Gravity as Information Written at a Boundary
Ram Kirpitch
Abstract
If gravity is information, then the gravitational potential must be an amount of information written at a boundary — and it must be written in a way that does not depend on what is on the other side. That is the informational form of the equivalence principle: a potential is universal, indifferent to the composition of what it acts on. This paper puts that requirement on a quantum processor and measures it.This is a follow-up to Gravity as Approach to a Rate–Resolution Bound (doi:10.5281/zenodo.21808410), and reports the hardware campaign that paper did not contain. In the parity–entanglement programme a crossing write changes a region's bias by ⟨P⟩ → ⟨P⟩|cos 2θ|, so that the amount of record it fixes, δS₂ = −ln|cos 2θ|, has the bias cancelled out of it: it is independent of the state of the region written on, and under Φ = −c²δS₂ it is the Newtonian potential, with θ = v_esc/2c. That independence, and not the mere existence of a correlation between regions, is what separates a gravitational potential from an arbitrary entanglement flow, and it is directly testable. We test it with an estimator that is decay-free by construction — the ratio R_ij(θ) = ⟨P⟩_i(θ)/⟨P⟩_j(θ) of two states carried through identical circuits, in which any depolarising factor cancels identically — and we test it against a deliberately introduced violation. On ibm_kingston, across fifteen independent measurements in two jobs, the sector-exchange write shows no drift: χ² = 12.77 on 15 degrees of freedom, p = 0.62, largest deviation 2.3σ. In the same jobs, and with the same estimator, a control write that is state-dependent by construction is resolved at p = 1.1×10⁻²⁰, and its drift is reproduced by a closed form with no fitted parameter at χ²/dof = 0.52, the largest predicted effect (16.74%) being recovered as 16.59%. The conserved membrane ⟨Z_A Z_B⟩ supplies a second, independent discriminator, dropping by 0.007 in the signal arm and by 0.071 in the control. We also report a measured correction to Eq. (3) of the parent paper: the identity ⟨P⟩ = Tr ρ_A² = e^−S₂ holds in the Gaussian sector but not in the qubit sector, where Tr ρ_A² = (1+⟨P⟩²)/2, confirmed tomographically to 0.3%. Exact density-matrix propagation shows the estimator cancels depolarising, dephasing, bit-flip and coherent write-angle error to machine precision, so a miscalibrated write can manufacture neither the null nor a violation; amplitude damping is the sole surviving channel, at 0.11γ. Every number is re-derived offline from the deposited counts by the accompanying script. Statements are labelled [P] proved, [K] known, [N] numerical or measured, [F] formal.