You may have seen headlines like “Mathematicians proved the existence of God”. I’ve seen it a couple of times. In this post, we will see two of such proofs: a very famous one by Godel (the first version of which appears in Godel’s papers from 1941) and another a little bit less famous by Hartshorne (published 1962).
We’ll use classical higher-order modal logic and first of all I have to explain what it means.
“Classical” in context of logic means that any statement is either true or false. If \(S\) is an arbitrary statement, it means that \(S\lor\neg S\) is true and this is an axiom. Here \(\lor\) means “or” and \(\neg\) means “not”. We refer to this as to the law of excluded middle.
We will also make use of symbol \(\land\) for “and”, \(\rightarrow\) for implication and \(\leftrightarrow\) for logical equivalence.
Two things about implication may be non-intuitive. First of all, according to laws of implication absolutely any statement follows from a false statement. For example, the following seemingly absurd statement is considered to be true: “\(1+2=14\) follows from the fact that \(2+3=1\)“. It may be uncomfortable to accept this kind of reasoning, but this is really a true implication. I’ll explain this in more details in one of the following blog posts, but for now, I can just say that the fact that we can derive arbitrary absurd from wrong statements doesn’t really affect anything. Since the assumption of implication is never true, such rule is never applied. It is still important to know this for understanding the following. We’ll refer to this as to paradoxical implication.
Let’s say that \(A\) and \(B\) are some statements. Statement \(A\rightarrow B\) is equivalent to \(\neg B\rightarrow \neg A\). You can either learn it in any basic logic text or just think about it for a bit and convince yourself that this equivalence is true. We’ll refer to this as to implication contrapositive rule.
If statement \(S\) depends on some object \(x\) and we know that such object always exists, we write it as \(\exists x. S(x)\). If statement \(S\) is true for arbitrary object, we write it as \(\forall x. S(x)\). The second classical assumption is that \(\neg\exists x.S(x)\leftrightarrow\forall x. \neg S(x)\). It is just an axiom and you can convince yourself that this makes sense by just exploring couple of examples.
In statement \(S(x)\), \(S\) is called predicate, while \(x\) is object (in a language of logic it is called “term”, but we will stick to more non-logician friendly terminology). First-order logic (the one which is studied in universities) allows us to apply quantifiers \(\exists\) and \(\forall\) for objects, but not for predicates themselves. We are also not allowed to apply predicates to other predicates. Higher-order logic differs from first-order logic in that it relaxes this requirement, so we may write statements like \(\forall S .\exists x. S(x)\).
Now the interesting part. Modal logic. As I said above, we assume that any statement is either true or false. Modal logic goes further and says that some of the statements can be necessarily true or false, and some of the statements are possibly true or false. We denote it with symbols \(\diamond\) and \(\Xi\) respectively:
- Statement is “just” true: \(S\).
- Statement is possibly true: \(\diamond S\).
- Statement is necessarily true: \(\Xi S\).
Unfortunately, I am not using a standard notation here. “Necessary truth” is traditionally denoted by a shallow square, but this symbol is unavailable in the tool which I use for writing formulas. I hope I’ll be able to fix it in the future and get back to this article to change it.
One way to think about these three modes of truthfulness is to assume that we have many possible worlds. They can be imaginary worlds or some physical alternative realities. We live in one of these worlds. Given this assumption we can interpret these formulas this way:
- \(S\): the statement is true in our world, but not necessarily in others.
- \(\diamond S\): the statement may be true in some of the worlds, but not necessarily in our world. What is important is that we can imagine the world where \(S\) is true.
- \(\Xi S\): the statement is true in all possible worlds we can imagine.
Modal logic has a number of axioms:
M1. \(\neg \Xi S \leftrightarrow \diamond \neg S\). If it is not true that something is true in each world, then there may exist a world where it is false. We can treat this first axiom as a definition of \(\diamond\) through \(\Xi\).
M2. \(\Xi S \rightarrow S\). If something is necessarily true, then it is true in our world as well.
M3. \((\Xi (S \rightarrow T)) \rightarrow ((\Xi S) \rightarrow (\Xi T))\). If in all worlds \(T\) follows from \(S\), then if in all worlds \(S\) is true we also know that in all worlds \(T\) is true as well.
M4. If \(S\) can be proven mathematically (which includes the case when \(S\) is an axiom), then \(\Xi S\).
M5. \(\diamond \Xi S \rightarrow \Xi S\). If \(\diamond\Xi S\) is true, then it means that there exists some world where it is known that \(\Xi S\). If we could transfer ourselves to that world, we would know that \(S\) is necessarily true. So we conclude that \(\Xi S\).
Application of implication contrapositive rule to M5 gives us formula \(\diamond \neg S \rightarrow \Xi \diamond \neg S\). Denoting \(\neg S\) by \(T\) we get an equivalent statement of the same axiom:
M5′. \(\diamond T \rightarrow \Xi \diamond T\)
Modal logic often uses other axioms as well, but we wouldn’t need them.
Now we are ready to prove the existence of God following Hartshorne. I’ll loosely follow the paper by Small here.
Let’s denote statement “God exists” by \(g\). Hartshorne assumed two axioms:
HA1. \(\diamond g\). God may exist (he assumed that even atheists can imagine the existence of God in some alternative reality).
HA2. \(\diamond\neg g \rightarrow \neg g\). It is quite a pessimistic assumption: from the fact that it is possible that God doesn’t exist in some other reality, we immediately conclude that it doesn’t exist in our own reality. (In fact, I am tricking you right now and abusing the initial axiom, but I’ll explain it a little bit later).
Now we are ready for the proof. Below if I don’t mention explicitly where we derive a formula from, it implies that we use the previous formula.
H1. First let’s substitute \(\neg g\) to M5′ axiom: \(\diamond\neg g \rightarrow \Xi\diamond \neg g\). Note that we are not making assumptions about God here, only using tautologies of modal logic.
H2. Substitute \(\Xi g\) to the law of excluded middle and use M1: \((\Xi g)\lor(\diamond \neg g)\).
H3. Applying implication H1 to H2 we are getting \((\Xi g) \lor (\Xi\diamond\neg g)\).
H4. Since we assume HA2 as an axiom, we can apply rule M4 to get \(\Xi(\diamond\neg g \rightarrow \neg g)\)
H5. By M3: \((\Xi\diamond\neg g) \rightarrow (\Xi\neg g)\)
H6. H5 and H3 together gives us \((\Xi g) \lor (\Xi\neg g)\). This eliminates the possibility of accidental God existence. Either God exists in all possible worlds or it cannot exist in any of them.
H7. Axiom M1 is equivalent to \(\neg\neg\diamond g\) (two negations cancel each other), to which we can apply M1 to get \(\neg\Xi\neg g\).
H8. From H7 we know that term \(\Xi\neg g\) in H6 cannot be true. So the only possibility for H6 to be true (and we know that it is true because we proved it) is that \(\Xi g\) is true. This shows that God necessarily exists which completes the proof.
Does this really prove that God exists? It depends if you agree with Hartshorne’s axioms and axioms of modal logic. In fact, this proof doesn’t even say what “God” is. It can be applied for any statement which matches the same set of axioms. For example, substitute “God doesn’t exist” for \(g\) and the same reasoning would apply for the proof that God doesn’t exist.
Also, it is interesting to note that axiom HA2 is equivalent to statement \(g \rightarrow \Xi g\) (using implication contrapositive and M1) which reads as “if God exists in our world, it must exist in any possible world”. This statement looks much stronger than the original axiom and seems to be less probable to be true, even though these two formulations are totally equivalent! Actually, Hartshorne used exactly this “less pessimistic” axiom, but I chose to change it to make it look less shaky at least initially.
Can this proof be considered mathematical? Once again it depends on what you call “mathematics”. It definitely doesn’t use any advanced mathematical concepts and is quite trivial. The only bit of mathematics here is the usage of formulas and a strict agreement on axioms and what logical derivations we consider as acceptable. This is what mathematics is actually is: it is just an agreement on rules which we use and a convenient language which helps our reasoning. Exactly the same argument could be repeated without any formulas, but then it would be much more difficult to follow. In fact, originally a kind of this proof was presented informally by Anselm in XI century and only in 1962 it was made formal by Hartshorne (although honestly, I find the relation to be quite loose).
Does this proof have any value? Of course! It is at least a nice exercise in modal logic and an interesting model you can think of. As was discussed in the post on models, even if some model is not correct, it is still able to give food for thoughts. In this particular case, a formal model made obvious how shaky Anselm’s argument was (which was clear anyway, but this model makes the proof and the vision of its weaknesses accessible to non-philosophers).
Now after a short warm-up with modal logic we are ready to discuss Godel’s proof. In comparison to Hartshorne’s proof, this proof also tries to explain what God actually is. Informally speaking, Godel assumes that each imaginable object has a set of properties and is uniquely described by that set (here I don’t assume any definition of “set” in a set-theoretic sense). This reflects what physicists do when they introduce some object. For example, in order to explain what “electron” is you have to specify its properties: its charge, mass, diameter, etc.
Each property is binary: an object can either have a property or not, it cannot have some degree of a property. If object \(x\) has property \(\phi\) we denote it as \(\phi(x)\). This is not limiting at all since properties can be arbitrarily precise (like “an object is 1.23 nanometers in diameter”), but just a consequence of the syntax we use. Also, some of the properties are “positive” and some are not. If \(\phi\) is positive we denote it as \(P(\phi)\). What “positive” means is not specified, but God is such an object that has all positive properties.
It is possible that for some property \(\phi\) there is no object which has this property (consider property “this object does not exist”). If this is not the case and there are some object \(x\) such that \(\phi(x)\) holds, we call \(\phi\) instantiable and \(x\) an instance for property \(\phi\).
In the following, I’ll follow the notation and statements order of Wikipedia.
Axiom 1. \((P(\phi)\land \Xi \forall x . \phi(x)\rightarrow \psi(x)) \rightarrow P(\psi)\)
Formula \(\forall x . \phi(x)\rightarrow \psi(x)\) means that property \(\psi\) follows from property \(\phi\): any object which has property \(\phi\) will also have property \(\psi\). We can also call \(\psi\) a subproperty of \(\phi\). As an example consider properties “Bigger than 1 meter” and “Bigger than 1 millimeter”. The second clearly follows from the first one. What Axiom 1 says is that from positive property only positive properties can follow. It cannot be the case that some positive property implies a negative one.
Axiom 2. \(P(\neg\phi)\leftrightarrow\neg P(\phi)\)
For any property \(\phi\) we have an opposite property denoted by \(\neg\phi\) which reads just as absence of \(\phi\). Axiom 2 says that if the presence of some property is a positive thing than absence of it is not and vise versa. It means that we cannot have neutral properties (or we just don’t consider such properties).
Theorem 1. For any positive property it may exist an object which has this property (or in other words, any positive property is instantiable): \(P(\phi)\rightarrow \diamond \exists x . \phi(x)\)
Proof: Let’s assume that property \(\phi\) is non-instantiable. Then statement \(\phi(x)\) is always false and any statement follows from it by paradoxical implication. So we have statement \(\forall x. \phi(x)\rightarrow \psi(x)\) for arbitrary property \(\psi\). This means that all of the properties are subproperties of \(\phi\) together with their opposites. If \(\phi\) was positive that would contradict Axiom 2. This concludes that no positive property can be non-instantiable. QED
Definition 1. \(G(x)\leftrightarrow \forall\phi . (P(\phi)\rightarrow \phi(x))\)
\(G\) is a property which we’ll call “God-like”. This definition says that a property of being God-like implies having all positive properties (and no negative properties, otherwise it would contradict Axiom 2).
Axiom 3. \(P(G)\) (being God-like is a positive property).
Theorem 2. \(\diamond \exists x . G(x)\) (it is possible that God-like object exists at least in some world).
Proof: This theorem is trivial since God-likeness is a positive property and we know that all positive properties are instantiable by Theorem 1. QED
Definition 2. \(\phi \;\mathrm{ess}\; x \leftrightarrow (\phi(x)\land (\forall \psi . \psi(x)\rightarrow \Xi \forall y . \phi(y)\rightarrow\psi(y)))\)
This defines the essence of an object. Informally speaking, the essence of an object is a property which determines an object uniquely up to a world. Let’s say that \(x\) and \(y\) are the same objects but in the different worlds. We cannot write \(x=y\) because these two objects are still not exactly the same—they live in different worlds. But if you had a way to bring this objects in the same world, they would become indistinguishable and effectively the same physical entity. In order to capture this idea, Godel introduced “essences”—properties which describe objects uniquely within the world.
Given this informal reasoning, you now can understand what the formal definition means. It just says that if some property \(\phi\) is the essence of an object \(x\), then any other object \(y\) which has the same essence \(\phi\), would have exactly the same set of properties in any world imaginable. As I said before, the idea here is that properties describe objects uniquely, but belonging to a world is not a property which can be used to identify objects across the worlds. Definition of essence does the trick.
Axiom 4. \(P(\phi)\rightarrow \Xi P(\phi)\) (if some property is positive, it is positive in all imaginable worlds).
Theorem 3. \(G(x) \rightarrow G\;\mathrm{ess}\; x\) (if \(x\) is God-like, then \(G\) is its essence)
Proof: We know by definition that being God-like determines all of the properties of an object uniquely (it is all positive properties). Since we know that positivity of a property is the same in all of the worlds (by Axiom 4), it means that all of the God-like objects have exactly the same properties in all of the worlds. This is what essence is all about (although it might require some reflection). QED
Definition 3. \(E(x) \leftrightarrow \forall \phi .( \phi \;\mathrm{ess}\; x \rightarrow \Xi \exists y . \phi(y))\)
Property \(E\) is called necessary existence. If object \(x\) necessarily exists, it means that in any imaginable world exists some object with the same essence. It is instructive to compare this with \(\Xi\) operator. While \(\Xi\) tells that some logical statement is true in every possible world, \(E\) makes a statement that some object exists in every possible world with exactly the essence?of a given object.
Axiom 5. \(P(E)\)
This is another controversial axiom which says that necessary existence is a positive property. One way to think about it is that positivity of a property is not about being “better”, but about being “greater” in a sense of being bigger, smarter, more powerful, containing more energy, having bigger speed, etc. These examples are arbitrary, nothing actually says that being “greater” cannot mean “smaller in size”, but if we continue this intuitive idea, then we can think that the more worlds the object can exist in, the greater it is. This is what this axiom is about.
Theorem 4. God exists (more precisely, in any world exists a God-like object): \(\Xi\exists x. G(x)\)
Proof: Since a God-like object has all positive properties, it also has the property of necessary existence, which gives us formula $$\forall x.G(x)\rightarrow E(x)$$
If we combine this with theorem 2 we get $$\diamond \exists x. G(x)\land E(x)$$
By definition of essence $$\diamond \exists x. G(x)\land \Xi\exists y.G(y)$$
If some object has some two properties, it definitely has one of them, so we get $$\diamond \exists x. \Xi\exists y.G(y)$$
or, since \(x\) is not used in this formula anymore, we drop it and get $$\diamond\Xi\exists y.G(y)$$
Applying axiom of modal logic M5 we get the statement of the theorem. QED
Did this prove that God really exists? Not really. At least not in a popular religious way. Next time when you see in a newspaper that “scientists proved that God exists”, you know what scientists really did and how it is related to the God which you or your neighbor worship. Most likely, not related at all. And at best it is not more than a language play by formal rules.
Once again, even though this proof doesn’t tell us much about the reality in which we live and that each of the axioms can be criticized, I find this proof is still quite elegant. The play with essences is smart and may be even used in absolutely unrelated modal logic derivations.
In order to get a better feeling of this proof and modal logic I suggest you to solve the following exercise:
Exercise. Let’s assume that the definition of essence is formulated as
$$\phi \;\mathrm{ess}\; x \leftrightarrow (\forall \psi . \psi(x)\rightarrow \Xi \forall y . \phi(y)\rightarrow\psi(y))$$
and all of the axioms are still the same. Prove that in this case, Godel’s axioms are self-contradictory. (Hint: start by proving that a non-instantiable property is the essence for any object and get a contradiction from this).