨AB\vee AB is a relation in T\mathscr{T}.

Consider two formative constructions (in T\mathscr{T}), one of which contains AA and the other BB. Consider the sequence of assemblies obtained by writing first the assemblies of the first construction, then the assemblies of the second construction, and finally ∨AB\vee AB. Since AA and BB are of the second species, it is immediately verified that this sequence is a formative construction of T\mathscr{T}. The assembly ∨AB\vee AB is of the second species, hence it is a relation in T\mathscr{T}.

The three following criteria are established similarly :

CF2. If AA is a relation in a theory T\mathscr{T}, then ¬A\neg A is a relation in T\mathscr{T}.

CF3. If AA is a relation in a theory T\mathscr{T}, and if xx is a letter, then τx(A)\tau_x(A) is a term in T\mathscr{T}.

CF4. If A1A_1, A2A_2, ..., AnA_n are terms in a theory T\mathscr{T}, and if ss is a relational (resp. substantific) sign of weight nn in T\mathscr{T}, then sA1A2...AnsA_1A_2...A_n is a relation (resp. a term) in T\mathscr{T}.

These criteria immediately imply the following :

CF5. If AA and BB are relations in a theory T\mathscr{T}, then ⟹AB\Longrightarrow AB is a relation in T\mathscr{T}.

CF6. Let A1A_1, A2A_2, ..., AnA_n be a formative construction in a theory T\mathscr{T}, and let xx and yy be letters. Suppose that yy does not appear in any AiA_i. Then (y∣x)A1(y|x) A_1, (y∣x)A2(y|x) A_2, ..., (y∣x)An(y|x) A_n is a formative construction in T\mathscr{T}.

To prove CF6, let Ai′A_i' be the assembly (y∣x)Ai(y|x)A_i. If AiA_i is a letter, then Ai′A_i' is a letter. If AiA_i is of the form ¬Aj\neg A_j, where AjA_j is an assembly of the second species which precedes AiA_i in the construction, then Ai′A_i' is identical with ¬Aj′\neg A_j' by CS5, and Aj′A_j' is an assembly of the second species. The reasoning is similar if AiA_i is of the form ∨AjAk\vee A_jA_k or sAj1Aj2…AjmsA_{j_1}A_{j_2}\ldots A_{j_m}, ss being a specific sign of T\mathscr{T}. If finally AiA_i is of the form τz(Aj)\tau_z(A_j), where AjA_j is an assembly of the second species which precedes AiA_i in the construction, there are various cases to consider:

(a) zz is a letter distinct from xx and yy. Then Ai′A_i' is identical with τz(Aj′)\tau_z(A_j') by CS4, and Aj′A_j' is an assembly of the second species.

(b) zz is identical with xx. Then AiA_i does not contain xx, hence Ai′A_i' is identical with AiA_i, that is to say with τx(Aj)\tau_x(A_j); since yy does not appear in AjA_j, τx(Aj)\tau_x(A_j) is identical with τy(Aj)\tau_y(A_j) by CS3.

(c) zz is identical with yy. Then AiA_i is the assembly τAj\tau A_j, because yy does not appear in AjA_j; therefore Ai′A_i' is the assembly τAj′\tau A_j', that is τu(Aj′)\tau_u(A_j'), where uu is a letter which does not appear in Aj′A_j'.

CF7. Let AA be a relation (resp. a term) in a theory T\mathscr{T}, and let xx and yy be letters. Then (y∣x)A(y|x)A is a relation (resp. a term) in T\mathscr{T}.

Let A1,A2,…,AnA_1,A_2,\ldots,A_n be a formative construction in which AA appears. We shall show step by step that, if AiA_i is a relation (resp. a term) then (y∣x)Ai(y|x)A_i, which we shall denote by Ai′A_i', is also a relation (resp. a term). Suppose that this point has been established for A1,A2,…,Ai−1A_1,A_2,\ldots,A_{i-1}; let us prove it for AiA_i. If AiA_i is a letter, then Ai′A_i' is a letter. If AiA_i is preceded in the construction by a relation AjA_j such that AiA_i is ¬Aj\neg A_j, then Ai′A_i' is identical with ¬Aj′\neg A_j' by CS5, and ¬Aj′\neg A_j' is a relation by CF2. The argument is similar if AiA_i is preceded by relations Aj,AkA_j,A_k such that AiA_i is ∨AjAk\vee A_jA_k, or if AiA_i is preceded by terms Aj1,…,AjmA_{j_1},\ldots,A_{j_m} such that AiA_i is sAj1⋯AjmsA_{j_1}\cdots A_{j_m}, where ss is a specific sign of T\mathscr{T} of weight mm. Finally, if AiA_i is preceded by a relation AjA_j such that AiA_i is τz(Aj)\tau_z(A_j), there are various cases to consider:

(a) zz is distinct from both xx and yy. Then Ai′A_i' is identical with τz(Aj′)\tau_z(A_j') by CS4, and we know already that Aj′A_j' is a relation; hence AiA_i is a term, by CF3.

(b) zz is identical with xx. Then AiA_i does not contain xx, therefore Ai′A_i' is identical with AiA_i, and consequently is a term.

(c) zz is identical with yy. Then let uu be a letter, distinct from both xx and yy, which does not appear in A1,A2,…,AjA_1,A_2,\ldots,A_j. By CF6, the sequence of assemblies (u∣y)A1,…,(u∣y)Aj(u|y)A_1,\ldots,(u|y)A_j, which we shall denote by A1′′,…,Aj′′A_1'',\ldots,A_j'', constitutes a formative construction in T\mathscr{T}. Since yy no longer appears in this new construction, (y∣x)A1′′,…,(y∣x)Aj′′(y|x)A_1'',\ldots,(y|x)A_j'' is a formative construction by CF6, so that (y∣x)Aj′′(y|x)A_j'' is a relation in T\mathscr{T}; consequently τu((y∣x)Aj′′)\tau_u((y|x)A_j'') is a term of T\mathscr{T}. But this term is identical with (y∣x)τu(Aj′′)(y|x)\tau_u(A_j'') by CS4, hence with (y∣x)τy(Aj)(y|x)\tau_y(A_j) by CS3, hence is identical with AiA_i.

CF8. Let AA be a relation (resp. a term) in a theory T\mathscr{T}, let xx be a letter and TT a term in T\mathscr{T}. Then (T∣x)A(T|x)A is a relation (resp. a term) in T\mathscr{T}.

Let A1,A2,…,AnA_1, A_2, \ldots, A_n be a formative construction in which AA appears. Let x1,x2,…,xpx_1, x_2, \ldots, x_p be the distinct letters which appear in TT. Let us associate with each letter xix_i a letter xi′x_i', distinct from each of the letters x1,…,xpx_1, \ldots, x_p and the letters which appear in A1,…,AnA_1, \ldots, A_n, in such a way that the letters x1′,…,xp′x_1', \ldots, x_p' are all distinct. The assembly

(x1′∣x1)(x2′∣x2)…(xp′∣xp)T (x_1'|x_1)(x_2'|x_2)\ldots(x_p'|x_p)T

is a term T′T' by CF7, and (T∣x)A(T|x)A is identical with

(x1∣x1′)(x2∣x2′)…(xp∣xp′)(T′∣x)A (x_1|x_1')(x_2|x_2')\ldots(x_p|x_p')(T'|x)A

by application of CS1. It is therefore enough to show that (T′∣x) A(T'|x)\,A is a relation (resp. a term); in other words, we may suppose from now on that the letters which appear in TT do not appear in A1,…,AnA_1, \ldots, A_n.

¶ We shall show step by step that, if AtA_t is a relation (resp. a term), then (T∣x) At(T|x)\,A_t, which we shall denote by At′A_t', is a relation (resp. a term). Suppose this point has been established for A1,A2,…,Ai−1A_1,A_2,\ldots,A_{i-1}, and let us prove it for AiA_i. If AiA_i is a letter, then Ai′A_i' is either the same letter or TT, and therefore a term. If AiA_i is of the form ¬Aj\neg A_j, where AjA_j is a relation which precedes AiA_i in construction, then AiA_i is identical with ¬Aj′\neg A_j' by CS5, and we know already that Aj′A_j' is a relation, hence Ai′A_i' is a relation by CF2. The proof is analogous if AiA_i is of the form ∨AjAk\vee A_jA_k, or sAj1…AjmsA_{j_1}\ldots A_{j_m}. Finally, if AiA_i is of the form τz(Aj)\tau_z(A_j), where AjA_j is a relation which precedes AiA_i in the construction, there are various cases to be considered:

(a) zz is distinct from xx and from the letters which appear in TT. Then Ai′A_i' is identical with τz(Aj′)\tau_z(A_j') by CS4, and we know already that Aj′A_j' is a relation; hence Ai′A_i' is a term by CF3.

(b) zz is identical with xx. Then AiA_i does not contain xx, hence Ai′A_i' is identical with AiA_i and consequently is a term.

(c) zz appears in TT. Then zz does not appear in AjA_j, so that Ai′A_i' is identical with τAj′\tau A_j'; hence Ai′A_i' is identical with τAj′\tau A_j'. Now, we know already that Aj′A_j' is a relation, and τAj′\tau A_j' is identical with τu(Aj)\tau_u(A_j), where uu is a letter which does not appear in AjA_j; it follows by CF3 that Ai′A_i' is a term.

Intuitively, if AA is a relation in T\mathscr{T}, which we may regard as expressing a property of an object xx, the assertion (B∣x) A(B|x)\,A amounts to saying that the object BB has this property. If AA is a term in T\mathscr{T}, it represents an object which depends in some way on the object denoted by xx; the term (B∣x) A(B|x)\,A represents what the object AA becomes when we take xx to be the object BB.

Exercises

See the exercises for § 1.

[^1]: The meaning of this expression will become clear as the chapter progresses. [^2]: For the intuitive meanings of these signs, see no. 3, Remark. [^3]: In accordance with what was said in no. 1, the phrase “xx and yy are distinct letters” is an abuse of language: it means that xx and yy denote distinct letters in the assemblies under consideration. [^4]: As was said above, it would be possible, for the development of present-day mathematical theories, to limit our consideration to specific signs of weight 2, and consequently to avoid using the expression “natural number nn” in the definition of a formative construction.

Transcribed from the printed edition, 2004, Springer. Not translated.