Hidden Costs When Quantum Computers Use Simple Error Models

In the paper's words

Quantum processors built from superconducting transmons operate as anharmonic oscillators with many energy levels, yet fault‑tolerant architectures model each device as a two‑level qubit subject to a discrete Pauli error set. This mismatch forces a continuous bosonic error model into a discrete representation, incurring hidden bookkeeping and physical costs. We analytically quantify four concrete overheads for a single transmon treated as a qudit of dimension d=12: (i) the minimal classical register entropy H_reg=log₂12=3.584963 bits, (ii) the binary‑encoding dimension overhead 4/3 with encoding efficiency η=0.896241, (iii) the entropy distortion introduced by Pauli twirling of an amplitude‑damping channel, Δ H=1.5×10^-3 bits at damping probability γ=10^-3 and the exact law Δ H=(3γ)/(2) for all γ, and (iv) the discarded Hilbert‑space weight for a thermal mode with mean occupation bar n=0.1, p_disc≈3.19×10^-13.

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

Numbers it reports

1. We analytically quantify four concrete overheads for a single transmon treated as a qudit of dimension $d=12$: (i) the minimal classical register entropy $H_{\mathrm{reg}}=\log_{2}12=3.584963$ bits, (ii) the binary‑encoding dimension overhead $4/3$ with encoding efficiency $\eta=0.896241$, (iii) the entropy distortion introduced by Pauli twirling of an amplitude‑damping channel, $\Delta H=1.5\times10^{-3}$ bits at damping probability $\gamma=10^{-3}$ and the exact law $\Delta H=\tfrac{3\gamma}{2}$ for all $\gamma$, and (iv) the discarded Hilbert‑space weight for a thermal mode with mean occupation $\bar n=0.1$, $p_{\mathrm{disc}}\approx3.19\times10^{-13}$.

   from Abstract

2. Derivation of the register entropy floor $H_{\mathrm{reg}}(d)=\log_{2}d$ for $d=12$ (Section 4.1).

   from 1. Introduction

3. Computation of binary‑encoding overhead and efficiency $\eta$ for $d=12$ (Section 4.2).

   from 1. Introduction

4. Exact calculation of entropy distortion $\Delta H$ under Pauli twirling, including the closed‑form law $\Delta H=\tfrac{3\gamma}{2}$ (Sections 4.3–4.4).

   from 1. Introduction

5. Register entropy: $$ H_{\mathrm{reg}}(12)=\log_{2}12 =\frac{\ln12}{\ln2} =\frac{2.484907}{0.693147} =3.584963\ \text{bits}. $$ Base truncation entropy: $$ S_{\mathrm{Base}}(12)=\ln\!\left(\frac{12}{2}\right)=\ln6=1.791759\ \text{nats}. $$ Conversion to bits using $\log_{2}e=1.442695$: $$ 1.791759\times1.442695=2.584963\ \text{bits}. $$ Thus the register‑entropy floor is $3.584963$ bits per mode per use, while the truncation of the bosonic tail contributes an additional $2.584963$ bits of entropy in the classical description.

   from 4.1 Register entropy, base truncation entropy, and register‑entropy floor

6. Dimension‑overhead ratio: $$ \frac{2^{m}}{d}=\frac{16}{12}=\frac{4}{3}\approx1.333333. $$ Encoding efficiency: $$ \eta=\frac{\log_{2}d}{m}=\frac{3.584963}{4}=0.896241. $$ Idle fraction: $$ 1-\eta=0.103759\approx10.4\%. $$

   from 4.2 Binary encoding overhead

How it is done

1. The bosonic Hilbert space of a single mode is $\mathcal{H}_{\mathrm{bos}}=\mathrm{span}\{|k\rangle\}_{k=0}^{\infty}$.

   from 3.1 Model of digitization

2. For a binary distribution $(p,1-p)$ we use the binary entropy $h_{2}(p)=-p\log_{2}p-(1-p)\log_{2}(1-p)$.

   from 3.2 Entropy quantities

3. All input numbers are stated with their source, and every arithmetic step is shown.

   from 4. Analysis

4. Larger cutoff $d=16$ (power of two).

   from Projections for larger design points

What would falsify it

1. Demonstrating experimentally that the environment entropy of the amplitude‑damping channel and its Pauli‑twirled surrogate are identical for all $\gamma$ would falsify the linear distortion law $\Delta H=\tfrac{3\gamma}{2}$.

   from 6.4 What would falsify the main claims

2. Constructing a binary encoding of a $d=12$ qudit that achieves $\eta\geq0.99$ and dimension‑overhead $\leq1.01$ would falsify the binary‑encoding inefficiency result.

   from 6.4 What would falsify the main claims

3. Measuring a discarded population for $\bar n=0.1$, $d=12$ that exceeds $10^{-9}$ would falsify the leakage estimate.

   from 6.4 What would falsify the main claims

Where the claim stops

1. The analysis does not include full fault‑tolerance lifecycle modeling, multi‑mode interactions, or hardware‑specific gate errors.

   from 6.1 Scope of the results

2. We do not model full fault‑tolerance lifecycles; the presented numbers can be added to broader resource budgets as bookkeeping terms.

   from 3.4 Accounting frame

3. Input 1: $d=12$ (modeling assumption).

   from 4.1 Register entropy, base truncation entropy, and register‑entropy floor

Where to read

1. Abstract182 words · 1 min
2. 1. Introduction223 words · 1 min
3. 2. Background and Related Work206 words · 1 min
4. 3. Methods161 words · 1 min
5. 4. Analysis322 words · 1 min
6. 5. Results201 words · 1 min
7. 6. Discussion320 words · 1 min
8. 7. Practical Implications and Engineering Pathways386 words · 2 min
9. 8. Conclusion121 words · 1 min
10. Version History211 words · 1 min
11. Provenance and AI Involvement225 words · 1 min