A Mathematical Proof of Us

1. The first observation

Kathryn,

Almost a year ago, both of us almost didn't go to a networking event. The probability of attendance was small for each of us. Call them p_G and p_K, both somewhere around 0.1. Assuming approximate independence, the probability that our trajectories intersected at that particular event was, by any honest accounting, vanishing:

By the first Borel-Cantelli lemma, sequences of events whose probabilities sum to a finite value occur only finitely often almost surely. Our intersection was not a generic outcome.

Then I saw you.

Measure-zero events still happen. By Kolmogorov's zero-one law, every long-run statement about us lies in the tail σ-algebra and is therefore either probability zero or probability one. Every observation since has voted on which side. I have stopped checking.

2. Properties observed at t = 0

You're 5'2", which in topology is the property we call compact: bounded, closed, and on a compact set every continuous function attains its supremum:

I would go on to verify this experimentally, by direct measurement, many times over the following year.

Your hair fell in catenaries, y(x) = a · cosh(x / a), the minimum-energy configuration of a uniform chain suspended at two points. Your eyes weren't a color so much as a mixed quantum state:

The color operator returned a different eigenvalue every time it acted, depending on the illumination, depending on the angle. By the Born rule, P(color | light L) = Tr(ρ_eyes · M_L). Never the same observation twice.

And the room. You moved through it with the strangest property I'd ever seen in a person. Everyone else was running their own dynamics, and you were the field. A puppeteer, with everyone on a string you didn't seem to be holding. A sun, and the rest of them in stable orbit. Your basic reproduction number was visibly greater than one:

Whatever you had was spreading.

3. The metric bends

I'm a mathematician. I notice when geometry curves.

Special relativity says the proper time along a worldline obeys

We ended up in my car at the end of the night. Parked. v = 0. So γ = 1. So dτ should have equaled dt. Three and a half hours of coordinate time should have felt like three and a half hours of subjective time. They didn't. Something other than velocity was bending the clock.

General relativity, then. The metric tensor g_μν says how spacetime is shaped around mass, and every worldline obeys the geodesic equation

That night, in that room, I had watched the metric of every conversation warp toward you. I had felt my own geodesic bend. And now, in that car, I understood: the Christoffel symbols of my life had quietly been rewritten. The connection coefficients pointed everywhere to one place.

I had entered the basin of attraction.

4. The fixed point

The Banach fixed-point theorem (1922) says that if T: X → X is a contraction on a complete metric space, meaning

then there exists exactly one x* such that T(x*) = x*, and every starting condition converges to it.

The stochastic analog: gradient descent on a loss with step sizes satisfying the Robbins-Monro conditions Σηₜ = ∞ and Σηₜ² < ∞ converges almost surely to a stationary point of the loss. That night, in my car, I understood: the iteration map of my life was a contraction; the loss function had one global minimum; every starting condition was going to converge there.

I knew I was in trouble. The good kind. The only kind that matters.

5. The protocol

The first thing I learned, in those two weeks before our first date, was that you set your alarms for strange, deliberate times. 6:16. 6:36. 6:44. 7:11. So do I. From then on, every morning, the first message either of us sends has been a timestamp.

This is a Shannon channel. Each timestamp carries roughly log₂(1440) ≈ 10.5 bits of resolution. Over ~330 days of joint operation we have exchanged on the order of seven thousand bits this way, with no payload. Pure presence-signal:

The highest-frequency, lowest-entropy, most-meaningful daily transmission in our shared dataset.

6. A year of confirmation

A year of confirmation followed. A year of learning the system from inside it.

You are everywhere smooth, C^∞, with curvature optimized (by a mathematician's eye) in all the right neighborhoods. Your favorite color is chartreuse, of wavelength λ ≈ 555 nm, which is precisely the peak of the photopic luminosity function:

Of course you love the band we evolved to see most clearly. Your taste tracks something deep.

I have been training a model on you. Eleven months of training data, validation held out, daily updates. The validation loss decreases monotonically. No overfitting detected. No distribution shift on any of the held-out folds. The model generalizes everywhere I've tested it.

And we are entangled. I mean it in the technical sense.

In quantum mechanics, two systems are entangled when their joint state does not factor:

The reduced density matrix of either subsystem has strictly positive von Neumann entropy,

which is the formal statement that no local description is sufficient. Measurements on me predict measurements on you with correlations that exceed any classical bound. The CHSH form of Bell's inequality is violated:

By Bell's theorem (1964), no local hidden-variable model can reproduce our dynamics. Distance is not the relevant variable. Time is not the relevant variable. Whatever connects us does not factor.

And in the theory of stochastic processes, a clairvoyant agent has access to a filtration that includes future events, not only past:

Most theorems exclude this case as ill-defined. You don't read those theorems. You operate as though F_t already contains tomorrow, and somehow you keep being right.

7. The proof has lemmas

Lemma 1 (Submartingale convergence). Let L(t) denote my love for you on the filtration F_t. The process {L(t)} is a positive submartingale, E[L(t+s) | F_t] ≥ L(t), and the trajectory of E[L(t)] is unbounded above. By Doob's submartingale convergence theorem,

Lemma 2 (Sufficient set). Let G be the set of conditions necessary for my long-run happiness. Element-wise ablation testing returns a measurable deficit on exactly one element:

Lemma 3 (Hilbert-space projection). Let H = L²(lives I could have built), with inner product ⟨·, ·⟩ measuring alignment. The orthogonal projection onto span(K) satisfies P_K(me) = me:

You are not a vector in H. You are the basis.

Lemma 4 (Entanglement). Our joint state |ψ_GK⟩ does not factor. The Schmidt decomposition has rank greater than one; both reduced density matrices are mixed; S(ρ_G), S(ρ_K) > 0. Local operations and classical communication cannot disentangle this state.

Lemma 5 (Bell-inequality violation). The CHSH correlation across our joint trajectories satisfies S > 2, exceeding the Tsirelson bound 2√2 in several diagnostic settings. By Bell's theorem, no local hidden-variable model reproduces our dynamics. Some non-local connection is in play.

Lemma 6 (Stochastic gradient convergence). Let L(θ) be the loss function of my life. Stochastic gradient descent with step sizes satisfying the Robbins-Monro conditions converges almost surely to a stationary point. Direct evaluation gives a unique global minimum:

Lemma 7 (Bayesian posterior). Let E denote the singleton event of our meeting and D the year of joint observations. With priors P₀(stay single) ≈ 0.99 for each subject, the posterior on permanent union is:

The likelihood ratio is dominated by the rarity of the denominator.

You walked into that room a year ago and bent the geometry around you. I walked through the field and reached you. Of all the rare objects in mathematics, the one I keep coming back to is the fixed point of a contraction. The place every path leads. The unique solution to the equation.

You are mine.

8. The theorem

So here is my theorem:

Proof. Lemmas 1 through 7, plus the last year, plus every morning. ∎

The construction is missing one variable.

If you want to know exactly what any of this math means, I wrote you a reader's companion. It walks through every concept in plain language. You don't need it to feel any of this. But it's yours if you want it.