Atomlyze user manual

Last revised 13 October 2026 (PKT). The Help link under each tool opens its section of this manual. The manual describes the version shown at the foot of every page, and the revision history at the end lists what changed and when.

Contents

1. What Atomlyze does
2. Your files stay on your computer
3. A first session
4. Habits that apply to every tool
5. Preparing calculation inputs
6. Reading calculation outputs
7. Spectra and densities of states
8. Trajectories and RMSD
9. Viewing structures
10. Wavefunctions: orbitals, charges, bond orders and excited states
11. Figures for papers and theses
12. When something does not work
13. What to cite
14. Revision history

1. What Atomlyze does

Atomlyze handles the routine parts of computational chemistry work. It writes input files for ORCA and Gaussian, pulls energies and properties out of their outputs, plots spectra and densities of states, computes RMSD along a trajectory, shows structures and orbitals in 3D and calculates atomic charges from a wavefunction.

It runs in a web browser, so there is nothing to install, and it behaves the same on Windows, macOS and Linux. Each of the eleven tools reports what it read from your file. When a file is ambiguous, the tool refuses it and says why rather than guessing.

Atomlyze does not run the calculations themselves. ORCA or Gaussian does that; Atomlyze prepares the inputs and helps you read and present the results.

2. Your files stay on your computer

Everything Atomlyze does happens inside your browser. Your files are never uploaded, nothing is stored on a server, and no usage statistics are collected. The first time you run a tool, the browser downloads the program code, a few tens of megabytes that are mostly the Python runtime, and keeps it for later visits. Once a page is open it keeps working if the network drops (tested in Chrome and Firefox; not yet in Safari).

If you work with unpublished results or confidential structures, closing the tab is enough to clear them.

3. A first session

1. Choose a tool. On a wide screen the tools are listed on the left; on a phone, use the Choose a tool list. They are grouped by task, such as “Read calculation outputs”, and each has its own address (for example atomlyze.com/tools/properties/) that you can bookmark.
2. Read the short notes under the tool’s title: what it reads, which conventions it follows, and what it refuses.
3. Choose your file with the file button, or drag it onto the panel.
4. Press Process file, or Generate input in the two input writers. A progress line shows what is happening.
5. Read the result and use the download buttons.

Try an example loads a small public or synthetic file, so you can see what a tool produces before using your own data.

Each tool shows the settings most people need. The rest are under Advanced options. The defaults suit most work, and the notes with every result list the values that were used.

4. Habits that apply to every tool

Charge and multiplicity are always yours to check. The input writers take them from the file when it states them; otherwise they suggest charge 0 and the lowest multiplicity for the electron count. Both stay visible and editable, and they are checked against the electron count. Pseudopotentials, scaling factors and program keywords are never filled in for you.

Units are stated. Coordinates are in ångström unless a format says otherwise, and every exported table carries its units. Energies are converted with the CODATA 2018 value 1 Eh = 27.211386245988 eV.

Missing values stay missing. If a property is not in the file, the table cell is empty, not zero.

“Normal termination” is not proof of convergence. Termination lines are reported as printed, nothing more.

Several files at once. The properties table takes up to 50 files, and the spectrum and RMSD tools up to 20. A file that cannot be read is listed with the reason, and the others are still processed.

Program versions. Outputs of Gaussian 16 and 09 and of ORCA 6 and 5 are read, and every tool was checked on files from each. Older versions, such as Gaussian 03 or ORCA 4, have not been tested; if one is refused or looks wrong, please report it.

Size limits. Each file can be up to 2 MB for the input writers, 25 MB for outputs and trajectories, and 150 MB for wavefunction files (.fchk, Molden). The hole–electron tool takes its excited-state output and orca_2json file up to 50 MB each. Where a tool takes several files, the total is also capped: 20 MB for a batch of structures, and 100 MB for property tables and RMSD. The wavefunction limit is generous because a TD-DFT checkpoint carries large arrays the tools do not need: 61 MB for a 46-atom dye. Each tool shows its limits next to the file button.

5. Preparing calculation inputs

ORCA input (Structure → ORCA input)

Writes an ORCA 6 input from an XYZ, MOL/SDF, PDB, GJF/COM or MOL2 file, or from a Gaussian or ORCA output (.log or .out; the program is recognized from the content). The input is named after your file, so water.gjf gives water.inp.

A GJF/COM file can be any Gaussian 16 input with Cartesian coordinates. Routes, several jobs separated by --Link1-- (the first job with coordinates is used), atom labels, isotopes, freeze codes and coordinates in bohr (Units=Bohr) are all read the way Gaussian reads them. Z-matrices, ghost atoms and geometries taken from a checkpoint file are refused, with the reason.

As soon as you choose a file, Atomlyze fills in the charge and multiplicity. If the file states them, as a GJF file or a program output does, those values are used. Otherwise the charge is 0 and the multiplicity is the lowest the electron count allows: 1 for an even count, 2 for an odd one. Change the charge and the multiplicity follows, until you type a multiplicity yourself. The lowest spin is only a starting point. O₂, carbenes and many transition-metal complexes have high-spin ground states, so enter the state you mean.

From an optimization output, Atomlyze takes the optimized structure, but only when the output shows that the optimization converged and any later geometry is the same one (a frequency job on the optimized structure, for example). For relaxed scans, unfinished optimizations and other files with several different geometries, you pick the geometry from a list.

• Method. The default, r2SCAN-3c, is reliable and cheap enough for most molecules. GFN2-xTB and B97-3c are faster. The list also has the other composite methods (PBEh-3c, B3LYP-3c, ωB97X-3c), hybrid functionals with a triple-ζ basis (including ωB97M-D4 and ωB97X-D4), and DLPNO-CCSD(T) for single points.
• Job type. Single point, optimization, frequency, optimization plus frequency, transition state (OptTS with an exact first Hessian, then frequencies), conformer search (GOAT, best with GFN2-xTB), excited states (TD-DFT), NMR shielding, or molecular dynamics.
• Solvent. None by default. With a solvent, C-PCM is used (SMD under Advanced options); GFN2-xTB uses its own ALPB model.
• Processes and memory. Set them to match your machine or queue.

Advanced options add SMD, tighter SCF or optimization convergence, the number of excited states, triplets, full TD-DFT instead of TDA, and room for your own keywords and input blocks, which are copied as written. A choice that does not apply to the job, such as tight optimization convergence for a single point, is named in the notes.

Choose several structure files to get a ZIP of inputs. A file that states its charge and multiplicity keeps them. The others take the charge you entered and their own lowest multiplicity, unless you type a multiplicity, which then applies to all of them. The listing shows the state used for each file, and the ZIP is offered only when every file converts.

Every method and job type was checked against the ORCA 6.1.1 manual and run with ORCA 6.1.1 on a small molecule. A method whose basis set does not cover an element in your structure is refused. As with any generated input, read it before you submit it.

Gaussian input (Structure → Gaussian input)

Reads the same files as the ORCA input writer, fills in charge and multiplicity the same way, and writes a Gaussian 16 input named after your file (water.xyz gives water.gjf). Several files give a ZIP of inputs, as with ORCA, and each input’s checkpoint (%chk) is named after its own file.

The defaults are B3LYP with Grimme’s D3(BJ) dispersion and the Def2SVP basis set. You can type any other method or basis set, with suggestions as you type. The job types are single point, optimization, frequency, optimization plus frequency, transition state (Opt=(TS,CalcFC) with frequencies), excited states (TD-DFT) and NMR shielding. A solvent uses PCM, or SMD under Advanced options. Functionals that already include dispersion, such as ωB97X-D, need the dispersion correction set to None.

Advanced options add tighter SCF and optimization convergence, the number of excited states, singlets and triplets (50-50), the Tamm–Dancoff approximation, extra route keywords, the title, and Gen/GenECP basis-set input with sections before or after it. Atomlyze makes sure the route stays on one line but does not judge the chemistry of what you type. Following the Gaussian manual, the title must be one line without the characters @ # ! - _ or \. Every job type and option was run with Gaussian 16 C.01.

6. Reading calculation outputs

Properties table (ORCA or Gaussian outputs → properties table)

Takes up to 50 outputs from ORCA, Gaussian or both, and recognizes each file’s program from its content. The table has a column per file and a row per property, showing the last value printed in each file. Select a value to see the line it was read from. Rows with no value in any file are left out, a missing value stays empty, and files that cannot be read are listed below the table with the reason.

The properties are the energies (final single point, SCF, Gibbs free energy, enthalpy, zero-point energy, entropy and temperature), the dipole moment, the frontier orbitals, the lowest vibrational frequency and the number of imaginary modes, the atoms and formula, charge, multiplicity, electron counts, ⟨S²⟩ for open shells, the number of converged optimizations and the termination lines. The table downloads as an Excel workbook or a CSV file, with one row per file under a header row and a unit row.

Two program details matter when you compare numbers:

• For Gaussian, the final single-point energy is the energy of the method you ran: for MP2, double-hybrid functionals, MP3, MP4, CCSD, CCSD(T) or QCISD(T) it is the correlated energy, not the SCF part that Gaussian prints first. The Energy Method row names it and the SCF Energy row keeps the SCF value. For TD-DFT runs the table also gives the total energy of the excited state of interest.
• For ORCA TD-DFT and TDA runs, ORCA’s final single-point energy includes the excitation energy of the chosen root; the table then also gives the ground-state SCF energy (in the SCF Energy row) and the root number.

When a file prints polarizabilities, a second table appears below the first (see the next section). It is also the second sheet of the workbook and has its own CSV download.

If NBO 7 ran in the job (Pop=NBO7 or Pop=NBO7Read in Gaussian, ! NBO in ORCA), an NBO analysis section lists the natural bond orbitals and the second-order delocalization energies E(2) as NBO printed them. Each orbital comes with its occupancy, energy and hybrids, for example “C1 36.63% sp2.06 · N3 63.37% sp1.47”. The E(2) list starts with the strongest delocalizations, such as the amide resonance n(N)→π*(C=O), and open-shell molecules get separate α and β lists. Atomlyze does not redo the NBO analysis; it reads NBO’s results and checks each E(2) against NBO’s own formula from the printed F(L,NL) and energy gap. Both tables download as CSV and are sheets of the workbook.

Polarizabilities (in the properties table)

Reads the polarizability (α) and first hyperpolarizability (β) tensors that Gaussian (Polar) or ORCA (%elprop Polar 1 Hyperpol 1 end) prints, and derives the quantities usually reported:

• the isotropic α and the anisotropy Δα;
• β_tot = |β|, the component β_vec along the dipole, and the EFISH value β∥ = (3/5) β_vec;
• the hyper-Rayleigh scattering β_HRS and the depolarization ratio DR, averaged exactly over molecular orientations;
• the second hyperpolarizability γ∥, when Gaussian prints it (Polar=Gamma).

Every value is given in atomic units, esu and SI, with CODATA 2018 factors. The first row of each file holds the static values. If Gaussian also computed frequency-dependent values (CPHF=RdFreq with a wavelength), each wavelength gets its own row: α(−ω;ω), the second-harmonic β(−2ω;ω,ω) with its β_HRS and DR, the electro-optic β(−ω;ω,0) as β tot (EOPE), and γ for second-harmonic generation and the dc-Kerr effect. These derived values agree with Multiwfn’s.

Where the program prints a derived quantity itself, Atomlyze recomputes it and shows whether the two agree. β keeps each program’s sign convention. Gaussian’s values omit a factor of 1/2, and ORCA and Multiwfn use the opposite sign to Gaussian; magnitudes, β_HRS and DR are unaffected. ORCA computes no γ.

Gaussian 09 prints the tensors in an older layout, which is read as well. If Gaussian 09 reoriented the molecule (that is, without Symmetry=None), β along the dipole and β∥ are left out, because the dipole and the tensors may not share a frame. The other values do not depend on the frame.

7. Spectra and densities of states

Every plot in this group downloads as SVG, PNG or TIFF, and the full curve as CSV. Section 11 covers the figure settings.

UV–Vis or ECD spectrum (Excited states → UV–Vis or ECD spectrum)

Reads the last excited-state section of a Gaussian or ORCA output and turns each transition into a band. UV–Vis absorption plots the molar absorption coefficient ε from the oscillator strengths; ECD plots Δε from the rotatory strengths. Both are in L mol⁻¹ cm⁻¹, against wavelength, energy or wavenumber. ECD uses the velocity gauge by default because it does not depend on where the coordinate origin is; the length gauge is under Advanced options. A molecule without chirality has no rotatory strength, and the notes say so when the ECD curve is flat.

Bands are always broadened in energy, by default as Gaussians with a half-width σ = 0.4 eV at 1/e of the maximum, the formula Gaussian publishes. Advanced options offer Lorentzian and pseudo-Voigt shapes (set by their full width at half maximum), an energy shift for every transition, the plot range and peak labels.

Several files are drawn over one common range. If they are conformers of one molecule at one level of theory, they can be combined into a Boltzmann-weighted spectrum using their electronic or Gibbs free energies, and the weights are listed. To compare with experiment, add a two-column text file (x in the plot’s unit, then y) to your selection. Atomlyze recognizes it and scales it to the calculated maximum unless you turn that off.

The results list every transition with its energy, wavelength, oscillator strength and rotatory strengths, and, for Gaussian and ORCA TDA outputs, its two main orbital pairs (for example HOMO → LUMO). The peak positions follow.

IR, Raman or VCD spectrum and modes (Frequencies → IR, Raman or VCD spectrum and modes)

Reads the last frequency table of a Gaussian or ORCA output.

• IR is plotted as the molar absorption coefficient ε (L mol⁻¹ cm⁻¹), with wavenumbers decreasing to the right and peaks pointing down, the way experimental IR spectra are usually drawn. Peak direction turns the peaks upwards. An experimental spectrum recorded in transmittance is converted to absorbance before it is overlaid. Under Advanced options, each band’s height can instead be the printed intensity in km/mol.
• Raman shows activities, or intensities calculated from them for your excitation wavelength and temperature. There is no default wavelength: enter the laser line you use.
• VCD shows Δε from the rotational strengths (Gaussian Freq=VCD, ORCA %freq doVCD true end).

Bands have a full width at half maximum of 15 cm⁻¹ by default. Imaginary modes are left out of the curve and counted in the notes. Frequencies are scaled only if you choose a factor, uniform or power-law (λ₁ ν^λ₂). Take it from the literature for your method and basis set; Atomlyze does not supply one.

Several files, conformer weighting, experimental spectra and peak labels work as for UV–Vis. The results list every mode, imaginary ones included, with its symmetry, frequency, IR intensity, Raman activity and VCD rotational strength in the program’s units. A value the output does not give is shown as a dash, never as zero.

The UV–Vis, ECD, IR, Raman and VCD curves were checked against Multiwfn and against ORCA’s own spectrum tool on real Gaussian and ORCA calculations.

Density of states (Orbital energies → density of states)

Reads the last orbital-energy printout of a Gaussian or ORCA output and broadens each orbital to unit area, so the curve is in orbitals per eV. The defaults follow Multiwfn: Gaussian broadening of 0.5 eV FWHM and a window from −0.8 to 0.2 Eh. For unrestricted calculations α points up and β down, or the two can be summed. A dashed line marks the HOMO and short sticks mark the orbital levels. Advanced options can put the HOMO at 0 eV, change the line shape or set the window.

A single wavefunction file (a Gaussian .fchk, or a Molden file, including ORCA’s) also gives the partial DOS. An output gives only the total DOS, because it does not contain the orbital coefficients and basis set; the notes say so. The partial DOS can be grouped, under Advanced options:

• by element (the default) or by atom;
• by shell type (s, p, d …), or by element and shell type;
• by fragments you define, such as ring: 1-6; O; 7,8 (atom numbers, ranges or element symbols, ; between fragments, an optional name before :).

Orbital compositions are Mulliken by default. SCPA compositions are never negative. Hirshfeld compositions cope better with diffuse basis functions but take longer, and need groups made of whole atoms. When the groups cover every atom once, the partial curves add up to the total.

The overlap DOS (OPDOS) between the first two groups is positive where they bond and negative where they are antibonding, and is drawn dashed. All these curves were checked against Multiwfn’s.

8. Trajectories and RMSD

Aligned RMSD (XYZ trajectory → aligned RMSD)

Computes the RMSD of each frame from the first after removing translation and rotation, with equal atom weights. The trajectory must keep one atom order throughout and use unwrapped coordinates in ångström; no periodic correction is applied.

Enter the time between frames (the arrows step by 0.5 fs, and any other positive value can be typed). ORCA MD trajectories (name_traj.xyz) carry the time of each frame, which is used instead, so a dump stride is handled correctly. You can inspect single frames on the plot, choose the frame range for the image and download every frame as CSV. Several trajectories can share one plot, each aligned to its own first frame.

9. Viewing structures

3D viewer (Structure or output → 3D viewer and XYZ)

Shows XYZ, MOL/SDF, PDB, GJF and MOL2 structures, Gaussian and ORCA outputs, multi-frame XYZ trajectories, and the structure stored in a Gaussian .fchk or a Molden file.

An output opens at its optimized geometry if the file shows that the optimization converged, and otherwise at its last geometry. Every geometry in the file becomes a frame: move the slider or press Play frames to step through an optimization or a trajectory. Each frame’s energy is shown with its difference from the lowest, next to a small energy profile.

If the output contains a frequency calculation, choose a mode under Vibration to animate it; imaginary modes are marked. The amplitude only scales the drawing, and the arrows show which way each atom moves. When the output lists IR intensities, Raman activities or VCD rotational strengths, the spectrum of the modes is drawn below the list as the IR/Raman/VCD tool draws it by default, with a choice between them when there is more than one. Click a band to animate its mode.

Download XYZ saves the structure on screen, or the current frame, in ångström and in the file’s atom order, with the frame and its energy on the comment line. Use it to take a geometry into another program. (The ORCA and Gaussian input writers read outputs directly.)

A few notes on file formats:

• MOL2 and PDB files from Open Babel, Multiwfn, Gaussian’s newzmat and similar programs are read as written. Open Babel 3.1 takes the first geometry of an ORCA output, not the optimized one, so use Download XYZ here for an optimized ORCA structure.
• MOL and SDF files may use the V2000 layout or the newer V3000 (written by Open Babel with -x3, RDKit, ChemDraw and Marvin). A V3000 structure is drawn from its atoms, with bonds assigned from distances.
• An SDF file holding several conformers of one molecule, from RDKit, Open Babel or CREST for example, opens as frames.
• Anything a file states beyond its coordinates is listed under the summary, such as the charge, multiplicity, route and title of a GJF file or the partial charges in a MOL2 file. These values are shown, not checked or used.

You can change the style (stick, ball and stick, space filling, line, or cartoon for proteins), the colouring (by element, chain, residue or atom order), the background and the atom labels.

Rotate with the mouse or the arrow keys, zoom with the wheel or +/−, and press R to fit the structure back into view. On a touch screen, double-tap to fit.

To measure, type two, three or four atom numbers and press Measure, or tick “Pick atoms by clicking” and click the atoms. Two atoms give a distance, three an angle, four a dihedral angle with the IUPAC sign convention. Recent measurements are kept in a list that you can download as CSV.

Images of the view are covered in section 11.

10. Wavefunctions: orbitals, charges, bond orders and excited states

These tools read a Gaussian formatted checkpoint (.fchk, made with formchk) or a Molden file. ORCA users can make a Molden file with:

orca_2mkl name -molden

which writes name.molden.input. ORCA’s Molden files follow their own conventions, and Atomlyze handles them.

Before showing anything, Atomlyze checks that it has read the basis set correctly: it computes the overlap matrix and requires the orbitals to be orthonormal. A file that fails is refused, with the size of the error. Molden files do not record charge or multiplicity, so these come from the orbital occupations. Natural orbitals with fractional occupations are refused.

Orbitals and densities (Wavefunction → orbitals and densities)

Pick an orbital from the list, which starts at the HOMO, and Atomlyze draws it as two surfaces at ± the isovalue (0.05 a.u. by default): blue for positive, vermilion for negative. A new isovalue redraws at once; another orbital is recomputed. The orbital downloads as a Gaussian cube file.

At the top of the list are the total electron density and, for open-shell molecules, the spin density (α − β, blue where α electrons dominate). The spin density starts at an isovalue of 0.005 e/bohr³. Both include all electrons and also download as cube files. Gaussian’s cubegen density=scf writes a valence-only density, so near the nuclei it differs from Atomlyze’s.

Advanced options change the grid spacing and the margin around the molecule. Charges, spin populations and bond orders are in the next tool.

Atomic charges and bond orders (Wavefunction → atomic charges and bond orders)

Gives seven kinds of atomic charge, and bond orders:

• Hirshfeld charges divide the electron density between atoms in proportion to free, neutral, spherically averaged atoms.
• CM5 charges apply the published correction of Marenich, Jerome, Cramer and Truhlar (2012) to the Hirshfeld charges. They were designed to reproduce molecular dipole moments, depend little on the basis set and make a good general choice.
• CHELPG and MK charges are fitted to the electrostatic potential around the molecule, on the standard grids of Breneman and Wiberg (CHELPG) and of Merz and Kollman (MK). They are often used for force fields and docking, but can be unstable for atoms buried inside a large molecule. The notes give the number of fitting points and how well the charges reproduce the potential. They agree with Multiwfn’s; Gaussian’s and ORCA’s own CHELPG and MK charges use slightly different grids and differ by about 0.002 e, and ORCA’s default CHELPG radii differ more.
• Mulliken charges are listed for comparison. They depend strongly on the basis set.
• Löwdin charges come from the density in an orthogonalized basis. They depend less on the basis set than Mulliken charges, although they still do not converge as it grows, and they match what ORCA prints.
• Natural (NPA) charges come from natural atomic orbitals, as in the NBO program. They depend little on the basis set and agree with NBO 7 to the digits it prints. The same analysis gives Wiberg bond indices (about 1 for a single bond, 2 for a double bond), listed below the table. NPA needs spherical basis functions (5D/7F, the default for def2 and Dunning basis sets) and elements up to argon; otherwise the notes say why it was skipped.
• Mayer bond orders come from the density and overlap matrices in the calculation’s own basis, and are listed beside the Wiberg indices for every pair where either is at least 0.1.

Open-shell molecules also get Hirshfeld, Mulliken, Löwdin and natural spin populations.

Free atoms for the Hirshfeld and CM5 charges. By default these are the spherically averaged atoms built into Multiwfn, computed close to the basis-set limit. Atomlyze’s Hirshfeld charges and spin populations agree with Multiwfn’s within 5×10⁻⁵ e on fine grids, and Gaussian’s, which use its own free atoms, are within a few 10⁻⁴ e for a typical organic molecule. ORCA and Q-Chem instead compute free atoms with the molecule’s basis set, so their Hirshfeld charges differ more, especially with small basis sets.

To use your own free atoms, add one file per element alongside the molecule’s: a neutral atom, in its ground-state multiplicity, at the same method and basis set, as .fchk or Molden. Atomlyze tells these files from the molecule’s by their single atom, averages each density over all directions, and names in the notes the file each element came from. Multiwfn’s own-atom option gives the same charges for hydrogen, carbon and nitrogen. For oxygen, fluorine, sulfur and chlorine it makes the atom spherical in a different way, and the charges differ by a few 10⁻³ e.

The density is integrated numerically on grids around each atom, and the notes report the integrated number of electrons, which should be within about 10⁻⁴ of the true count. Advanced options offer a finer grid (about three times slower) and the 1.2 × CM5 charges used with OPLS-AA-type force fields.

A 20-atom molecule with a small basis set takes about 20 seconds in a desktop browser. Larger molecules take longer, as these times show. They were measured in Chromium on an 8-core laptop (Intel Core Ultra 9 288V) with B3LYP/6-31G(d) wavefunctions:

Most of that time goes into fitting the electrostatic potential for the CHELPG and MK charges. Switch them off under Advanced options and the other charges arrive in a fraction of the time. The browser used up to about 2 GB of memory for the largest of these.

Wavefunctions with effective core potentials, as for heavy elements in def2 basis sets, get CHELPG, MK, Mulliken and Löwdin charges. Hirshfeld, CM5 and natural charges need all electrons, so they are skipped and the notes say why. Elements without a CHELPG or MK radius, such as iodine, take the UFF radius divided by 1.2, as Multiwfn does, and the notes name them.

Hole and electron of excited states (Excited states → hole–electron analysis)

For each excited state of a TD-DFT, TDA or CIS calculation, this tool shows where the excitation takes an electron from (the hole) and where it puts it (the electron), with the usual numbers for telling a local excitation from a charge-transfer one.

Choose two files from the same calculation together:

• the wavefunction: a Gaussian .fchk file, or a Molden file (for ORCA, name.molden.input from orca_2mkl name -molden);
• the output that lists the excitation coefficients: a Gaussian TD or CIS log, or an ORCA output from a TDA calculation (TDA is ORCA’s default).

For ORCA full TD-DFT (tda false) the output prints only the weights of the coefficients, so add a third file: name.json, written by orca_2json name.gbw when a file name.json.conf containing {"CIS": true, "CISNRoots": true} sits next to it. Atomlyze then reads the excitation and de-excitation coefficients from the JSON file, and the oscillator strengths from the output.

The order does not matter, because each file is recognized by its content. Atomlyze then checks that the files belong together: the orbital numbers must exist, the numbers of occupied orbitals must agree and the orbital energies in the output must match the wavefunction. A mismatched pair is refused, with the reason.

Programs print only the larger coefficients, and Atomlyze can use only what was printed. Each state therefore reports N, the share of the excitation that the printed coefficients capture; it should be close to 1. If a state captures less than 95 %, the notes say so. Print more coefficients and run the calculation again (IOp(9/40=3) in Gaussian, a smaller TPrint in ORCA’s %tddft block).

The table lists every state with:

• D, the distance between the centres of the hole and the electron. A large D means charge transfer.
• Sr, the overlap of hole and electron, between 0 and 1. Local excitations have large Sr.
• H, the average size of hole and electron.
• t = D − H_CT, where H_CT is the size along the transfer direction. Positive t means hole and electron are clearly separated. When D is essentially zero, as in a molecule with a centre of symmetry, the direction is undefined and t is shown as “—”.
• Δσ, the difference in size between electron and hole.

Each state also gets its natural transition orbitals (NTOs): the excitation rewritten as a few hole → particle orbital pairs, each with its share. A single pair near 100 % means one orbital picture describes the excitation well, even if many MO pairs contribute. The download includes the leading NTO pair of the chosen state as cube files.

Pick a state to see its main orbital pairs, the centres and sizes of hole and electron, and how much each atom contributes to each. The hole is drawn in blue and the electron in vermilion at the isovalue you set (0.002 a.u. by default). Changing the isovalue or the state redraws the picture without repeating the analysis.

For singlets, the notes put the oscillator strength recalculated from the printed coefficients next to the program’s own value. They differ when small coefficients were not printed, which affects the oscillator strength far more than the hole and electron.

The downloads are a spreadsheet of every state and its atom contributions, and a ZIP with the same tables as CSV and the hole and electron of the chosen state as cube files. Advanced options change the cube grid and margin, which affect only the pictures and cube files, and offer a finer integration grid for Sr and the atom contributions.

Open-shell (unrestricted) calculations work as well: Gaussian TD or CIS, ORCA TDA with UKS or UHF, and ORCA full TD-DFT with the name.json file described above. Each spin’s hole and electron are built from that spin’s orbitals and then added, orbital pairs are marked α or β, and ⟨S²⟩ is shown as printed so you can judge spin contamination. ORCA states carry the multiplicity ORCA assigns them, for example “Mult 2”.

The results agree with Multiwfn on the same files, with two deliberate differences. Atomlyze divides Sr by the captured norm N, so Multiwfn’s Sr equals Atomlyze’s Sr × N. And where D is zero, Atomlyze leaves t undefined, while Multiwfn prints a value based on an arbitrary direction. If you print coefficients down to 0.01, as recommended, Multiwfn drops the small ones from its cross terms to save time, and its indices can differ from Atomlyze’s by up to about 0.015 Å.

The definitions are those of Multiwfn and of Liu, Lu and Chen (Carbon, 2020); please cite them (section 13). A 20-atom molecule with a small basis set takes about half a minute in a desktop browser, and switching states about 15 seconds.

11. Figures for papers and theses

Plots and 3D images are made at their final printed size, so text and lines come out at the size the journal prints them.

Plots (spectra, DOS, RMSD). Open Figure options under the plot:

• Width: single column (8.5 cm, the default), 1.5 columns (11.4 cm) or double column (17.4 cm).
• Resolution for PNG and TIFF: 300, 600 (default) or 1200 dpi. Use 1200 dpi for line art if the journal asks for it.
• The legend can be switched off. A title can be put inside the figure, though most journals want it in the caption.

Choose SVG to edit the figure in Inkscape or Illustrator, or to save it as PDF or EPS from there, and PNG or TIFF for a finished raster image. Both raster files record their resolution, so they import at the right size; TIFF uses lossless LZW compression.

3D images. Image options in the viewer set the width and resolution in the same way. The image keeps the view and proportions you see on screen, so arrange and zoom the structure first. If your graphics card cannot make an image that large, the message under the buttons says so and gives the resolution it managed. “As shown on screen” saves the screen pixels as they are, which is enough for slides and web pages.

12. When something does not work

“Cannot read” or “Not read” with a reason. The reason points to the problem: a missing section, an unsupported keyword, a line number. Check the file against the tool’s notes. Atomlyze refuses a file it cannot read with certainty rather than give a doubtful result.

An output with several geometries. You are asked to choose one, so that a scan or an unfinished optimization is never reduced to its last step without your knowing.

The first run is slow. The browser is downloading the program code, which later runs reuse. If the download stalls, the message says so; try again, and the parts already received are kept.

The 3D view stays blank. The viewer needs WebGL. Try another browser, or check that hardware acceleration is on. Results and downloads work without the 3D view.

A wavefunction file is refused with “basis check failed”. The file uses a convention Atomlyze does not recognize. Please report it, with a non-confidential example if you can.

Browser diagnostics. Add #diagnostics to the address (for example atomlyze.com/#diagnostics) to show a Browser diagnostics panel at the bottom of the page. It checks the numerical libraries and a 100 MB file stream with generated data, and can scan one of your outputs. If a tool fails in an unusual browser, these checks help tell a browser limitation from a problem with the file.

Reporting a problem. Write to support@atomlyze.com with the tool, the version at the foot of the page, the message you saw and what you expected. A small file that shows the problem helps, if you are free to share it; please do not send confidential data.

13. What to cite

If Atomlyze helps your work, please cite it with the version you used, shown at the foot of every page:

F. Ullah, Atomlyze: a browser-based computational chemistry toolkit, version 2026.10.13, https://atomlyze.com.

The version is the release date (year.month.day). A journal article on Atomlyze is in preparation and will replace this reference once it is published. Please also cite the methods behind the numbers you report:

The notes of each tool name further sources where they apply, such as the program manuals that input keywords were checked against.

14. Revision history