The soundness theorem for a deductive system such as the
one we have been using states that every provable sequent
is valid according to the given definition of semantics
for the language. That is, if there is an interpretation
on which all formulae in X are true but a
conclusion C is not, then there is no way to
derive C from X by the rules of the
calculus.
Recall that every line of every natural deduction proof
represents a sequent: the assumption numbers on the left
refer to its premises, and the formula written in the
main part of the line is its conclusion. We wish to show
that every sequent capable of occurring in a proof is valid.
Soundness is proved rather laboriously by showing that
none of the rules is capable of introducing an invalid
sequent into a proof if all the previous lines represent
valid sequents. Trivially, the rule of assumptions cannot
produce an invalid sequent, since it always results in
something of the form A
TS A, so all derivations start
out valid. Now if there were a proof of an invalid
sequent, then that proof would have to have a first
invalid line. All lines above this would be valid, but at
this point the derivation would have gone wrong. The first
"bad" line would have to come from "good" lines by one of
the rules. Hence what has to be shown for soundness is
that this can never happen—that no rule can ever
transform valid inputs into an invalid output. This is a
matter of grinding through all the cases and is tedious
but not especially difficult.
Click to start grinding!
Case 1: Conjunction
|
|
X TS A |
Y TS B |
ANDI |
X, Y TS A AND B |
|
|
|
As usual, the easiest connective to deal with is
conjuinction. Could the first invalid sequent in our
hypothetical proof be the output of the
rule ANDE, taking us
from A
AND B to A? No, because
if the input
sequent X
TS A AND B is valid,
then any interpretation satisfying all of X
has to satisfy A
AND B and so by the truth table
for AND it has to
satisfy A in particular, meaning
that X
TS A is also valid. By the
same reasoning, so
is X
TS B.
The case of ANDI is
similar. Suppose X
STS A
and Y
STS B both hold, and consider
any interpretation satisfying everything
in X ∪ Y. Since
it satisfies X and Y, it satisfies
both A and B, and therefore (by truth tables
again) their conjunction A
AND B. Hence, if the inputs to a step
of ANDI are both valid, so is its
output.
|
Case 2: Implication
X TS A IMP B |
Y TS A |
IMPE |
X, Y TS B |
|
|
|
X, A
TS B |
IMPI |
X TS A
IMP B |
|
|
|
The rule IMPE is similarly
easy. If X STS A
IMP B
and Y
STS A,
and I is an interpretation
satisfying X
∪ Y, then I has to
satisfy both A
IMP B
and A, and thus
satisfies B as well; so
X, Y
STS B.
Now suppose X
TS A IMP B is invalid and
comes by a step of IMPI from a
sequent X, A
TS B . Since it is invalid, there
must be an interpretation I which
satisfies all formulae in X but fails to
satisfy A
IMP B. The only way for this to happen
is for I to satisfy A but not
satisfy B, which means that the input
sequent X, A
TS B . is likewise invalid.
|
Case 3: Negation
X TS A |
Y TS NOT A |
NOTE
|
X, Y TS BAD |
|
|
|
X, A
TS BAD |
NOTI |
X TS
NOT A |
|
|
|
|
X TS NOTNOTA |
NOTNOTE |
X TS A |
|
|
The rule NOTE cannot be the
first to introduce an invalid sequent into a proof, since
in order for its
output X, Y TS
BAD to be invalid, there has to be an
interpretation I which satisfies
everything in X
∪ Y. Obviously, I
cannot satisfy both A
and NOTA, so it also
provides a counter-example showing that one of the two
input sequents is invalid.
Nor can NOTI be the culprit,
since if its
input X, A TS
BAD is valid, there is no interpretation
satisfying all of X while also
satisfying A. But that means that any
interpretation which satisfies X must
satisfy NOTA, so the
output is also valid.
RAA does not require a separate argument, since it is
merely NOTE
and NOTI telescoped together, so
it is covered by the above
cases. NOTNOTE does need to be
noted, but it is trivial because its input and output
formulae are completely equivalent, so neither can follow
validly from X unless the other does.
|
Case 4: Disjunction
X TS A OR B |
Y, A
TS C |
Z, B
TS C |
ORE |
X, Y, Z
TS C |
|
|
|
|
|
ORI is unproblematic. By
inspection of the truth table, we see that
if X
STS A and I
satisfies X, then I satisfies the
disjunction A
OR B. The same holds if the input
formula is B.
The most complicated propositional rule is ORE. We may suppose all three of its
input sequents are valid. That is, X STS A OR B,
Y, A STS C, and
Z, B STS C.
We need to show on these suppositions that the output is valid:
X, Y, Z
STS C. Consider, then, any
interpretation giving the value 1 to every formula
in X, Y or Z. Since this makes
everything in X true it must make
A OR B true, in
which case by the truth table for 'or' it either
gives A the value 1 or gives B the value
1. If it gives A the value 1 then it gives 1 to
everything in Y and to A, so by the
second of our three suppositions it gives 1
to C. Similarly, if it gives 1 to B, by
the third supposition it gives 1 to C. Either
way C gets the value 1. In sum, any
interpretation verifying all of X, Y
and Z verifies C which is what we want.
|
Case 5: Universality
|
|
X
TS A |
ALLI |
m not in any formula
in X |
X TS
ALLv
Avm |
|
|
|
The inference encapsulated in the
rule ALLE, that what is
universally true is true in particular of whatever object
is named by a term t, looks obviously valid, but in
fact showing the semantic correctness
of ALLE is less simple than it
looks. Suppose, then, that there is an
interpretation I on which the output
sequent fails. That is, every formula in X is
satisfied by I
but Atv
is not. We wish to show that ALLv
A is also false for
interpretation I. To obtain the truth
value of ALLv A, we choose
a name m which does not occur in A and find
an m-variant of I in which
of Amv
is false. Unsurprisingly, the m-variant which
falsifies Amv
is I[m←δ(t)]
where δ(t) is the object denoted by t
in interpretation I. There are two cases
to consider: either m is a sub-term of t, or
it is not. If it is, then t does not occur
in A, so the two formulae
Atv
and Amv
are exactly the same except that one has m exactly
where the other has t, so since the two terms
denote the same thing, the values of the formulae are the
same. If m is not a sub-term of t,
then m is the only term
in Amv
whose denotation changes between I
and I
[m←δ(t)] , so again it
is clear that all subformulae
of Amv
have the same values
in I[m←δ(t)]
as the corresponding subformulae
of Atv
have in I. Since every formula
in X is true for I
but ALLv A is false
for I, the input sequent is invalid as
required.
Now consider the rule
ALLI. Suppose
m is a name which occurs in A but not in
any formula in X. Let v be a variable not
occurring in A. Suppose X does not semantically entail
ALLv Avmso
there is an interpretation I which is a
model of X but for which
ALLv Avmis
false. By the modelling condition for the universal
quantifier, this means that there is
some m-variant J
of I for which A is
false. Since m does not occur in X,
and I is a model of X, it
follows that J is also a model
of X. That is, there is an interpretation for
which everything in X is true but A is
false, showing that X does not (semantically)
entail A.
|
Case 6: Existence
X
TS SOMEv
Avm |
Y, A TS B |
SOMEE |
m not in B or in
any formula in Y |
X, Y TS
B |
|
|
|
X TS Atv |
SOMEI |
X TS SOMEv A |
|
|
|
SOMEI preserves validity in much
the same way as ALLE. Suppose the
input is valid: that is, X
STS Atv
and suppose every formula in X is true
for I. Choose m as
before. Since Atv
is true for I,
Amvis
true
for I[m←δ(t)]
which makes SOMEv A true
for I. Since this holds for any
interpretation satisfying X, we
have X STS SOMEv A.
For SOMEI, suppose the inputs are
valid, and consider any interpretation I
satisfying X
and Y. Since X
STS SOMEv
Avmthere
is some m-variant J
of I fopr which A is
true. Since m does not occur anywhere in Y,
all formulae in Y are true for J,
but Y, A
STS B, so B is also true
for J. The name m does not occur
in B, so the value of m is irrelevant
to B, which is therefore also true
in I.
|
To recapitulate, the upshot of all the above arguments is
that no rule can be responsible for the first invalid line
in any proof. Since there is no first invalid line,
there are no invalid lines at all. Hence every provable
sequent is valid.