Skip to main content
Sandbox Physics

FR13-C · A method needs more than one test

Map Electronic-Structure Errors Along a Bond

Does a small energy error also guarantee a good bond force? A force depends on how quickly the energy changes with separation. A small energy offset can have a steep slope, and an accurate-looking energy at one geometry can hide a poor dissociation curve. Compare energies and their derivatives across several geometries.

Compare several electronic methods against same-basis FCI along a fixed hydrogen stretching grid. Inspect energy errors, force errors and the occupations that signal a changing electronic state.

Interactive modelMap Electronic-Structure Errors Along a Bond
Selected energy error0.1830.183
Selected force error−0.0788-0.0788
Second natural occupation0.6440.644
Separation in bohr44

Saved PySCF calculations · STO-3G

Where each method begins to struggle

RHFUHFFCIMP2PBE · offline

A force tests the slope

Energy and force have different units. Read each comparison against its own axis.

Units, density and calculation detailsTr⁡(PS)=2\operatorname{Tr}(PS)=2

All total energies include nuclear repulsion. FCI solves the same finite Hamiltonian as HF. The saved calculation includes its geometry, method, density matrices and numerical diagnostics.

Learn how the calculation works, from orbitals to a broken bond

Physics tutorial

Map Electronic-Structure Errors Along a Bond

BackgroundAn electronic method predicts a whole potential-energy curve. Its slope sets the force on the nuclei, so one accurate energy cannot certify the curve.

Why it mattersComparing the same geometries and basis isolates differences between electronic methods.

Start with the essentials

Focus question
Does a small energy error also guarantee a good bond force?
One-sentence intuition
Force error is the negative slope of energy error along the same geometric coordinate.

Core mathematical model

Compare the slope as well as the height

ΔEm(R)=Em(R)−EFCI(R),ΔFm(R)=−d ΔEmdR\Delta E_m(R)=E_m(R)-E_{\rm FCI}(R),\qquad \Delta F_m(R)=-\frac{d\,\Delta E_m}{dR}

All methods use the same nuclear coordinate and include the same nuclear repulsion. Energy and force errors therefore answer different questions.

A controlled numerical derivative

F(R)≈−E(R+h)−E(R−h)2hF(R)\approx-\frac{E(R+h)-E(R-h)}{2h}

The reference record fixes the derivative step and stores the method and software version. Decreasing the step checks numerical sensitivity.

Common difficulties

One accurate point

Typical misconceptionA small energy error at equilibrium guarantees an accurate dissociation curve.

Better mental modelA whole curve can bend incorrectly despite passing through one accurate point. Forces and occupations reveal changes that a single energy misses.

Run the experiment

  1. 01

    Predict before stretching

    Choose Near the energy minimum, then Correlation grows. Say what you expect the density and the energy difference to do before changing the preset.

    What to observe: The molecular density separates into two lobes, and the restricted description loses accuracy as the bond stretches.
  2. 02

    Follow the second diagnostic

    Select each method in turn. Compare the energy-error map with the force-error graph at the same geometry.

    What to observe: A method can have a modest energy error and a noticeable force error. The latter follows the slope of the former.
  3. 03

    Save a reproducible comparison

    Use Save calculation at two separations. Compare the saved basis, geometry, density matrices and total energies.

    What to observe: Every record identifies the finite electronic model. Comparing it with a larger basis tests a separate source of error.

Related experiments and articles

Molecular electrons and breaking bonds · Reading path

Orbitals, self-consistency, correlation and dissociation

Articles on this path that use this experiment, followed by the other experiments they suggest.

  1. ArticleWhen a small correlation correction becomes too large
  2. ArticleWhat density functional theory changes
  3. ArticleAn accurate energy can still give the wrong bond force
  4. ArticleA fractional-spin test for a broken bond
FeedbackSomething broken, confusing, or missing on this page?Tell us
What kind of feedback?

Sent with this page’s address, its Lab ID, the site language and your device type. Please leave out personal details. Feedback is kept for up to 12 months.Privacy policy