The Exponential Arena: A Reconciled Quantitative Account of Hilbert-Space Growth with Qubit Number

In the paper's words

A recurring question in quantum information is what "space" the state of a quantum computer inhabits, and how that space grows as the number of qubits n increases. This paper treats the question quantitatively, formalizing the arena of an n-qubit register as the complex projective Hilbert space CP^d-1 with d = 2^n, and deriving the scaling of its real dimension, the classical memory required to store one state vector, the per-gate cost of exact state-vector simulation, the entanglement entropy available across bipartitions, and the sparsity of classically simulable structured subsets. All headline numbers are computed explicitly from stated inputs: a general $n = 50$ register requires 2⁵4 ≈ 1.80 × 10¹6 bytes ($16$ PiB) per stored state vector; a single gate touches 2⁵0 ≈ 1.13 × 10¹5 amplitudes; the central cut of a $100$-qubit maximally entangled state carries $50$ bits of entropy, and reproducing it with a matrix-product state requires bond dimension 2⁵0 and ≈ 2.5 × 10³2 parameters — worse than the raw state vector. [...]

The abstract, verbatim, and its opening sentence is the claim. Every line below is quoted from the text.

Numbers it reports

1. All headline numbers are computed explicitly from stated inputs: a general $n = 50$ register requires $2^{54} \approx 1.80 \times 10^{16}$ bytes ($16$ PiB) per stored state vector; a single gate touches $2^{50} \approx 1.13 \times 10^{15}$ amplitudes; the central cut of a $100$-qubit maximally entangled state carries $50$ bits of entropy, and reproducing it with a matrix-product state requires bond dimension $2^{50}$ and $\approx 2.5 \times 10^{32}$ parameters — worse than the raw state vector.

   from Abstract

2. This discipline matters because the literature is full of informal appeals to "$10^{100}$-dimensional Hilbert spaces" that are rarely checked against actual memory, time, and entropy budgets of concrete register sizes.

   from 1. Introduction

3. The cost of that approximation enters our memory budget in Section 4.2, where we adopt $16$ bytes per complex amplitude. [6] Polynomial-time computing over quadratic maps I: sampling in real algebraic sets (arXiv:cs/0403008v3).

   from 2. Background and Related Work

4. We engage with its spirit by treating classical simulation cost, state-space dimension, and entanglement entropy as geometric quantities of one and the same arena, rather than as incidental complexity measures. [12] QNFO: Spectral Benchmarking of Holographic Quantum Simulations (https://papers.qnfo.org/papers/spectral-benchmarking-of-holographic-quantum-simulations/).

   from 2. Background and Related Work

5. Benchmarking is where the exponential bites hardest in practice: verifying an $n$-qubit device against classical prediction is possible only while $2^n$ remains simulable, and Section 4.3's cost figures make that boundary explicit.

   from 2. Background and Related Work

6. Storing each amplitude in double precision costs $2 \times 64 = 128$ bits $= 16$ bytes, so the memory for one state vector is $$M_n = 16 \cdot 2^n \ \text{bytes}.$$ Definition 3.4 (Gate cost).

   from 3.1 The arena

How it is done

1. Our contribution is not a new theorem but a reconciled, fully explicit quantitative assessment: every number in Section 4 is derived from stated inputs with shown arithmetic, and every projection is labeled as such.

   from 1. Introduction

2. We use it as a data-volume benchmark: in Section 4.5 we compare the memory footprint of a single $n = 50$ state vector against the projected petabyte-scale data volume of such a survey, giving the exponential an operational feel. [4] A geometric interpretation of Milnor's triple linking numbers (arXiv:math/0110001v3).

   from 2. Background and Related Work

3. Following the reconciled-assessment methodology of [9], every quantitative claim below is either (i) derived in Section 4 from the definitions above with all arithmetic shown, or (ii) explicitly labeled a projection with stated assumptions and uncertainty bounds.

   from 3.2 Method of assessment

4. All inputs are stated here with sources: qubit counts $n$ are illustrative register sizes; the constants $64$ bits per double-precision float, $8$ bits per byte, $\ln 2 = 0.693147$, $\log_{10} 2 = 0.301030$, $1$ year $= 3.156 \times 10^7$ s, $1$ day $= 8.64 \times 10^4$ s are standard; the cosmological atom count $\sim 10^{80}$ is an order-of-magnitude input.

   from 4. Analysis

What would falsify it

1. Limitations, failure modes, and falsification criteria are discussed.

   from Abstract

2. Benchmarking is where the exponential bites hardest in practice: verifying an $n$-qubit device against classical prediction is possible only while $2^n$ remains simulable, and Section 4.3's cost figures make that boundary explicit.

   from 2. Background and Related Work

3. Hilbert-space dimensions: $D_{10} = 1024$; $D_{50} = 2^{50} = 10^{50 \times 0.301030} = 10^{15.0515} \approx 1.1259 \times 10^{15}$; $D_{100} = 2^{100} = 10^{30.1030} \approx 1.2677 \times 10^{30}$; $D_{300} = 2^{300} = 10^{90.3090} \approx 2.037 \times 10^{90}$, exceeding the baryon count of the observable universe ($\sim 10^{80}$) by ten orders of magnitude.

   from 4.1 Dimension of the arena

Where the claim stops

1. Useful quantum states, as in the structured optical targets of [9], do not sample the arena uniformly.

   from 6. Discussion

2. (i) The per-gate timings are projections resting on an assumed throughput $R = 10^{10}\ \mathrm{s^{-1}}$; real node performance varies by about a factor of $10$ in either direction, and algorithmic improvements (vectorization, gate fusion) change constants, not the $2^n$ scaling.

   from 6. Discussion

3. (ii) The survey comparison in Section 4.5 assumes $V_{\mathrm{survey}} = 10^{15}$ bytes; the ratio is only order-of-magnitude.

   from 6. Discussion

Where to read

1. Abstract233 words · 1 min
2. 1. Introduction321 words · 1 min
3. 2. Background and Related Work902 words · 4 min
4. 3. Methods415 words · 2 min
5. 4. Analysis1,105 words · 5 min
6. 5. Results199 words · 1 min
7. 6. Discussion461 words · 2 min
8. 7. Conclusion99 words · 1 min
9. Appendix A. Divergence report162 words · 1 min
10. Appendix B. Claim attribution166 words · 1 min