Jan von Plato From Axiomatic Logic to

Natural Deduction

Abstract. Recently discovered documents have shown how Gentzen had arrived at the
final form of natural deduction, namely by trying out a great number of alternative formu-
lations. What led him to natural deduction in the first place, other than the general idea
of studying “mathematical inference as it appears in practice,” is not indicated anywhere
in his publications or preserved manuscripts. It is suggested that formal work in axiomatic
logic lies behind the birth of Gentzen’s natural deduction, rather than any single decisive
influence in the work of others. Various axiomatizations are explored in turn, from the
classical axioms of Hilbert and Ackermann to the intuitionistic axiomatization of Heyting.

Keywords: Gentzen, Natural deduction, Axiomatic logic.

1. Introduction

After Gerhard Gentzen had presented the calculus of natural deduction in

his thesis of May 1933, published in two parts in 1934 and 1935, the system of
rules remained stable for some fifty years; Nothing better was ever proposed
that would have taken its place. There were in the 1970s some attempts at
changing the way open assumptions are closed, usually failed ones, and more
recently, a more uniform treatment of the elimination rules that leads to some
simplifications (as detailed in von Plato 2012, section 15). The discovery of
a few central sources, such as the handwritten version of Gentzen’s thesis
and a short manuscript with the title Fünf verschiedene Formen natürlicher
Kalküle (Five different forms of natural calculi) of the Fall of 1932, has
led to an understanding of how he had arrived at the final form of natural
deduction, namely by trying out a great number of alternative formulations.
What led him to natural deduction in the first place, other than an the
general idea of studying “mathematical inference as it appears in practice,” is
not indicated anywhere in his publications or preserved manuscripts. Thus,
the present somewhat speculative topic: What, besides Gentzen’s genius,
are the possible origins of the calculus of natural deduction?

Special Issue: Gentzen’s and Jaśkowski’s Heritage

80 Years of Natural Deduction and Sequent Calculi
Edited by Andrzej Indrzejczak

Studia Logica
DOI: 10.1007/s11225-014-9565-0 © Springer Science+Business Media Dordrecht 2014
2 J. von Plato

The leading idea of natural deduction is that actual reasoning is hypothet-

ical, based on temporary assumptions. Such hypothetical logical reasoning
had been the norm in logic since Aristotle’s times. However, logic got carried
along in the wake of the axiomatic mathematics that became fashionable in
the second half of the 19th century: Frege, Peano, Russell, and Hilbert and
Bernays had turned logic into an axiomatic discipline that got, little by little,
removed rather far away from its intuitive origins as a science of reasoning.
Logic became a field in which certain axiomatic calculi are studied, just as
one does in geometry: The tasks are a judicious choice of basic concepts,
axioms, and rules, and the study of the central foundational questions such
as the independence of the axioms, their consistency and completeness, and
the decision problem.
Gentzen distanced himself from the prevailing axiomatic form of logic,
apparently with no difficulty. His independence was one of a prodigious
youth, for he was merely 22 years old at the time he discovered natural
deduction. The new ideas were accompanied by an enormous amount of
formal work. My view is that the leading idea of natural deduction, inference
under assumptions, together with the formal work on axiomatic logic, is
responsible for the birth of Gentzen’s system of natural deduction, rather
than any single decisive influence in the work of others. I will explore in
turn the various possibilities and have a clear candidate, namely Heyting’s
axiomatic intuitionistic logic.

2. What Gentzen says and does about natural deduction

The handwritten thesis manuscript that I had the good fortune to find in
February 2005 has the same title as the published version, Untersuchungen
über das logische Schliessen, but it is mostly concerned with natural deduc-
tion against sequent calculus that is at the center of the final form of the
thesis. Sequent calculus is only outlined and there is not as much as an idea
of cut elimination, but natural deduction instead is fully developed, with
a detailed 13-page proof of normalization for intuitionistic predicate logic,
now published as Gentzen (2008). The “foundation and setting up of the
calculus N1I ” begins with:

The way logical inference is formalized in Russell, Hilbert, Heyting

(for intuitionistic inference) among others is rather far apart from the
way of inference as it is practised in reality (say, in number-theoretic
proofs). Notable formal advantages are won thereby. I shall for once
set up a formalism (“calculus N1I ”) that comes as close as possible
From Axiomatic Logic to Natural Deduction 3

to actual inference. One can assume that such a calculus has certain
advantages and I think I can maintain, on the basis of my further
results, that this is the case.

Next comes the basic idea of inference from assumptions:

Natural deduction, in contrast [to axiomatic logic], does not in gen-

eral start from logical axioms, but from assumptions . . . . Inference
proceeds further from these. The result is made independent of the
assumptions through a later inference.

Natural deduction is not just a system of rules, but it is founded on much

more general consideration of how any system of proofs from assumptions
has to be organized:

A proof consists of a number of propositions in an arrangement in

tree form. Its look is as follows, in an example:

A1 A2 A6 A11
A3 A4 A5 A10
A7 A8
A9
The tree form was in practice a novelty in Gentzen: The arrangement of
formal deductions had been invariably a linear succession of formulas, with
some device such as a numbering of the formulas, so that one could refer
to formulas deduced earlier. The first one, apparently, to depart from this
form and to present proofs in tree form had been Paul Hertz in his work of
1923, as suggested by Schroeder-Heister (2002). Gentzen’s first paper (1932)
treated these Hertz-systems and introduced tree derivations that stem from
a rule of cut that has two premisses. The work of Hertz was not widely read,
by which Gentzen’s work can be considered the one that made the use of
the tree form for formal derivations generally known. He himself continues
from the display of the above tree figure with the following remarks:

I deviate somewhat from actual reasoning by the requirement of an

arrangement in tree form, because 1. there is in actual reasoning a
linear succession of propositions and 2. it is common in continuing
further to use repeatedly a result already won, whereas the tree form
allows to use in continuation only once a proposition already proved.
Both deviations are obviously inessential, but they make it instead
easier to conceive the notion of a proof.
4 J. von Plato

Each of the horizontal lines represents an inference. The proposi-

tion directly below the line is the conclusion of the inference, and
those directly above the line its premisses. There can further belong
assumptions to an inference, these are certain uppermost proposi-
tions of the proof that stand above the conclusion of the inference
(indicated as belonging to the inference through a common number-
ing). (I.e. not necessarily directly above, but in general: uppermost
propositions of “proof threads” that pass through the inference line.
The last point is just an explanation of how a step of inference can close
assumptions, indicated by a numbering next to the inference line and above
the closed assumptions. The explanation continues with:
All uppermost propositions of a proof are assumptions, and each
belongs to exactly one inference.
All propositions that stand under an assumption (i.e., in the proof
thread determined by the assumption), but still above the inference
line to which the assumption belongs, are said to depend on the as-
sumption. The assumption itself included (so the inference makes the
propositions following it, beginning with its conclusion, independent
of the assumption in question.)
Gentzen met Stanislaw Jaśkowski in Münster in June 1936 and became ac-
quainted with the linear version of natural deduction of the latter, Jaśkowski
(1934). As Gentzen emphasizes in the above passage, the tree form displays
the open assumptions on which a given formula in a derivation depends, as
those topmost formulas of the upward threads that begin from the given
formula but are not yet closed at the point in which the formula is found.
This full display of the deductive dependencies in a proof is a crucial prop-
erty by which parts of derivations can be taken in isolation and composed
in new ways.
The method of permutation of parts of derivations is the most essen-
tial method in all of Gentzen’s central results: normalization for natural
deduction, cut elimination for sequent calculus, and consistency proofs for
arithmetic. The permutable parts of proofs are the subtrees determined by
a given formula. All of this is lost in the linear variant of natural deduction,
because the dependence of formula occurrences on assumptions is muddled
and the permutation of parts of proof blocked. It is no accident that no
results on the structure of formal derivations were won in the linear variant
comparable to normalization and the subformula property. It has instead
become the preferred approach in first introductions to natural deduction,
where the tree form is too abstract as a starting point.
From Axiomatic Logic to Natural Deduction 5

One possible source of natural deduction is offered by Heyting’s explana-

tions of the logical connectives in his article (1931) based on the talk at the
Königsberg meeting of September 1930, the one in which Gödel announced
his incompleteness result. There is towards the end of Heyting’s paper a
brief explanation of negation as “a proof procedure that leads to a contra-
diction.” Next, it is stated that a proof of a disjunction consists in a proof of
one of the disjuncts. It is not known if Gentzen had read Heyting’s remarks
of 1931.
There is a long set of notes by Gentzen with the title Die formale Er-
fassung des Begriffs der inhaltlichen Richtigkeit in der reinen Zahlentheorie,
Verhältnis zum Widerspuchsfreiheitsbeweis (The formal conception of the
notion of correctness in terms of content of pure number theory, relations to
proof of consistency) written mainly in October 1932. It contains explana-
tions quite similar to the well-known BHK-explanations, the following ones
dated 16 October 1932:
I explain the concept “B is provable from V1 . . . Vν ” as follows recur-
sively:1 (More correctly: “A consequence of”)
C&D is provable from the V ’s when and only when C is and D is,
C ∨ D is provable from the V ’s when and only when C is or D is,2
C → D is provable from the V ’s when and only when D is provable
from the V ’s and C, ←− is this still recursive?
x P x is provable from the V ’s when and only when P ν is provable
from the V ’s for arbitrary numbers ν,
Ex P x is provable from the V ’s when and only when there is a number
μ such that P μ is provable from the V ’s.3
¬C is provable from the V ’s when and only when the contradiction
1 = 1 is provable from the V ’s and C. I.e., ¬C ∼ C → O.
Nothing new.
These observations were made after Gentzen had put up the system of natu-
ral deduction. His aim was to define a notion of “correctness of intuitionistic
proofs in terms of their content” (inhaltliche Richtigkeit), to be used as a
basis for the proof of consistency of arithmetic.

1
The manuscript has in the list of assumptions an upper case V in Sütterlin script, an
old form of German handwriting, likely for Voraussetzung. Such script would have been
printed in Fraktur in the 1930s, but is here given in the Latin alphabet.
2
A blatant error: For a counterexample, let V be D ∨ C.
3
Again, an error if the assumptions contain existential formulas.
6 J. von Plato

It seems that part of the idea of a subformula property of normal deriva-

tions came from the requirement that the explanation of the logical oper-
ations be given in terms of their components, and especially, that the cor-
rectness of a proof should be somehow reduced back to the correctness of its
parts. For such an explanation to be well-founded, a subformula property
is needed.
As I will propose here, the sources of natural deduction are to be found
in the axiomatic logical tradition. There are at least four candidates to be
explored: 1. The first axiomatization to consider is the one of Hilbert and
Ackermann’s 1928 book Grundzüge der theoretischen Logik. These axioms
come from Principia Mathematica and are purely classical, with negation
and disjunction as primitive concepts in the propositional part. Hilbert used
them in his early work around 1920 and, strangely enough, in the 1928 book,
though rewritten in the latter with a defined notion of implication. 2. A sec-
ond axiom system is found in von Neumann’s 1927 article Zur Hilbertschen
Beweistheorie, explored here because of an intriguing footnote suggestive of
a system of logic based on rules rather than axioms. 3. Third, there are
the axioms of Hilbert and Bernays from the mid-1920s, in which an attempt
is made to have distinct groups of axioms for the different connectives and
quantifiers, similarly to the grouping of axioms in geometry into axioms of
incidence, order, etc. One aim was to single out the role of negation in logic.
4. Finally, there is Heyting’s axiomatization of intuitionistic logic in his 1930
article. The way from these four axiom systems to natural deduction will
now be studied in detail.

3. From the Hilbert-Ackermann axioms to natural deduction

In the handwritten thesis, Gentzen’ design was to first present the calculus
of natural deduction, then to show its equivalence to axiomatic logic, then
to establish normalization and the subformula property. Finally, the aim
was to extend all of this to arithmetic, which however failed as there is no
subformula property for derivations in a formal system of arithmetic.
The axiomatic system of logic in the thesis manuscript was that of Hilbert
and Ackermann’s book, abbreviated as H-A below. A remarkable aspect of
H-A is a clear awareness of the role of derivable rules (p. 24):

In the derivation of formulas, it is recommendable to collect into deri-

ved rules certain operations that repeat themselves often. . . the proof
of the rule consists in giving in a general form the procedure by which
the passage is carried through in each single case by the basic rules.
From Axiomatic Logic to Natural Deduction 7

Among such derivable rules, there is one by which A ∨ B can be concluded

if A has been proved, and another that is the rule of cut.
The way to the axioms in H-A is a rather tortuous one: First some
propositional equivalences are ascertained by showing that the equivalent
formulas in question have always the same truth values (p. 5). These are
gathered into rules of logic that bear an analogy to algebra, used for the
production of a conjunctive normal form (pp. 9–10). Finally, certain rules
of inference are shown to be correct (p. 11), written here in a modernized
notation: b1) A ∨ ¬A “is always correct,” b2) if A is correct A ∨ B is correct,
b3) if A is correct and B is correct, A&B is correct.
As mentioned, H-A presents a system of classical logic with negation and
disjunction as primitives in the propositional part. The axioms are written
with disjunctions and implications, the latter with the standard classical
definition A → B ≡ ¬A ∨ B. Gentzen takes into use a classical system
of natural deduction, called the N2-formalism that he then relates to the
H-A axiomatization. It is the classical ¬-&-∀-fragment of natural deduction
together with a rule of induction CI. The rules are:

Table 1. Gentzen’s classical natural calculus N2 .

AI AE UI UE
AB A&B A&B Pa (x)P x
A&B A B (x)P x Ph

DN TND CI
A
.. A.. P.a
.. .. ..
.
B ¬B ¬¬A P 1 P a
¬A A Ph

Schematic letters A, B, C, . . . are used in the rules, except when free and
bound variables appear and the writing becomes P a, P h, and so on. The
variable condition is that a is an eigenvariable in rules UI and CI . The latter,
as well as UE , have an arbitrary term h in the conclusion. Furthermore, an
eigenvariable must not occur in formula A in rule DN, in which the letters
should stand for something like ‘dilemma for negation.’
Gentzen now takes into use an axiomatic formulation of logic that he
calls the “H-A-formalism,” from Hilbert and Ackermann. The axioms are
(Hilbert and Ackermann 1928, p. 22 and 53):
8 J. von Plato

Table 2. The H-A axioms of Gentzen’s thesis ms.

a) A∨A → A
b) A → A∨B
c) A∨B → B∨A
d) (A → B) → (C ∨ A → C ∨ B)
e) (x)P x → P h
f) P h → (Ex)P x
The rules of inference are modus ponens, or FE in Gentzen’s notation, and
for the quantifiers the following two (ibid., p. 54):
Let B(x) be an arbitrary logical expression that depends on x, and
A one that doesn’t. If now A → B(x) is a correct formula, also
A → (x)B(x) is. One obtains similarly from a correct formula
B(x) → A the new (Ex)B(x) → A.
To prove that derivations in the axiomatic calculus of H-A can be reproduced
in N2, Gentzen makes the
Definition. A∨B is ¬.¬A&¬B, (Ex)P x is ¬(x)¬P x, A → B is ¬.A&¬B.
The axioms of H-A now read:
a) ¬...¬.¬A&¬A..&.¬A
b) ¬..A&¬¬.¬A&¬B
c) ¬...¬.¬A&¬B..&..¬¬.¬B&¬A
d) ¬{¬.A&¬B..&¬¬...¬.¬C&¬A..&..¬¬.¬C&¬B}
e) ¬.(x)P x&¬P h
f) ¬.P h&¬¬(x)¬P x
These are easily proved in N2. Gentzen gives as an example the derivation
of axiom b:
1
A&¬¬.¬A&¬B AE
1 ¬¬.¬A&¬B T N D
A&¬¬.¬A&¬B AE ¬A&¬B AE
A ¬A DN 1
Axiom b
The inference schemes of H-A can be reproduced in N2:
The inference scheme A ¬.A&¬B
B becomes
1
A ¬B
A&¬B ¬.A&¬B DN 1
¬¬B T N D
B
From Axiomatic Logic to Natural Deduction 9

¬.A&¬(x)P x
The generalization scheme becomes:
¬.A&¬P a
1
A&¬(x)P x 2
AE
A ¬P a AI
A&¬P a ¬.A&¬P a DN 2
1 ¬¬P a T N D
A&¬(x)P x P a UI
AE
¬(x)P x (x)P x
DN 1
¬.A&¬(x)P x

UI fulfils the variable condition, because a does not occur in ¬.A&¬(x)P x.

The scheme of existential instantiation is treated analogously and also
the induction axiom is derivable. By these results, the calculus N2 is at least
as strong as the standard axiomatic calculus of Hilbert and Ackermann. In
the other direction, derivations in N2 are reproduced in axiomatic logic as
follows:

One finds for each proposition A of the N2-proof all those assumptions
above it the respective DN’s (CI’s) of which still stand under A.
These are B1 , . . . Bϕ . Then one substitutes A by B1 & . . . &Bϕ → A.

If A is an assumption, A → A takes its place. The steps of inference of

natural deduction become:

Table 3. Gentzen’s translation from N2 to axiomatic logic.

AI AE UE TND
D→A E→B C → A&B C → (x)P x C → ¬¬A
D&E → A&B C →A C → Ph C →A
B
UI DN CI
C → Pa D&A → B E&A → ¬B D → P 1 E&P a → P a
C → (x)P x D&E → ¬A D&E → P h
Now it remains to show that these steps can be reproduced in axiomatic logic.
With AI , one proves first (D → A) → ((E → B) → (D&E → A&B)) to
be a theorem of axiomatic logic. Then the translation of AI in table 3 is
replaced by two implication eliminations:

(D → A) → ((E → B) → (D&E → A&B)) D → A

(E → B) → (D&E → A&B) E→B
D&E → A&B
10 J. von Plato

The pattern is the same for the rest of the translations.

It is a fair conclusion to say that axiomatic logic of the Hilbert-Acker-
mann variety contains some elements that are suggestive of the formalism of
natural deduction, such as the idea of derivable rules, including conjunction
introduction and cut. However, in the formal details, the two are far apart;
the reproduction of derivations in natural deduction within axiomatic logic,
especially, is burdensome and not suggestive of a bridge from the H-A axioms
to natural deduction.

4. From von Neumann’s axioms to natural deduction

Johann von Neumann’s 1927 paper On Hilbertian proof theory contains the
following interesting footnote (note 9, p. 38):

The possibility lies close of substituting the logical axioms by the

logical ways of inference that rely on the axioms. There would be in
place of a single syllogistic rule of inference several ones, by which the
(rather arbitrarily built) group of axioms I would disappear. We have
refrained from such a construction, because it departs too radically
from the usual one.

Gentzen studied in Berlin for the winter semester of 1930–31, and von Neu-
mann gave a course on proof theory. It is not known if Gentzen attended,
but he must have at least heard about the lectures. Here is one account (by
Carl Hempel):

I took a course there with von Neumann which dealt with Hilbert’s
attempt to prove the consistency of classical mathematics by finitary
means. I recall that in the middle of the course von Neumann came
in one day and announced that he had just received a paper from...
Kurt Gödel who showed that the objectives which Hilbert had in
mind and on which I had heard Hilbert’s course in Göttingen could
not be achieved at all. Von Neumann, therefore, dropped the pursuit
of this subject and devoted the rest of the course to the presentation
of Gödel’s results. The finding evoked an enormous excitement.

The axioms von Neumann refers to are as follows:

Table 5. The von Neumann axioms.

1. A → (B → A),
2. (A → (A → B)) → (A → B),
From Axiomatic Logic to Natural Deduction 11

3. (A → (B → C)) → (B → (A → C)),
4. (A → B) → ((B → C) → (A → C)),
5. A → (¬A → B),
6. (A → B) → ((¬A → B) → B).
The only rule of inference is implication elimination.
When the above axioms are actually used, the same thing happens as
before: Axiom instances are written that can be readily abbreviated into
corresponding rules of inference. From axiom 1, we get a rule of weakening:
A → (B → A) A A Wk
⊃E
B→A turns into B→A

For the second axiom, the axiomatic inference is:

(A → (A → B)) → (A → B) A → (A → B)
⊃E
A→B A ⊃E
B

Now there are two choices of a rule of contraction:

A → (A → B) A A → (A → B)
Ctr Ctr
B A→B

The first one combines contraction and implication elimination, by which

the second rule leads to a more flexible system.
Now the rest of the rules can be derived analogously from the axioms,
with choices similar to axiom 2. Axiom 3 corresponds to the exchange of the
order of the assumptions A and B in A → (B → C), into B → (A → C),
axiom 4 to a rule of cut that gives A → C from the premisses A → B
and B → C, axiom 5 to the rule of falsity elimination that gives B from
the premisses A and ¬A, and axiom 6 gives finally the conclusion B from
the premisses A → B and ¬A → B, by a rule of excluded middle (Tnd ).
Altogether, we have:
Table 6. From von Neumann’s axioms to rules.

A Wk A → (A → B) A → (B → C)
Ctr Exc
B→A A→B B → (A → C)
A → B B → C Cut A ¬A Efq A→B ¬A → B Tnd
A→C B B
The additional rule of inference is implication elimination. Now we come to
the big conceptual problem:
12 J. von Plato

If there are no axioms, how should derivations in the rule system begin?

At the time von Neumann wrote his paper, it had become clear that
axiomatic logic has to be extended from its original goal in Frege and Russell,
which was to grind out more logical truths from the axioms of logic that are
the given logical truths. Hilbert, Bernays, and Ackermann wanted to apply
logic in mathematics, by admitting mathematical axioms as premisses in
logical derivations. These axioms were not given truths, but hypothetical
elements that could turn out true in one application, and fail in another. In
the second part of the 1920s, there thus evolved the idea of derivations from
mathematical axioms. Hilbert and Ackermann’s book of 1928 demonstrates
clearly this change: The task of its §9 (pp. 18–22) is to extend the pure logic
of connectives to axiom systems:

Let there be given a finite number of axioms A1 , A2 , . . . An . The ques-

tion whether a given propositional formula C is a logical consequence
of these axioms can be always answered by the methods presented so
far. Such is the case if and only if A1 &A2 & . . . An → C is a logically
generally valid formula.

The passage shows a clear awareness of the need for a deduction theorem
when pure logic is extended to axioms that are treated as assumptions (Vor-
aussetzungen in Hilbert and Ackermann, see their p. 18). The 1928 book
was based on Hilbert’s lectures held almost ten years earlier, and elaborated
by Bernays in later lecture series. It should be clear that von Neumann,
a Göttingen student with the fastest mind of all, knew such things: He
lectured in fact on “Hilbertian proof theory” in Berlin at least in 1928/29
and 1930/31.
Whatever von Neumann’s rule system was, one of the first tasks it faced
was to show its equivalence to an axiomatic system. However, the only
way to arrive at the logical axioms is to apply a step that corresponds to a
deduction theorem. A notation is needed for what are assumptions, what
conclusions, and how proofs from assumptions are converted into proofs of
implications in pure logic. As an example, and following the lead of H-A in
the above quotation, we could derive axiom 1 by: Assume A. By rule Wk,
B → A follows. Therefore the implication A → (B → A) follows. A usable
formal system results if derivations from assumptions are written with some
explicit notation such as A1 , A2 , . . . , An C, with a deduction rule by which
from the premiss A1 , A2 , . . . , An , B C, the conclusion A1 , A2 , . . . , An
B → C can be drawn. The sequent notation, with a list of assumptions
at the left of a suitable marker for derivability, and a conclusion at right,
From Axiomatic Logic to Natural Deduction 13

was known well in Göttingen: Professor extraordinarius Paul Hertz had

written long papers on sequents that Bernays endorsed and got published in
the Mathematische Annalen. We could thus go on and define translations
between derivations in such a system and in Gentzen’s natural deduction,
to determine how close they are. However, this seems far-fetched unless
it should turn out some day that Gentzen indeed followed von Neumann’s
course and notes on the course are found. Otherwise there is little reason
to believe that von Neumann’s rule-based logic would have led Gentzen to
natural deduction.

5. From the Hilbert-Bernays axioms to natural deduction

The attempt at delimiting the role of negation led to a grouping of logical

axioms and to a list of axioms that can be found, with slight variations, in
several papers by Hilbert, Bernays, and Ackermann, and in the first volume
of the Grundlagen der Mathematik.
Table 4. The Hilbert-Bernays axioms.
1. A → (B → A),
2. (A → (A → B)) → (A → B),
3. (A → B) → ((B → C) → (A → C)),
4. A&B → A, A&B → B,
5. (A → B) → ((A → C) → (A → B&C)),
6. A → A ∨ B, B → A ∨ B,
7. (A → C) → ((B → C) → (A ∨ B → C)),
8. (A → B) → (¬B → ¬A),
9. A → ¬¬A,
10. ¬¬A → A.
For intuitionistic propositional logic, it is sufficient to leave out the last
axiom, a thing known to Bernays already in 1925.
The first three axioms concern implication. If the first axiom is derived
in natural deduction, there appears a vacuous discharge of formula B; it is
not used at all in the derivation of the consequent of the implication B ⊃ A:
1
A ⊃I
B→A ⊃I,1
A → (B → A)
14 J. von Plato

In terms of sequent calculus, vacuous discharge corresponds to a step of

weakening, and this is what axiom 1 does.
Axiom 2 corresponds to a multiple discharge of an assumption in natural
deduction, or a step of contraction in terms of sequent calculus. The third
axiom expresses the transitivity of implication, also rendered as the rule of
cut in sequent calculus.
Now we come to the axiom group for conjunction. The two axioms in 4
correspond to the rules of conjunction elimination, and axiom 5 to conjunc-
tion introduction. We can make these correspondences formal through the
following derivations:
3 1 2 1
A→B A ⊃E A→C A
⊃E
B C &I
B&C ⊃I,1
1 1
A&B &E A&B &E A → B&C ⊃I,2
A
1
B
2
(A → C) → (A → B&C)
⊃I,1 ⊃I,1 ⊃I,3
A&B → A A&B → B (A → B) → ((A → C) → (A → B&C))

The other direction, from axiomatic logic to natural deduction, is much more
suggestive of a direct connection: Assume there to be at hand a derivation
of A&B. To conclude A resp. B in axiomatic logic, a step with ⊃E is made:

A&B → A A&B ⊃E A&B → B A&B ⊃E

A B

The first premiss in both derivations is an instance of axiom 4. These two

steps are just like conjunction eliminations in natural deduction. For con-
junction introduction, the natural rule is not quite straightforward, because
conjunction appears in axiom 5 under an assumption A. The corresponding
axiom without such an assumption is B → (C → B&C) and it makes the
connection to the natural rule direct:

B → (C → B&C) B
⊃E
C → B&C C ⊃E
B&C

Looking at the axioms of disjunction, the connection to disjunction intro-

duction and elimination is straightforward. Say, the premisses of disjunction
elimination are A∨B plus a derivation of C from A and C from B. From the
latter two, we get A → C and B → C, respectively, so we have altogether:
From Axiomatic Logic to Natural Deduction 15

(A → C) → ((B → C) → (A ∨ B → C)) A → C
⊃E
(B → C) → (A ∨ B → C) B→C
⊃E
A∨B → C A ∨ B ⊃E
C

The uppermost leftmost assumption formula is the axiom, and the remaining
three assumptions A → C and B → C and A∨B came from the given of the
disjunction elimination. The pattern as a whole is suggestive of the scheme
of disjunction elimination.
It goes similarly for the rest of the Hilbert-Bernays axioms: Each of them
goes hand in hand with a scheme in which an axiom instance plus the premiss
or premisses of a rule of natural deduction gives by implication elimination
the conclusion of the natural rule. Hilbert’s last paper on proof theory, the
Proof of the excluded middle (Beweis des Tertium non datur) of 1931, goes
even one step further: In the axiomatic treatment, there are usually two rules
and two axioms for the quantifiers, but here even existential introduction is
A(a)
written as a rule with an inference line, in the form . Existential
(Ex)A(x)
elimination, in turn, is given as the following rule: “An expression (Ex)A(x)
can be replaced by A(η) in which η is a letter that does not yet occur.”
(p. 121). Nothing but the fact that the Hilbert-Bernays axioms figure in no
way in the early handwritten version of Gentzen’s thesis prevents us from
concluding: here is the way to natural deduction!

6. From Heyting’s intuitionistic axioms to natural deduction

Heyting’s axiomatization has eleven propositional axioms, not quite like

those of Principia Mathematica as has been often said, but more like the
Hilbert-Bernays axioms in which all connectives and quantifiers are present.
The original version of 1928 is lost, but the published version refers to a
paper by V. Glivenko of 1928 that has an elegant set of axioms.
Implication elimination was not the only propositional rule, but there
was also an explicit rule of conjunction introduction. It could be dispensed
with, but the proofs would then be ‘even more intricate’ (verwickelt). A little
reconstruction shows that this is the case:
Heyting’s first theorem is · a ∧ b ⊃ a, and the proof he gives is
[2.14] · a ⊃ · b ⊃ a : ⊃ :
[2.12] · a ∧ b ⊃ · b ⊃ a · ∧ b · [2.15] b [should be a]
16 J. von Plato

The notation uses abbreviations and the dot notation in place of parentheses,
with the axioms referred to by numbers in square brackets. When axioms
are written, they are indicated by a double turnstile, and a more detailed
rendering of Heyting’s derivation, by his own conventions, would be:
[2.14] · a ⊃ · b ⊃ a
[2.12] · a ⊃ · b ⊃ a : ⊃ : a ∧ b ⊃ · b ⊃ a · ∧b
·a∧b ⊃ ·b ⊃ a · ∧b
[2.15] · b ∧ · b ⊃ a · ⊃ a
The axioms in use here are:
[2.14] · b ⊃ · a ⊃ b
[2.12] · a ⊃ b · ⊃ · a ∧ c ⊃ b ∧ c
[2.15] · a ∧ · a ⊃ b · ⊃ b
A complete proof without abbreviations has eleven lines, the formulas writ-
ten for clarity with parentheses. I have organized it so that each rule instance
is preceded by its premisses in a determinate order. [1.3] is implication elim-
ination, [1.2] conjunction introduction, the rest are axiom instances:
1. [2.14] a ⊃ (b ⊃ a)
2. [2.12] (a ⊃ (b ⊃ a)) ⊃ (a ∧ b ⊃ (b ⊃ a) ∧ b))
3. [1.3] a ∧ b ⊃ (b ⊃ a) ∧ b
4. [2.11] (b ⊃ a) ∧ b ⊃ b ∧ (b ⊃ a)
5. [1.2] (a ∧ b ⊃ (b ⊃ a) ∧ b) ∧ ((b ⊃ a) ∧ b ⊃ b ∧ (b ⊃ a))
6. [2.13] ((a ∧ b ⊃ (b ⊃ a) ∧ b) ∧ ((b ⊃ a) ∧ b ⊃ b ∧ (b ⊃ a))
⊃ (a ∧ b ⊃ b ∧ (b ⊃ a))
7. [1.3] a ∧ b ⊃ b ∧ (b ⊃ a)
8. [2.15] b ∧ (b ⊃ a) ⊃ a
9. [1.2] (a ∧ b ⊃ b ∧ (b ⊃ a)) ∧ (b ∧ (b ⊃ a) ⊃ a)
10. [2.13] ((a∧b ⊃ b∧(b ⊃ a))∧(b ∧ (b ⊃ a) ⊃ a)) ⊃ (a ∧ b ⊃ a)
11. [1.3] a ∧ b ⊃ a
A reader who undertakes to turn Heyting’s proof outlines into formal deriva-
tions will soon notice two things:
1. There is so much work that one lets it be after a while.
2. The way to proceed instead becomes clear, because one starts to read
Heyting’s axioms and theorems in terms of their intuitive content.
From Axiomatic Logic to Natural Deduction 17

Formally, axioms or previously proved theorems of the forms a ⊃ b and a

lead to another theorem b by a step of implication elimination. In practice,
one sees implications a ⊃ b as rules by which b can be concluded whenever
a is at hand, as Heyting seems to have done, considering his very sketchy
derivations. The situation is particularly tempting if the major premiss is
an axiom such as a ⊃ a ∨ b. The step to deleting the axiom and implication
elimination, to conclude a ∨ b directly from a, is short, as in:
1. a ⊃ a ∨ b
2. a
3. [1.3] a ∨ b
It would suffice to delete line 1 and to change the reference to rule [1.3] into an
explicit disjunction introduction rule. Thus, with the experience of actually
using Heyting’s logical calculus for the construction of formal proofs, one
reads his first theorem a & b ⊃ a as: From a & b follows a. Axioms such as a ⊃
a ∨ b, b ⊃ a ∨ b, and (a ⊃ c) ⊃ ((b ⊃ c) ⊃ (a ∨ b ⊃ c)) turn into what are now
the familiar natural deduction rules of disjunction. With these observations,
we have arrived at the point in which Gentzen began his development of
natural deduction: He had studied carefully Heyting’s deductive machinery
and like anyone who does that, cut short the ever-repeating pattern that
combines an axiom instance A → B and the antecedent A to arrive at the
conclusion B: the same passage is effected more easily by a rule that gives
the conclusion B directly from the premiss A.

References
Gentzen, G. (1932) Über die Existenz unabhängiger Axiomensysteme zu unendlichen
Satzsysteme. Mathematische Annalen 107:329–250.
Gentzen, G. (1934-35) Untersuchungen über das logische Schliessen. Mathematische
Zeitschrift 39:176–210 and 405–431.
Gentzen, G. (2008) The normalization of derivations. The Bulletin of Symbolic Logic
14:245–257.
Glivenko, V. (1928) Sur la logique de M. Brouwer. Academie Royale de Belgique, Bulletin
de la Classe des Sciences 5:225–228.
Hertz, P. (1923) Über Axiomensysteme für beliebige Satzsysteme. Teil II. Mathematische
Annalen 89:76–102.
Heyting, A. (1930) Die formalen Regeln der intuitionistischen Logik. Sitzungsberichte
der Preussischen Akademie von Wissenschaften, Physikalisch-mathematische Klasse, pp.
42–56.
Heyting, A. (1931) Die intuitionistische Grundlegung der Mathematik. Erkenntnis 2:106–
115.
18 J. von Plato

Hilbert, D. (1931) Beweis des tertium non datur. Nachrichten von der Gesellschaft der
Wissenschaften zu Göttingen, mathematisch-philosophische Klasse, pp. 120–125.
Hilbert, D. and W. Ackermann (1928) Grundzüge der theoretischen Logik. Springer.
Hilbert, D. and P. Bernays (1934) Grundlagen der Mathematik I. Springer.
Jaśkowski, S. (1934) On the rules of supposition in formal logic, as reprinted in S. McCall
(ed.), Polish Logic 1920–1939, Oxford U. P. 1967, pp. 232–258.
von Neumann, J. (1927) Zur Hilbertschen Beweistheorie. Mathematische Zeitschrift 26:1–
46.
von Plato, J. (2012) Gentzen’s proof systems: byproducts in a program of genius. The
Bulletin of Symbolic Logic 18:313–367.
Schroeder-Heister, P. (2002) Resolution and the origins of structural reasoning: early
proof-theoretic ideas of Hertz and Gentzen. The Bulletin of Symbolic Logic 8:246–265.

Jan von Plato

University of Helsinki
Helsinki, Finland
jan.vonplato@helsinki.fi