FockMap

Chapter 5: Checking Our Answer

In this chapter, you’ll verify the Hamiltonian numerically and confirm equivalence across encodings.

In This Chapter

Exact Diagonalisation

Before diagonalising, it’s helpful to separate two workflows:

In other words: symbolic form is ideal for constructing and simplifying the model; matrix form is ideal for spectrum checks and numerical validation.

To verify the qubit Hamiltonian, we construct its $16 \times 16$ matrix representation. Each Pauli string $\sigma_\alpha$ corresponds to a known matrix (the tensor product of its single-qubit Pauli matrices). The full Hamiltonian matrix is:

\[H = \sum_\alpha c_\alpha \cdot \sigma_\alpha\]

where $c_\alpha$ are the 15 coefficients from Chapter 4.

Diagonalising this matrix gives 16 eigenvalues, grouped by particle-number sector (since the Hamiltonian conserves particle number):

Sector ($N_e$) Dimension Eigenvalues (Ha, electronic)
0 1 $0$
1 4 $-1.2563,\ -1.2563,\ -0.4719,\ -0.4719$
2 6 $-1.8573,\ -1.3390,\ -0.9032,\ -0.9032,\ -0.6753,\ 0.0$
3 4 $-1.7282,\ -1.7282,\ -0.9438,\ -0.9438$
4 1 $-2.2001$

The ground state of the $N_e = 2$ sector is $E_0^\text{el} = -1.8573$ Ha. Adding nuclear repulsion:

\[E_0^\text{total} = E_0^\text{el} + V_{nn} = -1.8573 + 0.7151 = -1.1422 \text{ Ha}\]

Comparison with Known Results

The Hartree–Fock energy (single determinant $\lvert1100\rangle$) is:

\[E_\text{HF} = 2h_{00} + [00 \mid 00] = 2(-1.2563) + 0.6745 = -1.8382 \text{ Ha (electronic)}\]

The Full CI correlation energy is:

\[E_\text{corr} = E_\text{FCI} - E_\text{HF} = -1.8573 - (-1.8382) = -0.0191 \text{ Ha} \approx -12.0 \text{ kcal/mol}\]

This correlation energy — about 1% of the total energy but ~12 kcal/mol — is precisely what makes quantum simulation valuable. It captures the effect of electron–electron correlation that the single-determinant Hartree–Fock approximation misses, and it is this quantity that determines chemical accuracy in reaction energies, barrier heights, and binding affinities.

Cross-Encoding Comparison

The same Hamiltonian encoded under all five transforms:

Encoding Terms Max Weight Avg Weight Identity Coeff (Ha)
Jordan–Wigner 15 4 2.13 $-1.0704$
Bravyi–Kitaev 15 4 2.40 $-1.0704$
Parity 15 4 2.27 $-1.0704$
Balanced Binary 15 4 2.27 $-1.0704$
Balanced Ternary 15 4 2.40 $-1.0704$

All encodings produce the same number of terms with the same identity coefficient — as they must, since the identity coefficient equals $\text{Tr}(\hat{H})/2^n$, which is invariant under any unitary change of basis.

At $n = 4$, the encodings are too small for the weight advantages of BK and tree encodings to manifest. At $n = 16$ (8 spatial orbitals), the max single-operator weight would be 16 for JW but only 5 for BK and 4 for balanced ternary — a significant reduction.

All five encodings produce the same eigenspectrum to machine precision ($\lvert\Delta\lambda\rvert < 5 \times 10^{-16}$), confirming that the encoding is a unitary change of basis that preserves the physics exactly.

For more details on the encodings and their scaling properties, see Beyond Jordan–Wigner.

The Compare Encodings interactive lab reproduces this comparison as executable F# code.

With validation done, we can now look at how this Hamiltonian feeds directly into practical quantum algorithms.


Previous: Chapter 4 — Building the H₂ Qubit Hamiltonian Next: Chapter 6 — What Comes Next